From 965f0c26ac6d0b927c5270a0b8bcec5335462702 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Pablo=20Costas=20S=C3=A1nchez?= Date: Tue, 14 Feb 2023 14:46:25 +0100 Subject: [PATCH] [RTI-13985] Update to zstd 1.5.4 (#38) * [RTI-13985] Update to zstd 1.5.4 and switch to submoduling the dependency * [RTI-13985] CI: checkout submodules too * Revert back to static pinned release --- README.md | 2 +- priv/zstd/.circleci/config.yml | 44 +- priv/zstd/.circleci/images/primary/Dockerfile | 2 +- priv/zstd/.cirrus.yml | 3 +- .../zstd/.github/ISSUE_TEMPLATE/bug_report.md | 4 +- priv/zstd/.github/dependabot.yml | 6 + .../zstd/.github/workflows/dev-long-tests.yml | 271 + .../.github/workflows/dev-short-tests.yml | 583 ++ priv/zstd/.github/workflows/generic-dev.yml | 225 - .../.github/workflows/generic-release.yml | 59 - priv/zstd/.github/workflows/linux-kernel.yml | 13 - priv/zstd/.github/workflows/main.yml | 30 - .../workflows/publish-release-artifacts.yml | 13 +- priv/zstd/.github/workflows/scorecards.yml | 64 + priv/zstd/.gitignore | 1 + priv/zstd/.travis.yml | 112 +- priv/zstd/CHANGELOG | 83 + priv/zstd/CONTRIBUTING.md | 133 +- priv/zstd/LICENSE | 8 +- priv/zstd/Makefile | 47 +- priv/zstd/Package.swift | 36 + priv/zstd/README.md | 73 +- priv/zstd/TESTING.md | 2 +- priv/zstd/appveyor.yml | 152 +- priv/zstd/build/VS2008/zstd/zstd.vcproj | 4 + .../build/VS2010/libzstd-dll/libzstd-dll.rc | 4 +- priv/zstd/build/VS2010/zstd/zstd.rc | 4 +- priv/zstd/build/VS2010/zstd/zstd.vcxproj | 1 + priv/zstd/build/VS_scripts/build.generic.cmd | 14 +- priv/zstd/build/cmake/CMakeLists.txt | 37 +- .../AddZstdCompilationFlags.cmake | 54 +- .../build/cmake/CMakeModules/JoinPaths.cmake | 23 + priv/zstd/build/cmake/README.md | 6 +- priv/zstd/build/cmake/contrib/CMakeLists.txt | 2 +- .../cmake/contrib/gen_html/CMakeLists.txt | 2 +- .../build/cmake/contrib/pzstd/CMakeLists.txt | 10 +- priv/zstd/build/cmake/lib/CMakeLists.txt | 53 +- priv/zstd/build/cmake/programs/CMakeLists.txt | 10 +- priv/zstd/build/cmake/tests/CMakeLists.txt | 23 +- .../build/meson/contrib/pzstd/meson.build | 3 +- priv/zstd/build/meson/lib/meson.build | 44 +- priv/zstd/build/meson/meson.build | 35 +- priv/zstd/build/meson/meson_options.txt | 4 +- priv/zstd/build/meson/programs/meson.build | 23 +- priv/zstd/build/meson/tests/meson.build | 125 +- priv/zstd/build/meson/tests/valgrindTest.py | 10 +- priv/zstd/build/single_file_libs/README.md | 4 +- .../single_file_libs/build_decoder_test.sh | 2 +- .../single_file_libs/build_library_test.sh | 4 +- priv/zstd/build/single_file_libs/combine.py | 234 + priv/zstd/build/single_file_libs/combine.sh | 60 +- .../create_single_file_decoder.sh | 9 +- .../create_single_file_library.sh | 9 +- priv/zstd/build/single_file_libs/zstd-in.c | 12 +- .../build/single_file_libs/zstddeclib-in.c | 12 +- priv/zstd/contrib/VS2005/zstd/zstd.vcproj | 4 + .../zstd/contrib/diagnose_corruption/Makefile | 2 +- .../diagnose_corruption/check_flipped_bits.c | 2 +- .../externalSequenceProducer/.gitignore | 2 + .../contrib/externalSequenceProducer/Makefile | 40 + .../externalSequenceProducer/README.md | 14 + .../contrib/externalSequenceProducer/main.c | 107 + .../sequence_producer.c | 80 + .../sequence_producer.h | 26 + .../contrib/freestanding_lib/freestanding.py | 37 +- priv/zstd/contrib/gen_html/Makefile | 2 +- priv/zstd/contrib/gen_html/gen_html.cpp | 2 +- priv/zstd/contrib/largeNbDicts/Makefile | 2 +- priv/zstd/contrib/largeNbDicts/README.md | 24 +- priv/zstd/contrib/largeNbDicts/largeNbDicts.c | 187 +- priv/zstd/contrib/linux-kernel/Makefile | 29 +- .../contrib/linux-kernel/decompress_sources.h | 10 +- priv/zstd/contrib/linux-kernel/linux.mk | 23 +- priv/zstd/contrib/linux-kernel/linux_zstd.h | 4 +- priv/zstd/contrib/linux-kernel/mem.h | 6 +- priv/zstd/contrib/linux-kernel/test/Makefile | 11 +- .../linux-kernel/test/include/asm/unaligned.h | 1 + .../test/include/linux/compiler.h | 4 +- .../linux-kernel/test/include/linux/errno.h | 2 +- .../linux-kernel/test/include/linux/kernel.h | 2 +- .../linux-kernel/test/include/linux/limits.h | 2 +- .../linux-kernel/test/include/linux/math64.h | 2 +- .../linux-kernel/test/include/linux/module.h | 4 +- .../linux-kernel/test/include/linux/printk.h | 2 +- .../linux-kernel/test/include/linux/stddef.h | 2 +- .../linux-kernel/test/include/linux/swab.h | 2 +- .../linux-kernel/test/include/linux/types.h | 2 +- .../linux-kernel/test/include/linux/xxhash.h | 13 +- .../contrib/linux-kernel/test/macro-test.sh | 2 +- .../contrib/linux-kernel/test/static_test.c | 14 +- priv/zstd/contrib/linux-kernel/test/test.c | 68 +- .../contrib/linux-kernel/zstd_common_module.c | 32 + .../linux-kernel/zstd_compress_module.c | 54 +- .../linux-kernel/zstd_decompress_module.c | 4 +- priv/zstd/contrib/linux-kernel/zstd_deps.h | 4 +- priv/zstd/contrib/match_finders/README.md | 8 +- priv/zstd/contrib/match_finders/zstd_edist.c | 24 +- priv/zstd/contrib/match_finders/zstd_edist.h | 10 +- priv/zstd/contrib/pzstd/ErrorHolder.h | 2 +- priv/zstd/contrib/pzstd/Logging.h | 2 +- priv/zstd/contrib/pzstd/Makefile | 18 +- priv/zstd/contrib/pzstd/Options.cpp | 4 +- priv/zstd/contrib/pzstd/Options.h | 2 +- priv/zstd/contrib/pzstd/Pzstd.cpp | 11 +- priv/zstd/contrib/pzstd/Pzstd.h | 2 +- priv/zstd/contrib/pzstd/README.md | 2 +- priv/zstd/contrib/pzstd/SkippableFrame.cpp | 2 +- priv/zstd/contrib/pzstd/SkippableFrame.h | 2 +- priv/zstd/contrib/pzstd/main.cpp | 2 +- priv/zstd/contrib/pzstd/test/OptionsTest.cpp | 2 +- priv/zstd/contrib/pzstd/test/PzstdTest.cpp | 2 +- priv/zstd/contrib/pzstd/test/RoundTrip.h | 2 +- .../zstd/contrib/pzstd/test/RoundTripTest.cpp | 2 +- priv/zstd/contrib/pzstd/utils/Buffer.h | 4 +- priv/zstd/contrib/pzstd/utils/FileSystem.h | 22 +- priv/zstd/contrib/pzstd/utils/Likely.h | 2 +- priv/zstd/contrib/pzstd/utils/Portability.h | 16 + priv/zstd/contrib/pzstd/utils/Range.h | 10 +- priv/zstd/contrib/pzstd/utils/ResourcePool.h | 2 +- priv/zstd/contrib/pzstd/utils/ScopeGuard.h | 4 +- priv/zstd/contrib/pzstd/utils/ThreadPool.h | 2 +- priv/zstd/contrib/pzstd/utils/WorkQueue.h | 2 +- .../contrib/pzstd/utils/test/BufferTest.cpp | 2 +- .../contrib/pzstd/utils/test/RangeTest.cpp | 2 +- .../pzstd/utils/test/ResourcePoolTest.cpp | 2 +- .../pzstd/utils/test/ScopeGuardTest.cpp | 2 +- .../pzstd/utils/test/ThreadPoolTest.cpp | 2 +- .../pzstd/utils/test/WorkQueueTest.cpp | 2 +- priv/zstd/contrib/recovery/Makefile | 2 +- .../zstd/contrib/recovery/recover_directory.c | 2 +- .../contrib/seekable_format/examples/Makefile | 2 +- .../examples/parallel_compression.c | 2 +- .../examples/parallel_processing.c | 2 +- .../examples/seekable_compression.c | 2 +- .../examples/seekable_decompression.c | 2 +- .../examples/seekable_decompression_mem.c | 2 +- .../contrib/seekable_format/tests/Makefile | 2 +- .../seekable_format/tests/seekable_tests.c | 34 + .../zstd_seekable_compression_format.md | 2 +- .../seekable_format/zstdseek_compress.c | 4 +- .../seekable_format/zstdseek_decompress.c | 59 +- priv/zstd/contrib/seqBench/Makefile | 58 + priv/zstd/contrib/seqBench/seqBench.c | 53 + priv/zstd/doc/README.md | 5 +- priv/zstd/doc/educational_decoder/Makefile | 2 +- priv/zstd/doc/educational_decoder/harness.c | 2 +- .../doc/educational_decoder/zstd_decompress.c | 6 +- .../doc/educational_decoder/zstd_decompress.h | 2 +- priv/zstd/doc/images/zstd_logo86.png | Bin 5963 -> 13069 bytes priv/zstd/doc/zstd_compression_format.md | 50 +- priv/zstd/doc/zstd_manual.html | 431 +- priv/zstd/examples/Makefile | 2 +- priv/zstd/examples/common.h | 34 +- priv/zstd/examples/dictionary_compression.c | 16 +- priv/zstd/examples/dictionary_decompression.c | 2 +- .../examples/multiple_simple_compression.c | 2 +- .../examples/multiple_streaming_compression.c | 2 +- priv/zstd/examples/simple_compression.c | 2 +- priv/zstd/examples/simple_decompression.c | 2 +- priv/zstd/examples/streaming_compression.c | 32 +- .../streaming_compression_thread_pool.c | 10 +- priv/zstd/examples/streaming_decompression.c | 2 +- priv/zstd/examples/streaming_memory_usage.c | 2 +- priv/zstd/lib/Makefile | 229 +- priv/zstd/lib/README.md | 17 +- priv/zstd/lib/common/bits.h | 175 + priv/zstd/lib/common/bitstream.h | 90 +- priv/zstd/lib/common/compiler.h | 157 +- priv/zstd/lib/common/cpu.h | 2 +- priv/zstd/lib/common/debug.c | 2 +- priv/zstd/lib/common/debug.h | 2 +- priv/zstd/lib/common/entropy_common.c | 52 +- priv/zstd/lib/common/error_private.c | 11 +- priv/zstd/lib/common/error_private.h | 81 +- priv/zstd/lib/common/fse.h | 91 +- priv/zstd/lib/common/fse_decompress.c | 108 +- priv/zstd/lib/common/huf.h | 253 +- priv/zstd/lib/common/mem.h | 103 +- priv/zstd/lib/common/pool.c | 53 +- priv/zstd/lib/common/pool.h | 12 +- priv/zstd/lib/common/portability_macros.h | 156 + priv/zstd/lib/common/threading.c | 82 +- priv/zstd/lib/common/threading.h | 15 +- priv/zstd/lib/common/xxhash.c | 814 +-- priv/zstd/lib/common/xxhash.h | 5735 ++++++++++++++++- priv/zstd/lib/common/zstd_common.c | 2 +- priv/zstd/lib/common/zstd_deps.h | 2 +- priv/zstd/lib/common/zstd_internal.h | 187 +- priv/zstd/lib/common/zstd_trace.h | 17 +- priv/zstd/lib/compress/clevels.h | 134 + priv/zstd/lib/compress/fse_compress.c | 219 +- priv/zstd/lib/compress/hist.c | 2 +- priv/zstd/lib/compress/hist.h | 2 +- priv/zstd/lib/compress/huf_compress.c | 978 ++- priv/zstd/lib/compress/zstd_compress.c | 2062 +++--- .../lib/compress/zstd_compress_internal.h | 453 +- .../lib/compress/zstd_compress_literals.c | 153 +- .../lib/compress/zstd_compress_literals.h | 22 +- .../lib/compress/zstd_compress_sequences.c | 35 +- .../lib/compress/zstd_compress_sequences.h | 2 +- .../lib/compress/zstd_compress_superblock.c | 51 +- .../lib/compress/zstd_compress_superblock.h | 2 +- priv/zstd/lib/compress/zstd_cwksp.h | 76 +- priv/zstd/lib/compress/zstd_double_fast.c | 501 +- priv/zstd/lib/compress/zstd_double_fast.h | 5 +- priv/zstd/lib/compress/zstd_fast.c | 978 ++- priv/zstd/lib/compress/zstd_fast.h | 5 +- priv/zstd/lib/compress/zstd_lazy.c | 1151 ++-- priv/zstd/lib/compress/zstd_lazy.h | 6 +- priv/zstd/lib/compress/zstd_ldm.c | 24 +- priv/zstd/lib/compress/zstd_ldm.h | 4 +- priv/zstd/lib/compress/zstd_ldm_geartab.h | 7 +- priv/zstd/lib/compress/zstd_opt.c | 519 +- priv/zstd/lib/compress/zstd_opt.h | 2 +- priv/zstd/lib/compress/zstdmt_compress.c | 123 +- priv/zstd/lib/compress/zstdmt_compress.h | 7 +- priv/zstd/lib/decompress/huf_decompress.c | 1569 +++-- .../lib/decompress/huf_decompress_amd64.S | 576 ++ priv/zstd/lib/decompress/zstd_ddict.c | 7 +- priv/zstd/lib/decompress/zstd_ddict.h | 2 +- priv/zstd/lib/decompress/zstd_decompress.c | 311 +- .../lib/decompress/zstd_decompress_block.c | 1191 +++- .../lib/decompress/zstd_decompress_block.h | 12 +- .../lib/decompress/zstd_decompress_internal.h | 43 +- priv/zstd/lib/deprecated/zbuff.h | 2 +- priv/zstd/lib/deprecated/zbuff_common.c | 2 +- priv/zstd/lib/deprecated/zbuff_compress.c | 28 +- priv/zstd/lib/deprecated/zbuff_decompress.c | 4 +- priv/zstd/lib/dictBuilder/cover.c | 45 +- priv/zstd/lib/dictBuilder/cover.h | 2 +- priv/zstd/lib/dictBuilder/fastcover.c | 19 +- priv/zstd/lib/dictBuilder/zdict.c | 155 +- priv/zstd/lib/dll/example/Makefile | 2 +- priv/zstd/lib/dll/example/README.md | 2 +- priv/zstd/lib/legacy/zstd_legacy.h | 9 +- priv/zstd/lib/legacy/zstd_v01.c | 69 +- priv/zstd/lib/legacy/zstd_v01.h | 2 +- priv/zstd/lib/legacy/zstd_v02.c | 93 +- priv/zstd/lib/legacy/zstd_v02.h | 2 +- priv/zstd/lib/legacy/zstd_v03.c | 97 +- priv/zstd/lib/legacy/zstd_v03.h | 2 +- priv/zstd/lib/legacy/zstd_v04.c | 92 +- priv/zstd/lib/legacy/zstd_v04.h | 2 +- priv/zstd/lib/legacy/zstd_v05.c | 112 +- priv/zstd/lib/legacy/zstd_v05.h | 2 +- priv/zstd/lib/legacy/zstd_v06.c | 121 +- priv/zstd/lib/legacy/zstd_v06.h | 2 +- priv/zstd/lib/legacy/zstd_v07.c | 109 +- priv/zstd/lib/legacy/zstd_v07.h | 2 +- priv/zstd/lib/libzstd.mk | 214 + priv/zstd/lib/libzstd.pc.in | 7 +- priv/zstd/lib/module.modulemap | 35 + priv/zstd/lib/zdict.h | 92 +- priv/zstd/lib/zstd.h | 866 ++- priv/zstd/lib/zstd_errors.h | 35 +- priv/zstd/programs/Makefile | 193 +- priv/zstd/programs/README.md | 8 +- priv/zstd/programs/benchfn.c | 8 +- priv/zstd/programs/benchfn.h | 4 +- priv/zstd/programs/benchzstd.c | 81 +- priv/zstd/programs/benchzstd.h | 4 +- priv/zstd/programs/datagen.c | 2 +- priv/zstd/programs/datagen.h | 2 +- priv/zstd/programs/dibio.c | 254 +- priv/zstd/programs/dibio.h | 8 +- priv/zstd/programs/fileio.c | 1648 ++--- priv/zstd/programs/fileio.h | 25 +- priv/zstd/programs/fileio_asyncio.c | 663 ++ priv/zstd/programs/fileio_asyncio.h | 203 + priv/zstd/programs/fileio_common.h | 125 + priv/zstd/programs/fileio_types.h | 74 + priv/zstd/programs/platform.h | 14 +- priv/zstd/programs/timefn.c | 139 +- priv/zstd/programs/timefn.h | 59 +- priv/zstd/programs/util.c | 503 +- priv/zstd/programs/util.h | 62 +- priv/zstd/programs/windres/verrsrc.h | 2 +- priv/zstd/programs/windres/zstd.rc | 4 +- priv/zstd/programs/zstd.1 | 406 +- priv/zstd/programs/zstd.1.md | 294 +- priv/zstd/programs/zstdcli.c | 584 +- priv/zstd/programs/zstdcli_trace.c | 2 +- priv/zstd/programs/zstdcli_trace.h | 2 +- priv/zstd/programs/zstdgrep.1 | 20 +- priv/zstd/programs/zstdgrep.1.md | 12 +- priv/zstd/programs/zstdless | 8 +- priv/zstd/programs/zstdless.1 | 13 +- priv/zstd/programs/zstdless.1.md | 6 +- priv/zstd/tests/.gitignore | 1 + priv/zstd/tests/DEPRECATED-test-zstd-speed.py | 2 +- priv/zstd/tests/Makefile | 140 +- priv/zstd/tests/README.md | 8 +- priv/zstd/tests/automated_benchmarking.py | 16 +- priv/zstd/tests/bigdict.c | 2 +- priv/zstd/tests/checkTag.c | 2 +- priv/zstd/tests/check_size.py | 31 + priv/zstd/tests/cli-tests/.gitignore | 6 + priv/zstd/tests/cli-tests/README.md | 258 + priv/zstd/tests/cli-tests/basic/help.sh | 10 + .../tests/cli-tests/basic/help.sh.stdout.glob | 34 + priv/zstd/tests/cli-tests/basic/memlimit.sh | 40 + .../cli-tests/basic/memlimit.sh.stderr.exact | 13 + .../cli-tests/basic/memlimit.sh.stdout.exact | 13 + priv/zstd/tests/cli-tests/basic/output_dir.sh | 7 + .../basic/output_dir.sh.stderr.exact | 2 + .../basic/output_dir.sh.stdout.exact | 2 + priv/zstd/tests/cli-tests/basic/version.sh | 6 + .../cli-tests/basic/version.sh.stdout.glob | 2 + priv/zstd/tests/cli-tests/bin/cmp_size | 44 + priv/zstd/tests/cli-tests/bin/datagen | 3 + priv/zstd/tests/cli-tests/bin/die | 4 + priv/zstd/tests/cli-tests/bin/println | 2 + priv/zstd/tests/cli-tests/bin/unzstd | 1 + priv/zstd/tests/cli-tests/bin/zstd | 9 + priv/zstd/tests/cli-tests/bin/zstdcat | 1 + priv/zstd/tests/cli-tests/bin/zstdgrep | 2 + priv/zstd/tests/cli-tests/bin/zstdless | 2 + priv/zstd/tests/cli-tests/cltools/setup | 6 + priv/zstd/tests/cli-tests/cltools/zstdgrep.sh | 8 + .../tests/cli-tests/cltools/zstdgrep.sh.exit | 1 + .../cltools/zstdgrep.sh.stderr.exact | 1 + .../cli-tests/cltools/zstdgrep.sh.stdout.glob | 4 + priv/zstd/tests/cli-tests/cltools/zstdless.sh | 10 + .../cltools/zstdless.sh.stderr.exact | 2 + .../cli-tests/cltools/zstdless.sh.stdout.glob | 5 + priv/zstd/tests/cli-tests/common/format.sh | 19 + priv/zstd/tests/cli-tests/common/mtime.sh | 13 + .../tests/cli-tests/common/permissions.sh | 18 + priv/zstd/tests/cli-tests/common/platform.sh | 37 + .../zstd/tests/cli-tests/compression/adapt.sh | 14 + .../zstd/tests/cli-tests/compression/basic.sh | 34 + .../compression/compress-literals.sh | 10 + .../tests/cli-tests/compression/format.sh | 16 + .../tests/cli-tests/compression/golden.sh | 12 + .../cli-tests/compression/gzip-compat.sh | 15 + .../tests/cli-tests/compression/levels.sh | 62 + .../compression/levels.sh.stderr.exact | 69 + .../compression/long-distance-matcher.sh | 7 + .../cli-tests/compression/multi-threaded.sh | 15 + .../multi-threaded.sh.stderr.exact | 11 + .../cli-tests/compression/multiple-files.sh | 21 + .../multiple-files.sh.stdout.exact | 12 + .../cli-tests/compression/row-match-finder.sh | 7 + priv/zstd/tests/cli-tests/compression/setup | 7 + .../cli-tests/compression/stream-size.sh | 7 + .../cli-tests/compression/verbose-wlog.sh | 11 + .../compression/verbose-wlog.sh.stderr.glob | 5 + .../compression/verbose-wlog.sh.stdout.glob | 5 + .../cli-tests/compression/window-resize.sh | 9 + .../window-resize.sh.stderr.ignore | 0 .../compression/window-resize.sh.stdout.glob | 3 + .../tests/cli-tests/decompression/golden.sh | 7 + .../cli-tests/decompression/pass-through.sh | 57 + .../pass-through.sh.stderr.exact | 11 + .../pass-through.sh.stdout.exact | 25 + .../cli-tests/dict-builder/empty-input.sh | 9 + .../dict-builder/empty-input.sh.stderr.exact | 1 + .../tests/cli-tests/dict-builder/no-inputs.sh | 3 + .../cli-tests/dict-builder/no-inputs.sh.exit | 1 + .../dict-builder/no-inputs.sh.stderr.exact | 5 + .../dictionaries/dictionary-mismatch.sh | 29 + .../dictionary-mismatch.sh.stderr.exact | 7 + .../tests/cli-tests/dictionaries/golden.sh | 9 + priv/zstd/tests/cli-tests/dictionaries/setup | 6 + .../tests/cli-tests/dictionaries/setup_once | 24 + .../file-stat/compress-file-to-file.sh | 9 + .../compress-file-to-file.sh.stderr.exact | 42 + .../file-stat/compress-file-to-stdout.sh | 8 + .../compress-file-to-stdout.sh.stderr.exact | 24 + .../file-stat/compress-stdin-to-file.sh | 8 + .../compress-stdin-to-file.sh.stderr.exact | 24 + .../file-stat/compress-stdin-to-stdout.sh | 8 + .../compress-stdin-to-stdout.sh.stderr.exact | 18 + .../file-stat/decompress-file-to-file.sh | 8 + .../decompress-file-to-file.sh.stderr.exact | 38 + .../file-stat/decompress-file-to-stdout.sh | 7 + .../decompress-file-to-stdout.sh.stderr.exact | 18 + .../file-stat/decompress-stdin-to-file.sh | 7 + .../decompress-stdin-to-file.sh.stderr.exact | 20 + .../file-stat/decompress-stdin-to-stdout.sh | 7 + ...decompress-stdin-to-stdout.sh.stderr.exact | 14 + .../tests/cli-tests/progress/no-progress.sh | 46 + .../progress/no-progress.sh.stderr.glob | 96 + .../zstd/tests/cli-tests/progress/progress.sh | 41 + .../progress/progress.sh.stderr.glob | 62 + priv/zstd/tests/cli-tests/run.py | 738 +++ priv/zstd/tests/cli-tests/zstd-symlinks/setup | 6 + .../tests/cli-tests/zstd-symlinks/zstdcat.sh | 12 + .../zstd-symlinks/zstdcat.sh.stdout.exact | 8 + priv/zstd/tests/datagencli.c | 4 +- priv/zstd/tests/decodecorpus.c | 99 +- priv/zstd/tests/external_matchfinder.c | 140 + priv/zstd/tests/external_matchfinder.h | 39 + priv/zstd/tests/fullbench.c | 63 +- priv/zstd/tests/fuzz/.gitignore | 2 + priv/zstd/tests/fuzz/Makefile | 68 +- priv/zstd/tests/fuzz/README.md | 14 + priv/zstd/tests/fuzz/block_decompress.c | 2 +- priv/zstd/tests/fuzz/block_round_trip.c | 2 +- .../tests/fuzz/decompress_dstSize_tooSmall.c | 2 +- priv/zstd/tests/fuzz/dictionary_decompress.c | 2 +- priv/zstd/tests/fuzz/dictionary_loader.c | 2 +- priv/zstd/tests/fuzz/dictionary_round_trip.c | 36 +- .../tests/fuzz/dictionary_stream_round_trip.c | 2 +- priv/zstd/tests/fuzz/fse_read_ncount.c | 2 +- priv/zstd/tests/fuzz/fuzz.h | 5 +- priv/zstd/tests/fuzz/fuzz.py | 6 +- priv/zstd/tests/fuzz/fuzz_data_producer.c | 8 +- priv/zstd/tests/fuzz/fuzz_data_producer.h | 5 +- priv/zstd/tests/fuzz/fuzz_helpers.c | 2 +- priv/zstd/tests/fuzz/fuzz_helpers.h | 2 +- priv/zstd/tests/fuzz/huf_decompress.c | 68 + priv/zstd/tests/fuzz/huf_round_trip.c | 137 + .../tests/fuzz/raw_dictionary_round_trip.c | 2 +- priv/zstd/tests/fuzz/regression_driver.c | 2 +- priv/zstd/tests/fuzz/seekable_roundtrip.c | 2 +- .../tests/fuzz/sequence_compression_api.c | 292 +- priv/zstd/tests/fuzz/simple_compress.c | 4 +- priv/zstd/tests/fuzz/simple_decompress.c | 2 +- priv/zstd/tests/fuzz/simple_round_trip.c | 72 +- priv/zstd/tests/fuzz/stream_decompress.c | 10 +- priv/zstd/tests/fuzz/stream_round_trip.c | 20 +- priv/zstd/tests/fuzz/zstd_frame_info.c | 2 +- priv/zstd/tests/fuzz/zstd_helpers.c | 48 +- priv/zstd/tests/fuzz/zstd_helpers.h | 3 +- priv/zstd/tests/fuzzer.c | 831 ++- .../large-literal-and-match-lengths | Bin 0 -> 199998 bytes priv/zstd/tests/gzip/Makefile | 2 +- priv/zstd/tests/gzip/gzip-env.sh | 2 +- priv/zstd/tests/gzip/helin-segv.sh | 2 +- priv/zstd/tests/gzip/help-version.sh | 2 +- priv/zstd/tests/gzip/hufts.sh | 2 +- priv/zstd/tests/gzip/init.sh | 2 +- priv/zstd/tests/gzip/keep.sh | 2 +- priv/zstd/tests/gzip/list.sh | 2 +- priv/zstd/tests/gzip/memcpy-abuse.sh | 2 +- priv/zstd/tests/gzip/mixed.sh | 2 +- priv/zstd/tests/gzip/null-suffix-clobber.sh | 2 +- priv/zstd/tests/gzip/stdin.sh | 2 +- priv/zstd/tests/gzip/test-driver.sh | 2 +- priv/zstd/tests/gzip/trailing-nul.sh | 2 +- priv/zstd/tests/gzip/unpack-invalid.sh | 2 +- priv/zstd/tests/gzip/z-suffix.sh | 2 +- priv/zstd/tests/gzip/zdiff.sh | 2 +- priv/zstd/tests/gzip/zgrep-context.sh | 2 +- priv/zstd/tests/gzip/zgrep-f.sh | 2 +- priv/zstd/tests/gzip/zgrep-signal.sh | 2 +- priv/zstd/tests/gzip/znew-k.sh | 2 +- priv/zstd/tests/invalidDictionaries.c | 2 +- priv/zstd/tests/legacy.c | 2 +- ...td_partial_builds.sh => libzstd_builds.sh} | 9 +- priv/zstd/tests/longmatch.c | 35 +- priv/zstd/tests/paramgrill.c | 50 +- priv/zstd/tests/playTests.sh | 276 +- priv/zstd/tests/poolTests.c | 2 +- priv/zstd/tests/rateLimiter.py | 2 +- priv/zstd/tests/regression/.gitignore | 1 + priv/zstd/tests/regression/Makefile | 5 +- priv/zstd/tests/regression/config.c | 14 +- priv/zstd/tests/regression/config.h | 2 +- priv/zstd/tests/regression/data.c | 2 +- priv/zstd/tests/regression/data.h | 2 +- priv/zstd/tests/regression/levels.h | 14 +- priv/zstd/tests/regression/method.c | 2 +- priv/zstd/tests/regression/method.h | 2 +- priv/zstd/tests/regression/result.c | 2 +- priv/zstd/tests/regression/result.h | 2 +- priv/zstd/tests/regression/results.csv | 2052 +++--- priv/zstd/tests/regression/test.c | 2 +- priv/zstd/tests/roundTripCrash.c | 2 +- priv/zstd/tests/seqgen.c | 2 +- priv/zstd/tests/seqgen.h | 2 +- priv/zstd/tests/test-license.py | 7 +- priv/zstd/tests/test-variants.sh | 115 + priv/zstd/tests/test-zstd-versions.py | 185 +- priv/zstd/tests/zstreamtest.c | 720 ++- priv/zstd/zlibWrapper/Makefile | 7 +- priv/zstd/zlibWrapper/README.md | 4 +- priv/zstd/zlibWrapper/examples/example.c | 26 +- .../zlibWrapper/examples/example_original.c | 26 +- priv/zstd/zlibWrapper/examples/fitblk.c | 2 +- .../zlibWrapper/examples/fitblk_original.c | 2 +- priv/zstd/zlibWrapper/examples/minigzip.c | 36 +- priv/zstd/zlibWrapper/examples/zwrapbench.c | 20 +- priv/zstd/zlibWrapper/gzclose.c | 2 +- priv/zstd/zlibWrapper/gzcompatibility.h | 16 +- priv/zstd/zlibWrapper/gzguts.h | 20 +- priv/zstd/zlibWrapper/gzlib.c | 6 +- priv/zstd/zlibWrapper/gzread.c | 20 +- priv/zstd/zlibWrapper/gzwrite.c | 10 +- priv/zstd/zlibWrapper/zstd_zlibwrapper.c | 95 +- priv/zstd/zlibWrapper/zstd_zlibwrapper.h | 5 +- 492 files changed, 28991 insertions(+), 11447 deletions(-) create mode 100644 priv/zstd/.github/dependabot.yml create mode 100644 priv/zstd/.github/workflows/dev-long-tests.yml create mode 100644 priv/zstd/.github/workflows/dev-short-tests.yml delete mode 100644 priv/zstd/.github/workflows/generic-dev.yml delete mode 100644 priv/zstd/.github/workflows/generic-release.yml delete mode 100644 priv/zstd/.github/workflows/linux-kernel.yml delete mode 100644 priv/zstd/.github/workflows/main.yml create mode 100644 priv/zstd/.github/workflows/scorecards.yml create mode 100644 priv/zstd/Package.swift create mode 100644 priv/zstd/build/cmake/CMakeModules/JoinPaths.cmake create mode 100755 priv/zstd/build/single_file_libs/combine.py create mode 100644 priv/zstd/contrib/externalSequenceProducer/.gitignore create mode 100644 priv/zstd/contrib/externalSequenceProducer/Makefile create mode 100644 priv/zstd/contrib/externalSequenceProducer/README.md create mode 100644 priv/zstd/contrib/externalSequenceProducer/main.c create mode 100644 priv/zstd/contrib/externalSequenceProducer/sequence_producer.c create mode 100644 priv/zstd/contrib/externalSequenceProducer/sequence_producer.h create mode 100644 priv/zstd/contrib/linux-kernel/zstd_common_module.c create mode 100644 priv/zstd/contrib/pzstd/utils/Portability.h create mode 100644 priv/zstd/contrib/seqBench/Makefile create mode 100644 priv/zstd/contrib/seqBench/seqBench.c create mode 100644 priv/zstd/lib/common/bits.h create mode 100644 priv/zstd/lib/common/portability_macros.h create mode 100644 priv/zstd/lib/compress/clevels.h create mode 100644 priv/zstd/lib/decompress/huf_decompress_amd64.S create mode 100644 priv/zstd/lib/libzstd.mk create mode 100644 priv/zstd/lib/module.modulemap create mode 100644 priv/zstd/programs/fileio_asyncio.c create mode 100644 priv/zstd/programs/fileio_asyncio.h create mode 100644 priv/zstd/programs/fileio_common.h create mode 100644 priv/zstd/programs/fileio_types.h create mode 100755 priv/zstd/tests/check_size.py create mode 100644 priv/zstd/tests/cli-tests/.gitignore create mode 100644 priv/zstd/tests/cli-tests/README.md create mode 100755 priv/zstd/tests/cli-tests/basic/help.sh create mode 100644 priv/zstd/tests/cli-tests/basic/help.sh.stdout.glob create mode 100755 priv/zstd/tests/cli-tests/basic/memlimit.sh create mode 100644 priv/zstd/tests/cli-tests/basic/memlimit.sh.stderr.exact create mode 100644 priv/zstd/tests/cli-tests/basic/memlimit.sh.stdout.exact create mode 100755 priv/zstd/tests/cli-tests/basic/output_dir.sh create mode 100644 priv/zstd/tests/cli-tests/basic/output_dir.sh.stderr.exact create mode 100644 priv/zstd/tests/cli-tests/basic/output_dir.sh.stdout.exact create mode 100755 priv/zstd/tests/cli-tests/basic/version.sh create mode 100644 priv/zstd/tests/cli-tests/basic/version.sh.stdout.glob create mode 100755 priv/zstd/tests/cli-tests/bin/cmp_size create mode 100755 priv/zstd/tests/cli-tests/bin/datagen create mode 100755 priv/zstd/tests/cli-tests/bin/die create mode 100755 priv/zstd/tests/cli-tests/bin/println create mode 120000 priv/zstd/tests/cli-tests/bin/unzstd create mode 100755 priv/zstd/tests/cli-tests/bin/zstd create mode 120000 priv/zstd/tests/cli-tests/bin/zstdcat create mode 100755 priv/zstd/tests/cli-tests/bin/zstdgrep create mode 100755 priv/zstd/tests/cli-tests/bin/zstdless create mode 100755 priv/zstd/tests/cli-tests/cltools/setup create mode 100755 priv/zstd/tests/cli-tests/cltools/zstdgrep.sh create mode 100644 priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.exit create mode 100644 priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stderr.exact create mode 100644 priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stdout.glob create mode 100755 priv/zstd/tests/cli-tests/cltools/zstdless.sh create mode 100644 priv/zstd/tests/cli-tests/cltools/zstdless.sh.stderr.exact create mode 100644 priv/zstd/tests/cli-tests/cltools/zstdless.sh.stdout.glob create mode 100644 priv/zstd/tests/cli-tests/common/format.sh create mode 100644 priv/zstd/tests/cli-tests/common/mtime.sh create mode 100644 priv/zstd/tests/cli-tests/common/permissions.sh create mode 100644 priv/zstd/tests/cli-tests/common/platform.sh create mode 100755 priv/zstd/tests/cli-tests/compression/adapt.sh create mode 100755 priv/zstd/tests/cli-tests/compression/basic.sh create mode 100755 priv/zstd/tests/cli-tests/compression/compress-literals.sh create mode 100755 priv/zstd/tests/cli-tests/compression/format.sh create mode 100755 priv/zstd/tests/cli-tests/compression/golden.sh create mode 100755 priv/zstd/tests/cli-tests/compression/gzip-compat.sh create mode 100755 priv/zstd/tests/cli-tests/compression/levels.sh create mode 100644 priv/zstd/tests/cli-tests/compression/levels.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/compression/long-distance-matcher.sh create mode 100755 priv/zstd/tests/cli-tests/compression/multi-threaded.sh create mode 100644 priv/zstd/tests/cli-tests/compression/multi-threaded.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/compression/multiple-files.sh create mode 100644 priv/zstd/tests/cli-tests/compression/multiple-files.sh.stdout.exact create mode 100755 priv/zstd/tests/cli-tests/compression/row-match-finder.sh create mode 100755 priv/zstd/tests/cli-tests/compression/setup create mode 100755 priv/zstd/tests/cli-tests/compression/stream-size.sh create mode 100755 priv/zstd/tests/cli-tests/compression/verbose-wlog.sh create mode 100644 priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stderr.glob create mode 100644 priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stdout.glob create mode 100755 priv/zstd/tests/cli-tests/compression/window-resize.sh create mode 100644 priv/zstd/tests/cli-tests/compression/window-resize.sh.stderr.ignore create mode 100644 priv/zstd/tests/cli-tests/compression/window-resize.sh.stdout.glob create mode 100755 priv/zstd/tests/cli-tests/decompression/golden.sh create mode 100755 priv/zstd/tests/cli-tests/decompression/pass-through.sh create mode 100644 priv/zstd/tests/cli-tests/decompression/pass-through.sh.stderr.exact create mode 100644 priv/zstd/tests/cli-tests/decompression/pass-through.sh.stdout.exact create mode 100755 priv/zstd/tests/cli-tests/dict-builder/empty-input.sh create mode 100644 priv/zstd/tests/cli-tests/dict-builder/empty-input.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh create mode 100644 priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.exit create mode 100644 priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh create mode 100644 priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/dictionaries/golden.sh create mode 100755 priv/zstd/tests/cli-tests/dictionaries/setup create mode 100755 priv/zstd/tests/cli-tests/dictionaries/setup_once create mode 100755 priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh create mode 100644 priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh.stderr.exact create mode 100755 priv/zstd/tests/cli-tests/progress/no-progress.sh create mode 100644 priv/zstd/tests/cli-tests/progress/no-progress.sh.stderr.glob create mode 100755 priv/zstd/tests/cli-tests/progress/progress.sh create mode 100644 priv/zstd/tests/cli-tests/progress/progress.sh.stderr.glob create mode 100755 priv/zstd/tests/cli-tests/run.py create mode 100755 priv/zstd/tests/cli-tests/zstd-symlinks/setup create mode 100755 priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh create mode 100644 priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh.stdout.exact create mode 100644 priv/zstd/tests/external_matchfinder.c create mode 100644 priv/zstd/tests/external_matchfinder.h create mode 100644 priv/zstd/tests/fuzz/huf_decompress.c create mode 100644 priv/zstd/tests/fuzz/huf_round_trip.c create mode 100644 priv/zstd/tests/golden-compression/large-literal-and-match-lengths rename priv/zstd/tests/{libzstd_partial_builds.sh => libzstd_builds.sh} (92%) create mode 100755 priv/zstd/tests/test-variants.sh diff --git a/README.md b/README.md index c87bec3..9287a9f 100644 --- a/README.md +++ b/README.md @@ -7,7 +7,7 @@ Zstd binding for Erlang/Elixir http://facebook.github.io/zstd/ -The current version uses [v1.5.0](https://github.com/facebook/zstd/releases/tag/v1.5.0) +The current version uses [v1.5.4](https://github.com/facebook/zstd/releases/tag/v1.5.4) Usage ----- diff --git a/priv/zstd/.circleci/config.yml b/priv/zstd/.circleci/config.yml index c634737..621ac49 100644 --- a/priv/zstd/.circleci/config.yml +++ b/priv/zstd/.circleci/config.yml @@ -13,13 +13,13 @@ jobs: name: Test command: | ./tests/test-license.py - cc -v; CFLAGS="-O0 -Werror -pedantic" make all && make clean - make c99build ; make clean - make c11build ; make clean - make aarch64build ; make clean + cc -v + CFLAGS="-O0 -Werror -pedantic" make allmost; make clean + make c99build; make clean + make c11build; make clean make -j regressiontest; make clean - make shortest ; make clean - make cxxtest ; make clean + make shortest; make clean + make cxxtest; make clean # the second half of the jobs are in this test short-tests-1: docker: @@ -34,6 +34,7 @@ jobs: make ppc64build V=1; make clean make ppcbuild V=1; make clean make armbuild V=1; make clean + make aarch64build V=1; make clean make -C tests test-legacy test-longmatch; make clean make -C lib libzstd-nomt; make clean # This step should only be run in a cron job @@ -84,31 +85,10 @@ workflows: commit: jobs: # Run the tests in parallel - - short-tests-0: - filters: - tags: - only: /.*/ - - short-tests-1: - filters: - tags: - only: /.*/ - # Create a branch called regression and set it to dev to force a - # regression test run - - regression-test: - filters: - branches: - only: - - regression - # Only run on release tags. - - publish-github-release: - requires: - - short-tests-0 - - short-tests-1 - filters: - branches: - ignore: /.*/ - tags: - only: /^v\d+\.\d+\.\d+$/ + - short-tests-0 + - short-tests-1 + - regression-test + nightly: triggers: - schedule: @@ -120,7 +100,7 @@ workflows: - dev - master jobs: - # Run daily long regression tests + # Run daily regression tests - regression-test diff --git a/priv/zstd/.circleci/images/primary/Dockerfile b/priv/zstd/.circleci/images/primary/Dockerfile index dd80041..bfcb6af 100644 --- a/priv/zstd/.circleci/images/primary/Dockerfile +++ b/priv/zstd/.circleci/images/primary/Dockerfile @@ -1,4 +1,4 @@ -FROM circleci/buildpack-deps:bionic +FROM circleci/buildpack-deps:focal RUN sudo dpkg --add-architecture i386 RUN sudo apt-get -y -qq update diff --git a/priv/zstd/.cirrus.yml b/priv/zstd/.cirrus.yml index fe17aac..27ca65e 100644 --- a/priv/zstd/.cirrus.yml +++ b/priv/zstd/.cirrus.yml @@ -2,9 +2,8 @@ task: name: FreeBSD (shortest) freebsd_instance: matrix: + image_family: freebsd-13-0 image_family: freebsd-12-2 - # The stable 11.3 image causes "Agent is not responding" so use a snapshot - image_family: freebsd-11-3-snap install_script: pkg install -y gmake coreutils script: | MOREFLAGS="-Werror" gmake -j all diff --git a/priv/zstd/.github/ISSUE_TEMPLATE/bug_report.md b/priv/zstd/.github/ISSUE_TEMPLATE/bug_report.md index bacdac2..755a46a 100644 --- a/priv/zstd/.github/ISSUE_TEMPLATE/bug_report.md +++ b/priv/zstd/.github/ISSUE_TEMPLATE/bug_report.md @@ -26,8 +26,8 @@ If applicable, add screenshots and charts to help explain your problem. **Desktop (please complete the following information):** - OS: [e.g. Mac] - Version [e.g. 22] - - Compiler [e.g gcc] - - Flags [e.g O2] + - Compiler [e.g. gcc] + - Flags [e.g. O2] - Other relevant hardware specs [e.g. Dual-core] - Build system [e.g. Makefile] diff --git a/priv/zstd/.github/dependabot.yml b/priv/zstd/.github/dependabot.yml new file mode 100644 index 0000000..5ace460 --- /dev/null +++ b/priv/zstd/.github/dependabot.yml @@ -0,0 +1,6 @@ +version: 2 +updates: + - package-ecosystem: "github-actions" + directory: "/" + schedule: + interval: "weekly" diff --git a/priv/zstd/.github/workflows/dev-long-tests.yml b/priv/zstd/.github/workflows/dev-long-tests.yml new file mode 100644 index 0000000..1c8c9ec --- /dev/null +++ b/priv/zstd/.github/workflows/dev-long-tests.yml @@ -0,0 +1,271 @@ +name: dev-long-tests +# Tests longer than 10mn + +concurrency: + group: long-${{ github.ref }} + cancel-in-progress: true + +on: + pull_request: + branches: [ dev, release, actionsTest ] + +jobs: + make-all: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: make all + run: make all + + # lasts ~24mn + make-test: + runs-on: ubuntu-latest + env: + DEVNULLRIGHTS: 1 + READFROMBLOCKDEVICE: 1 + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: make test + run: make test + + # lasts ~26mn + make-test-osx: + runs-on: macos-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: OS-X test + run: make test # make -c lib all doesn't work because of the fact that it's not a tty + + no-intrinsics-fuzztest: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: no intrinsics fuzztest + run: MOREFLAGS="-DZSTD_NO_INTRINSICS" make -C tests fuzztest + + tsan-zstreamtest: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: thread sanitizer zstreamtest + run: CC=clang ZSTREAM_TESTTIME=-T3mn make tsan-test-zstream + + ubsan-zstreamtest: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: undefined behavior sanitizer zstreamtest + run: CC=clang make uasan-test-zstream + + # lasts ~15mn + tsan-fuzztest: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: thread sanitizer fuzztest + run: CC=clang make tsan-fuzztest + + + big-tests-zstreamtest32: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: zstream tests in 32bit mode, with big tests + run: | + sudo apt-get -qqq update + make libc6install + CC=clang make -C tests test-zstream32 FUZZER_FLAGS="--big-tests" + + # lasts ~23mn + gcc-8-asan-ubsan-testzstd: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: gcc-8 + ASan + UBSan + Test Zstd + # See https://askubuntu.com/a/1428822 + run: | + echo "deb [arch=amd64] http://archive.ubuntu.com/ubuntu focal main universe" | sudo tee -a /etc/apt/sources.list + sudo apt-get -qqq update + make gcc8install + CC=gcc-8 make -j uasan-test-zstd + msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=${{matrix.toolset}} + /t:Clean,Build /p:Platform=${{matrix.platform}} /p:Configuration=${{matrix.configuration}} + + # This tests that we don't accidently grow the size too much. + # If the size grows intentionally, you can raise these numbers. + # But we do need to think about binary size, since it is a concern. + libzstd-size: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: libzstd size test + run: | + make clean && make -j -C lib libzstd && ./tests/check_size.py lib/libzstd.so 1100000 + make clean && make -j -C lib libzstd ZSTD_LIB_COMPRESSION=0 ZSTD_LIB_DICTBUILDER=0 && ./tests/check_size.py lib/libzstd.so 400000 + make clean && make -j -C lib libzstd ZSTD_LIB_MINIFY=1 && ./tests/check_size.py lib/libzstd.so 300000 + make clean && make -j -C lib libzstd ZSTD_LIB_MINIFY=1 ZSTD_LIB_COMPRESSION=0 ZSTD_LIB_DICTBUILDER=0 && ./tests/check_size.py lib/libzstd.so 80000 + + minimal-decompressor-macros: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: minimal decompressor macros + run: | + make clean && make -j all ZSTD_LIB_MINIFY=1 MOREFLAGS="-Werror" + make clean && make check ZSTD_LIB_MINIFY=1 MOREFLAGS="-Werror" + make clean && make -j all MOREFLAGS="-Werror -DHUF_FORCE_DECOMPRESS_X1 -DZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT" + make clean && make check MOREFLAGS="-Werror -DHUF_FORCE_DECOMPRESS_X1 -DZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT" + make clean && make -j all MOREFLAGS="-Werror -DHUF_FORCE_DECOMPRESS_X2 -DZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG" + make clean && make check MOREFLAGS="-Werror -DHUF_FORCE_DECOMPRESS_X2 -DZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG" + make clean && make -j all MOREFLAGS="-Werror -DZSTD_NO_INLINE -DZSTD_STRIP_ERROR_STRINGS" + make clean && make check MOREFLAGS="-Werror -DZSTD_NO_INLINE -DZSTD_STRIP_ERROR_STRINGS" + + dynamic-bmi2: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: dynamic bmi2 tests + run: | + make clean && make -j check MOREFLAGS="-O0 -Werror -mbmi2" + make clean && make -j check MOREFLAGS="-O0 -Werror -DDYNAMIC_BMI2=1" + make clean && make -j check MOREFLAGS="-O0 -Werror -DDYNAMIC_BMI2=1 -mbmi2" + make clean && make -j check MOREFLAGS="-O0 -Werror -DDYNAMIC_BMI2=0" + make clean && make -j check MOREFLAGS="-O0 -Werror -DDYNAMIC_BMI2=0 -mbmi2" + + test-variants: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: make all variants & validate + run: | + make -j -C programs allVariants MOREFLAGS=-O0 + ./tests/test-variants.sh + + + qemu-consistency: + name: QEMU ${{ matrix.name }} + runs-on: ubuntu-20.04 + strategy: + fail-fast: false # 'false' means Don't stop matrix workflows even if some matrix failed. + matrix: + include: [ + { name: ARM, xcc_pkg: gcc-arm-linux-gnueabi, xcc: arm-linux-gnueabi-gcc, xemu_pkg: qemu-system-arm, xemu: qemu-arm-static }, + { name: ARM64, xcc_pkg: gcc-aarch64-linux-gnu, xcc: aarch64-linux-gnu-gcc, xemu_pkg: qemu-system-arm, xemu: qemu-aarch64-static }, + { name: PPC, xcc_pkg: gcc-powerpc-linux-gnu, xcc: powerpc-linux-gnu-gcc, xemu_pkg: qemu-system-ppc, xemu: qemu-ppc-static }, + { name: PPC64LE, xcc_pkg: gcc-powerpc64le-linux-gnu, xcc: powerpc64le-linux-gnu-gcc, xemu_pkg: qemu-system-ppc, xemu: qemu-ppc64le-static }, + { name: S390X, xcc_pkg: gcc-s390x-linux-gnu, xcc: s390x-linux-gnu-gcc, xemu_pkg: qemu-system-s390x, xemu: qemu-s390x-static }, + { name: MIPS, xcc_pkg: gcc-mips-linux-gnu, xcc: mips-linux-gnu-gcc, xemu_pkg: qemu-system-mips, xemu: qemu-mips-static }, + { name: M68K, xcc_pkg: gcc-m68k-linux-gnu, xcc: m68k-linux-gnu-gcc, xemu_pkg: qemu-system-m68k, xemu: qemu-m68k-static }, + ] + env: # Set environment variables + XCC: ${{ matrix.xcc }} + XEMU: ${{ matrix.xemu }} + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: apt update & install + run: | + sudo apt-get update + sudo apt-get install gcc-multilib g++-multilib qemu-utils qemu-user-static + sudo apt-get install ${{ matrix.xcc_pkg }} ${{ matrix.xemu_pkg }} + - name: Environment info + run: | + echo && which $XCC + echo && $XCC --version + echo && $XCC -v # Show built-in specs + echo && which $XEMU + echo && $XEMU --version + - name: ARM + if: ${{ matrix.name == 'ARM' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + - name: ARM64 + if: ${{ matrix.name == 'ARM64' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + - name: PPC + if: ${{ matrix.name == 'PPC' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + - name: PPC64LE + if: ${{ matrix.name == 'PPC64LE' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + - name: S390X + if: ${{ matrix.name == 'S390X' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + - name: MIPS + if: ${{ matrix.name == 'MIPS' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + - name: M68K + if: ${{ matrix.name == 'M68K' }} + run: | + LDFLAGS="-static" CC=$XCC QEMU_SYS=$XEMU make clean check + + mingw-short-test: + runs-on: windows-latest + strategy: + fail-fast: false # 'false' means Don't stop matrix workflows even if some matrix failed. + matrix: + include: [ + { compiler: gcc, msystem: MINGW32, cflags: "-Werror"}, + { compiler: gcc, msystem: MINGW64, cflags: "-Werror"}, + { compiler: clang, msystem: MINGW64, cflags: "--target=x86_64-w64-mingw32 -Werror -Wconversion -Wno-sign-conversion -Wno-unused-command-line-argument"}, + ] + defaults: + run: + shell: msys2 {0} + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - uses: msys2/setup-msys2@v2 + with: + msystem: ${{ matrix.msystem }} + install: make diffutils + update: true + # Based on https://ariya.io/2020/07/on-github-actions-with-msys2 + - name: install mingw gcc i686 + if: ${{ (matrix.msystem == 'MINGW32') && (matrix.compiler == 'gcc') }} + run: pacman --noconfirm -S mingw-w64-i686-gcc + - name: install mingw gcc x86_64 + if: ${{ (matrix.msystem == 'MINGW64') && (matrix.compiler == 'gcc') }} + run: pacman --noconfirm -S mingw-w64-x86_64-gcc + - name: install mingw clang i686 + if: ${{ (matrix.msystem == 'MINGW32') && (matrix.compiler == 'clang') }} + run: pacman --noconfirm -S mingw-w64-i686-clang + - name: install mingw clang x86_64 + if: ${{ (matrix.msystem == 'MINGW64') && (matrix.compiler == 'clang') }} + run: pacman --noconfirm -S mingw-w64-x86_64-clang + - name: run mingw tests + run: | + make -v + export CC=${{ matrix.compiler }} + $CC --version + CFLAGS="${{ matrix.cflags }}" make -j allzstd + echo "Testing $CC ${{ matrix.msystem }}" + make clean + MSYS="" make check + + visual-runtime-tests: + runs-on: windows-latest + strategy: + matrix: + platform: [x64, Win32] + configuration: [Release] + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: Add MSBuild to PATH + uses: microsoft/setup-msbuild@v1.3 + - name: Build and run tests + working-directory: ${{env.GITHUB_WORKSPACE}} + env: + ZSTD_BIN: ./zstd.exe + DATAGEN_BIN: ./datagen.exe + # See https://docs.microsoft.com/visualstudio/msbuild/msbuild-command-line-reference + run: | + msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v142 /t:Clean,Build /p:Platform=${{matrix.platform}} /p:Configuration=${{matrix.configuration}} + COPY build\VS2010\bin\${{matrix.platform}}_${{matrix.configuration}}\*.exe tests\ + CD tests + sh -e playTests.sh + .\fuzzer.exe -T2m + + # Following instructions at: https://github.com/marketplace/actions/install-cygwin-action + cygwin-tests: + runs-on: windows-latest + steps: + - run: git config --global core.autocrlf input + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - uses: cygwin/cygwin-install-action@f5e0f048310c425e84bc789f493a828c6dc80a25 # tag=master + with: + platform: x86_64 + packages: >- + autoconf, + automake, + gcc-g++, + make, + mingw64-x86_64-gcc-g++, + patch, + perl + - name: cygwin tests + shell: C:\cygwin\bin\bash.exe --noprofile --norc -eo pipefail '{0}' + run: >- + export PATH=/usr/bin:$(cygpath ${SYSTEMROOT})/system32 && + export CFLAGS="-Werror -O1" && + ls && + make -j allzstd && + make -C tests fuzzer && + ./tests/fuzzer.exe -v -T1m + + intel-cet-compatibility: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: Build Zstd + run: | + make -j zstd V=1 + readelf -n zstd + - name: Get Intel SDE + run: | + curl -LO https://downloadmirror.intel.com/684899/sde-external-9.0.0-2021-11-07-lin.tar.xz + tar xJvf sde-external-9.0.0-2021-11-07-lin.tar.xz + - name: Configure Permissions + run: | + echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope + - name: Run Under SDE + run: | + sde-external-9.0.0-2021-11-07-lin/sde -cet -cet-raise 0 -cet-endbr-exe -cet-stderr -cet-abort -- ./zstd -b3 + + pkg-config: + runs-on: ubuntu-latest + container: + image: debian:testing + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: Install dependencies + run: | + apt -y update + apt -y install --no-install-recommends gcc libc6-dev make pkg-config + - name: Build and install + run: make -C lib install + - name: Test pkg-config + run: | + cc -Wall -Wextra -Wpedantic -Werror -o simple examples/simple_compression.c $(pkg-config --cflags --libs libzstd) + ./simple LICENSE + + versions-compatibility: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: Versions Compatibility Test + run: | + make -C tests versionsTest + + clangbuild: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: make clangbuild + run: | + make clangbuild + + clang-pgo: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: Build PGO Zstd with Clang + env: + CC: clang-14 + LLVM_PROFDATA: llvm-profdata-14 + run: | + make -C programs zstd-pgo + ./programs/zstd -b + + gcc-pgo: + runs-on: ubuntu-latest + steps: + - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 + - name: Build PGO Zstd with GCC + env: + CC: gcc + run: | + make -C programs zstd-pgo + ./programs/zstd -b + + +# For reference : icc tests +# icc tests are currently failing on Github Actions, likely to issues during installation stage +# To be fixed later +# +# icc: +# name: icc-check +# runs-on: ubuntu-latest +# steps: +# - name: install icc +# run: | +# export DEBIAN_FRONTEND=noninteractive +# sudo apt-get -qqq update +# sudo apt-get install -y wget build-essential pkg-config cmake ca-certificates gnupg +# sudo wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2023.PUB +# sudo apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS-2023.PUB +# sudo add-apt-repository "deb https://apt.repos.intel.com/oneapi all main" +# sudo apt-get update +# sudo apt-get install -y intel-basekit intel-hpckit +# - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # tag=v3 +# - name: make check +# run: | +# make CC=/opt/intel/oneapi/compiler/latest/linux/bin/intel64/icc check diff --git a/priv/zstd/.github/workflows/generic-dev.yml b/priv/zstd/.github/workflows/generic-dev.yml deleted file mode 100644 index ae4ee7e..0000000 --- a/priv/zstd/.github/workflows/generic-dev.yml +++ /dev/null @@ -1,225 +0,0 @@ -name: generic-dev - -on: - pull_request: - branches: [ dev, release, actionsTest ] - -jobs: - -# Dev PR jobs that still have to be migrated from travis -# -# versionTag (only on release tags) -# valgrindTest (keeps failing for some reason. need investigation) -# staticAnalyze (need trusty so need self-hosted) -# pcc-fuzz: (need trusty so need self-hosted) -# min-decomp-macros (flakey) -# -# setting up self-hosted is pretty straightforward, but -# I need admins permissions to the repo for that it looks like -# So I'm tabling that for now -# -# The release branch exclusive jobs will be in a separate -# workflow file (the osx tests and meson build that is) - - benchmarking: - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v2 - - name: make benchmarking - run: make benchmarking - - test: - runs-on: ubuntu-latest - env: - DEVNULLRIGHTS: 1 - READFROMBLOCKDEVICE: 1 - steps: - - uses: actions/checkout@v2 - - name: make test - run: make test - - check-32bit: # designed to catch https://github.com/facebook/zstd/issues/2428 - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v2 - - name: make check on 32-bit - run: | - sudo apt update - APT_PACKAGES="gcc-multilib" make apt-install - CFLAGS="-m32 -O1 -fstack-protector" make check V=1 - - gcc-7-libzstd: - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v2 - - name: gcc-7 + libzstdmt compilation - run: | - make gcc7install - CC=gcc-7 CFLAGS=-Werror make -j all - make clean - LDFLAGS=-Wl,--no-undefined make -C lib libzstd-mt - - # candidate test (to check) : underlink test - # LDFLAGS=-Wl,--no-undefined : will make the linker fail if dll is underlinked - - gcc-8-asan-ubsan-testzstd: - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v2 - - name: gcc-8 + ASan + UBSan + Test Zstd - run: | - make gcc8install - CC=gcc-8 CFLAGS="-Werror" make -j all - make clean - CC=gcc-8 make -j uasan-test-zstd git push origin ``` @@ -68,8 +68,8 @@ Our contribution process works in three main stages: ``` 2. Code Review and CI tests * Ensure CI tests pass: - * Before sharing anything to the community, make sure that all CI tests pass on your local fork. - See our section on setting up your CI environment for more information on how to do this. + * Before sharing anything to the community, create a pull request in your own fork against the dev branch + and make sure that all GitHub Actions CI tests pass. See the Continuous Integration section below for more information. * Ensure that static analysis passes on your development machine. See the Static Analysis section below to see how to do this. * Create a pull request: @@ -104,7 +104,7 @@ Our contribution process works in three main stages: issue at hand, then please indicate this by requesting that an issue be closed by commenting. * Just because your changes have been merged does not mean the topic or larger issue is complete. Remember that the change must make it to an official zstd release for it to be meaningful. We recommend - that contributers track the activity on their pull request and corresponding issue(s) page(s) until + that contributors track the activity on their pull request and corresponding issue(s) page(s) until their change makes it to the next release of zstd. Users will often discover bugs in your code or suggest ways to refine and improve your initial changes even after the pull request is merged. @@ -134,11 +134,47 @@ It can be useful to look at additional static analyzers once in a while (and we - Static analyzers are full of false positive. The signal to noise ratio is actually pretty low. - A good CI policy is "zero-warning tolerance". That means that all issues must be solved, including false positives. This quickly becomes a tedious workload. - Multiple static analyzers will feature multiple kind of false positives, sometimes applying to the same code but in different ways leading to : - + torteous code, trying to please multiple constraints, hurting readability and therefore maintenance. Sometimes, such complexity introduce other more subtle bugs, that are just out of scope of the analyzers. + + tortuous code, trying to please multiple constraints, hurting readability and therefore maintenance. Sometimes, such complexity introduce other more subtle bugs, that are just out of scope of the analyzers. + sometimes, these constraints are mutually exclusive : if one try to solve one, the other static analyzer will complain, they can't be both happy at the same time. - As if that was not enough, the list of false positives change with each version. It's hard enough to follow one static analyzer, but multiple ones with their own update agenda, this quickly becomes a massive velocity reducer. -This is different from running a static analyzer once in a while, looking at the output, and __cherry picking__ a few warnings that seem helpful, either because they detected a genuine risk of bug, or because it helps expressing the code in a way which is more readable or more difficult to misuse. These kind of reports can be useful, and are accepted. +This is different from running a static analyzer once in a while, looking at the output, and __cherry picking__ a few warnings that seem helpful, either because they detected a genuine risk of bug, or because it helps expressing the code in a way which is more readable or more difficult to misuse. These kinds of reports can be useful, and are accepted. + +## Continuous Integration +CI tests run every time a pull request (PR) is created or updated. The exact tests +that get run will depend on the destination branch you specify. Some tests take +longer to run than others. Currently, our CI is set up to run a short +series of tests when creating a PR to the dev branch and a longer series of tests +when creating a PR to the release branch. You can look in the configuration files +of the respective CI platform for more information on what gets run when. + +Most people will just want to create a PR with the destination set to their local dev +branch of zstd. You can then find the status of the tests on the PR's page. You can also +re-run tests and cancel running tests from the PR page or from the respective CI's dashboard. + +Almost all of zstd's CI runs on GitHub Actions (configured at `.github/workflows`), which will automatically run on PRs to your +own fork. A small number of tests run on other services (e.g. Travis CI, Circle CI, Appveyor). +These require work to set up on your local fork, and (at least for Travis CI) cost money. +Therefore, if the PR on your local fork passes GitHub Actions, feel free to submit a PR +against the main repo. + +### Third-party CI +A small number of tests cannot run on GitHub Actions, or have yet to be migrated. +For these, we use a variety of third-party services (listed below). It is not necessary to set +these up on your fork in order to contribute to zstd; however, we do link to instructions for those +who want earlier signal. + +| Service | Purpose | Setup Links | Config Path | +|-----------|------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------|------------------------| +| Travis CI | Used for testing on non-x86 architectures such as PowerPC | https://docs.travis-ci.com/user/tutorial/#to-get-started-with-travis-ci-using-github
https://github.com/marketplace/travis-ci | `.travis.yml` | +| AppVeyor | Used for some Windows testing (e.g. cygwin, mingw) | https://www.appveyor.com/blog/2018/10/02/github-apps-integration/
https://github.com/marketplace/appveyor | `appveyor.yml` | +| Cirrus CI | Used for testing on FreeBSD | https://github.com/marketplace/cirrus-ci/ | `.cirrus.yml` | +| Circle CI | Historically was used to provide faster signal,
but we may be able to migrate these to Github Actions | https://circleci.com/docs/2.0/getting-started/#setting-up-circleci
https://youtu.be/Js3hMUsSZ2c
https://circleci.com/docs/2.0/enable-checks/ | `.circleci/config.yml` | + +Note: the instructions linked above mostly cover how to set up a repository with CI from scratch. +The general idea should be the same for setting up CI on your fork of zstd, but you may have to +follow slightly different steps. In particular, please ignore any instructions related to setting up +config files (since zstd already has configs for each of these services). ## Performance Performance is extremely important for zstd and we only merge pull requests whose performance @@ -161,7 +197,7 @@ something subtle merged is extensive benchmarking. You will be doing us a great take the time to run extensive, long-duration, and potentially cross-(os, platform, process, etc) benchmarks on your end before submitting a PR. Of course, you will not be able to benchmark your changes on every single processor and os out there (and neither will we) but do that best -you can:) We've adding some things to think about when benchmarking below in the Benchmarking +you can:) We've added some things to think about when benchmarking below in the Benchmarking Performance section which might be helpful for you. 3. Optimizing performance for a certain OS, processor vendor, compiler, or network system is a perfectly legitimate thing to do as long as it does not harm the overall performance health of Zstd. @@ -237,7 +273,7 @@ for that options you have just provided. If you want to look at the internals of benchmarking script works, you can check out programs/benchzstd.c For example: say you have made a change that you believe improves the speed of zstd level 1. The -very first thing you should use to asses whether you actually achieved any sort of improvement +very first thing you should use to assess whether you actually achieved any sort of improvement is `zstd -b`. You might try to do something like this. Note: you can use the `-i` option to specify a running time for your benchmark in seconds (default is 3 seconds). Usually, the longer the running time, the more stable your results will be. @@ -263,24 +299,24 @@ this method of evaluation will not be sufficient. ### Profiling There are a number of great profilers out there. We're going to briefly mention how you can profile your code using `instruments` on mac, `perf` on linux and `visual studio profiler` -on windows. +on Windows. Say you have an idea for a change that you think will provide some good performance gains for level 1 compression on Zstd. Typically this means, you have identified a section of code that you think can be made to run faster. The first thing you will want to do is make sure that the piece of code is actually taking up -a notable amount of time to run. It is usually not worth optimzing something which accounts for less than +a notable amount of time to run. It is usually not worth optimizing something which accounts for less than 0.0001% of the total running time. Luckily, there are tools to help with this. Profilers will let you see how much time your code spends inside a particular function. -If your target code snippit is only part of a function, it might be worth trying to -isolate that snippit by moving it to its own function (this is usually not necessary but +If your target code snippet is only part of a function, it might be worth trying to +isolate that snippet by moving it to its own function (this is usually not necessary but might be). -Most profilers (including the profilers dicusssed below) will generate a call graph of -functions for you. Your goal will be to find your function of interest in this call grapch -and then inspect the time spent inside of it. You might also want to to look at the -annotated assembly which most profilers will provide you with. +Most profilers (including the profilers discussed below) will generate a call graph of +functions for you. Your goal will be to find your function of interest in this call graph +and then inspect the time spent inside of it. You might also want to look at the annotated +assembly which most profilers will provide you with. #### Instruments We will once again consider the scenario where you think you've identified a piece of code @@ -294,23 +330,23 @@ Instruments. * You will want a benchmark that runs for at least a few seconds (5 seconds will usually be long enough). This way the profiler will have something to work with and you will have ample time to attach your profiler to this process:) - * I will just use benchzstd as my bencharmking script for this example: + * I will just use benchzstd as my benchmarmking script for this example: ``` $ zstd -b1 -i5 # this will run for 5 seconds ``` 5. Once you run your benchmarking script, switch back over to instruments and attach your process to the time profiler. You can do this by: * Clicking on the `All Processes` drop down in the top left of the toolbar. - * Selecting your process from the dropdown. In my case, it is just going to be labled + * Selecting your process from the dropdown. In my case, it is just going to be labeled `zstd` * Hitting the bright red record circle button on the top left of the toolbar -6. You profiler will now start collecting metrics from your bencharking script. Once +6. You profiler will now start collecting metrics from your benchmarking script. Once you think you have collected enough samples (usually this is the case after 3 seconds of recording), stop your profiler. 7. Make sure that in toolbar of the bottom window, `profile` is selected. 8. You should be able to see your call graph. * If you don't see the call graph or an incomplete call graph, make sure you have compiled - zstd and your benchmarking scripg using debug flags. On mac and linux, this just means + zstd and your benchmarking script using debug flags. On mac and linux, this just means you will have to supply the `-g` flag alone with your build script. You might also have to provide the `-fno-omit-frame-pointer` flag 9. Dig down the graph to find your function call and then inspect it by double clicking @@ -329,7 +365,7 @@ Some general notes on perf: counter statistics. Perf uses a high resolution timer and this is likely one of the first things your team will run when assessing your PR. * Perf has a long list of hardware counters that can be viewed with `perf --list`. -When measuring optimizations, something worth trying is to make sure the handware +When measuring optimizations, something worth trying is to make sure the hardware counters you expect to be impacted by your change are in fact being so. For example, if you expect the L1 cache misses to decrease with your change, you can look at the counter `L1-dcache-load-misses` @@ -339,57 +375,6 @@ counter `L1-dcache-load-misses` TODO - -## Setting up continuous integration (CI) on your fork -Zstd uses a number of different continuous integration (CI) tools to ensure that new changes -are well tested before they make it to an official release. Specifically, we use the platforms -travis-ci, circle-ci, and appveyor. - -Changes cannot be merged into the main dev branch unless they pass all of our CI tests. -The easiest way to run these CI tests on your own before submitting a PR to our dev branch -is to configure your personal fork of zstd with each of the CI platforms. Below, you'll find -instructions for doing this. - -### travis-ci -Follow these steps to link travis-ci with your github fork of zstd - -1. Make sure you are logged into your github account -2. Go to https://travis-ci.org/ -3. Click 'Sign in with Github' on the top right -4. Click 'Authorize travis-ci' -5. Click 'Activate all repositories using Github Apps' -6. Select 'Only select repositories' and select your fork of zstd from the drop down -7. Click 'Approve and Install' -8. Click 'Sign in with Github' again. This time, it will be for travis-pro (which will let you view your tests on the web dashboard) -9. Click 'Authorize travis-pro' -10. You should have travis set up on your fork now. - -### circle-ci -TODO - -### appveyor -Follow these steps to link circle-ci with your girhub fork of zstd - -1. Make sure you are logged into your github account -2. Go to https://www.appveyor.com/ -3. Click 'Sign in' on the top right -4. Select 'Github' on the left panel -5. Click 'Authorize appveyor' -6. You might be asked to select which repositories you want to give appveyor permission to. Select your fork of zstd if you're prompted -7. You should have appveyor set up on your fork now. - -### General notes on CI -CI tests run every time a pull request (PR) is created or updated. The exact tests -that get run will depend on the destination branch you specify. Some tests take -longer to run than others. Currently, our CI is set up to run a short -series of tests when creating a PR to the dev branch and a longer series of tests -when creating a PR to the release branch. You can look in the configuration files -of the respective CI platform for more information on what gets run when. - -Most people will just want to create a PR with the destination set to their local dev -branch of zstd. You can then find the status of the tests on the PR's page. You can also -re-run tests and cancel running tests from the PR page or from the respective CI's dashboard. - ## Issues We use GitHub issues to track public bugs. Please ensure your description is clear and has sufficient instructions to be able to reproduce the issue. @@ -470,7 +455,7 @@ This design requirement is fundamental to preserve the portability of the code b Any variable that can be `const` (aka. read-only) **must** be `const`. Any pointer which content will not be modified must be `const`. This property is then controlled at compiler level. - `const` variables are an important signal to readers that this variable isn’t modified. + `const` variables are an important signal to readers that this variable isn't modified. Conversely, non-const variables are a signal to readers to watch out for modifications later on in the function. * If a function must be inlined, mention it explicitly, using project's own portable macros, such as `FORCE_INLINE_ATTR`, diff --git a/priv/zstd/LICENSE b/priv/zstd/LICENSE index a793a80..7580028 100644 --- a/priv/zstd/LICENSE +++ b/priv/zstd/LICENSE @@ -2,7 +2,7 @@ BSD License For Zstandard software -Copyright (c) 2016-present, Facebook, Inc. All rights reserved. +Copyright (c) Meta Platforms, Inc. and affiliates. All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: @@ -14,9 +14,9 @@ are permitted provided that the following conditions are met: this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. - * Neither the name Facebook nor the names of its contributors may be used to - endorse or promote products derived from this software without specific - prior written permission. + * Neither the name Facebook, nor Meta, nor the names of its contributors may + be used to endorse or promote products derived from this software without + specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED diff --git a/priv/zstd/Makefile b/priv/zstd/Makefile index c1908f0..a7890a5 100644 --- a/priv/zstd/Makefile +++ b/priv/zstd/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2015-2021, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -29,6 +29,7 @@ VOID = /dev/null # fail on other tested distros (ubuntu, debian) even # without manually specifying the TARGET_SYSTEM. TARGET_SYSTEM ?= $(OS) +CP ?= cp ifneq (,$(filter Windows%,$(TARGET_SYSTEM))) EXT =.exe @@ -69,7 +70,7 @@ zstd zstd-release: .PHONY: zstdmt zstdmt: $(Q)$(MAKE) -C $(PRGDIR) $@ - $(Q)cp $(PRGDIR)/zstd$(EXT) ./zstdmt$(EXT) + $(Q)$(CP) $(PRGDIR)/zstd$(EXT) ./zstdmt$(EXT) .PHONY: zlibwrapper zlibwrapper: lib @@ -122,6 +123,7 @@ contrib: lib $(MAKE) -C contrib/seekable_format/examples all $(MAKE) -C contrib/seekable_format/tests test $(MAKE) -C contrib/largeNbDicts all + $(MAKE) -C contrib/externalSequenceProducer all cd build/single_file_libs/ ; ./build_decoder_test.sh cd build/single_file_libs/ ; ./build_library_test.sh @@ -141,6 +143,7 @@ clean: $(Q)$(MAKE) -C contrib/seekable_format/examples $@ > $(VOID) $(Q)$(MAKE) -C contrib/seekable_format/tests $@ > $(VOID) $(Q)$(MAKE) -C contrib/largeNbDicts $@ > $(VOID) + $(Q)$(MAKE) -C contrib/externalSequenceProducer $@ > $(VOID) $(Q)$(RM) zstd$(EXT) zstdmt$(EXT) tmp* $(Q)$(RM) -r lz4 @echo Cleaning completed @@ -148,13 +151,15 @@ clean: #------------------------------------------------------------------------------ # make install is validated only for Linux, macOS, Hurd and some BSD targets #------------------------------------------------------------------------------ -ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD DragonFly NetBSD MSYS_NT Haiku)) +ifneq (,$(filter $(shell uname),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD DragonFly NetBSD MSYS_NT Haiku AIX)) HOST_OS = POSIX +MKDIR ?= mkdir -p + HAVE_COLORNEVER = $(shell echo a | egrep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0) EGREP_OPTIONS ?= -ifeq ($HAVE_COLORNEVER, 1) +ifeq ($(HAVE_COLORNEVER), 1) EGREP_OPTIONS += --color=never endif EGREP = egrep $(EGREP_OPTIONS) @@ -217,7 +222,7 @@ armbuild: clean CC=arm-linux-gnueabi-gcc CFLAGS="-Werror" $(MAKE) allzstd aarch64build: clean - CC=aarch64-linux-gnu-gcc CFLAGS="-Werror" $(MAKE) allzstd + CC=aarch64-linux-gnu-gcc CFLAGS="-Werror -O0" $(MAKE) allzstd ppcbuild: clean CC=powerpc-linux-gnu-gcc CFLAGS="-m32 -Wno-attributes -Werror" $(MAKE) -j allzstd @@ -284,6 +289,22 @@ uasanregressiontest: msanregressiontest: $(MAKE) -C $(FUZZDIR) regressiontest CC=clang CXX=clang++ CFLAGS="-O3 -fsanitize=memory" CXXFLAGS="-O3 -fsanitize=memory" +update_regressionResults : REGRESS_RESULTS_DIR := /tmp/regress_results_dir/ +update_regressionResults: + $(MAKE) -C programs zstd + $(MAKE) -C tests/regression test + $(RM) -rf $(REGRESS_RESULTS_DIR) + $(MKDIR) $(REGRESS_RESULTS_DIR) + ./tests/regression/test \ + --cache tests/regression/cache \ + --output $(REGRESS_RESULTS_DIR)/results.csv \ + --zstd programs/zstd + echo "Showing results differences" + ! diff tests/regression/results.csv $(REGRESS_RESULTS_DIR)/results.csv + echo "Updating results.csv" + $(CP) $(REGRESS_RESULTS_DIR)/results.csv tests/regression/results.csv + + # run UBsan with -fsanitize-recover=pointer-overflow # this only works with recent compilers such as gcc 8+ usan: clean @@ -315,6 +336,8 @@ tsan-%: clean .PHONY: apt-install apt-install: + # TODO: uncomment once issue 3011 is resolved and remove hack from Github Actions .yml + # sudo apt-get update sudo apt-get -yq --no-install-suggests --no-install-recommends --force-yes install $(APT_PACKAGES) .PHONY: apt-add-repo @@ -372,7 +395,7 @@ ifneq (,$(filter $(HOST_OS),MSYS POSIX)) cmakebuild: cmake --version $(RM) -r $(BUILDIR)/cmake/build - mkdir $(BUILDIR)/cmake/build + $(MKDIR) $(BUILDIR)/cmake/build cd $(BUILDIR)/cmake/build; cmake -DCMAKE_INSTALL_PREFIX:PATH=~/install_test_dir $(CMAKE_PARAMS) .. $(MAKE) -C $(BUILDIR)/cmake/build -j4; $(MAKE) -C $(BUILDIR)/cmake/build install; @@ -381,23 +404,23 @@ cmakebuild: c89build: clean $(CC) -v - CFLAGS="-std=c89 -Werror" $(MAKE) allmost # will fail, due to missing support for `long long` + CFLAGS="-std=c89 -Werror -Wno-attributes -Wpedantic -Wno-long-long -Wno-variadic-macros -O0" $(MAKE) lib zstd gnu90build: clean $(CC) -v - CFLAGS="-std=gnu90 -Werror" $(MAKE) allmost + CFLAGS="-std=gnu90 -Werror -O0" $(MAKE) allmost c99build: clean $(CC) -v - CFLAGS="-std=c99 -Werror" $(MAKE) allmost + CFLAGS="-std=c99 -Werror -O0" $(MAKE) allmost gnu99build: clean $(CC) -v - CFLAGS="-std=gnu99 -Werror" $(MAKE) allmost + CFLAGS="-std=gnu99 -Werror -O0" $(MAKE) allmost c11build: clean $(CC) -v - CFLAGS="-std=c11 -Werror" $(MAKE) allmost + CFLAGS="-std=c11 -Werror -O0" $(MAKE) allmost bmix64build: clean $(CC) -v @@ -416,5 +439,5 @@ bmi32build: clean staticAnalyze: SCANBUILD ?= scan-build staticAnalyze: $(CC) -v - CC=$(CC) CPPFLAGS=-g $(SCANBUILD) --status-bugs -v $(MAKE) allzstd examples contrib + CC=$(CC) CPPFLAGS=-g $(SCANBUILD) --status-bugs -v $(MAKE) zstd endif diff --git a/priv/zstd/Package.swift b/priv/zstd/Package.swift new file mode 100644 index 0000000..97f2c6a --- /dev/null +++ b/priv/zstd/Package.swift @@ -0,0 +1,36 @@ +// swift-tools-version:5.0 +// The swift-tools-version declares the minimum version of Swift required to build this package. + +import PackageDescription + +let package = Package( + name: "zstd", + platforms: [ + .macOS(.v10_10), .iOS(.v9), .tvOS(.v9) + ], + products: [ + // Products define the executables and libraries a package produces, and make them visible to other packages. + .library( + name: "libzstd", + targets: [ "libzstd" ]) + ], + dependencies: [ + // Dependencies declare other packages that this package depends on. + // .package(url: /* package url */, from: "1.0.0"), + ], + targets: [ + // Targets are the basic building blocks of a package. A target can define a module or a test suite. + // Targets can depend on other targets in this package, and on products in packages this package depends on. + .target( + name: "libzstd", + path: "lib", + sources: [ "common", "compress", "decompress", "dictBuilder" ], + publicHeadersPath: ".", + cSettings: [ + .headerSearchPath(".") + ]) + ], + swiftLanguageVersions: [.v5], + cLanguageStandard: .gnu11, + cxxLanguageStandard: .gnucxx14 +) diff --git a/priv/zstd/README.md b/priv/zstd/README.md index dcca766..6bcf757 100644 --- a/priv/zstd/README.md +++ b/priv/zstd/README.md @@ -4,10 +4,11 @@ __Zstandard__, or `zstd` as short version, is a fast lossless compression algori targeting real-time compression scenarios at zlib-level and better compression ratios. It's backed by a very fast entropy stage, provided by [Huff0 and FSE library](https://github.com/Cyan4973/FiniteStateEntropy). -The project is provided as an open-source dual [BSD](LICENSE) and [GPLv2](COPYING) licensed **C** library, +Zstandard's format is stable and documented in [RFC8878](https://datatracker.ietf.org/doc/html/rfc8878). Multiple independent implementations are already available. +This repository represents the reference implementation, provided as an open-source dual [BSD](LICENSE) and [GPLv2](COPYING) licensed **C** library, and a command line utility producing and decoding `.zst`, `.gz`, `.xz` and `.lz4` files. Should your project require another programming language, -a list of known ports and bindings is provided on [Zstandard homepage](http://www.zstd.net/#other-languages). +a list of known ports and bindings is provided on [Zstandard homepage](https://facebook.github.io/zstd/#other-languages). **Development branch status:** @@ -17,8 +18,8 @@ a list of known ports and bindings is provided on [Zstandard homepage](http://ww [![Build status][CirrusDevBadge]][CirrusLink] [![Fuzzing Status][OSSFuzzBadge]][OSSFuzzLink] -[travisDevBadge]: https://travis-ci.org/facebook/zstd.svg?branch=dev "Continuous Integration test suite" -[travisLink]: https://travis-ci.org/facebook/zstd +[travisDevBadge]: https://api.travis-ci.com/facebook/zstd.svg?branch=dev "Continuous Integration test suite" +[travisLink]: https://travis-ci.com/facebook/zstd [AppveyorDevBadge]: https://ci.appveyor.com/api/projects/status/xt38wbdxjk5mrbem/branch/dev?svg=true "Windows test suite" [AppveyorLink]: https://ci.appveyor.com/project/YannCollet/zstd-p0yf0 [CircleDevBadge]: https://circleci.com/gh/facebook/zstd/tree/dev.svg?style=shield "Short test suite" @@ -31,37 +32,36 @@ a list of known ports and bindings is provided on [Zstandard homepage](http://ww ## Benchmarks For reference, several fast compression algorithms were tested and compared -on a server running Arch Linux (`Linux version 5.5.11-arch1-1`), -with a Core i9-9900K CPU @ 5.0GHz, +on a desktop running Ubuntu 20.04 (`Linux 5.11.0-41-generic`), +with a Core i7-9700K CPU @ 4.9GHz, using [lzbench], an open-source in-memory benchmark by @inikep compiled with [gcc] 9.3.0, on the [Silesia compression corpus]. [lzbench]: https://github.com/inikep/lzbench -[Silesia compression corpus]: http://sun.aei.polsl.pl/~sdeor/index.php?page=silesia +[Silesia compression corpus]: https://sun.aei.polsl.pl//~sdeor/index.php?page=silesia [gcc]: https://gcc.gnu.org/ | Compressor name | Ratio | Compression| Decompress.| | --------------- | ------| -----------| ---------- | -| **zstd 1.4.5 -1** | 2.884 | 500 MB/s | 1660 MB/s | -| zlib 1.2.11 -1 | 2.743 | 90 MB/s | 400 MB/s | -| brotli 1.0.7 -0 | 2.703 | 400 MB/s | 450 MB/s | -| **zstd 1.4.5 --fast=1** | 2.434 | 570 MB/s | 2200 MB/s | -| **zstd 1.4.5 --fast=3** | 2.312 | 640 MB/s | 2300 MB/s | -| quicklz 1.5.0 -1 | 2.238 | 560 MB/s | 710 MB/s | -| **zstd 1.4.5 --fast=5** | 2.178 | 700 MB/s | 2420 MB/s | -| lzo1x 2.10 -1 | 2.106 | 690 MB/s | 820 MB/s | -| lz4 1.9.2 | 2.101 | 740 MB/s | 4530 MB/s | -| **zstd 1.4.5 --fast=7** | 2.096 | 750 MB/s | 2480 MB/s | -| lzf 3.6 -1 | 2.077 | 410 MB/s | 860 MB/s | -| snappy 1.1.8 | 2.073 | 560 MB/s | 1790 MB/s | - -[zlib]: http://www.zlib.net/ -[LZ4]: http://www.lz4.org/ +| **zstd 1.5.1 -1** | 2.887 | 530 MB/s | 1700 MB/s | +| [zlib] 1.2.11 -1 | 2.743 | 95 MB/s | 400 MB/s | +| brotli 1.0.9 -0 | 2.702 | 395 MB/s | 450 MB/s | +| **zstd 1.5.1 --fast=1** | 2.437 | 600 MB/s | 2150 MB/s | +| **zstd 1.5.1 --fast=3** | 2.239 | 670 MB/s | 2250 MB/s | +| quicklz 1.5.0 -1 | 2.238 | 540 MB/s | 760 MB/s | +| **zstd 1.5.1 --fast=4** | 2.148 | 710 MB/s | 2300 MB/s | +| lzo1x 2.10 -1 | 2.106 | 660 MB/s | 845 MB/s | +| [lz4] 1.9.3 | 2.101 | 740 MB/s | 4500 MB/s | +| lzf 3.6 -1 | 2.077 | 410 MB/s | 830 MB/s | +| snappy 1.1.9 | 2.073 | 550 MB/s | 1750 MB/s | + +[zlib]: https://www.zlib.net/ +[lz4]: https://lz4.github.io/lz4/ The negative compression levels, specified with `--fast=#`, -offer faster compression and decompression speed in exchange for some loss in -compression ratio compared to level 1, as seen in the table above. +offer faster compression and decompression speed +at the cost of compression ratio (compared to level 1). Zstd can also offer stronger compression ratios at the cost of compression speed. Speed vs Compression trade-off is configurable by small increments. @@ -124,14 +124,27 @@ Dictionary gains are mostly effective in the first few KB. Then, the compression ## Build instructions +`make` is the officially maintained build system of this project. +All other build systems are "compatible" and 3rd-party maintained, +they may feature small differences in advanced options. +When your system allows it, prefer using `make` to build `zstd` and `libzstd`. + ### Makefile If your system is compatible with standard `make` (or `gmake`), invoking `make` in root directory will generate `zstd` cli in root directory. +It will also create `libzstd` into `lib/`. Other available options include: - `make install` : create and install zstd cli, library and man pages -- `make check` : create and run `zstd`, tests its behavior on local platform +- `make check` : create and run `zstd`, test its behavior on local platform + +The `Makefile` follows the [GNU Standard Makefile conventions](https://www.gnu.org/prep/standards/html_node/Makefile-Conventions.html), +allowing staged install, standard flags, directory variables and command variables. + +For advanced use cases, specialized compilation flags which control binary generation +are documented in [`lib/README.md`](lib/README.md#modular-build) for the `libzstd` library +and in [`programs/README.md`](programs/README.md#compilation-variables) for the `zstd` CLI. ### cmake @@ -178,13 +191,15 @@ The output binary will be in `buck-out/gen/programs/`. ## Testing -You can run quick local smoke tests by executing the `playTest.sh` script from the `src/tests` directory. -Two env variables `$ZSTD_BIN` and `$DATAGEN_BIN` are needed for the test script to locate the zstd and datagen binary. -For information on CI testing, please refer to TESTING.md +You can run quick local smoke tests by running `make check`. +If you can't use `make`, execute the `playTest.sh` script from the `src/tests` directory. +Two env variables `$ZSTD_BIN` and `$DATAGEN_BIN` are needed for the test script to locate the `zstd` and `datagen` binary. +For information on CI testing, please refer to `TESTING.md`. ## Status -Zstandard is currently deployed within Facebook. It is used continuously to compress large amounts of data in multiple formats and use cases. +Zstandard is currently deployed within Facebook and many other large cloud infrastructures. +It is run continuously to compress large amounts of data in multiple formats and use cases. Zstandard is considered safe for production environments. ## License diff --git a/priv/zstd/TESTING.md b/priv/zstd/TESTING.md index 32b133b..df842cc 100644 --- a/priv/zstd/TESTING.md +++ b/priv/zstd/TESTING.md @@ -22,7 +22,7 @@ They consist of the following tests: - `tests/playTests.sh --test-large-data` - Fuzzer tests: `tests/fuzzer.c`, `tests/zstreamtest.c`, and `tests/decodecorpus.c` - `tests/zstreamtest.c` under Tsan (streaming mode, including multithreaded mode) -- Valgrind Test (`make -C tests valgrindTest`) (testing CLI and fuzzer under valgrind) +- Valgrind Test (`make -C tests test-valgrind`) (testing CLI and fuzzer under `valgrind`) - Fuzzer tests (see above) on ARM, AArch64, PowerPC, and PowerPC64 Long Tests diff --git a/priv/zstd/appveyor.yml b/priv/zstd/appveyor.yml index c6ab786..c58ef91 100644 --- a/priv/zstd/appveyor.yml +++ b/priv/zstd/appveyor.yml @@ -23,34 +23,6 @@ SCRIPT: "make allzstd MOREFLAGS=-static" ARTIFACT: "true" BUILD: "true" - - COMPILER: "clang" - HOST: "mingw" - PLATFORM: "x64" - SCRIPT: "MOREFLAGS='--target=x86_64-w64-mingw32 -Werror -Wconversion -Wno-sign-conversion' make -j allzstd V=1" - BUILD: "true" - - - COMPILER: "gcc" - HOST: "mingw" - PLATFORM: "x64" - SCRIPT: "" - TEST: "cmake" - - - COMPILER: "visual" - HOST: "visual" - PLATFORM: "x64" - CONFIGURATION: "Debug" - - COMPILER: "visual" - HOST: "visual" - PLATFORM: "Win32" - CONFIGURATION: "Debug" - - COMPILER: "visual" - HOST: "visual" - PLATFORM: "x64" - CONFIGURATION: "Release" - - COMPILER: "visual" - HOST: "visual" - PLATFORM: "Win32" - CONFIGURATION: "Release" - COMPILER: "clang-cl" HOST: "cmake-visual" @@ -113,56 +85,6 @@ appveyor PushArtifact zstd-win-release-%PLATFORM%.zip ) ) - - if [%HOST%]==[visual] ( - ECHO *** && - ECHO *** Building Visual Studio 2008 %PLATFORM%\%CONFIGURATION% in %APPVEYOR_BUILD_FOLDER% && - ECHO *** && - msbuild "build\VS2008\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v90 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2008\bin\%PLATFORM%\%CONFIGURATION%\*.exe && - MD5sum build/VS2008/bin/%PLATFORM%/%CONFIGURATION%/*.exe && - COPY build\VS2008\bin\%PLATFORM%\%CONFIGURATION%\fuzzer.exe tests\fuzzer_VS2008_%PLATFORM%_%CONFIGURATION%.exe && - ECHO *** && - ECHO *** Building Visual Studio 2010 %PLATFORM%\%CONFIGURATION% && - ECHO *** && - msbuild "build\VS2010\zstd.sln" %ADDITIONALPARAM% /m /verbosity:minimal /property:PlatformToolset=v100 /p:ForceImportBeforeCppTargets=%APPVEYOR_BUILD_FOLDER%\build\VS2010\CompileAsCpp.props /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - msbuild "build\VS2010\zstd.sln" %ADDITIONALPARAM% /m /verbosity:minimal /property:PlatformToolset=v100 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\fuzzer.exe tests\fuzzer_VS2010_%PLATFORM%_%CONFIGURATION%.exe && - ECHO *** && - ECHO *** Building Visual Studio 2012 %PLATFORM%\%CONFIGURATION% && - ECHO *** && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v110 /p:ForceImportBeforeCppTargets=%APPVEYOR_BUILD_FOLDER%\build\VS2010\CompileAsCpp.props /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v110 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\fuzzer.exe tests\fuzzer_VS2012_%PLATFORM%_%CONFIGURATION%.exe && - ECHO *** && - ECHO *** Building Visual Studio 2013 %PLATFORM%\%CONFIGURATION% && - ECHO *** && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v120 /p:ForceImportBeforeCppTargets=%APPVEYOR_BUILD_FOLDER%\build\VS2010\CompileAsCpp.props /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v120 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\fuzzer.exe tests\fuzzer_VS2013_%PLATFORM%_%CONFIGURATION%.exe && - ECHO *** && - ECHO *** Building Visual Studio 2015 %PLATFORM%\%CONFIGURATION% && - ECHO *** && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v140 /p:ForceImportBeforeCppTargets=%APPVEYOR_BUILD_FOLDER%\build\VS2010\CompileAsCpp.props /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v140 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\fuzzer.exe tests\fuzzer_VS2015_%PLATFORM%_%CONFIGURATION%.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe tests\ - ) - if [%HOST%]==[cmake-visual] ( ECHO *** && ECHO *** Building %CMAKE_GENERATOR% ^(%CMAKE_GENERATOR_TOOLSET%^) %PLATFORM%\%CONFIGURATION% && @@ -186,18 +108,6 @@ cd ..\..\.. && make clean ) - - SET "FUZZERTEST=-T30s" - - if [%HOST%]==[visual] if [%CONFIGURATION%]==[Release] ( - CD tests && - SET ZSTD_BIN=./zstd.exe&& - SET DATAGEN_BIN=./datagen.exe&& - sh -e playTests.sh --test-large-data && - fullbench.exe -i1 && - fullbench.exe -i1 -P0 && - fuzzer_VS2012_%PLATFORM%_Release.exe %FUZZERTEST% && - fuzzer_VS2013_%PLATFORM%_Release.exe %FUZZERTEST% && - fuzzer_VS2015_%PLATFORM%_Release.exe %FUZZERTEST% - ) # The following tests are for regular pushes @@ -208,22 +118,6 @@ version: 1.0.{build} environment: matrix: - - COMPILER: "gcc" - HOST: "cygwin" - PLATFORM: "x64" - - COMPILER: "gcc" - HOST: "mingw" - PLATFORM: "x64" - SCRIPT: "CFLAGS=-Werror make -j allzstd DEBUGLEVEL=2" - - COMPILER: "gcc" - HOST: "mingw" - PLATFORM: "x86" - SCRIPT: "CFLAGS=-Werror make -j allzstd" - - COMPILER: "clang" - HOST: "mingw" - PLATFORM: "x64" - SCRIPT: "CFLAGS='--target=x86_64-w64-mingw32 -Werror -Wconversion -Wno-sign-conversion' make -j allzstd V=1" - - COMPILER: "visual" HOST: "visual" PLATFORM: "x64" @@ -241,6 +135,10 @@ PLATFORM: "Win32" CONFIGURATION: "Release" + - COMPILER: "gcc" + HOST: "cygwin" + PLATFORM: "x64" + - COMPILER: "clang-cl" HOST: "cmake-visual" PLATFORM: "x64" @@ -260,12 +158,6 @@ cmake,^ make ) - - if [%HOST%]==[mingw] ( - SET "PATH_MINGW32=C:\mingw-w64\i686-6.3.0-posix-dwarf-rt_v5-rev1\mingw32\bin" && - SET "PATH_MINGW64=C:\mingw-w64\x86_64-6.3.0-posix-seh-rt_v5-rev1\mingw64\bin" && - COPY C:\msys64\usr\bin\make.exe C:\mingw-w64\i686-6.3.0-posix-dwarf-rt_v5-rev1\mingw32\bin\make.exe && - COPY C:\msys64\usr\bin\make.exe C:\mingw-w64\x86_64-6.3.0-posix-seh-rt_v5-rev1\mingw64\bin\make.exe - ) - IF [%HOST%]==[visual] IF [%PLATFORM%]==[x64] ( SET ADDITIONALPARAM=/p:LibraryPath="C:\Program Files\Microsoft SDKs\Windows\v7.1\lib\x64;c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\lib\amd64;C:\Program Files (x86)\Microsoft Visual Studio 10.0\;C:\Program Files (x86)\Microsoft Visual Studio 10.0\lib\amd64;" ) @@ -283,30 +175,6 @@ ctest -V -L Medium; " ) - - if [%HOST%]==[mingw] ( - ( if [%PLATFORM%]==[x64] ( - SET "PATH=%PATH_MINGW64%;%PATH_ORIGINAL%" - ) else if [%PLATFORM%]==[x86] ( - SET "PATH=%PATH_MINGW32%;%PATH_ORIGINAL%" - ) ) && - make -v && - sh -c "%COMPILER% -v" && - set "CC=%COMPILER%" && - sh -c "%SCRIPT%" - ) - - if [%HOST%]==[visual] ( - ECHO *** && - ECHO *** Building Visual Studio 2015 %PLATFORM%\%CONFIGURATION% && - ECHO *** && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v140 /p:ForceImportBeforeCppTargets=%APPVEYOR_BUILD_FOLDER%\build\VS2010\CompileAsCpp.props /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v140 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && - DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && - MD5sum build/VS2010/bin/%PLATFORM%_%CONFIGURATION%/*.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\fuzzer.exe tests\fuzzer_VS2015_%PLATFORM%_%CONFIGURATION%.exe && - COPY build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe tests\ - ) - if [%HOST%]==[cmake-visual] ( ECHO *** && ECHO *** Building %CMAKE_GENERATOR% ^(%CMAKE_GENERATOR_TOOLSET%^) %PLATFORM%\%CONFIGURATION% && @@ -316,12 +184,22 @@ POPD && ECHO *** ) + - if [%HOST%]==[visual] ( + ECHO *** && + ECHO *** Building Visual Studio 2012 %PLATFORM%\%CONFIGURATION% && + ECHO *** && + msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v110 /p:ForceImportBeforeCppTargets=%APPVEYOR_BUILD_FOLDER%\build\VS2010\CompileAsCpp.props /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && + DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe && + msbuild "build\VS2010\zstd.sln" /m /verbosity:minimal /property:PlatformToolset=v110 /t:Clean,Build /p:Platform=%PLATFORM% /p:Configuration=%CONFIGURATION% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll" && + DIR build\VS2010\bin\%PLATFORM%_%CONFIGURATION%\*.exe + ) test_script: - ECHO Testing %COMPILER% %PLATFORM% %CONFIGURATION% + - SET "FUZZERTEST=-T10s" - if [%HOST%]==[mingw] ( set "CC=%COMPILER%" && make clean && make check - ) + ) \ No newline at end of file diff --git a/priv/zstd/build/VS2008/zstd/zstd.vcproj b/priv/zstd/build/VS2008/zstd/zstd.vcproj index c7eec57..91f2bda 100644 --- a/priv/zstd/build/VS2008/zstd/zstd.vcproj +++ b/priv/zstd/build/VS2008/zstd/zstd.vcproj @@ -384,6 +384,10 @@ RelativePath="..\..\..\programs\fileio.c" > + + diff --git a/priv/zstd/build/VS2010/libzstd-dll/libzstd-dll.rc b/priv/zstd/build/VS2010/libzstd-dll/libzstd-dll.rc index ee9f562..13e8746 100644 --- a/priv/zstd/build/VS2010/libzstd-dll/libzstd-dll.rc +++ b/priv/zstd/build/VS2010/libzstd-dll/libzstd-dll.rc @@ -32,11 +32,11 @@ BEGIN BEGIN BLOCK "040904B0" BEGIN - VALUE "CompanyName", "Yann Collet, Facebook, Inc." + VALUE "CompanyName", "Meta Platforms, Inc." VALUE "FileDescription", "Zstandard - Fast and efficient compression algorithm" VALUE "FileVersion", ZSTD_VERSION_STRING VALUE "InternalName", "libzstd.dll" - VALUE "LegalCopyright", "Copyright (c) 2013-present, Yann Collet, Facebook, Inc." + VALUE "LegalCopyright", "Copyright (c) Meta Platforms, Inc. and affiliates." VALUE "OriginalFilename", "libzstd.dll" VALUE "ProductName", "Zstandard" VALUE "ProductVersion", ZSTD_VERSION_STRING diff --git a/priv/zstd/build/VS2010/zstd/zstd.rc b/priv/zstd/build/VS2010/zstd/zstd.rc index f5e4047..a2118c2 100644 --- a/priv/zstd/build/VS2010/zstd/zstd.rc +++ b/priv/zstd/build/VS2010/zstd/zstd.rc @@ -32,11 +32,11 @@ BEGIN BEGIN BLOCK "040904B0" BEGIN - VALUE "CompanyName", "Yann Collet, Facebook, Inc." + VALUE "CompanyName", "Meta Platforms, Inc." VALUE "FileDescription", "Zstandard - Fast and efficient compression algorithm" VALUE "FileVersion", ZSTD_VERSION_STRING VALUE "InternalName", "zstd.exe" - VALUE "LegalCopyright", "Copyright (c) 2013-present, Yann Collet, Facebook, Inc." + VALUE "LegalCopyright", "Copyright (c) Meta Platforms, Inc. and affiliates." VALUE "OriginalFilename", "zstd.exe" VALUE "ProductName", "Zstandard" VALUE "ProductVersion", ZSTD_VERSION_STRING diff --git a/priv/zstd/build/VS2010/zstd/zstd.vcxproj b/priv/zstd/build/VS2010/zstd/zstd.vcxproj index 46e22f4..8ab239d 100644 --- a/priv/zstd/build/VS2010/zstd/zstd.vcxproj +++ b/priv/zstd/build/VS2010/zstd/zstd.vcxproj @@ -62,6 +62,7 @@ + diff --git a/priv/zstd/build/VS_scripts/build.generic.cmd b/priv/zstd/build/VS_scripts/build.generic.cmd index a7ca4d0..b24e6ed 100644 --- a/priv/zstd/build/VS_scripts/build.generic.cmd +++ b/priv/zstd/build/VS_scripts/build.generic.cmd @@ -19,10 +19,10 @@ GOTO build :display_help echo Syntax: build.generic.cmd msbuild_version msbuild_platform msbuild_configuration msbuild_toolset -echo msbuild_version: VS installed version (VS2012, VS2013, VS2015, VS2017, ...) +echo msbuild_version: VS installed version (VS2012, VS2013, VS2015, VS2017, VS2019, ...) echo msbuild_platform: Platform (x64 or Win32) echo msbuild_configuration: VS configuration (Release or Debug) -echo msbuild_toolset: Platform Toolset (v100, v110, v120, v140, v141) +echo msbuild_toolset: Platform Toolset (v100, v110, v120, v140, v141, v142, ...) EXIT /B 1 @@ -43,6 +43,16 @@ IF %msbuild_version% == VS2017 ( IF EXIST %msbuild_vs2017enterprise% SET msbuild=%msbuild_vs2017enterprise% ) +:: VS2019 +SET msbuild_vs2019community="%programfiles(x86)%\Microsoft Visual Studio\2019\Community\MSBuild\Current\Bin\MSBuild.exe" +SET msbuild_vs2019professional="%programfiles(x86)%\Microsoft Visual Studio\2019\Professional\MSBuild\Current\Bin\MSBuild.exe" +SET msbuild_vs2019enterprise="%programfiles(x86)%\Microsoft Visual Studio\2019\Enterprise\MSBuild\Current\Bin\MSBuild.exe" +IF %msbuild_version% == VS2019 ( + IF EXIST %msbuild_vs2019community% SET msbuild=%msbuild_vs2019community% + IF EXIST %msbuild_vs2019professional% SET msbuild=%msbuild_vs2019professional% + IF EXIST %msbuild_vs2019enterprise% SET msbuild=%msbuild_vs2019enterprise% +) + SET project="%~p0\..\VS2010\zstd.sln" SET msbuild_params=/verbosity:minimal /nologo /t:Clean,Build /p:Platform=%msbuild_platform% /p:Configuration=%msbuild_configuration% diff --git a/priv/zstd/build/cmake/CMakeLists.txt b/priv/zstd/build/cmake/CMakeLists.txt index a050577..0bffc87 100644 --- a/priv/zstd/build/cmake/CMakeLists.txt +++ b/priv/zstd/build/cmake/CMakeLists.txt @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2016-present, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -8,18 +8,18 @@ # ################################################################ cmake_minimum_required(VERSION 2.8.12 FATAL_ERROR) - -# As of 2018-12-26 ZSTD has been validated to build with cmake version 3.13.2 new policies. -# Set and use the newest cmake policies that are validated to work -set(ZSTD_MAX_VALIDATED_CMAKE_MAJOR_VERSION "3") + +# As of 2018-12-26 ZSTD has been validated to build with cmake version 3.13.2 new policies. +# Set and use the newest cmake policies that are validated to work +set(ZSTD_MAX_VALIDATED_CMAKE_MAJOR_VERSION "3") set(ZSTD_MAX_VALIDATED_CMAKE_MINOR_VERSION "13") #Policies never changed at PATCH level if("${CMAKE_MAJOR_VERSION}" LESS 3) - set(ZSTD_CMAKE_POLICY_VERSION "${CMAKE_VERSION}") -elseif( "${ZSTD_MAX_VALIDATED_CMAKE_MAJOR_VERSION}" EQUAL "${CMAKE_MAJOR_VERSION}" AND + set(ZSTD_CMAKE_POLICY_VERSION "${CMAKE_VERSION}") +elseif( "${ZSTD_MAX_VALIDATED_CMAKE_MAJOR_VERSION}" EQUAL "${CMAKE_MAJOR_VERSION}" AND "${ZSTD_MAX_VALIDATED_CMAKE_MINOR_VERSION}" GREATER "${CMAKE_MINOR_VERSION}") - set(ZSTD_CMAKE_POLICY_VERSION "${CMAKE_VERSION}") -else() - set(ZSTD_CMAKE_POLICY_VERSION "${ZSTD_MAX_VALIDATED_CMAKE_MAJOR_VERSION}.${ZSTD_MAX_VALIDATED_CMAKE_MINOR_VERSION}.0") + set(ZSTD_CMAKE_POLICY_VERSION "${CMAKE_VERSION}") +else() + set(ZSTD_CMAKE_POLICY_VERSION "${ZSTD_MAX_VALIDATED_CMAKE_MAJOR_VERSION}.${ZSTD_MAX_VALIDATED_CMAKE_MINOR_VERSION}.0") endif() cmake_policy(VERSION ${ZSTD_CMAKE_POLICY_VERSION}) @@ -40,16 +40,18 @@ if( CMAKE_MAJOR_VERSION LESS 3 ) set(PROJECT_VERSION_PATCH ${zstd_VERSION_PATCH}) set(PROJECT_VERSION "${zstd_VERSION_MAJOR}.${zstd_VERSION_MINOR}.${zstd_VERSION_PATCH}") enable_language(C) # Main library is in C + enable_language(ASM) # And ASM enable_language(CXX) # Testing contributed code also utilizes CXX else() project(zstd VERSION "${zstd_VERSION_MAJOR}.${zstd_VERSION_MINOR}.${zstd_VERSION_PATCH}" LANGUAGES C # Main library is in C + ASM # And ASM CXX # Testing contributed code also utilizes CXX ) endif() message(STATUS "ZSTD VERSION: ${zstd_VERSION}") -set(zstd_HOMEPAGE_URL "http://www.zstd.net") +set(zstd_HOMEPAGE_URL "https://facebook.github.io/zstd") set(zstd_DESCRIPTION "Zstandard is a real-time compression algorithm, providing high compression ratios.") # Set a default build type if none was specified @@ -82,18 +84,25 @@ message(STATUS "CMAKE_INSTALL_LIBDIR: ${CMAKE_INSTALL_LIBDIR}") #----------------------------------------------------------------------------- # Legacy support -option(ZSTD_LEGACY_SUPPORT "LEGACY SUPPORT" OFF) +option(ZSTD_LEGACY_SUPPORT "LEGACY SUPPORT" ON) if (ZSTD_LEGACY_SUPPORT) message(STATUS "ZSTD_LEGACY_SUPPORT defined!") - add_definitions(-DZSTD_LEGACY_SUPPORT=5) + set(ZSTD_LEGACY_LEVEL 5 CACHE STRING "") + add_definitions(-DZSTD_LEGACY_SUPPORT=${ZSTD_LEGACY_LEVEL}) else () message(STATUS "ZSTD_LEGACY_SUPPORT not defined!") add_definitions(-DZSTD_LEGACY_SUPPORT=0) endif () +if (ANDROID) + set(ZSTD_MULTITHREAD_SUPPORT_DEFAULT OFF) +else() + set(ZSTD_MULTITHREAD_SUPPORT_DEFAULT ON) +endif() + # Multi-threading support -option(ZSTD_MULTITHREAD_SUPPORT "MULTITHREADING SUPPORT" ON) +option(ZSTD_MULTITHREAD_SUPPORT "MULTITHREADING SUPPORT" ${ZSTD_MULTITHREAD_SUPPORT_DEFAULT}) if (ZSTD_MULTITHREAD_SUPPORT) message(STATUS "ZSTD_MULTITHREAD_SUPPORT is enabled") diff --git a/priv/zstd/build/cmake/CMakeModules/AddZstdCompilationFlags.cmake b/priv/zstd/build/cmake/CMakeModules/AddZstdCompilationFlags.cmake index e23b9d6..0265349 100644 --- a/priv/zstd/build/cmake/CMakeModules/AddZstdCompilationFlags.cmake +++ b/priv/zstd/build/cmake/CMakeModules/AddZstdCompilationFlags.cmake @@ -1,7 +1,8 @@ include(CheckCXXCompilerFlag) include(CheckCCompilerFlag) +include(CheckLinkerFlag) -function(EnableCompilerFlag _flag _C _CXX) +function(EnableCompilerFlag _flag _C _CXX _LD) string(REGEX REPLACE "\\+" "PLUS" varname "${_flag}") string(REGEX REPLACE "[^A-Za-z0-9]+" "_" varname "${varname}") string(REGEX REPLACE "^_+" "" varname "${varname}") @@ -18,43 +19,58 @@ function(EnableCompilerFlag _flag _C _CXX) set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${_flag}" PARENT_SCOPE) endif () endif () + if (_LD) + CHECK_LINKER_FLAG(C ${_flag} LD_FLAG_${varname}) + if (LD_FLAG_${varname}) + set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${_flag}" PARENT_SCOPE) + set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${_flag}" PARENT_SCOPE) + endif () + endif () endfunction() macro(ADD_ZSTD_COMPILATION_FLAGS) if (CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang" OR MINGW) #Not only UNIX but also WIN32 for MinGW - #Set c++11 by default - EnableCompilerFlag("-std=c++11" false true) - #Set c99 by default - EnableCompilerFlag("-std=c99" true false) + # It's possible to select the exact standard used for compilation. + # It's not necessary, but can be employed for specific purposes. + # Note that zstd source code is compatible with both C++98 and above + # and C-gnu90 (c90 + long long + variadic macros ) and above + # EnableCompilerFlag("-std=c++11" false true) # Set C++ compilation to c++11 standard + # EnableCompilerFlag("-std=c99" true false) # Set C compiation to c99 standard if (CMAKE_CXX_COMPILER_ID MATCHES "Clang" AND MSVC) # clang-cl normally maps -Wall to -Weverything. - EnableCompilerFlag("/clang:-Wall" true true) + EnableCompilerFlag("/clang:-Wall" true true false) else () - EnableCompilerFlag("-Wall" true true) + EnableCompilerFlag("-Wall" true true false) endif () - EnableCompilerFlag("-Wextra" true true) - EnableCompilerFlag("-Wundef" true true) - EnableCompilerFlag("-Wshadow" true true) - EnableCompilerFlag("-Wcast-align" true true) - EnableCompilerFlag("-Wcast-qual" true true) - EnableCompilerFlag("-Wstrict-prototypes" true false) + EnableCompilerFlag("-Wextra" true true false) + EnableCompilerFlag("-Wundef" true true false) + EnableCompilerFlag("-Wshadow" true true false) + EnableCompilerFlag("-Wcast-align" true true false) + EnableCompilerFlag("-Wcast-qual" true true false) + EnableCompilerFlag("-Wstrict-prototypes" true false false) # Enable asserts in Debug mode if (CMAKE_BUILD_TYPE MATCHES "Debug") - EnableCompilerFlag("-DDEBUGLEVEL=1" true true) + EnableCompilerFlag("-DDEBUGLEVEL=1" true true false) endif () + # Add noexecstack flags + # LDFLAGS + EnableCompilerFlag("-z noexecstack" false false true) + # CFLAGS & CXXFLAGS + EnableCompilerFlag("-Qunused-arguments" true true false) + EnableCompilerFlag("-Wa,--noexecstack" true true false) elseif (MSVC) # Add specific compilation flags for Windows Visual set(ACTIVATE_MULTITHREADED_COMPILATION "ON" CACHE BOOL "activate multi-threaded compilation (/MP flag)") if (CMAKE_GENERATOR MATCHES "Visual Studio" AND ACTIVATE_MULTITHREADED_COMPILATION) - EnableCompilerFlag("/MP" true true) + EnableCompilerFlag("/MP" true true false) endif () - + # UNICODE SUPPORT - EnableCompilerFlag("/D_UNICODE" true true) - EnableCompilerFlag("/DUNICODE" true true) + EnableCompilerFlag("/D_UNICODE" true true false) + EnableCompilerFlag("/DUNICODE" true true false) # Enable asserts in Debug mode if (CMAKE_BUILD_TYPE MATCHES "Debug") - EnableCompilerFlag("/DDEBUGLEVEL=1" true true) + EnableCompilerFlag("/DDEBUGLEVEL=1" true true false) endif () endif () diff --git a/priv/zstd/build/cmake/CMakeModules/JoinPaths.cmake b/priv/zstd/build/cmake/CMakeModules/JoinPaths.cmake new file mode 100644 index 0000000..c68d91b --- /dev/null +++ b/priv/zstd/build/cmake/CMakeModules/JoinPaths.cmake @@ -0,0 +1,23 @@ +# This module provides function for joining paths +# known from most languages +# +# SPDX-License-Identifier: (MIT OR CC0-1.0) +# Copyright 2020 Jan Tojnar +# https://github.com/jtojnar/cmake-snips +# +# Modelled after Python’s os.path.join +# https://docs.python.org/3.7/library/os.path.html#os.path.join +# Windows not supported +function(join_paths joined_path first_path_segment) + set(temp_path "${first_path_segment}") + foreach(current_segment IN LISTS ARGN) + if(NOT ("${current_segment}" STREQUAL "")) + if(IS_ABSOLUTE "${current_segment}") + set(temp_path "${current_segment}") + else() + set(temp_path "${temp_path}/${current_segment}") + endif() + endif() + endforeach() + set(${joined_path} "${temp_path}" PARENT_SCOPE) +endfunction() diff --git a/priv/zstd/build/cmake/README.md b/priv/zstd/build/cmake/README.md index 73b30dc..a460dd1 100644 --- a/priv/zstd/build/cmake/README.md +++ b/priv/zstd/build/cmake/README.md @@ -1,13 +1,13 @@ # Cmake contributions Contributions to the cmake build configurations are welcome. Please -use case sensitivity that matches modern (ie. cmake version 2.6 and above) +use case sensitivity that matches modern (i.e. cmake version 2.6 and above) conventions of using lower-case for commands, and upper-case for variables. ## How to build -As cmake doesn't support command like `cmake clean`, it's recommended to perform a "out of source build". +As cmake doesn't support command like `cmake clean`, it's recommended to perform an "out of source build". To do this, you can create a new directory and build in it: ```sh cd build/cmake @@ -37,7 +37,7 @@ cmake -LH .. Bool options can be set to `ON/OFF` with `-D[option]=[ON/OFF]`. You can configure cmake options like this: ```sh cd build/cmake/builddir -cmake -DZSTD_BUILD_TESTS=ON -DZSTD_LEGACY_SUPPORT=ON .. +cmake -DZSTD_BUILD_TESTS=ON -DZSTD_LEGACY_SUPPORT=OFF .. make ``` diff --git a/priv/zstd/build/cmake/contrib/CMakeLists.txt b/priv/zstd/build/cmake/contrib/CMakeLists.txt index f7631d0..8df2a17 100644 --- a/priv/zstd/build/cmake/contrib/CMakeLists.txt +++ b/priv/zstd/build/cmake/contrib/CMakeLists.txt @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2016-present, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/build/cmake/contrib/gen_html/CMakeLists.txt b/priv/zstd/build/cmake/contrib/gen_html/CMakeLists.txt index 8fdd611..d1ff6c6 100644 --- a/priv/zstd/build/cmake/contrib/gen_html/CMakeLists.txt +++ b/priv/zstd/build/cmake/contrib/gen_html/CMakeLists.txt @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2015-present, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/build/cmake/contrib/pzstd/CMakeLists.txt b/priv/zstd/build/cmake/contrib/pzstd/CMakeLists.txt index 5c30a91..f7098fa 100644 --- a/priv/zstd/build/cmake/contrib/pzstd/CMakeLists.txt +++ b/priv/zstd/build/cmake/contrib/pzstd/CMakeLists.txt @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2016-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -21,10 +21,16 @@ add_executable(pzstd ${PROGRAMS_DIR}/util.c ${PZSTD_DIR}/main.cpp ${PZSTD_DIR}/O set_property(TARGET pzstd APPEND PROPERTY COMPILE_DEFINITIONS "NDEBUG") set_property(TARGET pzstd APPEND PROPERTY COMPILE_OPTIONS "-Wno-shadow") +if (ZSTD_BUILD_SHARED) + set(ZSTD_LIB libzstd_shared) +else() + set(ZSTD_LIB libzstd_static) +endif() + set(THREADS_PREFER_PTHREAD_FLAG ON) find_package(Threads REQUIRED) if (CMAKE_USE_PTHREADS_INIT) - target_link_libraries(pzstd libzstd_shared ${CMAKE_THREAD_LIBS_INIT}) + target_link_libraries(pzstd ${ZSTD_LIB} ${CMAKE_THREAD_LIBS_INIT}) else() message(SEND_ERROR "ZSTD currently does not support thread libraries other than pthreads") endif() diff --git a/priv/zstd/build/cmake/lib/CMakeLists.txt b/priv/zstd/build/cmake/lib/CMakeLists.txt index 5f75665..3034958 100644 --- a/priv/zstd/build/cmake/lib/CMakeLists.txt +++ b/priv/zstd/build/cmake/lib/CMakeLists.txt @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2015-present, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -7,7 +7,7 @@ # in the COPYING file in the root directory of this source tree). # ################################################################ -project(libzstd C) +project(libzstd C ASM) set(CMAKE_INCLUDE_CURRENT_DIR TRUE) option(ZSTD_BUILD_STATIC "BUILD STATIC LIBRARIES" ON) @@ -22,7 +22,12 @@ include_directories(${LIBRARY_DIR} ${LIBRARY_DIR}/common) file(GLOB CommonSources ${LIBRARY_DIR}/common/*.c) file(GLOB CompressSources ${LIBRARY_DIR}/compress/*.c) -file(GLOB DecompressSources ${LIBRARY_DIR}/decompress/*.c) +if (MSVC) + file(GLOB DecompressSources ${LIBRARY_DIR}/decompress/*.c) + add_compile_options(-DZSTD_DISABLE_ASM) +else () + file(GLOB DecompressSources ${LIBRARY_DIR}/decompress/*.c ${LIBRARY_DIR}/decompress/*.S) +endif () file(GLOB DictBuilderSources ${LIBRARY_DIR}/dictBuilder/*.c) set(Sources @@ -72,6 +77,12 @@ if (MSVC) set(PlatformDependResources ${MSVC_RESOURCE_DIR}/libzstd-dll.rc) endif () +# Explicitly set the language to C for all files, including ASM files. +# Our assembly expects to be compiled by a C compiler, and is only enabled for +# __GNUC__ compatible compilers. Otherwise all the ASM code is disabled by +# macros. +set_source_files_properties(${Sources} PROPERTIES LANGUAGE C) + # Split project to static and shared libraries build set(library_targets) if (ZSTD_BUILD_SHARED) @@ -106,7 +117,7 @@ if (MSVC) endif () # With MSVC static library needs to be renamed to avoid conflict with import library -if (MSVC) +if (MSVC OR (WIN32 AND CMAKE_CXX_COMPILER_ID STREQUAL "Clang" AND NOT MINGW)) set(STATIC_LIBRARY_BASE_NAME zstd_static) else () set(STATIC_LIBRARY_BASE_NAME zstd) @@ -130,30 +141,17 @@ if (ZSTD_BUILD_STATIC) OUTPUT_NAME ${STATIC_LIBRARY_BASE_NAME}) endif () -if (UNIX OR MINGW) - # pkg-config - set(PREFIX "${CMAKE_INSTALL_PREFIX}") - set(EXEC_PREFIX "\${prefix}") - set(LIBDIR "${CMAKE_INSTALL_FULL_LIBDIR}") - set(INCLUDEDIR "${CMAKE_INSTALL_FULL_INCLUDEDIR}") - set(VERSION "${zstd_VERSION}") - - string(LENGTH "${PREFIX}" PREFIX_LENGTH) - string(SUBSTRING "${LIBDIR}" 0 ${PREFIX_LENGTH} LIBDIR_PREFIX) - string(SUBSTRING "${LIBDIR}" ${PREFIX_LENGTH} -1 LIBDIR_SUFFIX) - string(SUBSTRING "${INCLUDEDIR}" 0 ${PREFIX_LENGTH} INCLUDEDIR_PREFIX) - string(SUBSTRING "${INCLUDEDIR}" ${PREFIX_LENGTH} -1 INCLUDEDIR_SUFFIX) - - if ("${INCLUDEDIR_PREFIX}" STREQUAL "${PREFIX}") - set(INCLUDEDIR "\${prefix}${INCLUDEDIR_SUFFIX}") - endif() - if ("${LIBDIR_PREFIX}" STREQUAL "${PREFIX}") - set(LIBDIR "\${exec_prefix}${LIBDIR_SUFFIX}") - endif() +# pkg-config +include(JoinPaths) # can be replaced by cmake_path(APPEND) in CMake 3.20 +set(PREFIX "${CMAKE_INSTALL_PREFIX}") +set(EXEC_PREFIX "\${prefix}") +join_paths(LIBDIR "\${exec_prefix}" "${CMAKE_INSTALL_LIBDIR}") +join_paths(INCLUDEDIR "\${prefix}" "${CMAKE_INSTALL_INCLUDEDIR}") +set(LIBS_PRIVATE "${THREADS_LIBS}") +set(VERSION "${zstd_VERSION}") - configure_file("${LIBRARY_DIR}/libzstd.pc.in" "${CMAKE_CURRENT_BINARY_DIR}/libzstd.pc" @ONLY) - install(FILES "${CMAKE_CURRENT_BINARY_DIR}/libzstd.pc" DESTINATION "${CMAKE_INSTALL_LIBDIR}/pkgconfig") -endif () +configure_file("${LIBRARY_DIR}/libzstd.pc.in" "${CMAKE_CURRENT_BINARY_DIR}/libzstd.pc" @ONLY) +install(FILES "${CMAKE_CURRENT_BINARY_DIR}/libzstd.pc" DESTINATION "${CMAKE_INSTALL_LIBDIR}/pkgconfig") # install target install(FILES @@ -168,6 +166,7 @@ install(TARGETS ${library_targets} ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}" LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}" RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}" + BUNDLE DESTINATION "${CMAKE_INSTALL_BINDIR}" ) # uninstall target diff --git a/priv/zstd/build/cmake/programs/CMakeLists.txt b/priv/zstd/build/cmake/programs/CMakeLists.txt index f1d1277..58d998e 100644 --- a/priv/zstd/build/cmake/programs/CMakeLists.txt +++ b/priv/zstd/build/cmake/programs/CMakeLists.txt @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2015-present, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -32,12 +32,14 @@ if (MSVC) set(PlatformDependResources ${MSVC_RESOURCE_DIR}/zstd.rc) endif () -add_executable(zstd ${PROGRAMS_DIR}/zstdcli.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${PROGRAMS_DIR}/fileio.c ${PROGRAMS_DIR}/benchfn.c ${PROGRAMS_DIR}/benchzstd.c ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/dibio.c ${PROGRAMS_DIR}/zstdcli_trace.c ${PlatformDependResources}) +add_executable(zstd ${PROGRAMS_DIR}/zstdcli.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${PROGRAMS_DIR}/fileio.c ${PROGRAMS_DIR}/fileio_asyncio.c ${PROGRAMS_DIR}/benchfn.c ${PROGRAMS_DIR}/benchzstd.c ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/dibio.c ${PROGRAMS_DIR}/zstdcli_trace.c ${PlatformDependResources}) target_link_libraries(zstd ${PROGRAMS_ZSTD_LINK_TARGET}) if (CMAKE_SYSTEM_NAME MATCHES "(Solaris|SunOS)") target_link_libraries(zstd rt) endif () -install(TARGETS zstd RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}") +install(TARGETS zstd + RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}" + BUNDLE DESTINATION "${CMAKE_INSTALL_BINDIR}") if (UNIX) add_custom_target(zstdcat ALL ${CMAKE_COMMAND} -E create_symlink zstd zstdcat DEPENDS zstd COMMENT "Creating zstdcat symlink") @@ -73,7 +75,7 @@ if (UNIX) ${CMAKE_CURRENT_BINARY_DIR}/zstdless.1 DESTINATION "${MAN_INSTALL_DIR}") - add_executable(zstd-frugal ${PROGRAMS_DIR}/zstdcli.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${PROGRAMS_DIR}/fileio.c) + add_executable(zstd-frugal ${PROGRAMS_DIR}/zstdcli.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${PROGRAMS_DIR}/fileio.c ${PROGRAMS_DIR}/fileio_asyncio.c) target_link_libraries(zstd-frugal ${PROGRAMS_ZSTD_LINK_TARGET}) set_property(TARGET zstd-frugal APPEND PROPERTY COMPILE_DEFINITIONS "ZSTD_NOBENCH;ZSTD_NODICT;ZSTD_NOTRACE") endif () diff --git a/priv/zstd/build/cmake/tests/CMakeLists.txt b/priv/zstd/build/cmake/tests/CMakeLists.txt index 8bba6ea..250f050 100644 --- a/priv/zstd/build/cmake/tests/CMakeLists.txt +++ b/priv/zstd/build/cmake/tests/CMakeLists.txt @@ -1,6 +1,6 @@ # ################################################################ # zstd - Makefile -# Copyright (C) Yann Collet 2014-present +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # BSD license @@ -27,7 +27,7 @@ # SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. # # You can contact the author at : -# - zstd homepage : http://www.zstd.net/ +# - zstd homepage : https://facebook.github.io/zstd/ # ################################################################ project(tests) @@ -57,7 +57,9 @@ target_link_libraries(datagen libzstd_static) # fullbench # add_executable(fullbench ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${PROGRAMS_DIR}/benchfn.c ${PROGRAMS_DIR}/benchzstd.c ${TESTS_DIR}/fullbench.c) -set_property(TARGET fullbench APPEND PROPERTY COMPILE_OPTIONS "-Wno-deprecated-declarations") +if (NOT MSVC) + target_compile_options(fullbench PRIVATE "-Wno-deprecated-declarations") +endif() target_link_libraries(fullbench libzstd_static) add_test(NAME fullbench COMMAND fullbench ${ZSTD_FULLBENCH_FLAGS}) @@ -65,7 +67,9 @@ add_test(NAME fullbench COMMAND fullbench ${ZSTD_FULLBENCH_FLAGS}) # fuzzer # add_executable(fuzzer ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${TESTS_DIR}/fuzzer.c) -set_property(TARGET fuzzer APPEND PROPERTY COMPILE_OPTIONS "-Wno-deprecated-declarations") +if (NOT MSVC) + target_compile_options(fuzzer PRIVATE "-Wno-deprecated-declarations") +endif() target_link_libraries(fuzzer libzstd_static) AddTestFlagsOption(ZSTD_FUZZER_FLAGS "$ENV{FUZZERTEST} $ENV{FUZZER_FLAGS}" "Semicolon-separated list of flags to pass to the fuzzer test (see `fuzzer -h` for usage)") @@ -77,8 +81,10 @@ add_test(NAME fuzzer COMMAND fuzzer ${ZSTD_FUZZER_FLAGS}) # # zstreamtest # -add_executable(zstreamtest ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${TESTS_DIR}/seqgen.c ${TESTS_DIR}/zstreamtest.c) -set_property(TARGET zstreamtest APPEND PROPERTY COMPILE_OPTIONS "-Wno-deprecated-declarations") +add_executable(zstreamtest ${PROGRAMS_DIR}/datagen.c ${PROGRAMS_DIR}/util.c ${PROGRAMS_DIR}/timefn.c ${TESTS_DIR}/seqgen.c ${TESTS_DIR}/zstreamtest.c ${TESTS_DIR}/external_matchfinder.c) +if (NOT MSVC) + target_compile_options(zstreamtest PRIVATE "-Wno-deprecated-declarations") +endif() target_link_libraries(zstreamtest libzstd_static) AddTestFlagsOption(ZSTD_ZSTREAM_FLAGS "$ENV{ZSTREAM_TESTTIME} $ENV{FUZZER_FLAGS}" "Semicolon-separated list of flags to pass to the zstreamtest test (see `zstreamtest -h` for usage)") @@ -90,13 +96,14 @@ add_test(NAME zstreamtest COMMAND zstreamtest ${ZSTD_ZSTREAM_FLAGS}) AddTestFlagsOption(ZSTD_PLAYTESTS_FLAGS "$ENV{PLAYTESTS_FLAGS}" "Semicolon-separated list of flags to pass to the playTests.sh test") add_test(NAME playTests COMMAND sh -c "\"${TESTS_DIR}/playTests.sh\" ${ZSTD_PLAYTESTS_FLAGS}") -if (ZSTD_BUILD_PROGRAMS) +find_program(UNAME uname) # Run script only in unix shell environments +if (ZSTD_BUILD_PROGRAMS AND UNAME) set_property(TEST playTests APPEND PROPERTY ENVIRONMENT "ZSTD_BIN=$" "DATAGEN_BIN=$" ) else() - message(STATUS "Disabling playTests.sh test because ZSTD_BUILD_PROGRAMS is not enabled") + message(STATUS "Disabling playTests.sh test because requirements not met") set_tests_properties(playTests PROPERTIES DISABLED YES) endif() diff --git a/priv/zstd/build/meson/contrib/pzstd/meson.build b/priv/zstd/build/meson/contrib/pzstd/meson.build index dcf2136..b95dc79 100644 --- a/priv/zstd/build/meson/contrib/pzstd/meson.build +++ b/priv/zstd/build/meson/contrib/pzstd/meson.build @@ -18,7 +18,8 @@ pzstd_sources = [join_paths(zstd_rootdir, 'programs/util.c'), join_paths(zstd_rootdir, 'contrib/pzstd/SkippableFrame.cpp')] pzstd = executable('pzstd', pzstd_sources, - cpp_args: [ '-DNDEBUG', '-Wno-shadow', '-pedantic', '-Wno-deprecated-declarations' ], + cpp_args: pzstd_warning_flags, include_directories: pzstd_includes, dependencies: [ libzstd_dep, thread_dep ], + override_options: ['b_ndebug=true'], install: true) diff --git a/priv/zstd/build/meson/lib/meson.build b/priv/zstd/build/meson/lib/meson.build index 5cc9fee..68db2ca 100644 --- a/priv/zstd/build/meson/lib/meson.build +++ b/priv/zstd/build/meson/lib/meson.build @@ -45,6 +45,15 @@ libzstd_sources = [join_paths(zstd_rootdir, 'lib/common/entropy_common.c'), join_paths(zstd_rootdir, 'lib/dictBuilder/divsufsort.c'), join_paths(zstd_rootdir, 'lib/dictBuilder/zdict.c')] +# really we need anything that defines __GNUC__ as that is what ZSTD_ASM_SUPPORTED is gated on +# but these are the two compilers that are supported in tree and actually handle this correctly +# Otherwise, explicitly disable assembly. +if [compiler_gcc, compiler_clang].contains(cc_id) + libzstd_sources += join_paths(zstd_rootdir, 'lib/decompress/huf_decompress_amd64.S') +else + add_project_arguments('-DZSTD_DISABLE_ASM', language: 'c') +endif + # Explicit define legacy support add_project_arguments('-DZSTD_LEGACY_SUPPORT=@0@'.format(legacy_level), language: 'c') @@ -74,7 +83,8 @@ libzstd_c_args = [] if cc_id == compiler_msvc if default_library_type != 'static' libzstd_sources += [windows_mod.compile_resources( - join_paths(zstd_rootdir, 'build/VS2010/libzstd-dll/libzstd-dll.rc'))] + join_paths(zstd_rootdir, 'build/VS2010/libzstd-dll/libzstd-dll.rc'), + include_directories: libzstd_includes)] libzstd_c_args += ['-DZSTD_DLL_EXPORT=1', '-DZSTD_HEAPMODE=0', '-D_CONSOLE', @@ -108,6 +118,7 @@ libzstd = library('zstd', libzstd_sources, include_directories: libzstd_includes, c_args: libzstd_c_args, + gnu_symbol_visibility: 'hidden', dependencies: libzstd_deps, install: true, version: zstd_libversion) @@ -115,12 +126,41 @@ libzstd = library('zstd', libzstd_dep = declare_dependency(link_with: libzstd, include_directories: libzstd_includes) +# we link to both: +# - the shared library (for public symbols) +# - the static library (for private symbols) +# +# this is needed because internally private symbols are used all the time, and +# -fvisibility=hidden means those cannot be found +if get_option('default_library') == 'static' + libzstd_static = libzstd + libzstd_internal_dep = libzstd_dep +else + if get_option('default_library') == 'shared' + libzstd_static = static_library('zstd_objlib', + objects: libzstd.extract_all_objects(recursive: true), + build_by_default: false) + else + libzstd_static = libzstd.get_static_lib() + endif + + if cc_id == compiler_msvc + # msvc does not actually support linking to both, but errors out with: + # error LNK2005: ZSTD_ already defined in zstd.lib(zstd-1.dll) + libzstd_internal_dep = declare_dependency(link_with: libzstd_static) + else + libzstd_internal_dep = declare_dependency(link_with: libzstd, + # the static library must be linked after the shared one + dependencies: declare_dependency(link_with: libzstd_static)) + endif +endif + pkgconfig.generate(libzstd, name: 'libzstd', filebase: 'libzstd', description: 'fast lossless compression algorithm library', version: zstd_libversion, - url: 'http://www.zstd.net/') + url: 'https://facebook.github.io/zstd/') install_headers(join_paths(zstd_rootdir, 'lib/zstd.h'), join_paths(zstd_rootdir, 'lib/zdict.h'), diff --git a/priv/zstd/build/meson/meson.build b/priv/zstd/build/meson/meson.build index 2a425b2..576dc44 100644 --- a/priv/zstd/build/meson/meson.build +++ b/priv/zstd/build/meson/meson.build @@ -12,12 +12,21 @@ project('zstd', ['c', 'cpp'], license: ['BSD', 'GPLv2'], default_options : [ - 'c_std=gnu99', + # There shouldn't be any need to force a C standard convention for zstd + # but in case one would want that anyway, this can be done here. + # 'c_std=gnu99', + # c++11 standard is useful for pzstd 'cpp_std=c++11', - 'buildtype=release' + 'buildtype=release', + 'warning_level=3', + # -Wdocumentation does not actually pass, nor do the test binaries, + # so this isn't safe + #'werror=true' ], - version: 'DUMMY', - meson_version: '>=0.47.0') + version: run_command( + find_program('GetZstdLibraryVersion.py'), '../../lib/zstd.h', + check: true).stdout().strip(), + meson_version: '>=0.50.0') cc = meson.get_compiler('c') cxx = meson.get_compiler('cpp') @@ -38,16 +47,6 @@ compiler_msvc = 'msvc' zstd_version = meson.project_version() -zstd_h_file = join_paths(meson.current_source_dir(), '../../lib/zstd.h') -GetZstdLibraryVersion_py = find_program('GetZstdLibraryVersion.py', native : true) -r = run_command(GetZstdLibraryVersion_py, zstd_h_file) -if r.returncode() == 0 - zstd_version = r.stdout().strip() - message('Project version is now: @0@'.format(zstd_version)) -else - error('Cannot find project version in @0@'.format(zstd_h_file)) -endif - zstd_libversion = zstd_version # ============================================================================= @@ -88,7 +87,7 @@ feature_lz4 = get_option('lz4') # Dependencies # ============================================================================= -libm_dep = cc.find_library('m', required: bin_tests) +libm_dep = cc.find_library('m', required: false) thread_dep = dependency('threads', required: feature_multi_thread) use_multi_thread = thread_dep.found() # Arguments in dependency should be equivalent to those passed to pkg-config @@ -105,11 +104,11 @@ use_lz4 = lz4_dep.found() add_project_arguments('-DXXH_NAMESPACE=ZSTD_', language: ['c']) +pzstd_warning_flags = [] if [compiler_gcc, compiler_clang].contains(cc_id) - common_warning_flags = [ '-Wextra', '-Wundef', '-Wshadow', '-Wcast-align', '-Wcast-qual' ] + common_warning_flags = [ '-Wundef', '-Wshadow', '-Wcast-align', '-Wcast-qual' ] + pzstd_warning_flags = ['-Wno-shadow', '-Wno-deprecated-declarations'] if cc_id == compiler_clang - # Should use Meson's own --werror build option - #common_warning_flags += '-Werror' common_warning_flags += ['-Wconversion', '-Wno-sign-conversion', '-Wdocumentation'] endif cc_compile_flags = cc.get_supported_arguments(common_warning_flags + ['-Wstrict-prototypes']) diff --git a/priv/zstd/build/meson/meson_options.txt b/priv/zstd/build/meson/meson_options.txt index 90a81c5..f35cd5f 100644 --- a/priv/zstd/build/meson/meson_options.txt +++ b/priv/zstd/build/meson/meson_options.txt @@ -10,11 +10,11 @@ # Read guidelines from https://wiki.gnome.org/Initiatives/GnomeGoals/MesonPorting -option('legacy_level', type: 'integer', min: 0, max: 7, value: '5', +option('legacy_level', type: 'integer', min: 0, max: 7, value: 5, description: 'Support any legacy format: 7 to 1 for v0.7+ to v0.1+') option('debug_level', type: 'integer', min: 0, max: 9, value: 1, description: 'Enable run-time debug. See lib/common/debug.h') -option('backtrace', type: 'boolean', value: false, +option('backtrace', type: 'feature', value: 'disabled', description: 'Display a stack backtrace when execution generates a runtime exception') option('static_runtime', type: 'boolean', value: false, description: 'Link to static run-time libraries on MSVC') diff --git a/priv/zstd/build/meson/programs/meson.build b/priv/zstd/build/meson/programs/meson.build index d255627..e611dc3 100644 --- a/priv/zstd/build/meson/programs/meson.build +++ b/priv/zstd/build/meson/programs/meson.build @@ -14,18 +14,26 @@ zstd_programs_sources = [join_paths(zstd_rootdir, 'programs/zstdcli.c'), join_paths(zstd_rootdir, 'programs/util.c'), join_paths(zstd_rootdir, 'programs/timefn.c'), join_paths(zstd_rootdir, 'programs/fileio.c'), + join_paths(zstd_rootdir, 'programs/fileio_asyncio.c'), join_paths(zstd_rootdir, 'programs/benchfn.c'), join_paths(zstd_rootdir, 'programs/benchzstd.c'), join_paths(zstd_rootdir, 'programs/datagen.c'), join_paths(zstd_rootdir, 'programs/dibio.c'), join_paths(zstd_rootdir, 'programs/zstdcli_trace.c')] +zstd_deps = [ libzstd_internal_dep ] zstd_c_args = libzstd_debug_cflags + +zstd_frugal_deps = [ libzstd_internal_dep ] +zstd_frugal_c_args = [ '-DZSTD_NOBENCH', '-DZSTD_NODICT', '-DZSTD_NOTRACE' ] + if use_multi_thread + zstd_deps += [ thread_dep ] zstd_c_args += [ '-DZSTD_MULTITHREAD' ] + zstd_frugal_deps += [ thread_dep ] + zstd_frugal_c_args += [ '-DZSTD_MULTITHREAD' ] endif -zstd_deps = [ libzstd_dep ] if use_zlib zstd_deps += [ zlib_dep ] zstd_c_args += [ '-DZSTD_GZCOMPRESS', '-DZSTD_GZDECOMPRESS' ] @@ -43,7 +51,8 @@ endif export_dynamic_on_windows = false # explicit backtrace enable/disable for Linux & Darwin -if not use_backtrace +have_execinfo = cc.has_header('execinfo.h', required: use_backtrace) +if not have_execinfo zstd_c_args += '-DBACKTRACE_ENABLE=0' elif use_debug and host_machine_os == os_windows # MinGW target zstd_c_args += '-DBACKTRACE_ENABLE=1' @@ -53,7 +62,8 @@ endif if cc_id == compiler_msvc if default_library_type != 'static' zstd_programs_sources += [windows_mod.compile_resources( - join_paths(zstd_rootdir, 'build/VS2010/zstd/zstd.rc'))] + join_paths(zstd_rootdir, 'build/VS2010/zstd/zstd.rc'), + include_directories: libzstd_includes)] endif endif @@ -67,14 +77,15 @@ zstd = executable('zstd', zstd_frugal_sources = [join_paths(zstd_rootdir, 'programs/zstdcli.c'), join_paths(zstd_rootdir, 'programs/timefn.c'), join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'programs/fileio.c')] + join_paths(zstd_rootdir, 'programs/fileio.c'), + join_paths(zstd_rootdir, 'programs/fileio_asyncio.c')] # Minimal target, with only zstd compression and decompression. # No bench. No legacy. executable('zstd-frugal', zstd_frugal_sources, - dependencies: libzstd_dep, - c_args: [ '-DZSTD_NOBENCH', '-DZSTD_NODICT', '-DZSTD_NOTRACE' ], + dependencies: zstd_frugal_deps, + c_args: zstd_frugal_c_args, install: true) install_data(join_paths(zstd_rootdir, 'programs/zstdgrep'), diff --git a/priv/zstd/build/meson/tests/meson.build b/priv/zstd/build/meson/tests/meson.build index 1b23363..e70b734 100644 --- a/priv/zstd/build/meson/tests/meson.build +++ b/priv/zstd/build/meson/tests/meson.build @@ -21,7 +21,6 @@ FUZZER_FLAGS = ['--no-big-tests'] FUZZERTEST = '-T200s' ZSTREAM_TESTTIME = '-T90s' DECODECORPUS_TESTTIME = '-T30' -ZSTDRTTEST = ['--test-large-data'] # ============================================================================= # Executables @@ -29,64 +28,64 @@ ZSTDRTTEST = ['--test-large-data'] test_includes = [ include_directories(join_paths(zstd_rootdir, 'programs')) ] -datagen_sources = [join_paths(zstd_rootdir, 'programs/datagen.c'), - join_paths(zstd_rootdir, 'tests/datagencli.c')] +testcommon_sources = [join_paths(zstd_rootdir, 'programs/datagen.c'), + join_paths(zstd_rootdir, 'programs/util.c'), + join_paths(zstd_rootdir, 'programs/timefn.c'), + join_paths(zstd_rootdir, 'programs/benchfn.c'), + join_paths(zstd_rootdir, 'programs/benchzstd.c')] + +testcommon = static_library('testcommon', + testcommon_sources, + # needed due to use of private symbol + -fvisibility=hidden + link_with: libzstd_static) + +testcommon_dep = declare_dependency(link_with: testcommon, + dependencies: libzstd_deps, + include_directories: libzstd_includes) + +datagen_sources = [join_paths(zstd_rootdir, 'tests/datagencli.c')] datagen = executable('datagen', datagen_sources, c_args: [ '-DNDEBUG' ], include_directories: test_includes, - dependencies: libzstd_dep, + dependencies: testcommon_dep, install: false) -fullbench_sources = [join_paths(zstd_rootdir, 'programs/datagen.c'), - join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'programs/timefn.c'), - join_paths(zstd_rootdir, 'programs/benchfn.c'), - join_paths(zstd_rootdir, 'programs/benchzstd.c'), - join_paths(zstd_rootdir, 'tests/fullbench.c')] +fullbench_sources = [join_paths(zstd_rootdir, 'tests/fullbench.c')] fullbench = executable('fullbench', fullbench_sources, include_directories: test_includes, - dependencies: libzstd_dep, + dependencies: testcommon_dep, install: false) -fuzzer_sources = [join_paths(zstd_rootdir, 'programs/datagen.c'), - join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'programs/timefn.c'), - join_paths(zstd_rootdir, 'tests/fuzzer.c')] +fuzzer_sources = [join_paths(zstd_rootdir, 'tests/fuzzer.c')] fuzzer = executable('fuzzer', fuzzer_sources, include_directories: test_includes, - dependencies: [ libzstd_dep, thread_dep ], + dependencies: [ testcommon_dep, thread_dep ], install: false) -zstreamtest_sources = [join_paths(zstd_rootdir, 'programs/datagen.c'), - join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'programs/timefn.c'), +zstreamtest_sources = [ join_paths(zstd_rootdir, 'tests/seqgen.c'), - join_paths(zstd_rootdir, 'tests/zstreamtest.c')] + join_paths(zstd_rootdir, 'tests/zstreamtest.c'), + join_paths(zstd_rootdir, 'tests/external_matchfinder.c')] zstreamtest = executable('zstreamtest', zstreamtest_sources, include_directories: test_includes, - dependencies: libzstd_dep, + dependencies: testcommon_dep, install: false) -paramgrill_sources = [join_paths(zstd_rootdir, 'programs/benchfn.c'), - join_paths(zstd_rootdir, 'programs/timefn.c'), - join_paths(zstd_rootdir, 'programs/benchzstd.c'), - join_paths(zstd_rootdir, 'programs/datagen.c'), - join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'tests/paramgrill.c')] +paramgrill_sources = [join_paths(zstd_rootdir, 'tests/paramgrill.c')] paramgrill = executable('paramgrill', paramgrill_sources, include_directories: test_includes, - dependencies: [ libzstd_dep, libm_dep ], + dependencies: [ testcommon_dep, libm_dep ], install: false) roundTripCrash_sources = [join_paths(zstd_rootdir, 'tests/roundTripCrash.c')] roundTripCrash = executable('roundTripCrash', roundTripCrash_sources, - dependencies: [ libzstd_dep ], + dependencies: [ testcommon_dep ], install: false) longmatch_sources = [join_paths(zstd_rootdir, 'tests/longmatch.c')] @@ -111,26 +110,18 @@ if 0 < legacy_level and legacy_level <= 4 install: false) endif -decodecorpus_sources = [join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'programs/timefn.c'), - join_paths(zstd_rootdir, 'tests/decodecorpus.c')] +decodecorpus_sources = [join_paths(zstd_rootdir, 'tests/decodecorpus.c')] decodecorpus = executable('decodecorpus', decodecorpus_sources, include_directories: test_includes, - dependencies: [ libzstd_dep, libm_dep ], + dependencies: [ testcommon_dep, libm_dep ], install: false) -poolTests_sources = [join_paths(zstd_rootdir, 'programs/util.c'), - join_paths(zstd_rootdir, 'programs/timefn.c'), - join_paths(zstd_rootdir, 'tests/poolTests.c'), - join_paths(zstd_rootdir, 'lib/common/pool.c'), - join_paths(zstd_rootdir, 'lib/common/threading.c'), - join_paths(zstd_rootdir, 'lib/common/zstd_common.c'), - join_paths(zstd_rootdir, 'lib/common/error_private.c')] +poolTests_sources = [join_paths(zstd_rootdir, 'tests/poolTests.c')] poolTests = executable('poolTests', poolTests_sources, include_directories: test_includes, - dependencies: [ libzstd_dep, thread_dep ], + dependencies: [ testcommon_dep, thread_dep ], install: false) checkTag_sources = [join_paths(zstd_rootdir, 'tests/checkTag.c')] @@ -144,24 +135,38 @@ checkTag = executable('checkTag', # ============================================================================= if tests_supported_oses.contains(host_machine_os) - valgrind_prog = find_program('valgrind', ['/usr/bin/valgrind'], required: true) + valgrind_prog = find_program('valgrind', ['/usr/bin/valgrind'], required: false) valgrindTest_py = files('valgrindTest.py') - test('valgrindTest', - valgrindTest_py, - args: [valgrind_prog.path(), zstd, datagen, fuzzer, fullbench], - depends: [zstd, datagen, fuzzer, fullbench], - timeout: 600) # Timeout should work on HDD drive + if valgrind_prog.found() + test('valgrindTest', + valgrindTest_py, + args: [valgrind_prog.path(), zstd, datagen, fuzzer, fullbench], + depends: [zstd, datagen, fuzzer, fullbench], + timeout: 600) # Timeout should work on HDD drive + endif endif if host_machine_os != os_windows playTests_sh = find_program(join_paths(zstd_rootdir, 'tests/playTests.sh'), required: true) - test('test-zstd', - playTests_sh, - args: ZSTDRTTEST, - env: ['ZSTD_BIN=' + zstd.full_path(), 'DATAGEN_BIN=./datagen'], - depends: [datagen], - workdir: meson.current_build_dir(), - timeout: 2800) # Timeout should work on HDD drive + + # add slow tests only if the meson version is new enough to support + # test setups with default-excluded suites + if meson.version().version_compare('>=0.57.0') + matrix = {'fast': [], 'slow': ['--test-large-data']} + else + matrix = {'fast': []} + endif + + foreach suite, opt: matrix + test('test-zstd-'+suite, + playTests_sh, + args: opt, + env: ['ZSTD_BIN=' + zstd.full_path(), 'DATAGEN_BIN=./datagen'], + depends: [datagen], + suite: suite, + workdir: meson.current_build_dir(), + timeout: 2800) # Timeout should work on HDD drive + endforeach endif test('test-fullbench-1', @@ -186,10 +191,6 @@ test('test-zstream-1', zstreamtest, args: ['-v', ZSTREAM_TESTTIME] + FUZZER_FLAGS, timeout: 240) -test('test-zstream-2', - zstreamtest, - args: ['-mt', '-t1', ZSTREAM_TESTTIME] + FUZZER_FLAGS, - timeout: 120) test('test-zstream-3', zstreamtest, args: ['--newapi', '-t1', ZSTREAM_TESTTIME] + FUZZER_FLAGS, @@ -204,3 +205,11 @@ test('test-decodecorpus', args: ['-t', DECODECORPUS_TESTTIME], timeout: 60) test('test-poolTests', poolTests) # should be fast + +if meson.version().version_compare('>=0.57.0') + add_test_setup('fast', + is_default: true, + exclude_suites: ['slow']) + add_test_setup('slow', + exclude_suites: ['fast']) +endif diff --git a/priv/zstd/build/meson/tests/valgrindTest.py b/priv/zstd/build/meson/tests/valgrindTest.py index 218f745..05d8487 100644 --- a/priv/zstd/build/meson/tests/valgrindTest.py +++ b/priv/zstd/build/meson/tests/valgrindTest.py @@ -21,7 +21,7 @@ def valgrindTest(valgrind, datagen, fuzzer, zstd, fullbench): if subprocess.call([*VALGRIND_ARGS, zstd], stdout=subprocess.DEVNULL) == 0: - raise subprocess.CalledProcessError('zstd without argument should have failed') + raise subprocess.SubprocessError('zstd without argument should have failed') with subprocess.Popen([datagen, '-g80'], stdout=subprocess.PIPE) as p1, \ subprocess.Popen([*VALGRIND_ARGS, zstd, '-', '-c'], @@ -30,7 +30,7 @@ def valgrindTest(valgrind, datagen, fuzzer, zstd, fullbench): p1.stdout.close() # Allow p1 to receive a SIGPIPE if p2 exits. p2.communicate() if p2.returncode != 0: - raise subprocess.CalledProcessError() + raise subprocess.SubprocessError() with subprocess.Popen([datagen, '-g16KB'], stdout=subprocess.PIPE) as p1, \ subprocess.Popen([*VALGRIND_ARGS, zstd, '-vf', '-', '-c'], @@ -39,7 +39,7 @@ def valgrindTest(valgrind, datagen, fuzzer, zstd, fullbench): p1.stdout.close() p2.communicate() if p2.returncode != 0: - raise subprocess.CalledProcessError() + raise subprocess.SubprocessError() with tempfile.NamedTemporaryFile() as tmp_fd: with subprocess.Popen([datagen, '-g2930KB'], stdout=subprocess.PIPE) as p1, \ @@ -48,7 +48,7 @@ def valgrindTest(valgrind, datagen, fuzzer, zstd, fullbench): p1.stdout.close() p2.communicate() if p2.returncode != 0: - raise subprocess.CalledProcessError() + raise subprocess.SubprocessError() subprocess.check_call([*VALGRIND_ARGS, zstd, '-vdf', tmp_fd.name, '-c'], stdout=subprocess.DEVNULL) @@ -60,7 +60,7 @@ def valgrindTest(valgrind, datagen, fuzzer, zstd, fullbench): p1.stdout.close() p2.communicate() if p2.returncode != 0: - raise subprocess.CalledProcessError() + raise subprocess.SubprocessError() subprocess.check_call([*VALGRIND_ARGS, fuzzer, '-T1mn', '-t1']) subprocess.check_call([*VALGRIND_ARGS, fullbench, '-i1']) diff --git a/priv/zstd/build/single_file_libs/README.md b/priv/zstd/build/single_file_libs/README.md index 1705b76..64c973a 100644 --- a/priv/zstd/build/single_file_libs/README.md +++ b/priv/zstd/build/single_file_libs/README.md @@ -12,7 +12,7 @@ This is the most common use case. The decompression library is small, adding, fo Create `zstddeclib.c` from the Zstd source using: ``` cd zstd/build/single_file_libs -./combine.sh -r ../../lib -o zstddeclib.c zstddeclib-in.c +python3 combine.py -r ../../lib -x legacy/zstd_legacy.h -o zstddeclib.c zstddeclib-in.c ``` Then add the resulting file to your project (see the [example files](examples)). @@ -26,7 +26,7 @@ The same tool can amalgamate the entire Zstd library for ease of adding both com Create `zstd.c` from the Zstd source using: ``` cd zstd/build/single_file_libs -./combine.sh -r ../../lib -o zstd.c zstd-in.c +python3 combine.py -r ../../lib -x legacy/zstd_legacy.h -k zstd.h -o zstd.c zstd-in.c ``` It's possible to create a compressor-only library but since the decompressor is so small in comparison this doesn't bring much of a gain (but for the curious, simply remove the files in the _decompress_ section at the end of `zstd-in.c`). diff --git a/priv/zstd/build/single_file_libs/build_decoder_test.sh b/priv/zstd/build/single_file_libs/build_decoder_test.sh index 48d017f..c4ca55f 100755 --- a/priv/zstd/build/single_file_libs/build_decoder_test.sh +++ b/priv/zstd/build/single_file_libs/build_decoder_test.sh @@ -11,7 +11,7 @@ IN_FILES="examples/emscripten.c" # Emscripten build using emcc. emscripten_emcc_build() { - # Compile the the same example as above + # Compile the same example as above CC_FLAGS="-Wall -Wextra -Wshadow -Werror -Os -g0 -flto" emcc $CC_FLAGS -s WASM=1 -I. -o $OUT_WASM $IN_FILES # Did compilation work? diff --git a/priv/zstd/build/single_file_libs/build_library_test.sh b/priv/zstd/build/single_file_libs/build_library_test.sh index 7fb9965..f4ba109 100755 --- a/priv/zstd/build/single_file_libs/build_library_test.sh +++ b/priv/zstd/build/single_file_libs/build_library_test.sh @@ -14,7 +14,7 @@ IN_FILES="zstd.c examples/roundtrip.c" # Emscripten build using emcc. emscripten_emcc_build() { - # Compile the the same example as above + # Compile the same example as above CC_FLAGS="-Wall -Wextra -Wshadow -Werror -Os -g0 -flto" emcc $CC_FLAGS -s WASM=1 -I. -o $OUT_WASM $IN_FILES # Did compilation work? @@ -69,7 +69,7 @@ fi echo "Single file library creation script: PASSED" # Copy the header to here (for the tests) -cp "$ZSTD_SRC_ROOT/zstd.h" zstd.h +cp "$ZSTD_SRC_ROOT/zstd.h" examples/zstd.h # Compile the generated output cc -Wall -Wextra -Werror -Wshadow -pthread -I. -Os -g0 -o $OUT_FILE zstd.c examples/roundtrip.c diff --git a/priv/zstd/build/single_file_libs/combine.py b/priv/zstd/build/single_file_libs/combine.py new file mode 100755 index 0000000..771dd20 --- /dev/null +++ b/priv/zstd/build/single_file_libs/combine.py @@ -0,0 +1,234 @@ +#!/usr/bin/env python3 + +# Tool to bundle multiple C/C++ source files, inlining any includes. +# +# Note: there are two types of exclusion options: the '-x' flag, which besides +# excluding a file also adds an #error directive in place of the #include, and +# the '-k' flag, which keeps the #include and doesn't inline the file. The +# intended use cases are: '-x' for files that would normally be #if'd out, so +# features that 100% won't be used in the amalgamated file, for which every +# occurrence adds the error, and '-k' for headers that we wish to manually +# include, such as a project's public API, for which occurrences after the first +# are removed. +# +# Todo: the error handling could be better, which currently throws and halts +# (which is functional just not very friendly). +# +# Author: Carl Woffenden, Numfum GmbH (this script is released under a CC0 license/Public Domain) + +import argparse, re, sys + +from pathlib import Path +from typing import Any, List, Optional, Pattern, Set, TextIO + +# Set of file roots when searching (equivalent to -I paths for the compiler). +roots: Set[Path] = set() + +# Set of (canonical) file Path objects to exclude from inlining (and not only +# exclude but to add a compiler error directive when they're encountered). +excludes: Set[Path] = set() + +# Set of (canonical) file Path objects to keep as include directives. +keeps: Set[Path] = set() + +# Whether to keep the #pragma once directives (unlikely, since this will result +# in a warning, but the option is there). +keep_pragma: bool = False + +# Destination file object (or stdout if no output file was supplied). +destn: TextIO = sys.stdout + +# Set of file Path objects previously inlined (and to ignore if reencountering). +found: Set[Path] = set() + +# Compiled regex Pattern to handle "#pragma once" in various formats: +# +# #pragma once +# #pragma once +# # pragma once +# #pragma once +# #pragma once // comment +# +# Ignoring commented versions, same as include_regex. +# +pragma_regex: Pattern = re.compile(r'^\s*#\s*pragma\s*once\s*') + +# Compiled regex Pattern to handle the following type of file includes: +# +# #include "file" +# #include "file" +# # include "file" +# #include "file" +# #include "file" // comment +# #include "file" // comment with quote " +# +# And all combinations of, as well as ignoring the following: +# +# #include +# //#include "file" +# /*#include "file"*/ +# +# We don't try to catch errors since the compiler will do this (and the code is +# expected to be valid before processing) and we don't care what follows the +# file (whether it's a valid comment or not, since anything after the quoted +# string is ignored) +# +include_regex: Pattern = re.compile(r'^\s*#\s*include\s*"(.+?)"') + +# Simple tests to prove include_regex's cases. +# +def test_match_include() -> bool: + if (include_regex.match('#include "file"') and + include_regex.match(' #include "file"') and + include_regex.match('# include "file"') and + include_regex.match('#include "file"') and + include_regex.match('#include "file" // comment')): + if (not include_regex.match('#include ') and + not include_regex.match('//#include "file"') and + not include_regex.match('/*#include "file"*/')): + found = include_regex.match('#include "file" // "') + if (found and found.group(1) == 'file'): + print('#include match valid') + return True + return False + +# Simple tests to prove pragma_regex's cases. +# +def test_match_pragma() -> bool: + if (pragma_regex.match('#pragma once') and + pragma_regex.match(' #pragma once') and + pragma_regex.match('# pragma once') and + pragma_regex.match('#pragma once') and + pragma_regex.match('#pragma once // comment')): + if (not pragma_regex.match('//#pragma once') and + not pragma_regex.match('/*#pragma once*/')): + print('#pragma once match valid') + return True + return False + +# Finds 'file'. First the list of 'root' paths are searched, followed by the +# currently processing file's 'parent' path, returning a valid Path in +# canonical form. If no match is found None is returned. +# +def resolve_include(file: str, parent: Optional[Path] = None) -> Optional[Path]: + for root in roots: + found = root.joinpath(file).resolve() + if (found.is_file()): + return found + if (parent): + found = parent.joinpath(file).resolve(); + else: + found = Path(file) + if (found.is_file()): + return found + return None + +# Helper to resolve lists of files. 'file_list' is passed in from the arguments +# and each entry resolved to its canonical path (like any include entry, either +# from the list of root paths or the owning file's 'parent', which in this case +# is case is the input file). The results are stored in 'resolved'. +# +def resolve_excluded_files(file_list: Optional[List[str]], resolved: Set[Path], parent: Optional[Path] = None) -> None: + if (file_list): + for filename in file_list: + found = resolve_include(filename, parent) + if (found): + resolved.add(found) + else: + error_line(f'Warning: excluded file not found: {filename}') + +# Writes 'line' to the open 'destn' (or stdout). +# +def write_line(line: str) -> None: + print(line, file=destn) + +# Logs 'line' to stderr. This is also used for general notifications that we +# don't want to go to stdout (so the source can be piped). +# +def error_line(line: Any) -> None: + print(line, file=sys.stderr) + +# Inline the contents of 'file' (with any of its includes also inlined, etc.). +# +# Note: text encoding errors are ignored and replaced with ? when reading the +# input files. This isn't ideal, but it's more than likely in the comments than +# code and a) the text editor has probably also failed to read the same content, +# and b) the compiler probably did too. +# +def add_file(file: Path, file_name: str = None) -> None: + if (file.is_file()): + if (not file_name): + file_name = file.name + error_line(f'Processing: {file_name}') + with file.open('r', errors='replace') as opened: + for line in opened: + line = line.rstrip('\n') + match_include = include_regex.match(line); + if (match_include): + # We have a quoted include directive so grab the file + inc_name = match_include.group(1) + resolved = resolve_include(inc_name, file.parent) + if (resolved): + if (resolved in excludes): + # The file was excluded so error if the compiler uses it + write_line(f'#error Using excluded file: {inc_name} (re-amalgamate source to fix)') + error_line(f'Excluding: {inc_name}') + else: + if (resolved not in found): + # The file was not previously encountered + found.add(resolved) + if (resolved in keeps): + # But the include was flagged to keep as included + write_line(f'/**** *NOT* inlining {inc_name} ****/') + write_line(line) + error_line(f'Not inlining: {inc_name}') + else: + # The file was neither excluded nor seen before so inline it + write_line(f'/**** start inlining {inc_name} ****/') + add_file(resolved, inc_name) + write_line(f'/**** ended inlining {inc_name} ****/') + else: + write_line(f'/**** skipping file: {inc_name} ****/') + else: + # The include file didn't resolve to a file + write_line(f'#error Unable to find: {inc_name}') + error_line(f'Error: Unable to find: {inc_name}') + else: + # Skip any 'pragma once' directives, otherwise write the source line + if (keep_pragma or not pragma_regex.match(line)): + write_line(line) + else: + error_line(f'Error: Invalid file: {file}') + +# Start here +parser = argparse.ArgumentParser(description='Amalgamate Tool', epilog=f'example: {sys.argv[0]} -r ../my/path -r ../other/path -o out.c in.c') +parser.add_argument('-r', '--root', action='append', type=Path, help='file root search path') +parser.add_argument('-x', '--exclude', action='append', help='file to completely exclude from inlining') +parser.add_argument('-k', '--keep', action='append', help='file to exclude from inlining but keep the include directive') +parser.add_argument('-p', '--pragma', action='store_true', default=False, help='keep any "#pragma once" directives (removed by default)') +parser.add_argument('-o', '--output', type=argparse.FileType('w'), help='output file (otherwise stdout)') +parser.add_argument('input', type=Path, help='input file') +args = parser.parse_args() + +# Fail early on an invalid input (and store it so we don't recurse) +args.input = args.input.resolve(strict=True) +found.add(args.input) + +# Resolve all of the root paths upfront (we'll halt here on invalid roots) +if (args.root): + for path in args.root: + roots.add(path.resolve(strict=True)) + +# The remaining params: so resolve the excluded files and #pragma once directive +resolve_excluded_files(args.exclude, excludes, args.input.parent) +resolve_excluded_files(args.keep, keeps, args.input.parent) +keep_pragma = args.pragma; + +# Then recursively process the input file +try: + if (args.output): + destn = args.output + add_file(args.input) +finally: + if (destn): + destn.close() diff --git a/priv/zstd/build/single_file_libs/combine.sh b/priv/zstd/build/single_file_libs/combine.sh index 8eac4f9..222c2c3 100755 --- a/priv/zstd/build/single_file_libs/combine.sh +++ b/priv/zstd/build/single_file_libs/combine.sh @@ -2,11 +2,7 @@ # Tool to bundle multiple C/C++ source files, inlining any includes. # -# Note: this POSIX-compliant script is many times slower than the original bash -# implementation (due to the grep calls) but it runs and works everywhere. -# # TODO: ROOTS, FOUND, etc., as arrays (since they fail on paths with spaces) -# TODO: revert to Bash-only regex (the grep ones being too slow) # # Author: Carl Woffenden, Numfum GmbH (this script is released under a CC0 license/Public Domain) @@ -52,13 +48,54 @@ test_deps() { fi } +# Test if glob pattern $1 matches subject $2 (see fnmatch(3)) +fnmatch() { + case "$2" in + $1) + return 0 + ;; + esac + return 1 +} + +# Test if line $1 is local include directive +is_include_line() { + fnmatch "*#*include*" "$1" || return 1 + printf "%s\n" "$1" | grep -Eq '^\s*#\s*include\s*".+"' +} + +# Test if line $1 is pragma once directive +is_pragma_once_line() { + fnmatch "*#*pragma*once*" "$1" || return 1 + printf "%s\n" "$1" | grep -Eq '^\s*#\s*pragma\s*once\s*' +} + # Tests if list $1 has item $2 (returning zero on a match) +# (originally used grep -Eq "(^|\s*)$2(\$|\s*)) +readonly list_FS="$IFS" list_has_item() { - if echo "$1" | grep -Eq "(^|\s*)$2(\$|\s*)"; then - return 0 - else - return 1 + # Re: escaping glob pattern special characters in item string: + # + # bash (tested 3.2.57, 5.1.4), dash (tested 0.5.10.2), NetBSD /bin/sh + # (tested 8.2), and Solaris /bin/sh (tested 11.4) require escaping + # backslashes in a bracket expression despite POSIX specifying that + # backslash loses significance in a bracket expression. + # + # Conversely, neither FreeBSD /bin/sh (tested 12.2) nor OpenBSD /bin/sh + # (tested 7.1) obey backslash-escaping in case statement patterns even + # outside bracket expressions, so escape special characters using bracket + # expressions. + # + # Solaris /bin/sh (tested 11.4) requires vertical bar (|) to be escaped. + # + # All accommodations should behave as expected under strict POSIX semantics. + if fnmatch "*[\\*?[|]*" "$2"; then + set -- "$1" "$(printf '%s\n' "$2" | sed -e 's/[*?[|]/[&]/g; s/[\]/[\\&]/g')" fi + for item_P in "*[$list_FS]$2[$list_FS]*" "*[$list_FS]$2" "$2[$list_FS]*" "$2"; do + fnmatch "${item_P}" "$1" && return 0 + done + return 1 } # Adds a new line with the supplied arguments to $DESTN (or stdout) @@ -124,13 +161,13 @@ add_file() { # Read the file local line= while IFS= read -r line; do - if echo "$line" | grep -Eq '^\s*#\s*include\s*".+"'; then + if is_include_line "$line"; then # We have an include directive so strip the (first) file local inc=$(echo "$line" | grep -Eo '".*"' | sed -E 's/"([^"]+)"/\1/' | head -1) local res_inc="$(resolve_include "$srcdir" "$inc")" if list_has_item "$XINCS" "$inc"; then # The file was excluded so error if the source attempts to use it - write_line "#error Using excluded file: $inc" + write_line "#error Using excluded file: $inc (re-amalgamate source to fix)" log_line "Excluding: $inc" else if ! list_has_item "$FOUND" "$res_inc"; then @@ -155,7 +192,7 @@ add_file() { # Skip any 'pragma once' directives, otherwise write the source line local write=$PONCE if [ $write -eq 0 ]; then - if echo "$line" | grep -Eqv '^\s*#\s*pragma\s*once\s*'; then + if ! is_pragma_once_line "$line"; then write=1 fi fi @@ -200,6 +237,7 @@ if [ -n "$1" ]; then printf "" > "$DESTN" fi test_deps + log_line "Processing using the slower shell script; this might take a while" add_file "$1" else echo "Input file not found: \"$1\"" diff --git a/priv/zstd/build/single_file_libs/create_single_file_decoder.sh b/priv/zstd/build/single_file_libs/create_single_file_decoder.sh index b5f5613..3c0c577 100755 --- a/priv/zstd/build/single_file_libs/create_single_file_decoder.sh +++ b/priv/zstd/build/single_file_libs/create_single_file_decoder.sh @@ -4,8 +4,13 @@ ZSTD_SRC_ROOT="../../lib" # Amalgamate the sources -echo "Amalgamating files... this can take a while" -./combine.sh -r "$ZSTD_SRC_ROOT" -o zstddeclib.c zstddeclib-in.c +echo "Amalgamating files..." +# Using the faster Python script if we have 3.8 or higher +if python3 -c 'import sys; assert sys.version_info >= (3,8)' 2>/dev/null; then + ./combine.py -r "$ZSTD_SRC_ROOT" -x legacy/zstd_legacy.h -o zstddeclib.c zstddeclib-in.c +else + ./combine.sh -r "$ZSTD_SRC_ROOT" -x legacy/zstd_legacy.h -o zstddeclib.c zstddeclib-in.c +fi # Did combining work? if [ $? -ne 0 ]; then echo "Combine script: FAILED" diff --git a/priv/zstd/build/single_file_libs/create_single_file_library.sh b/priv/zstd/build/single_file_libs/create_single_file_library.sh index 6f38526..a6f71f0 100755 --- a/priv/zstd/build/single_file_libs/create_single_file_library.sh +++ b/priv/zstd/build/single_file_libs/create_single_file_library.sh @@ -4,8 +4,13 @@ ZSTD_SRC_ROOT="../../lib" # Amalgamate the sources -echo "Amalgamating files... this can take a while" -./combine.sh -r "$ZSTD_SRC_ROOT" -o zstd.c zstd-in.c +echo "Amalgamating files..." +# Using the faster Python script if we have 3.8 or higher +if python3 -c 'import sys; assert sys.version_info >= (3,8)' 2>/dev/null; then + ./combine.py -r "$ZSTD_SRC_ROOT" -x legacy/zstd_legacy.h -o zstd.c zstd-in.c +else + ./combine.sh -r "$ZSTD_SRC_ROOT" -x legacy/zstd_legacy.h -o zstd.c zstd-in.c +fi # Did combining work? if [ $? -ne 0 ]; then echo "Combine script: FAILED" diff --git a/priv/zstd/build/single_file_libs/zstd-in.c b/priv/zstd/build/single_file_libs/zstd-in.c index 1b27953..e6fca9e 100644 --- a/priv/zstd/build/single_file_libs/zstd-in.c +++ b/priv/zstd/build/single_file_libs/zstd-in.c @@ -4,11 +4,11 @@ * * Generate using: * \code - * combine.sh -r ../../lib -o zstd.c zstd-in.c + * python combine.py -r ../../lib -x legacy/zstd_legacy.h -o zstd.c zstd-in.c * \endcode */ /* - * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -25,9 +25,13 @@ * Note: MEM_MODULE stops xxhash redefining BYTE, U16, etc., which are also * defined in mem.h (breaking C99 compatibility). * - * Note: the undefs for xxHash allow Zstd's implementation to coinside with with + * Note: the undefs for xxHash allow Zstd's implementation to coincide with * standalone xxHash usage (with global defines). * + * Note: if you enable ZSTD_LEGACY_SUPPORT the combine.py script will need + * re-running without the "-x legacy/zstd_legacy.h" option (it excludes the + * legacy support at the source level). + * * Note: multithreading is enabled for all platforms apart from Emscripten. */ #define DEBUGLEVEL 0 @@ -43,6 +47,8 @@ #define ZSTD_MULTITHREAD #endif #define ZSTD_TRACE 0 +/* TODO: Can't amalgamate ASM function */ +#define ZSTD_DISABLE_ASM 1 /* Include zstd_deps.h first with all the options we need enabled. */ #define ZSTD_DEPS_NEED_MALLOC diff --git a/priv/zstd/build/single_file_libs/zstddeclib-in.c b/priv/zstd/build/single_file_libs/zstddeclib-in.c index 019d9c2..8d9c1f5 100644 --- a/priv/zstd/build/single_file_libs/zstddeclib-in.c +++ b/priv/zstd/build/single_file_libs/zstddeclib-in.c @@ -4,11 +4,11 @@ * * Generate using: * \code - * combine.sh -r ../../lib -o zstddeclib.c zstddeclib-in.c + * python combine.py -r ../../lib -x legacy/zstd_legacy.h -o zstddeclib.c zstddeclib-in.c * \endcode */ /* - * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -25,8 +25,12 @@ * Note: MEM_MODULE stops xxhash redefining BYTE, U16, etc., which are also * defined in mem.h (breaking C99 compatibility). * - * Note: the undefs for xxHash allow Zstd's implementation to coinside with with + * Note: the undefs for xxHash allow Zstd's implementation to coincide with * standalone xxHash usage (with global defines). + * + * Note: if you enable ZSTD_LEGACY_SUPPORT the combine.py script will need + * re-running without the "-x legacy/zstd_legacy.h" option (it excludes the + * legacy support at the source level). */ #define DEBUGLEVEL 0 #define MEM_MODULE @@ -39,6 +43,8 @@ #define ZSTD_LEGACY_SUPPORT 0 #define ZSTD_STRIP_ERROR_STRINGS #define ZSTD_TRACE 0 +/* TODO: Can't amalgamate ASM function */ +#define ZSTD_DISABLE_ASM 1 /* Include zstd_deps.h first with all the options we need enabled. */ #define ZSTD_DEPS_NEED_MALLOC diff --git a/priv/zstd/contrib/VS2005/zstd/zstd.vcproj b/priv/zstd/contrib/VS2005/zstd/zstd.vcproj index 78645d1..e37ebee 100644 --- a/priv/zstd/contrib/VS2005/zstd/zstd.vcproj +++ b/priv/zstd/contrib/VS2005/zstd/zstd.vcproj @@ -363,6 +363,10 @@ RelativePath="..\..\..\programs\fileio.c" > + + diff --git a/priv/zstd/contrib/diagnose_corruption/Makefile b/priv/zstd/contrib/diagnose_corruption/Makefile index a21a002..ecc9e63 100644 --- a/priv/zstd/contrib/diagnose_corruption/Makefile +++ b/priv/zstd/contrib/diagnose_corruption/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2019-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/diagnose_corruption/check_flipped_bits.c b/priv/zstd/contrib/diagnose_corruption/check_flipped_bits.c index cc40ab8..09ddd46 100644 --- a/priv/zstd/contrib/diagnose_corruption/check_flipped_bits.c +++ b/priv/zstd/contrib/diagnose_corruption/check_flipped_bits.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/externalSequenceProducer/.gitignore b/priv/zstd/contrib/externalSequenceProducer/.gitignore new file mode 100644 index 0000000..147710a --- /dev/null +++ b/priv/zstd/contrib/externalSequenceProducer/.gitignore @@ -0,0 +1,2 @@ +# build artifacts +externalSequenceProducer diff --git a/priv/zstd/contrib/externalSequenceProducer/Makefile b/priv/zstd/contrib/externalSequenceProducer/Makefile new file mode 100644 index 0000000..0591ae0 --- /dev/null +++ b/priv/zstd/contrib/externalSequenceProducer/Makefile @@ -0,0 +1,40 @@ +# ################################################################ +# Copyright (c) Yann Collet, Meta Platforms, Inc. +# All rights reserved. +# +# This source code is licensed under both the BSD-style license (found in the +# LICENSE file in the root directory of this source tree) and the GPLv2 (found +# in the COPYING file in the root directory of this source tree). +# ################################################################ + +PROGDIR = ../../programs +LIBDIR = ../../lib + +LIBZSTD = $(LIBDIR)/libzstd.a + +CPPFLAGS+= -I$(LIBDIR) -I$(LIBDIR)/compress -I$(LIBDIR)/common + +CFLAGS ?= -O3 +CFLAGS += -std=gnu99 +DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \ + -Wstrict-aliasing=1 -Wswitch-enum \ + -Wstrict-prototypes -Wundef -Wpointer-arith \ + -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \ + -Wredundant-decls +CFLAGS += $(DEBUGFLAGS) $(MOREFLAGS) + +default: externalSequenceProducer + +all: externalSequenceProducer + +externalSequenceProducer: sequence_producer.c main.c $(LIBZSTD) + $(CC) $(CPPFLAGS) $(CFLAGS) $^ $(LDFLAGS) -o $@ + +.PHONY: $(LIBZSTD) +$(LIBZSTD): + $(MAKE) -C $(LIBDIR) libzstd.a CFLAGS="$(CFLAGS)" + +clean: + $(RM) *.o + $(MAKE) -C $(LIBDIR) clean > /dev/null + $(RM) externalSequenceProducer diff --git a/priv/zstd/contrib/externalSequenceProducer/README.md b/priv/zstd/contrib/externalSequenceProducer/README.md new file mode 100644 index 0000000..c16a170 --- /dev/null +++ b/priv/zstd/contrib/externalSequenceProducer/README.md @@ -0,0 +1,14 @@ +externalSequenceProducer +===================== + +`externalSequenceProducer` is a test tool for the Block-Level Sequence Producer API. +It demonstrates how to use the API to perform a simple round-trip test. + +A sample sequence producer is provided in sequence_producer.c, but the user can swap +this out with a different one if desired. The sample sequence producer implements +LZ parsing with a 1KB hashtable. Dictionary-based parsing is not currently supported. + +Command line : +``` +externalSequenceProducer filename +``` diff --git a/priv/zstd/contrib/externalSequenceProducer/main.c b/priv/zstd/contrib/externalSequenceProducer/main.c new file mode 100644 index 0000000..e67e295 --- /dev/null +++ b/priv/zstd/contrib/externalSequenceProducer/main.c @@ -0,0 +1,107 @@ +/* + * Copyright (c) Yann Collet, Meta Platforms, Inc. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + * You may select, at your option, one of the above-listed licenses. + */ + +#include +#include +#include +#include + +#define ZSTD_STATIC_LINKING_ONLY +#include "zstd.h" +#include "zstd_errors.h" +#include "sequence_producer.h" // simpleSequenceProducer + +#define CHECK(res) \ +do { \ + if (ZSTD_isError(res)) { \ + printf("ERROR: %s\n", ZSTD_getErrorName(res)); \ + return 1; \ + } \ +} while (0) \ + +int main(int argc, char *argv[]) { + if (argc != 2) { + printf("Usage: externalSequenceProducer \n"); + return 1; + } + + ZSTD_CCtx* const zc = ZSTD_createCCtx(); + + int simpleSequenceProducerState = 0xdeadbeef; + + // Here is the crucial bit of code! + ZSTD_registerSequenceProducer( + zc, + &simpleSequenceProducerState, + simpleSequenceProducer + ); + + { + size_t const res = ZSTD_CCtx_setParameter(zc, ZSTD_c_enableSeqProducerFallback, 1); + CHECK(res); + } + + FILE *f = fopen(argv[1], "rb"); + assert(f); + { + int const ret = fseek(f, 0, SEEK_END); + assert(ret == 0); + } + size_t const srcSize = ftell(f); + { + int const ret = fseek(f, 0, SEEK_SET); + assert(ret == 0); + } + + char* const src = malloc(srcSize + 1); + assert(src); + { + size_t const ret = fread(src, srcSize, 1, f); + assert(ret == 1); + int const ret2 = fclose(f); + assert(ret2 == 0); + } + + size_t const dstSize = ZSTD_compressBound(srcSize); + char* const dst = malloc(dstSize); + assert(dst); + + size_t const cSize = ZSTD_compress2(zc, dst, dstSize, src, srcSize); + CHECK(cSize); + + char* const val = malloc(srcSize); + assert(val); + + { + size_t const res = ZSTD_decompress(val, srcSize, dst, cSize); + CHECK(res); + } + + if (memcmp(src, val, srcSize) == 0) { + printf("Compression and decompression were successful!\n"); + printf("Original size: %lu\n", srcSize); + printf("Compressed size: %lu\n", cSize); + } else { + printf("ERROR: input and validation buffers don't match!\n"); + for (size_t i = 0; i < srcSize; i++) { + if (src[i] != val[i]) { + printf("First bad index: %zu\n", i); + break; + } + } + return 1; + } + + ZSTD_freeCCtx(zc); + free(src); + free(dst); + free(val); + return 0; +} diff --git a/priv/zstd/contrib/externalSequenceProducer/sequence_producer.c b/priv/zstd/contrib/externalSequenceProducer/sequence_producer.c new file mode 100644 index 0000000..60a2f95 --- /dev/null +++ b/priv/zstd/contrib/externalSequenceProducer/sequence_producer.c @@ -0,0 +1,80 @@ +/* + * Copyright (c) Yann Collet, Meta Platforms, Inc. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + * You may select, at your option, one of the above-listed licenses. + */ + +#include "zstd_compress_internal.h" +#include "sequence_producer.h" + +#define HSIZE 1024 +static U32 const HLOG = 10; +static U32 const MLS = 4; +static U32 const BADIDX = 0xffffffff; + +size_t simpleSequenceProducer( + void* sequenceProducerState, + ZSTD_Sequence* outSeqs, size_t outSeqsCapacity, + const void* src, size_t srcSize, + const void* dict, size_t dictSize, + int compressionLevel, + size_t windowSize +) { + const BYTE* const istart = (const BYTE*)src; + const BYTE* const iend = istart + srcSize; + const BYTE* ip = istart; + const BYTE* anchor = istart; + size_t seqCount = 0; + U32 hashTable[HSIZE]; + + (void)sequenceProducerState; + (void)dict; + (void)dictSize; + (void)outSeqsCapacity; + (void)compressionLevel; + + { int i; + for (i=0; i < HSIZE; i++) { + hashTable[i] = BADIDX; + } } + + while (ip + MLS < iend) { + size_t const hash = ZSTD_hashPtr(ip, HLOG, MLS); + U32 const matchIndex = hashTable[hash]; + hashTable[hash] = (U32)(ip - istart); + + if (matchIndex != BADIDX) { + const BYTE* const match = istart + matchIndex; + U32 const matchLen = (U32)ZSTD_count(ip, match, iend); + if (matchLen >= ZSTD_MINMATCH_MIN) { + U32 const litLen = (U32)(ip - anchor); + U32 const offset = (U32)(ip - match); + ZSTD_Sequence const seq = { + offset, litLen, matchLen, 0 + }; + + /* Note: it's crucial to stay within the window size! */ + if (offset <= windowSize) { + outSeqs[seqCount++] = seq; + ip += matchLen; + anchor = ip; + continue; + } + } + } + + ip++; + } + + { ZSTD_Sequence const finalSeq = { + 0, (U32)(iend - anchor), 0, 0 + }; + outSeqs[seqCount++] = finalSeq; + } + + return seqCount; +} diff --git a/priv/zstd/contrib/externalSequenceProducer/sequence_producer.h b/priv/zstd/contrib/externalSequenceProducer/sequence_producer.h new file mode 100644 index 0000000..19f9982 --- /dev/null +++ b/priv/zstd/contrib/externalSequenceProducer/sequence_producer.h @@ -0,0 +1,26 @@ +/* + * Copyright (c) Yann Collet, Meta Platforms, Inc. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + * You may select, at your option, one of the above-listed licenses. + */ + +#ifndef MATCHFINDER_H +#define MATCHFINDER_H + +#define ZSTD_STATIC_LINKING_ONLY +#include "zstd.h" + +size_t simpleSequenceProducer( + void* sequenceProducerState, + ZSTD_Sequence* outSeqs, size_t outSeqsCapacity, + const void* src, size_t srcSize, + const void* dict, size_t dictSize, + int compressionLevel, + size_t windowSize +); + +#endif diff --git a/priv/zstd/contrib/freestanding_lib/freestanding.py b/priv/zstd/contrib/freestanding_lib/freestanding.py index 1971687..df69832 100755 --- a/priv/zstd/contrib/freestanding_lib/freestanding.py +++ b/priv/zstd/contrib/freestanding_lib/freestanding.py @@ -1,6 +1,6 @@ #!/usr/bin/env python3 # ################################################################ -# Copyright (c) 2021-2021, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -27,7 +27,6 @@ "common/pool.h", "common/threading.c", "common/threading.h", - "common/zstd_trace.c", "common/zstd_trace.h", "compress/zstdmt_compress.h", "compress/zstdmt_compress.c", @@ -341,7 +340,7 @@ def _preprocess_once(self): if macro2 is not None and not resolved: assert ifdef and defined and op == '&&' and cmp is not None - # If the statment is true, but we have a single value check, then + # If the statement is true, but we have a single value check, then # check the value. defined_value = self._defs[macro] are_ints = True @@ -432,7 +431,7 @@ def __init__( external_xxhash: bool, xxh64_state: Optional[str], xxh64_prefix: Optional[str], rewritten_includes: [(str, str)], defs: [(str, Optional[str])], replaces: [(str, str)], - undefs: [str], excludes: [str], seds: [str], + undefs: [str], excludes: [str], seds: [str], spdx: bool, ): self._zstd_deps = zstd_deps self._mem = mem @@ -447,6 +446,7 @@ def __init__( self._undefs = undefs self._excludes = excludes self._seds = seds + self._spdx = spdx def _dst_lib_file_paths(self): """ @@ -461,7 +461,8 @@ def _log(self, *args, **kwargs): print(*args, **kwargs) def _copy_file(self, lib_path): - if not (lib_path.endswith(".c") or lib_path.endswith(".h")): + suffixes = [".c", ".h", ".S"] + if not any((lib_path.endswith(suffix) for suffix in suffixes)): return if lib_path in SKIPPED_FILES: self._log(f"\tSkipping file: {lib_path}") @@ -640,6 +641,27 @@ def _process_seds(self): for sed in self._seds: self._process_sed(sed) + def _process_spdx(self): + if not self._spdx: + return + self._log("Processing spdx") + SPDX_C = "// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause\n" + SPDX_H_S = "/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */\n" + for filepath in self._dst_lib_file_paths(): + file = FileLines(filepath) + if file.lines[0] == SPDX_C or file.lines[0] == SPDX_H_S: + continue + for line in file.lines: + if "SPDX-License-Identifier" in line: + raise RuntimeError(f"Unexpected SPDX license identifier: {file.filename} {repr(line)}") + if file.filename.endswith(".c"): + file.lines.insert(0, SPDX_C) + elif file.filename.endswith(".h") or file.filename.endswith(".S"): + file.lines.insert(0, SPDX_H_S) + else: + raise RuntimeError(f"Unexpected file extension: {file.filename}") + file.write() + def go(self): @@ -651,6 +673,7 @@ def go(self): self._rewrite_includes() self._replace_xxh64_prefix() self._process_seds() + self._process_spdx() def parse_optional_pair(defines: [str]) -> [(str, Optional[str])]: @@ -689,8 +712,9 @@ def main(name, args): parser.add_argument("--xxh64-prefix", default=None, help="Alternate XXH64 function prefix (excluding _) e.g. --xxh64-prefix=xxh64") parser.add_argument("--rewrite-include", default=[], dest="rewritten_includes", action="append", help="Rewrite an include REGEX=NEW (e.g. '=')") parser.add_argument("--sed", default=[], dest="seds", action="append", help="Apply a sed replacement. Format: `s/REGEX/FORMAT/[g]`. REGEX is a Python regex. FORMAT is a Python format string formatted by the regex dict.") + parser.add_argument("--spdx", action="store_true", help="Add SPDX License Identifiers") parser.add_argument("-D", "--define", default=[], dest="defs", action="append", help="Pre-define this macro (can be passed multiple times)") - parser.add_argument("-U", "--undefine", default=[], dest="undefs", action="append", help="Pre-undefine this macro (can be passed mutliple times)") + parser.add_argument("-U", "--undefine", default=[], dest="undefs", action="append", help="Pre-undefine this macro (can be passed multiple times)") parser.add_argument("-R", "--replace", default=[], dest="replaces", action="append", help="Pre-define this macro and replace the first ifndef block with its definition") parser.add_argument("-E", "--exclude", default=[], dest="excludes", action="append", help="Exclude all lines between 'BEGIN ' and 'END '") args = parser.parse_args(args) @@ -743,6 +767,7 @@ def main(name, args): args.undefs, args.excludes, args.seds, + args.spdx, ).go() if __name__ == "__main__": diff --git a/priv/zstd/contrib/gen_html/Makefile b/priv/zstd/contrib/gen_html/Makefile index 425f266..26e8703 100644 --- a/priv/zstd/contrib/gen_html/Makefile +++ b/priv/zstd/contrib/gen_html/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2016-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/gen_html/gen_html.cpp b/priv/zstd/contrib/gen_html/gen_html.cpp index 90d5b21..1da5879 100644 --- a/priv/zstd/contrib/gen_html/gen_html.cpp +++ b/priv/zstd/contrib/gen_html/gen_html.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Przemyslaw Skibinski, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/largeNbDicts/Makefile b/priv/zstd/contrib/largeNbDicts/Makefile index 4c055b0..40734e6 100644 --- a/priv/zstd/contrib/largeNbDicts/Makefile +++ b/priv/zstd/contrib/largeNbDicts/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2018-present, Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/largeNbDicts/README.md b/priv/zstd/contrib/largeNbDicts/README.md index f29bcdf..010102c 100644 --- a/priv/zstd/contrib/largeNbDicts/README.md +++ b/priv/zstd/contrib/largeNbDicts/README.md @@ -14,12 +14,20 @@ Command line : ``` largeNbDicts [Options] filename(s) -Options : --r : recursively load all files in subdirectories (default: off) --B# : split input into blocks of size # (default: no split) --# : use compression level # (default: 3) --D # : use # as a dictionary (default: create one) --i# : nb benchmark rounds (default: 6) ---nbDicts=# : set nb of dictionaries to # (default: one per block) --h : help (this text) +Options : +-z : benchmark compression (default) +-d : benchmark decompression +-r : recursively load all files in subdirectories (default: off) +-B# : split input into blocks of size # (default: no split) +-# : use compression level # (default: 3) +-D # : use # as a dictionary (default: create one) +-i# : nb benchmark rounds (default: 6) +--nbBlocks=#: use # blocks for bench (default: one per file) +--nbDicts=# : create # dictionaries for bench (default: one per block) +-h : help (this text) + +Advanced Options (see zstd.h for documentation) : +--dedicated-dict-search +--dict-content-type=# +--dict-attach-pref=# ``` diff --git a/priv/zstd/contrib/largeNbDicts/largeNbDicts.c b/priv/zstd/contrib/largeNbDicts/largeNbDicts.c index ddbb3e0..eeaaf71 100644 --- a/priv/zstd/contrib/largeNbDicts/largeNbDicts.c +++ b/priv/zstd/contrib/largeNbDicts/largeNbDicts.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-present, Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -19,7 +19,7 @@ /*--- Dependencies ---*/ #include /* size_t */ -#include /* malloc, free, abort */ +#include /* malloc, free, abort, qsort*/ #include /* fprintf */ #include /* UINT_MAX */ #include /* assert */ @@ -39,6 +39,7 @@ #define BLOCKSIZE_DEFAULT 0 /* no slicing into blocks */ #define DICTSIZE (4 KB) #define CLEVEL_DEFAULT 3 +#define DICT_LOAD_METHOD ZSTD_dlm_byCopy #define BENCH_TIME_DEFAULT_S 6 #define RUN_TIME_DEFAULT_MS 1000 @@ -156,19 +157,6 @@ createDictionaryBuffer(const char* dictionaryName, } } -static ZSTD_CDict* createCDictForDedicatedDictSearch(const void* dict, size_t dictSize, int compressionLevel) -{ - ZSTD_CCtx_params* params = ZSTD_createCCtxParams(); - ZSTD_CCtxParams_init(params, compressionLevel); - ZSTD_CCtxParams_setParameter(params, ZSTD_c_enableDedicatedDictSearch, 1); - ZSTD_CCtxParams_setParameter(params, ZSTD_c_compressionLevel, compressionLevel); - - ZSTD_CDict* cdict = ZSTD_createCDict_advanced2(dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto, params, ZSTD_defaultCMem); - - ZSTD_freeCCtxParams(params); - return cdict; -} - /*! BMK_loadFiles() : * Loads `buffer`, with content from files listed within `fileNamesTable`. * Fills `buffer` entirely. @@ -461,14 +449,12 @@ static void freeCDictCollection(cdict_collection_t cdictc) } /* returns .buffers=NULL if operation fails */ -static cdict_collection_t createCDictCollection(const void* dictBuffer, size_t dictSize, size_t nbCDict, int cLevel, int dedicatedDictSearch) +static cdict_collection_t createCDictCollection(const void* dictBuffer, size_t dictSize, size_t nbCDict, ZSTD_dictContentType_e dictContentType, ZSTD_CCtx_params* cctxParams) { ZSTD_CDict** const cdicts = malloc(nbCDict * sizeof(ZSTD_CDict*)); if (cdicts==NULL) return kNullCDictCollection; for (size_t dictNb=0; dictNb < nbCDict; dictNb++) { - cdicts[dictNb] = dedicatedDictSearch ? - createCDictForDedicatedDictSearch(dictBuffer, dictSize, cLevel) : - ZSTD_createCDict(dictBuffer, dictSize, cLevel); + cdicts[dictNb] = ZSTD_createCDict_advanced2(dictBuffer, dictSize, DICT_LOAD_METHOD, dictContentType, cctxParams, ZSTD_defaultCMem); CONTROL(cdicts[dictNb] != NULL); } cdict_collection_t cdictc; @@ -591,11 +577,13 @@ typedef struct { cdict_collection_t dictionaries; } compressInstructions; -compressInstructions createCompressInstructions(cdict_collection_t dictionaries) +compressInstructions createCompressInstructions(cdict_collection_t dictionaries, ZSTD_CCtx_params* cctxParams) { compressInstructions ci; ci.cctx = ZSTD_createCCtx(); CONTROL(ci.cctx != NULL); + if (cctxParams) + ZSTD_CCtx_setParametersUsingCCtxParams(ci.cctx, cctxParams); ci.nbDicts = dictionaries.nbCDict; ci.dictNb = 0; ci.dictionaries = dictionaries; @@ -636,10 +624,10 @@ size_t compress(const void* src, size_t srcSize, void* dst, size_t dstCapacity, compressInstructions* const ci = (compressInstructions*) payload; (void)dstCapacity; - ZSTD_compress_usingCDict(ci->cctx, - dst, srcSize, - src, srcSize, - ci->dictionaries.cdicts[ci->dictNb]); + ZSTD_CCtx_refCDict(ci->cctx, ci->dictionaries.cdicts[ci->dictNb]); + ZSTD_compress2(ci->cctx, + dst, srcSize, + src, srcSize); ci->dictNb = ci->dictNb + 1; if (ci->dictNb >= ci->nbDicts) ci->dictNb = 0; @@ -663,25 +651,55 @@ size_t decompress(const void* src, size_t srcSize, void* dst, size_t dstCapacity return result; } +typedef enum { + fastest = 0, + median = 1, +} metricAggregatePref_e; -static int benchMem(slice_collection_t dstBlocks, - slice_collection_t srcBlocks, +/* compareFunction() : + * Sort input in decreasing order when used with qsort() */ +int compareFunction(const void *a, const void *b) +{ + double x = *(const double *)a; + double y = *(const double *)b; + if (x < y) + return 1; + else if (x > y) + return -1; + return 0; +} + +double aggregateData(double *data, size_t size, + metricAggregatePref_e metricAggregatePref) +{ + qsort(data, size, sizeof(*data), compareFunction); + if (metricAggregatePref == fastest) + return data[0]; + else /* median */ + return (data[(size - 1) / 2] + data[size / 2]) / 2; +} + +static int benchMem(slice_collection_t dstBlocks, slice_collection_t srcBlocks, ddict_collection_t ddictionaries, - cdict_collection_t cdictionaries, - unsigned nbRounds, int benchCompression) + cdict_collection_t cdictionaries, unsigned nbRounds, + int benchCompression, const char *exeName, + ZSTD_CCtx_params *cctxParams, + metricAggregatePref_e metricAggregatePref) { assert(dstBlocks.nbSlices == srcBlocks.nbSlices); + if (benchCompression) assert(cctxParams); unsigned const ms_per_round = RUN_TIME_DEFAULT_MS; unsigned const total_time_ms = nbRounds * ms_per_round; - double bestSpeed = 0.; + double *const speedPerRound = (double *)malloc(nbRounds * sizeof(double)); BMK_timedFnState_t* const benchState = BMK_createTimedFnState(total_time_ms, ms_per_round); decompressInstructions di = createDecompressInstructions(ddictionaries); - compressInstructions ci = createCompressInstructions(cdictionaries); + compressInstructions ci = + createCompressInstructions(cdictionaries, cctxParams); void* payload = benchCompression ? (void*)&ci : (void*)&di; BMK_benchParams_t const bp = { .benchFn = benchCompression ? compress : decompress, @@ -697,6 +715,7 @@ static int benchMem(slice_collection_t dstBlocks, .blockResults = NULL }; + size_t roundNb = 0; for (;;) { BMK_runOutcome_t const outcome = BMK_benchTimedFn(benchState, bp); CONTROL(BMK_isSuccessful_runOutcome(outcome)); @@ -706,16 +725,57 @@ static int benchMem(slice_collection_t dstBlocks, double const dTime_sec = (double)dTime_ns / 1000000000; size_t const srcSize = result.sumOfReturn; double const speed_MBps = (double)srcSize / dTime_sec / (1 MB); - if (speed_MBps > bestSpeed) bestSpeed = speed_MBps; + speedPerRound[roundNb] = speed_MBps; if (benchCompression) - DISPLAY("Compression Speed : %.1f MB/s \r", bestSpeed); + DISPLAY("Compression Speed : %.1f MB/s \r", speed_MBps); else - DISPLAY("Decompression Speed : %.1f MB/s \r", bestSpeed); + DISPLAY("Decompression Speed : %.1f MB/s \r", speed_MBps); fflush(stdout); if (BMK_isCompleted_TimedFn(benchState)) break; + roundNb++; } DISPLAY("\n"); + /* BMK_benchTimedFn may not run exactly nbRounds iterations */ + double speedAggregated = + aggregateData(speedPerRound, roundNb + 1, metricAggregatePref); + if (metricAggregatePref == fastest) + DISPLAY("Fastest Speed : %.1f MB/s \n", speedAggregated); + else + DISPLAY("Median Speed : %.1f MB/s \n", speedAggregated); + + char* csvFileName = malloc(strlen(exeName) + 5); + strcpy(csvFileName, exeName); + strcat(csvFileName, ".csv"); + FILE* csvFile = fopen(csvFileName, "r"); + if (!csvFile) { + csvFile = fopen(csvFileName, "wt"); + assert(csvFile); + fprintf(csvFile, "%s\n", exeName); + /* Print table headers */ + fprintf( + csvFile, + "Compression/Decompression,Level,nbDicts,dictAttachPref,metricAggregatePref,Speed\n"); + } else { + fclose(csvFile); + csvFile = fopen(csvFileName, "at"); + assert(csvFile); + } + + int cLevel = -1; + int dictAttachPref = -1; + if (benchCompression) { + ZSTD_CCtxParams_getParameter(cctxParams, ZSTD_c_compressionLevel, + &cLevel); + ZSTD_CCtxParams_getParameter(cctxParams, ZSTD_c_forceAttachDict, + &dictAttachPref); + } + fprintf(csvFile, "%s,%d,%ld,%d,%d,%.1f\n", + benchCompression ? "Compression" : "Decompression", cLevel, + benchCompression ? ci.nbDicts : di.nbDicts, dictAttachPref, + metricAggregatePref, speedAggregated); + fclose(csvFile); + free(csvFileName); freeDecompressInstructions(di); freeCompressInstructions(ci); @@ -730,12 +790,11 @@ static int benchMem(slice_collection_t dstBlocks, * dictionary : optional (can be NULL), file to load as dictionary, * if none provided : will be calculated on the fly by the program. * @return : 0 is success, 1+ otherwise */ -int bench(const char** fileNameTable, unsigned nbFiles, - const char* dictionary, - size_t blockSize, int clevel, - unsigned nbDictMax, unsigned nbBlocks, +int bench(const char **fileNameTable, unsigned nbFiles, const char *dictionary, + size_t blockSize, int clevel, unsigned nbDictMax, unsigned nbBlocks, unsigned nbRounds, int benchCompression, - int dedicatedDictSearch) + ZSTD_dictContentType_e dictContentType, ZSTD_CCtx_params *cctxParams, + const char *exeName, metricAggregatePref_e metricAggregatePref) { int result = 0; @@ -786,20 +845,18 @@ int bench(const char** fileNameTable, unsigned nbFiles, /* dictionary determination */ buffer_t const dictBuffer = createDictionaryBuffer(dictionary, srcs.buffer.ptr, - srcs.slices.capacities, srcs.slices.nbSlices, + srcSlices.capacities, srcSlices.nbSlices, DICTSIZE); CONTROL(dictBuffer.ptr != NULL); - ZSTD_CDict* const cdict = dedicatedDictSearch ? - createCDictForDedicatedDictSearch(dictBuffer.ptr, dictBuffer.size, clevel) : - ZSTD_createCDict(dictBuffer.ptr, dictBuffer.size, clevel); + ZSTD_CDict* const cdict = ZSTD_createCDict_advanced2(dictBuffer.ptr, dictBuffer.size, DICT_LOAD_METHOD, dictContentType, cctxParams, ZSTD_defaultCMem); CONTROL(cdict != NULL); size_t const cTotalSizeNoDict = compressBlocks(NULL, dstSlices, srcSlices, NULL, clevel); CONTROL(cTotalSizeNoDict != 0); DISPLAYLEVEL(3, "compressing at level %u without dictionary : Ratio=%.2f (%u bytes) \n", clevel, - (double)totalSrcSlicesSize / cTotalSizeNoDict, (unsigned)cTotalSizeNoDict); + (double)totalSrcSlicesSize / (double)cTotalSizeNoDict, (unsigned)cTotalSizeNoDict); size_t* const cSizes = malloc(nbBlocks * sizeof(size_t)); CONTROL(cSizes != NULL); @@ -808,21 +865,21 @@ int bench(const char** fileNameTable, unsigned nbFiles, CONTROL(cTotalSize != 0); DISPLAYLEVEL(3, "compressed using a %u bytes dictionary : Ratio=%.2f (%u bytes) \n", (unsigned)dictBuffer.size, - (double)totalSrcSlicesSize / cTotalSize, (unsigned)cTotalSize); + (double)totalSrcSlicesSize / (double)cTotalSize, (unsigned)cTotalSize); /* now dstSlices contain the real compressed size of each block, instead of the maximum capacity */ shrinkSizes(dstSlices, cSizes); unsigned const nbDicts = nbDictMax ? nbDictMax : nbBlocks; - cdict_collection_t const cdictionaries = createCDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts, clevel, dedicatedDictSearch); + cdict_collection_t const cdictionaries = createCDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts, dictContentType, cctxParams); CONTROL(cdictionaries.cdicts != NULL); ddict_collection_t const ddictionaries = createDDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts); CONTROL(ddictionaries.ddicts != NULL); if (benchCompression) { - size_t const dictMem = ZSTD_estimateCDictSize(dictBuffer.size, ZSTD_dlm_byCopy); + size_t const dictMem = ZSTD_sizeof_CDict(cdictionaries.cdicts[0]); size_t const allDictMem = dictMem * nbDicts; DISPLAYLEVEL(3, "generating %u dictionaries, using %.1f MB of memory \n", nbDicts, (double)allDictMem / (1 MB)); @@ -832,11 +889,13 @@ int bench(const char** fileNameTable, unsigned nbFiles, buffer_collection_t resultCollection = createBufferCollection_fromSliceCollection(srcSlices); CONTROL(resultCollection.buffer.ptr != NULL); - result = benchMem(dstSlices, resultCollection.slices, ddictionaries, cdictionaries, nbRounds, benchCompression); + result = benchMem(dstSlices, resultCollection.slices, ddictionaries, + cdictionaries, nbRounds, benchCompression, exeName, + cctxParams, metricAggregatePref); freeBufferCollection(resultCollection); } else { - size_t const dictMem = ZSTD_estimateDDictSize(dictBuffer.size, ZSTD_dlm_byCopy); + size_t const dictMem = ZSTD_estimateDDictSize(dictBuffer.size, DICT_LOAD_METHOD); size_t const allDictMem = dictMem * nbDicts; DISPLAYLEVEL(3, "generating %u dictionaries, using %.1f MB of memory \n", nbDicts, (double)allDictMem / (1 MB)); @@ -846,7 +905,9 @@ int bench(const char** fileNameTable, unsigned nbFiles, buffer_collection_t resultCollection = createBufferCollection_fromSliceCollectionSizes(srcSlices); CONTROL(resultCollection.buffer.ptr != NULL); - result = benchMem(resultCollection.slices, dstSlices, ddictionaries, cdictionaries, nbRounds, benchCompression); + result = benchMem(resultCollection.slices, dstSlices, ddictionaries, + cdictionaries, nbRounds, benchCompression, exeName, + NULL, metricAggregatePref); freeBufferCollection(resultCollection); } @@ -924,9 +985,15 @@ int usage(const char* exeName) DISPLAY ("-# : use compression level # (default: %u) \n", CLEVEL_DEFAULT); DISPLAY ("-D # : use # as a dictionary (default: create one) \n"); DISPLAY ("-i# : nb benchmark rounds (default: %u) \n", BENCH_TIME_DEFAULT_S); + DISPLAY ("-p# : print speed for all rounds 0=fastest 1=median (default: 0) \n"); DISPLAY ("--nbBlocks=#: use # blocks for bench (default: one per file) \n"); DISPLAY ("--nbDicts=# : create # dictionaries for bench (default: one per block) \n"); DISPLAY ("-h : help (this text) \n"); + DISPLAY (" \n"); + DISPLAY ("Advanced Options (see zstd.h for documentation) : \n"); + DISPLAY ("--dedicated-dict-search\n"); + DISPLAY ("--dict-content-type=#\n"); + DISPLAY ("--dict-attach-pref=#\n"); return 0; } @@ -956,6 +1023,10 @@ int main (int argc, const char** argv) size_t blockSize = BLOCKSIZE_DEFAULT; unsigned nbDicts = 0; /* determine nbDicts automatically: 1 dictionary per block */ unsigned nbBlocks = 0; /* determine nbBlocks automatically, from source and blockSize */ + ZSTD_dictContentType_e dictContentType = ZSTD_dct_auto; + ZSTD_dictAttachPref_e dictAttachPref = ZSTD_dictDefaultAttach; + ZSTD_paramSwitch_e prefetchCDictTables = ZSTD_ps_auto; + metricAggregatePref_e metricAggregatePref = fastest; for (int argNb = 1; argNb < argc ; argNb++) { const char* argument = argv[argNb]; @@ -965,6 +1036,7 @@ int main (int argc, const char** argv) if (!strcmp(argument, "-r")) { recursiveMode = 1; continue; } if (!strcmp(argument, "-D")) { argNb++; assert(argNb < argc); dictionary = argv[argNb]; continue; } if (longCommandWArg(&argument, "-i")) { nbRounds = readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "-p")) { metricAggregatePref = (int)readU32FromChar(&argument); continue;} if (longCommandWArg(&argument, "--dictionary=")) { dictionary = argument; continue; } if (longCommandWArg(&argument, "-B")) { blockSize = readU32FromChar(&argument); continue; } if (longCommandWArg(&argument, "--blockSize=")) { blockSize = readU32FromChar(&argument); continue; } @@ -972,6 +1044,9 @@ int main (int argc, const char** argv) if (longCommandWArg(&argument, "--nbBlocks=")) { nbBlocks = readU32FromChar(&argument); continue; } if (longCommandWArg(&argument, "--clevel=")) { cLevel = (int)readU32FromChar(&argument); continue; } if (longCommandWArg(&argument, "--dedicated-dict-search")) { dedicatedDictSearch = 1; continue; } + if (longCommandWArg(&argument, "--dict-content-type=")) { dictContentType = (int)readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--dict-attach-pref=")) { dictAttachPref = (int)readU32FromChar(&argument); continue; } + if (longCommandWArg(&argument, "--prefetch-cdict-tables=")) { prefetchCDictTables = (int)readU32FromChar(&argument); continue; } if (longCommandWArg(&argument, "-")) { cLevel = (int)readU32FromChar(&argument); continue; } /* anything that's not a command is a filename */ nameTable[nameIdx++] = argument; @@ -989,10 +1064,22 @@ int main (int argc, const char** argv) nameTable = NULL; /* UTIL_createFileNamesTable() takes ownership of nameTable */ } - int result = bench(filenameTable->fileNames, (unsigned)filenameTable->tableSize, dictionary, blockSize, cLevel, nbDicts, nbBlocks, nbRounds, benchCompression, dedicatedDictSearch); + ZSTD_CCtx_params* cctxParams = ZSTD_createCCtxParams(); + ZSTD_CCtxParams_init(cctxParams, cLevel); + ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_enableDedicatedDictSearch, dedicatedDictSearch); + ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_nbWorkers, 0); + ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_forceAttachDict, dictAttachPref); + ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_prefetchCDictTables, prefetchCDictTables); + + int result = + bench(filenameTable->fileNames, (unsigned)filenameTable->tableSize, + dictionary, blockSize, cLevel, nbDicts, nbBlocks, nbRounds, + benchCompression, dictContentType, cctxParams, exeName, + metricAggregatePref); UTIL_freeFileNamesTable(filenameTable); free(nameTable); + ZSTD_freeCCtxParams(cctxParams); return result; } diff --git a/priv/zstd/contrib/linux-kernel/Makefile b/priv/zstd/contrib/linux-kernel/Makefile index c391df7..63dd15d 100644 --- a/priv/zstd/contrib/linux-kernel/Makefile +++ b/priv/zstd/contrib/linux-kernel/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -26,20 +26,20 @@ libzstd: --rewrite-include '"(\.\./)?zstd_errors.h"=' \ --sed 's,/\*\*\*,/* *,g' \ --sed 's,/\*\*,/*,g' \ + --spdx \ -DZSTD_NO_INTRINSICS \ -DZSTD_NO_UNUSED_FUNCTIONS \ -DZSTD_LEGACY_SUPPORT=0 \ -DZSTD_STATIC_LINKING_ONLY \ -DFSE_STATIC_LINKING_ONLY \ - -DHUF_STATIC_LINKING_ONLY \ -DXXH_STATIC_LINKING_ONLY \ - -DMEM_FORCE_MEMORY_ACCESS=0 \ -D__GNUC__ \ + -D__linux__=1 \ -DSTATIC_BMI2=0 \ -DZSTD_ADDRESS_SANITIZER=0 \ -DZSTD_MEMORY_SANITIZER=0 \ + -DZSTD_DATAFLOW_SANITIZER=0 \ -DZSTD_COMPRESS_HEAPMODE=1 \ - -UZSTD_NO_INLINE \ -UNO_PREFETCH \ -U__cplusplus \ -UZSTD_DLL_EXPORT \ @@ -48,14 +48,19 @@ libzstd: -UZSTD_MULTITHREAD \ -U_MSC_VER \ -U_WIN32 \ - -RZSTDLIB_VISIBILITY= \ - -RZSTDERRORLIB_VISIBILITY= \ + -RZSTDLIB_VISIBLE= \ + -RZSTDERRORLIB_VISIBLE= \ + -RZSTD_FALLTHROUGH=fallthrough \ -DZSTD_HAVE_WEAK_SYMBOLS=0 \ -DZSTD_TRACE=0 \ - -DZSTD_NO_TRACE + -DZSTD_NO_TRACE \ + -DZSTD_DISABLE_ASM \ + -DZSTD_LINUX_KERNEL + rm linux/lib/zstd/decompress/huf_decompress_amd64.S mv linux/lib/zstd/zstd.h linux/include/linux/zstd_lib.h mv linux/lib/zstd/zstd_errors.h linux/include/linux/ cp linux_zstd.h linux/include/linux/zstd.h + cp zstd_common_module.c linux/lib/zstd cp zstd_compress_module.c linux/lib/zstd cp zstd_decompress_module.c linux/lib/zstd cp decompress_sources.h linux/lib/zstd @@ -86,10 +91,18 @@ import-upstream: rm $(LINUX)/lib/zstd/common/xxhash.* rm $(LINUX)/lib/zstd/compress/zstdmt_* +DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \ + -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \ + -Wstrict-prototypes -Wundef -Wpointer-arith \ + -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \ + -Wredundant-decls -Wmissing-prototypes -Wc++-compat \ + -Wimplicit-fallthrough + .PHONY: test test: libzstd - $(MAKE) -C test run-test CFLAGS="-O3 $(CFLAGS)" -j + $(MAKE) -C test run-test CFLAGS="-O3 $(CFLAGS) $(DEBUGFLAGS) -Werror" -j .PHONY: clean clean: $(RM) -rf linux + $(MAKE) -C test clean diff --git a/priv/zstd/contrib/linux-kernel/decompress_sources.h b/priv/zstd/contrib/linux-kernel/decompress_sources.h index f35bef0..8a47eb2 100644 --- a/priv/zstd/contrib/linux-kernel/decompress_sources.h +++ b/priv/zstd/contrib/linux-kernel/decompress_sources.h @@ -1,6 +1,6 @@ -/* SPDX-License-Identifier: GPL-2.0-only */ +/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -16,6 +16,12 @@ * decompression. */ +/* + * Disable the ASM Huffman implementation because we need to + * include all the sources. + */ +#define ZSTD_DISABLE_ASM 1 + #include "common/debug.c" #include "common/entropy_common.c" #include "common/error_private.c" diff --git a/priv/zstd/contrib/linux-kernel/linux.mk b/priv/zstd/contrib/linux-kernel/linux.mk index 19485e3..464c410 100644 --- a/priv/zstd/contrib/linux-kernel/linux.mk +++ b/priv/zstd/contrib/linux-kernel/linux.mk @@ -1,6 +1,6 @@ -# SPDX-License-Identifier: GPL-2.0-only +# SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause # ################################################################ -# Copyright (c) Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -10,16 +10,10 @@ # ################################################################ obj-$(CONFIG_ZSTD_COMPRESS) += zstd_compress.o obj-$(CONFIG_ZSTD_DECOMPRESS) += zstd_decompress.o - -ccflags-y += -O3 +obj-$(CONFIG_ZSTD_COMMON) += zstd_common.o zstd_compress-y := \ zstd_compress_module.o \ - common/debug.o \ - common/entropy_common.o \ - common/error_private.o \ - common/fse_decompress.o \ - common/zstd_common.o \ compress/fse_compress.o \ compress/hist.o \ compress/huf_compress.o \ @@ -35,12 +29,15 @@ zstd_compress-y := \ zstd_decompress-y := \ zstd_decompress_module.o \ + decompress/huf_decompress.o \ + decompress/zstd_ddict.o \ + decompress/zstd_decompress.o \ + decompress/zstd_decompress_block.o \ + +zstd_common-y := \ + zstd_common_module.o \ common/debug.o \ common/entropy_common.o \ common/error_private.o \ common/fse_decompress.o \ common/zstd_common.o \ - decompress/huf_decompress.o \ - decompress/zstd_ddict.o \ - decompress/zstd_decompress.o \ - decompress/zstd_decompress_block.o \ diff --git a/priv/zstd/contrib/linux-kernel/linux_zstd.h b/priv/zstd/contrib/linux-kernel/linux_zstd.h index 446ecab..f109d49 100644 --- a/priv/zstd/contrib/linux-kernel/linux_zstd.h +++ b/priv/zstd/contrib/linux-kernel/linux_zstd.h @@ -1,6 +1,6 @@ -/* SPDX-License-Identifier: GPL-2.0-only */ +/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */ /* - * Copyright (c) Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/mem.h b/priv/zstd/contrib/linux-kernel/mem.h index 4b5db57..a723182 100644 --- a/priv/zstd/contrib/linux-kernel/mem.h +++ b/priv/zstd/contrib/linux-kernel/mem.h @@ -1,6 +1,6 @@ -/* SPDX-License-Identifier: GPL-2.0-only */ +/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */ /* - * Copyright (c) Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -30,6 +30,8 @@ * Basic Types *****************************************************************/ typedef uint8_t BYTE; +typedef uint8_t U8; +typedef int8_t S8; typedef uint16_t U16; typedef int16_t S16; typedef uint32_t U32; diff --git a/priv/zstd/contrib/linux-kernel/test/Makefile b/priv/zstd/contrib/linux-kernel/test/Makefile index 2908839..67b55e6 100644 --- a/priv/zstd/contrib/linux-kernel/test/Makefile +++ b/priv/zstd/contrib/linux-kernel/test/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -18,9 +18,13 @@ CPPFLAGS += -DZSTD_ASAN_DONT_POISON_WORKSPACE LINUX_ZSTD_MODULE := $(wildcard $(LINUX_ZSTDLIB)/*.c) LINUX_ZSTD_COMMON := $(wildcard $(LINUX_ZSTDLIB)/common/*.c) LINUX_ZSTD_COMPRESS := $(wildcard $(LINUX_ZSTDLIB)/compress/*.c) -LINUX_ZSTD_DECOMPRESS := $(wildcard $(LINUX_ZSTDLIB)/decompress/*.c) +LINUX_ZSTD_DECOMPRESS := $(wildcard $(LINUX_ZSTDLIB)/decompress/*.c $(LINUX_ZSTDLIB)/decompress/*.S) LINUX_ZSTD_FILES := $(LINUX_ZSTD_MODULE) $(LINUX_ZSTD_COMMON) $(LINUX_ZSTD_COMPRESS) $(LINUX_ZSTD_DECOMPRESS) -LINUX_ZSTD_OBJECTS := $(LINUX_ZSTD_FILES:.c=.o) +LINUX_ZSTD_OBJECTS0 := $(LINUX_ZSTD_FILES:.c=.o) +LINUX_ZSTD_OBJECTS := $(LINUX_ZSTD_OBJECTS0:.S=.o) + +%.o: %.S + $(COMPILE.S) $(OUTPUT_OPTION) $< liblinuxzstd.a: $(LINUX_ZSTD_OBJECTS) $(AR) $(ARFLAGS) $@ $^ @@ -41,4 +45,5 @@ clean: $(RM) -f $(LINUX_ZSTDLIB)/*.o $(RM) -f $(LINUX_ZSTDLIB)/**/*.o $(RM) -f *.o *.a + $(RM) -f static_test $(RM) -f test diff --git a/priv/zstd/contrib/linux-kernel/test/include/asm/unaligned.h b/priv/zstd/contrib/linux-kernel/test/include/asm/unaligned.h index 02c2d74..86ec4ca 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/asm/unaligned.h +++ b/priv/zstd/contrib/linux-kernel/test/include/asm/unaligned.h @@ -20,6 +20,7 @@ static unsigned _isLittleEndian(void) { const union { uint32_t u; uint8_t c[4]; } one = { 1 }; assert(_IS_LITTLE_ENDIAN == one.c[0]); + (void)one; return _IS_LITTLE_ENDIAN; } diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/compiler.h b/priv/zstd/contrib/linux-kernel/test/include/linux/compiler.h index ea3422e..988ce4a 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/compiler.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/compiler.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -18,4 +18,6 @@ #define noinline __attribute__((noinline)) #endif +#define fallthrough __attribute__((__fallthrough__)) + #endif diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/errno.h b/priv/zstd/contrib/linux-kernel/test/include/linux/errno.h index b247522..b4bdcba 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/errno.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/errno.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/kernel.h b/priv/zstd/contrib/linux-kernel/test/include/linux/kernel.h index 1f702ab..a4d791c 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/kernel.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/kernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/limits.h b/priv/zstd/contrib/linux-kernel/test/include/linux/limits.h index db9c099..574aa7b 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/limits.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/limits.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/math64.h b/priv/zstd/contrib/linux-kernel/test/include/linux/math64.h index 8eefa2d..7f6713e 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/math64.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/math64.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/module.h b/priv/zstd/contrib/linux-kernel/test/include/linux/module.h index be6d20d..06ef56f 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/module.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/module.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -12,6 +12,8 @@ #define EXPORT_SYMBOL(symbol) \ void* __##symbol = symbol +#define EXPORT_SYMBOL_GPL(symbol) \ + void* __##symbol = symbol #define MODULE_LICENSE(license) #define MODULE_DESCRIPTION(description) diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/printk.h b/priv/zstd/contrib/linux-kernel/test/include/linux/printk.h index eab08e0..92a2527 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/printk.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/printk.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/stddef.h b/priv/zstd/contrib/linux-kernel/test/include/linux/stddef.h index 8538eb3..15c7408 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/stddef.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/stddef.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/swab.h b/priv/zstd/contrib/linux-kernel/test/include/linux/swab.h index 783046b..2b48b43 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/swab.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/swab.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/types.h b/priv/zstd/contrib/linux-kernel/test/include/linux/types.h index 459a457..b413db6 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/types.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/types.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/test/include/linux/xxhash.h b/priv/zstd/contrib/linux-kernel/test/include/linux/xxhash.h index 0a43bb2..d41cbd9 100644 --- a/priv/zstd/contrib/linux-kernel/test/include/linux/xxhash.h +++ b/priv/zstd/contrib/linux-kernel/test/include/linux/xxhash.h @@ -2,7 +2,7 @@ * xxHash - Extremely Fast Hash algorithm * Copyright (C) 2012-2016, Yann Collet. * - * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php) + * BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php) * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are @@ -124,11 +124,10 @@ XXH_API uint64_t xxh64(const void *input, size_t length, uint64_t seed); static inline unsigned long xxhash(const void *input, size_t length, uint64_t seed) { -#if BITS_PER_LONG == 64 - return xxh64(input, length, seed); -#else - return xxh32(input, length, seed); -#endif + if (sizeof(size_t) == 8) + return xxh64(input, length, seed); + else + return xxh32(input, length, seed); } /*-**************************** @@ -261,7 +260,7 @@ XXH_API void xxh64_copy_state(struct xxh64_state *dst, const struct xxh64_state * xxHash - Extremely Fast Hash algorithm * Copyright (C) 2012-2016, Yann Collet. * - * BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php) + * BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php) * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are diff --git a/priv/zstd/contrib/linux-kernel/test/macro-test.sh b/priv/zstd/contrib/linux-kernel/test/macro-test.sh index c688ac0..9ea84aa 100755 --- a/priv/zstd/contrib/linux-kernel/test/macro-test.sh +++ b/priv/zstd/contrib/linux-kernel/test/macro-test.sh @@ -36,9 +36,9 @@ test_not_present "ZSTD_NO_INTRINSICS" test_not_present "ZSTD_NO_UNUSED_FUNCTIONS" test_not_present "ZSTD_LEGACY_SUPPORT" test_not_present "STATIC_BMI2" -test_not_present "ZSTD_NO_INLINE" test_not_present "ZSTD_DLL_EXPORT" test_not_present "ZSTD_DLL_IMPORT" test_not_present "__ICCARM__" test_not_present "_MSC_VER" test_not_present "_WIN32" +test_not_present "__linux__" diff --git a/priv/zstd/contrib/linux-kernel/test/static_test.c b/priv/zstd/contrib/linux-kernel/test/static_test.c index 50c594c..ba4a420 100644 --- a/priv/zstd/contrib/linux-kernel/test/static_test.c +++ b/priv/zstd/contrib/linux-kernel/test/static_test.c @@ -1,5 +1,5 @@ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -28,17 +28,19 @@ static const char kEmptyZstdFrame[] = { 0x28, 0xb5, 0x2f, 0xfd, 0x24, 0x00, 0x01, 0x00, 0x00, 0x99, 0xe9, 0xd8, 0x51 }; -static void test_decompress_unzstd() { +static void test_decompress_unzstd(void) { fprintf(stderr, "Testing decompress unzstd... "); { size_t const wkspSize = zstd_dctx_workspace_bound(); void* wksp = malloc(wkspSize); - CONTROL(wksp != NULL); ZSTD_DCtx* dctx = zstd_init_dctx(wksp, wkspSize); + CONTROL(wksp != NULL); CONTROL(dctx != NULL); - size_t const dSize = zstd_decompress_dctx(dctx, NULL, 0, kEmptyZstdFrame, sizeof(kEmptyZstdFrame)); - CONTROL(!zstd_is_error(dSize)); - CONTROL(dSize == 0); + { + size_t const dSize = zstd_decompress_dctx(dctx, NULL, 0, kEmptyZstdFrame, sizeof(kEmptyZstdFrame)); + CONTROL(!zstd_is_error(dSize)); + CONTROL(dSize == 0); + } free(wksp); } fprintf(stderr, "Ok\n"); diff --git a/priv/zstd/contrib/linux-kernel/test/test.c b/priv/zstd/contrib/linux-kernel/test/test.c index 9064be7..0f4ba3f 100644 --- a/priv/zstd/contrib/linux-kernel/test/test.c +++ b/priv/zstd/contrib/linux-kernel/test/test.c @@ -1,5 +1,5 @@ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -30,15 +30,15 @@ typedef struct { size_t compSize; } test_data_t; -test_data_t create_test_data(void) { +static test_data_t create_test_data(void) { test_data_t data; data.dataSize = 128 * 1024; - data.data = malloc(data.dataSize); + data.data = (char*)malloc(data.dataSize); CONTROL(data.data != NULL); - data.data2 = malloc(data.dataSize); + data.data2 = (char*)malloc(data.dataSize); CONTROL(data.data2 != NULL); data.compSize = zstd_compress_bound(data.dataSize); - data.comp = malloc(data.compSize); + data.comp = (char*)malloc(data.compSize); CONTROL(data.comp != NULL); memset(data.data, 0, data.dataSize); return data; @@ -54,26 +54,27 @@ static void free_test_data(test_data_t const *data) { #define MAX(a, b) ((a) > (b) ? (a) : (b)) static void test_btrfs(test_data_t const *data) { - fprintf(stderr, "testing btrfs use cases... "); size_t const size = MIN(data->dataSize, 128 * 1024); + fprintf(stderr, "testing btrfs use cases... "); for (int level = -1; level < 16; ++level) { zstd_parameters params = zstd_get_params(level, size); - CONTROL(params.cParams.windowLog <= 17); size_t const workspaceSize = MAX(zstd_cstream_workspace_bound(¶ms.cParams), zstd_dstream_workspace_bound(size)); void *workspace = malloc(workspaceSize); - CONTROL(workspace != NULL); char const *ip = data->data; char const *iend = ip + size; char *op = data->comp; char *oend = op + data->compSize; + + CONTROL(params.cParams.windowLog <= 17); + CONTROL(workspace != NULL); { zstd_cstream *cctx = zstd_init_cstream(¶ms, size, workspace, workspaceSize); - CONTROL(cctx != NULL); zstd_out_buffer out = {NULL, 0, 0}; zstd_in_buffer in = {NULL, 0, 0}; + CONTROL(cctx != NULL); for (;;) { if (in.pos == in.size) { in.src = ip; @@ -108,9 +109,9 @@ static void test_btrfs(test_data_t const *data) { oend = op + size; { zstd_dstream *dctx = zstd_init_dstream(1ULL << params.cParams.windowLog, workspace, workspaceSize); - CONTROL(dctx != NULL); zstd_out_buffer out = {NULL, 0, 0}; zstd_in_buffer in = {NULL, 0, 0}; + CONTROL(dctx != NULL); for (;;) { if (in.pos == in.size) { in.src = ip; @@ -125,15 +126,16 @@ static void test_btrfs(test_data_t const *data) { out.pos = 0; op += out.size; } - - size_t const ret = zstd_decompress_stream(dctx, &out, &in); - CONTROL(!zstd_is_error(ret)); - if (ret == 0) { - break; + { + size_t const ret = zstd_decompress_stream(dctx, &out, &in); + CONTROL(!zstd_is_error(ret)); + if (ret == 0) { + break; + } } } } - CONTROL(op - data->data2 == data->dataSize); + CONTROL((size_t)(op - data->data2) == data->dataSize); CONTROL(!memcmp(data->data, data->data2, data->dataSize)); free(workspace); } @@ -141,14 +143,14 @@ static void test_btrfs(test_data_t const *data) { } static void test_decompress_unzstd(test_data_t const *data) { - fprintf(stderr, "Testing decompress unzstd... "); size_t cSize; + fprintf(stderr, "Testing decompress unzstd... "); { zstd_parameters params = zstd_get_params(19, 0); size_t const wkspSize = zstd_cctx_workspace_bound(¶ms.cParams); void* wksp = malloc(wkspSize); - CONTROL(wksp != NULL); zstd_cctx* cctx = zstd_init_cctx(wksp, wkspSize); + CONTROL(wksp != NULL); CONTROL(cctx != NULL); cSize = zstd_compress_cctx(cctx, data->comp, data->compSize, data->data, data->dataSize, ¶ms); CONTROL(!zstd_is_error(cSize)); @@ -157,19 +159,21 @@ static void test_decompress_unzstd(test_data_t const *data) { { size_t const wkspSize = zstd_dctx_workspace_bound(); void* wksp = malloc(wkspSize); - CONTROL(wksp != NULL); zstd_dctx* dctx = zstd_init_dctx(wksp, wkspSize); + CONTROL(wksp != NULL); CONTROL(dctx != NULL); - size_t const dSize = zstd_decompress_dctx(dctx, data->data2, data->dataSize, data->comp, cSize); - CONTROL(!zstd_is_error(dSize)); - CONTROL(dSize == data->dataSize); + { + size_t const dSize = zstd_decompress_dctx(dctx, data->data2, data->dataSize, data->comp, cSize); + CONTROL(!zstd_is_error(dSize)); + CONTROL(dSize == data->dataSize); + } CONTROL(!memcmp(data->data, data->data2, data->dataSize)); free(wksp); } fprintf(stderr, "Ok\n"); } -static void test_f2fs() { +static void test_f2fs(void) { fprintf(stderr, "testing f2fs uses... "); CONTROL(zstd_min_clevel() < 0); CONTROL(zstd_max_clevel() == 22); @@ -182,26 +186,32 @@ static void __attribute__((noinline)) use(void *x) { asm volatile("" : "+r"(x)); } -static void __attribute__((noinline)) set_stack() { +static void __attribute__((noinline)) fill_stack(void) { + memset(g_stack, 0x33, 8192); +} + +static void __attribute__((noinline)) set_stack(void) { char stack[8192]; g_stack = stack; - memset(g_stack, 0x33, 8192); use(g_stack); } -static void __attribute__((noinline)) check_stack() { +static void __attribute__((noinline)) check_stack(void) { size_t cleanStack = 0; while (cleanStack < 8192 && g_stack[cleanStack] == 0x33) { ++cleanStack; } - size_t const stackSize = 8192 - cleanStack; - fprintf(stderr, "Maximum stack size: %zu\n", stackSize); - CONTROL(stackSize <= 2048 + 512); + { + size_t const stackSize = 8192 - cleanStack; + fprintf(stderr, "Maximum stack size: %zu\n", stackSize); + CONTROL(stackSize <= 2048 + 512); + } } static void test_stack_usage(test_data_t const *data) { set_stack(); + fill_stack(); test_f2fs(); test_btrfs(data); test_decompress_unzstd(data); diff --git a/priv/zstd/contrib/linux-kernel/zstd_common_module.c b/priv/zstd/contrib/linux-kernel/zstd_common_module.c new file mode 100644 index 0000000..2fead39 --- /dev/null +++ b/priv/zstd/contrib/linux-kernel/zstd_common_module.c @@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause +/* + * Copyright (c) Meta Platforms, Inc. and affiliates. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + * You may select, at your option, one of the above-listed licenses. + */ + +#include + +#include "common/huf.h" +#include "common/fse.h" +#include "common/zstd_internal.h" + +// Export symbols shared by compress and decompress into a common module + +#undef ZSTD_isError /* defined within zstd_internal.h */ +EXPORT_SYMBOL_GPL(FSE_readNCount); +EXPORT_SYMBOL_GPL(HUF_readStats); +EXPORT_SYMBOL_GPL(HUF_readStats_wksp); +EXPORT_SYMBOL_GPL(ZSTD_isError); +EXPORT_SYMBOL_GPL(ZSTD_getErrorName); +EXPORT_SYMBOL_GPL(ZSTD_getErrorCode); +EXPORT_SYMBOL_GPL(ZSTD_customMalloc); +EXPORT_SYMBOL_GPL(ZSTD_customCalloc); +EXPORT_SYMBOL_GPL(ZSTD_customFree); + +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_DESCRIPTION("Zstd Common"); diff --git a/priv/zstd/contrib/linux-kernel/zstd_compress_module.c b/priv/zstd/contrib/linux-kernel/zstd_compress_module.c index 37d08ff..8ecf432 100644 --- a/priv/zstd/contrib/linux-kernel/zstd_compress_module.c +++ b/priv/zstd/contrib/linux-kernel/zstd_compress_module.c @@ -1,6 +1,6 @@ -// SPDX-License-Identifier: GPL-2.0-only +// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -17,6 +17,43 @@ #include "common/zstd_deps.h" #include "common/zstd_internal.h" +#define ZSTD_FORWARD_IF_ERR(ret) \ + do { \ + size_t const __ret = (ret); \ + if (ZSTD_isError(__ret)) \ + return __ret; \ + } while (0) + +static size_t zstd_cctx_init(zstd_cctx *cctx, const zstd_parameters *parameters, + unsigned long long pledged_src_size) +{ + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_reset( + cctx, ZSTD_reset_session_and_parameters)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setPledgedSrcSize( + cctx, pledged_src_size)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_windowLog, parameters->cParams.windowLog)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_hashLog, parameters->cParams.hashLog)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_chainLog, parameters->cParams.chainLog)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_searchLog, parameters->cParams.searchLog)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_minMatch, parameters->cParams.minMatch)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_targetLength, parameters->cParams.targetLength)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_strategy, parameters->cParams.strategy)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_contentSizeFlag, parameters->fParams.contentSizeFlag)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_checksumFlag, parameters->fParams.checksumFlag)); + ZSTD_FORWARD_IF_ERR(ZSTD_CCtx_setParameter( + cctx, ZSTD_c_dictIDFlag, !parameters->fParams.noDictIDFlag)); + return 0; +} + int zstd_min_clevel(void) { return ZSTD_minCLevel(); @@ -59,7 +96,8 @@ EXPORT_SYMBOL(zstd_init_cctx); size_t zstd_compress_cctx(zstd_cctx *cctx, void *dst, size_t dst_capacity, const void *src, size_t src_size, const zstd_parameters *parameters) { - return ZSTD_compress_advanced(cctx, dst, dst_capacity, src, src_size, NULL, 0, *parameters); + ZSTD_FORWARD_IF_ERR(zstd_cctx_init(cctx, parameters, src_size)); + return ZSTD_compress2(cctx, dst, dst_capacity, src, src_size); } EXPORT_SYMBOL(zstd_compress_cctx); @@ -73,7 +111,6 @@ zstd_cstream *zstd_init_cstream(const zstd_parameters *parameters, unsigned long long pledged_src_size, void *workspace, size_t workspace_size) { zstd_cstream *cstream; - size_t ret; if (workspace == NULL) return NULL; @@ -86,8 +123,7 @@ zstd_cstream *zstd_init_cstream(const zstd_parameters *parameters, if (pledged_src_size == 0) pledged_src_size = ZSTD_CONTENTSIZE_UNKNOWN; - ret = ZSTD_initCStream_advanced(cstream, NULL, 0, *parameters, pledged_src_size); - if (ZSTD_isError(ret)) + if (ZSTD_isError(zstd_cctx_init(cstream, parameters, pledged_src_size))) return NULL; return cstream; @@ -97,7 +133,11 @@ EXPORT_SYMBOL(zstd_init_cstream); size_t zstd_reset_cstream(zstd_cstream *cstream, unsigned long long pledged_src_size) { - return ZSTD_resetCStream(cstream, pledged_src_size); + if (pledged_src_size == 0) + pledged_src_size = ZSTD_CONTENTSIZE_UNKNOWN; + ZSTD_FORWARD_IF_ERR( ZSTD_CCtx_reset(cstream, ZSTD_reset_session_only) ); + ZSTD_FORWARD_IF_ERR( ZSTD_CCtx_setPledgedSrcSize(cstream, pledged_src_size) ); + return 0; } EXPORT_SYMBOL(zstd_reset_cstream); diff --git a/priv/zstd/contrib/linux-kernel/zstd_decompress_module.c b/priv/zstd/contrib/linux-kernel/zstd_decompress_module.c index 15005cd..eb1c49e 100644 --- a/priv/zstd/contrib/linux-kernel/zstd_decompress_module.c +++ b/priv/zstd/contrib/linux-kernel/zstd_decompress_module.c @@ -1,6 +1,6 @@ -// SPDX-License-Identifier: GPL-2.0-only +// SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/linux-kernel/zstd_deps.h b/priv/zstd/contrib/linux-kernel/zstd_deps.h index 853b724..9251614 100644 --- a/priv/zstd/contrib/linux-kernel/zstd_deps.h +++ b/priv/zstd/contrib/linux-kernel/zstd_deps.h @@ -1,6 +1,6 @@ -/* SPDX-License-Identifier: GPL-2.0-only */ +/* SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause */ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/match_finders/README.md b/priv/zstd/contrib/match_finders/README.md index 0f4a3b1..54055c3 100644 --- a/priv/zstd/contrib/match_finders/README.md +++ b/priv/zstd/contrib/match_finders/README.md @@ -14,7 +14,7 @@ * files would be small relative to the size of the file. * * Various 'diffing' algorithms utilize this notion of edit distance and - * the corrensponding concept of a minimal edit script between two + * the corresponding concept of a minimal edit script between two * sequences to identify the regions within two files where they differ. * The core algorithm used in this match finder is described in: * @@ -28,12 +28,12 @@ * * Note: after some experimentation, this approach proved to not provide enough * utility to justify the additional CPU used in finding matches. The one area - * where this approach consistenly outperforms Zstandard even on level 19 is - * when compressing small files (<10 KB) using a equally small dictionary that + * where this approach consistently outperforms Zstandard even on level 19 is + * when compressing small files (<10 KB) using an equally small dictionary that * is very similar to the source file. For the use case that this was intended, * (large similar files) this approach by itself took 5-10X longer than zstd-19 and * generally resulted in 2-3X larger files. The core advantage that zstd-19 has - * over this appraoch for match finding is the overlapping matches. This approach + * over this approach for match finding is the overlapping matches. This approach * cannot find any. * * I'm leaving this in the contrib section in case this ever becomes interesting diff --git a/priv/zstd/contrib/match_finders/zstd_edist.c b/priv/zstd/contrib/match_finders/zstd_edist.c index aab545f..d685cdd 100644 --- a/priv/zstd/contrib/match_finders/zstd_edist.c +++ b/priv/zstd/contrib/match_finders/zstd_edist.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -12,7 +12,7 @@ * Dependencies ***************************************/ -/* Currently relies on qsort when combining contiguous matches. This can probabily +/* Currently relies on qsort when combining contiguous matches. This can probably * be avoided but would require changes to the algorithm. The qsort is far from * the bottleneck in this algorithm even for medium sized files so it's probably * not worth trying to address */ @@ -26,7 +26,7 @@ * Constants ***************************************/ -/* Just a sential for the entires of the diagnomal matrix */ +/* Just a sential for the entries of the diagonal matrix */ #define ZSTD_EDIST_DIAG_MAX (S32)(1 << 30) /* How large should a snake be to be considered a 'big' snake. @@ -39,7 +39,7 @@ #define ZSTD_EDIST_SNAKE_ITER_THRESH 200 /* After how many iterations should be just give up and take - * the best availabe edit script for this round */ + * the best available edit script for this round */ #define ZSTD_EDIST_EXPENSIVE_THRESH 1024 /*-************************************* @@ -57,10 +57,10 @@ typedef struct { const BYTE* src; size_t dictSize; size_t srcSize; - S32* forwardDiag; /* Entires of the forward diagonal stored here */ - S32* backwardDiag; /* Entires of the backward diagonal stored here. + S32* forwardDiag; /* Entries of the forward diagonal stored here */ + S32* backwardDiag; /* Entries of the backward diagonal stored here. * Note: this buffer and the 'forwardDiag' buffer - * are contiguous. See the ZSTD_eDist_genSequences */ + * are contiguous. See the ZSTD_eDist_genSequences */ ZSTD_eDist_match* matches; /* Accumulate matches of length 1 in this buffer. * In a subsequence post-processing step, we combine * contiguous matches. */ @@ -192,7 +192,7 @@ static void ZSTD_eDist_diag(ZSTD_eDist_state* state, if (!useHeuristics) continue; - /* Everything under this point is a heuritic. Using these will + /* Everything under this point is a heuristic. Using these will * substantially speed up the match finding. In some cases, taking * the total match finding time from several minutes to seconds. * Of course, the caveat is that the edit script found may no longer @@ -366,8 +366,8 @@ static int ZSTD_eDist_compare(ZSTD_eDist_state* state, } } else if (srcLow == srcHigh) { while (dictLow < dictHigh) { - /* Reaching this point means deleteing dict[dictLow] from - * the current positino of dict */ + /* Reaching this point means deleting dict[dictLow] from + * the current position of dict */ dictLow++; } } else { @@ -395,7 +395,7 @@ static int ZSTD_eDist_matchComp(const void* p, const void* q) } /* The matches from the approach above will all be of the form - * (dictIdx, srcIdx, 1). this method combines contiguous matches + * (dictIdx, srcIdx, 1). This method combines contiguous matches * of length MINMATCH or greater. Matches less than MINMATCH * are discarded */ static void ZSTD_eDist_combineMatches(ZSTD_eDist_state* state) @@ -460,7 +460,7 @@ static size_t ZSTD_eDist_convertMatchesToSequences(ZSTD_Sequence* sequences, } /*-************************************* -* Interal utils +* Internal utils ***************************************/ static size_t ZSTD_eDist_hamingDist(const BYTE* const a, diff --git a/priv/zstd/contrib/match_finders/zstd_edist.h b/priv/zstd/contrib/match_finders/zstd_edist.h index c775a49..c739e2a 100644 --- a/priv/zstd/contrib/match_finders/zstd_edist.h +++ b/priv/zstd/contrib/match_finders/zstd_edist.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Yann Collet, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -21,7 +21,7 @@ * files would be small relative to the size of the file. * * Various 'diffing' algorithms utilize this notion of edit distance and - * the corrensponding concept of a minimal edit script between two + * the corresponding concept of a minimal edit script between two * sequences to identify the regions within two files where they differ. * The core algorithm used in this match finder is described in: * @@ -35,12 +35,12 @@ * * Note: after some experimentation, this approach proved to not provide enough * utility to justify the additional CPU used in finding matches. The one area - * where this approach consistenly outperforms Zstandard even on level 19 is - * when compressing small files (<10 KB) using a equally small dictionary that + * where this approach consistently outperforms Zstandard even on level 19 is + * when compressing small files (<10 KB) using an equally small dictionary that * is very similar to the source file. For the use case that this was intended, * (large similar files) this approach by itself took 5-10X longer than zstd-19 and * generally resulted in 2-3X larger files. The core advantage that zstd-19 has - * over this appraoch for match finding is the overlapping matches. This approach + * over this approach for match finding is the overlapping matches. This approach * cannot find any. * * I'm leaving this in the contrib section in case this ever becomes interesting diff --git a/priv/zstd/contrib/pzstd/ErrorHolder.h b/priv/zstd/contrib/pzstd/ErrorHolder.h index 829651c..2c2797e 100644 --- a/priv/zstd/contrib/pzstd/ErrorHolder.h +++ b/priv/zstd/contrib/pzstd/ErrorHolder.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/Logging.h b/priv/zstd/contrib/pzstd/Logging.h index beb160b..84a08d2 100644 --- a/priv/zstd/contrib/pzstd/Logging.h +++ b/priv/zstd/contrib/pzstd/Logging.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/Makefile b/priv/zstd/contrib/pzstd/Makefile index 25265e7..830053c 100644 --- a/priv/zstd/contrib/pzstd/Makefile +++ b/priv/zstd/contrib/pzstd/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2016-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the @@ -57,19 +57,6 @@ LD_COMMAND = $(CXX) $^ $(ALL_LDFLAGS) $(LIBS) -pthread -o $@ CC_COMMAND = $(CC) $(DEPFLAGS) $(ALL_CFLAGS) -c $< -o $@ CXX_COMMAND = $(CXX) $(DEPFLAGS) $(ALL_CXXFLAGS) -c $< -o $@ -# Get a list of all zstd files so we rebuild the static library when we need to -ZSTDCOMMON_FILES := $(wildcard $(ZSTDDIR)/common/*.c) \ - $(wildcard $(ZSTDDIR)/common/*.h) -ZSTDCOMP_FILES := $(wildcard $(ZSTDDIR)/compress/*.c) \ - $(wildcard $(ZSTDDIR)/compress/*.h) -ZSTDDECOMP_FILES := $(wildcard $(ZSTDDIR)/decompress/*.c) \ - $(wildcard $(ZSTDDIR)/decompress/*.h) -ZSTDPROG_FILES := $(wildcard $(PROGDIR)/*.c) \ - $(wildcard $(PROGDIR)/*.h) -ZSTD_FILES := $(wildcard $(ZSTDDIR)/*.h) \ - $(ZSTDDECOMP_FILES) $(ZSTDCOMMON_FILES) $(ZSTDCOMP_FILES) \ - $(ZSTDPROG_FILES) - # List all the pzstd source files so we can determine their dependencies PZSTD_SRCS := $(wildcard *.cpp) PZSTD_TESTS := $(wildcard test/*.cpp) @@ -189,7 +176,8 @@ roundtrip: test/RoundTripTest$(EXT) # Use the static library that zstd builds for simplicity and # so we get the compiler options correct -$(ZSTDDIR)/libzstd.a: $(ZSTD_FILES) +.PHONY: $(ZSTDDIR)/libzstd.a +$(ZSTDDIR)/libzstd.a: CFLAGS="$(ALL_CFLAGS)" LDFLAGS="$(ALL_LDFLAGS)" $(MAKE) -C $(ZSTDDIR) libzstd.a # Rules to build the tests diff --git a/priv/zstd/contrib/pzstd/Options.cpp b/priv/zstd/contrib/pzstd/Options.cpp index 3729222..a77824e 100644 --- a/priv/zstd/contrib/pzstd/Options.cpp +++ b/priv/zstd/contrib/pzstd/Options.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -87,7 +87,7 @@ void usage() { std::fprintf(stderr, " -V, --version : display version number and exit\n"); std::fprintf(stderr, " -v, --verbose : verbose mode; specify multiple times to increase log level (default:2)\n"); std::fprintf(stderr, " -q, --quiet : suppress warnings; specify twice to suppress errors too\n"); - std::fprintf(stderr, " -c, --stdout : force write to standard output, even if it is the console\n"); + std::fprintf(stderr, " -c, --stdout : write to standard output (even if it is the console)\n"); #ifdef UTIL_HAS_CREATEFILELIST std::fprintf(stderr, " -r : operate recursively on directories\n"); #endif diff --git a/priv/zstd/contrib/pzstd/Options.h b/priv/zstd/contrib/pzstd/Options.h index 924543a..92c18a3 100644 --- a/priv/zstd/contrib/pzstd/Options.h +++ b/priv/zstd/contrib/pzstd/Options.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/Pzstd.cpp b/priv/zstd/contrib/pzstd/Pzstd.cpp index 2c09bda..67b9419 100644 --- a/priv/zstd/contrib/pzstd/Pzstd.cpp +++ b/priv/zstd/contrib/pzstd/Pzstd.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -10,11 +10,13 @@ #include "Pzstd.h" #include "SkippableFrame.h" #include "utils/FileSystem.h" +#include "utils/Portability.h" #include "utils/Range.h" #include "utils/ScopeGuard.h" #include "utils/ThreadPool.h" #include "utils/WorkQueue.h" +#include #include #include #include @@ -336,6 +338,10 @@ static size_t calculateStep( const ZSTD_parameters ¶ms) { (void)size; (void)numThreads; + // Not validated to work correctly for window logs > 23. + // It will definitely fail if windowLog + 2 is >= 4GB because + // the skippable frame can only store sizes up to 4GB. + assert(params.cParams.windowLog <= 23); return size_t{1} << (params.cParams.windowLog + 2); } @@ -587,7 +593,8 @@ std::uint64_t writeFile( // start writing before compression is done because we need to know the // compressed size. // Wait for the compressed size to be available and write skippable frame - SkippableFrame frame(out->size()); + assert(uint64_t(out->size()) < uint64_t(1) << 32); + SkippableFrame frame(uint32_t(out->size())); if (!writeData(frame.data(), outputFd)) { errorHolder.setError("Failed to write output"); return bytesWritten; diff --git a/priv/zstd/contrib/pzstd/Pzstd.h b/priv/zstd/contrib/pzstd/Pzstd.h index c667c88..3645e59 100644 --- a/priv/zstd/contrib/pzstd/Pzstd.h +++ b/priv/zstd/contrib/pzstd/Pzstd.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/README.md b/priv/zstd/contrib/pzstd/README.md index 84d9458..bc8f831 100644 --- a/priv/zstd/contrib/pzstd/README.md +++ b/priv/zstd/contrib/pzstd/README.md @@ -31,7 +31,7 @@ If this number is not suitable, during compilation you can define `PZSTD_NUM_THR ## Benchmarks -As a reference, PZstandard and Pigz were compared on an Intel Core i7 @ 3.1 GHz, each using 4 threads, with the [Silesia compression corpus](http://sun.aei.polsl.pl/~sdeor/index.php?page=silesia). +As a reference, PZstandard and Pigz were compared on an Intel Core i7 @ 3.1 GHz, each using 4 threads, with the [Silesia compression corpus](https://sun.aei.polsl.pl//~sdeor/index.php?page=silesia). Compression Speed vs Ratio with 4 Threads | Decompression Speed with 4 Threads ------------------------------------------|----------------------------------- diff --git a/priv/zstd/contrib/pzstd/SkippableFrame.cpp b/priv/zstd/contrib/pzstd/SkippableFrame.cpp index 769866d..3bea4eb 100644 --- a/priv/zstd/contrib/pzstd/SkippableFrame.cpp +++ b/priv/zstd/contrib/pzstd/SkippableFrame.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/SkippableFrame.h b/priv/zstd/contrib/pzstd/SkippableFrame.h index 60deed0..817415e 100644 --- a/priv/zstd/contrib/pzstd/SkippableFrame.h +++ b/priv/zstd/contrib/pzstd/SkippableFrame.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/main.cpp b/priv/zstd/contrib/pzstd/main.cpp index b93f043..422b4a5 100644 --- a/priv/zstd/contrib/pzstd/main.cpp +++ b/priv/zstd/contrib/pzstd/main.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/test/OptionsTest.cpp b/priv/zstd/contrib/pzstd/test/OptionsTest.cpp index e601148..91e3975 100644 --- a/priv/zstd/contrib/pzstd/test/OptionsTest.cpp +++ b/priv/zstd/contrib/pzstd/test/OptionsTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/test/PzstdTest.cpp b/priv/zstd/contrib/pzstd/test/PzstdTest.cpp index 5c7d663..75453f5 100644 --- a/priv/zstd/contrib/pzstd/test/PzstdTest.cpp +++ b/priv/zstd/contrib/pzstd/test/PzstdTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/test/RoundTrip.h b/priv/zstd/contrib/pzstd/test/RoundTrip.h index c6364ec..f777622 100644 --- a/priv/zstd/contrib/pzstd/test/RoundTrip.h +++ b/priv/zstd/contrib/pzstd/test/RoundTrip.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/test/RoundTripTest.cpp b/priv/zstd/contrib/pzstd/test/RoundTripTest.cpp index 36af067..c37646d 100644 --- a/priv/zstd/contrib/pzstd/test/RoundTripTest.cpp +++ b/priv/zstd/contrib/pzstd/test/RoundTripTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/Buffer.h b/priv/zstd/contrib/pzstd/utils/Buffer.h index f69c3b4..a85f770 100644 --- a/priv/zstd/contrib/pzstd/utils/Buffer.h +++ b/priv/zstd/contrib/pzstd/utils/Buffer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -45,7 +45,7 @@ class Buffer { : buffer_(buffer), range_(data) {} Buffer(Buffer&&) = default; - Buffer& operator=(Buffer&&) & = default; + Buffer& operator=(Buffer&&) = default; /** * Splits the data into two pieces: [begin, begin + n), [begin + n, end). diff --git a/priv/zstd/contrib/pzstd/utils/FileSystem.h b/priv/zstd/contrib/pzstd/utils/FileSystem.h index 3cfbe86..8d57d05 100644 --- a/priv/zstd/contrib/pzstd/utils/FileSystem.h +++ b/priv/zstd/contrib/pzstd/utils/FileSystem.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -8,16 +8,18 @@ */ #pragma once +#include "utils/Portability.h" #include "utils/Range.h" #include #include #include +#include #include // A small subset of `std::filesystem`. // `std::filesystem` should be a drop in replacement. -// See http://en.cppreference.com/w/cpp/filesystem for documentation. +// See https://en.cppreference.com/w/cpp/filesystem for documentation. namespace pzstd { @@ -28,7 +30,7 @@ typedef struct ::_stat64 file_status; typedef struct ::stat file_status; #endif -/// http://en.cppreference.com/w/cpp/filesystem/status +/// https://en.cppreference.com/w/cpp/filesystem/status inline file_status status(StringPiece path, std::error_code& ec) noexcept { file_status status; #if defined(_MSC_VER) @@ -44,7 +46,7 @@ inline file_status status(StringPiece path, std::error_code& ec) noexcept { return status; } -/// http://en.cppreference.com/w/cpp/filesystem/is_regular_file +/// https://en.cppreference.com/w/cpp/filesystem/is_regular_file inline bool is_regular_file(file_status status) noexcept { #if defined(S_ISREG) return S_ISREG(status.st_mode); @@ -55,12 +57,12 @@ inline bool is_regular_file(file_status status) noexcept { #endif } -/// http://en.cppreference.com/w/cpp/filesystem/is_regular_file +/// https://en.cppreference.com/w/cpp/filesystem/is_regular_file inline bool is_regular_file(StringPiece path, std::error_code& ec) noexcept { return is_regular_file(status(path, ec)); } -/// http://en.cppreference.com/w/cpp/filesystem/is_directory +/// https://en.cppreference.com/w/cpp/filesystem/is_directory inline bool is_directory(file_status status) noexcept { #if defined(S_ISDIR) return S_ISDIR(status.st_mode); @@ -71,22 +73,22 @@ inline bool is_directory(file_status status) noexcept { #endif } -/// http://en.cppreference.com/w/cpp/filesystem/is_directory +/// https://en.cppreference.com/w/cpp/filesystem/is_directory inline bool is_directory(StringPiece path, std::error_code& ec) noexcept { return is_directory(status(path, ec)); } -/// http://en.cppreference.com/w/cpp/filesystem/file_size +/// https://en.cppreference.com/w/cpp/filesystem/file_size inline std::uintmax_t file_size( StringPiece path, std::error_code& ec) noexcept { auto stat = status(path, ec); if (ec) { - return -1; + return std::numeric_limits::max(); } if (!is_regular_file(stat)) { ec.assign(ENOTSUP, std::generic_category()); - return -1; + return std::numeric_limits::max(); } ec.clear(); return stat.st_size; diff --git a/priv/zstd/contrib/pzstd/utils/Likely.h b/priv/zstd/contrib/pzstd/utils/Likely.h index 7cea8da..52243a6 100644 --- a/priv/zstd/contrib/pzstd/utils/Likely.h +++ b/priv/zstd/contrib/pzstd/utils/Likely.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/Portability.h b/priv/zstd/contrib/pzstd/utils/Portability.h new file mode 100644 index 0000000..ef1f86e --- /dev/null +++ b/priv/zstd/contrib/pzstd/utils/Portability.h @@ -0,0 +1,16 @@ +/* + * Copyright (c) Meta Platforms, Inc. and affiliates. + * All rights reserved. + * + * This source code is licensed under both the BSD-style license (found in the + * LICENSE file in the root directory of this source tree) and the GPLv2 (found + * in the COPYING file in the root directory of this source tree). + */ + +#pragma once + +#include + +// Required for windows, which defines min/max, but we want the std:: version. +#undef min +#undef max diff --git a/priv/zstd/contrib/pzstd/utils/Range.h b/priv/zstd/contrib/pzstd/utils/Range.h index fedb5d7..0fd8f9f 100644 --- a/priv/zstd/contrib/pzstd/utils/Range.h +++ b/priv/zstd/contrib/pzstd/utils/Range.h @@ -1,12 +1,12 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the * LICENSE file in the root directory of this source tree) and the GPLv2 (found * in the COPYING file in the root directory of this source tree). */ - + /** * A subset of `folly/Range.h`. * All code copied verbatim modulo formatting @@ -14,7 +14,9 @@ #pragma once #include "utils/Likely.h" +#include "utils/Portability.h" +#include #include #include #include @@ -83,8 +85,8 @@ class Range { Range(const Range&) = default; Range(Range&&) = default; - Range& operator=(const Range&) & = default; - Range& operator=(Range&&) & = default; + Range& operator=(const Range&) = default; + Range& operator=(Range&&) = default; constexpr size_type size() const { return e_ - b_; diff --git a/priv/zstd/contrib/pzstd/utils/ResourcePool.h b/priv/zstd/contrib/pzstd/utils/ResourcePool.h index 8dfcdd7..7c4bb62 100644 --- a/priv/zstd/contrib/pzstd/utils/ResourcePool.h +++ b/priv/zstd/contrib/pzstd/utils/ResourcePool.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/ScopeGuard.h b/priv/zstd/contrib/pzstd/utils/ScopeGuard.h index 31768f4..911fd98 100644 --- a/priv/zstd/contrib/pzstd/utils/ScopeGuard.h +++ b/priv/zstd/contrib/pzstd/utils/ScopeGuard.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -15,7 +15,7 @@ namespace pzstd { /** * Dismissable scope guard. * `Function` must be callable and take no parameters. - * Unless `dissmiss()` is called, the callable is executed upon destruction of + * Unless `dismiss()` is called, the callable is executed upon destruction of * `ScopeGuard`. * * Example: diff --git a/priv/zstd/contrib/pzstd/utils/ThreadPool.h b/priv/zstd/contrib/pzstd/utils/ThreadPool.h index 8ece8e0..a087d7c 100644 --- a/priv/zstd/contrib/pzstd/utils/ThreadPool.h +++ b/priv/zstd/contrib/pzstd/utils/ThreadPool.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/WorkQueue.h b/priv/zstd/contrib/pzstd/utils/WorkQueue.h index 1d14d92..d7947b8 100644 --- a/priv/zstd/contrib/pzstd/utils/WorkQueue.h +++ b/priv/zstd/contrib/pzstd/utils/WorkQueue.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/test/BufferTest.cpp b/priv/zstd/contrib/pzstd/utils/test/BufferTest.cpp index fbba74e..58bf08d 100644 --- a/priv/zstd/contrib/pzstd/utils/test/BufferTest.cpp +++ b/priv/zstd/contrib/pzstd/utils/test/BufferTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/test/RangeTest.cpp b/priv/zstd/contrib/pzstd/utils/test/RangeTest.cpp index 755b50f..8b7dee2 100644 --- a/priv/zstd/contrib/pzstd/utils/test/RangeTest.cpp +++ b/priv/zstd/contrib/pzstd/utils/test/RangeTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/test/ResourcePoolTest.cpp b/priv/zstd/contrib/pzstd/utils/test/ResourcePoolTest.cpp index 6fe1451..750ee08 100644 --- a/priv/zstd/contrib/pzstd/utils/test/ResourcePoolTest.cpp +++ b/priv/zstd/contrib/pzstd/utils/test/ResourcePoolTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/test/ScopeGuardTest.cpp b/priv/zstd/contrib/pzstd/utils/test/ScopeGuardTest.cpp index 7bc624d..0f77cdf 100644 --- a/priv/zstd/contrib/pzstd/utils/test/ScopeGuardTest.cpp +++ b/priv/zstd/contrib/pzstd/utils/test/ScopeGuardTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/test/ThreadPoolTest.cpp b/priv/zstd/contrib/pzstd/utils/test/ThreadPoolTest.cpp index 703fd4c..a01052e 100644 --- a/priv/zstd/contrib/pzstd/utils/test/ThreadPoolTest.cpp +++ b/priv/zstd/contrib/pzstd/utils/test/ThreadPoolTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/pzstd/utils/test/WorkQueueTest.cpp b/priv/zstd/contrib/pzstd/utils/test/WorkQueueTest.cpp index 14cf773..16600bb 100644 --- a/priv/zstd/contrib/pzstd/utils/test/WorkQueueTest.cpp +++ b/priv/zstd/contrib/pzstd/utils/test/WorkQueueTest.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/recovery/Makefile b/priv/zstd/contrib/recovery/Makefile index 9a9f4f2..be6ea4b 100644 --- a/priv/zstd/contrib/recovery/Makefile +++ b/priv/zstd/contrib/recovery/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2019-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/recovery/recover_directory.c b/priv/zstd/contrib/recovery/recover_directory.c index 13f83fd..b9bd7ab 100644 --- a/priv/zstd/contrib/recovery/recover_directory.c +++ b/priv/zstd/contrib/recovery/recover_directory.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2016-2021, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/examples/Makefile b/priv/zstd/contrib/seekable_format/examples/Makefile index 9df6b75..fcc0458 100644 --- a/priv/zstd/contrib/seekable_format/examples/Makefile +++ b/priv/zstd/contrib/seekable_format/examples/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2017-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/examples/parallel_compression.c b/priv/zstd/contrib/seekable_format/examples/parallel_compression.c index 4118b0a..0ec9fbd 100644 --- a/priv/zstd/contrib/seekable_format/examples/parallel_compression.c +++ b/priv/zstd/contrib/seekable_format/examples/parallel_compression.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/examples/parallel_processing.c b/priv/zstd/contrib/seekable_format/examples/parallel_processing.c index 36226b4..b1709db 100644 --- a/priv/zstd/contrib/seekable_format/examples/parallel_processing.c +++ b/priv/zstd/contrib/seekable_format/examples/parallel_processing.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/examples/seekable_compression.c b/priv/zstd/contrib/seekable_format/examples/seekable_compression.c index 9a331a8..182b46f 100644 --- a/priv/zstd/contrib/seekable_format/examples/seekable_compression.c +++ b/priv/zstd/contrib/seekable_format/examples/seekable_compression.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/examples/seekable_decompression.c b/priv/zstd/contrib/seekable_format/examples/seekable_decompression.c index e9e2013..2c4f3ba 100644 --- a/priv/zstd/contrib/seekable_format/examples/seekable_decompression.c +++ b/priv/zstd/contrib/seekable_format/examples/seekable_decompression.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/examples/seekable_decompression_mem.c b/priv/zstd/contrib/seekable_format/examples/seekable_decompression_mem.c index e7b1c65..44a06fb 100644 --- a/priv/zstd/contrib/seekable_format/examples/seekable_decompression_mem.c +++ b/priv/zstd/contrib/seekable_format/examples/seekable_decompression_mem.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/tests/Makefile b/priv/zstd/contrib/seekable_format/tests/Makefile index d51deb3..a81f222 100644 --- a/priv/zstd/contrib/seekable_format/tests/Makefile +++ b/priv/zstd/contrib/seekable_format/tests/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) 2017-present, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/contrib/seekable_format/tests/seekable_tests.c b/priv/zstd/contrib/seekable_format/tests/seekable_tests.c index a482638..1bb2d0e 100644 --- a/priv/zstd/contrib/seekable_format/tests/seekable_tests.c +++ b/priv/zstd/contrib/seekable_format/tests/seekable_tests.c @@ -186,6 +186,40 @@ int main(int argc, const char** argv) } printf("Success!\n"); + + printf("Test %u - check ZSTD magic in compressing empty string: ", testNb++); + { // compressing empty string should return a zstd header + size_t const capacity = 255; + char* inBuffer = malloc(capacity); + assert(inBuffer != NULL); + inBuffer[0] = '\0'; + void* const outBuffer = malloc(capacity); + assert(outBuffer != NULL); + + ZSTD_seekable_CStream *s = ZSTD_seekable_createCStream(); + ZSTD_seekable_initCStream(s, 1, 1, 255); + + ZSTD_inBuffer input = { .src=inBuffer, .pos=0, .size=0 }; + ZSTD_outBuffer output = { .dst=outBuffer, .pos=0, .size=capacity }; + + ZSTD_seekable_compressStream(s, &output, &input); + ZSTD_seekable_endStream(s, &output); + + if((((char*)output.dst)[0] != '\x28') | (((char*)output.dst)[1] != '\xb5') | (((char*)output.dst)[2] != '\x2f') | (((char*)output.dst)[3] != '\xfd')) { + printf("%#02x %#02x %#02x %#02x\n", ((char*)output.dst)[0], ((char*)output.dst)[1] , ((char*)output.dst)[2] , ((char*)output.dst)[3] ); + + free(inBuffer); + free(outBuffer); + ZSTD_seekable_freeCStream(s); + goto _test_error; + } + + free(inBuffer); + free(outBuffer); + ZSTD_seekable_freeCStream(s); + } + printf("Success!\n"); + /* TODO: Add more tests */ printf("Finished tests\n"); return 0; diff --git a/priv/zstd/contrib/seekable_format/zstd_seekable_compression_format.md b/priv/zstd/contrib/seekable_format/zstd_seekable_compression_format.md index 55aebfd..7bd0790 100644 --- a/priv/zstd/contrib/seekable_format/zstd_seekable_compression_format.md +++ b/priv/zstd/contrib/seekable_format/zstd_seekable_compression_format.md @@ -2,7 +2,7 @@ ### Notices -Copyright (c) 2017-present Facebook, Inc. +Copyright (c) Meta Platforms, Inc. and affiliates. Permission is granted to copy and distribute this document for any purpose and without charge, diff --git a/priv/zstd/contrib/seekable_format/zstdseek_compress.c b/priv/zstd/contrib/seekable_format/zstdseek_compress.c index 242bd2a..113f6f9 100644 --- a/priv/zstd/contrib/seekable_format/zstdseek_compress.c +++ b/priv/zstd/contrib/seekable_format/zstdseek_compress.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -350,7 +350,7 @@ size_t ZSTD_seekable_writeSeekTable(ZSTD_frameLog* fl, ZSTD_outBuffer* output) size_t ZSTD_seekable_endStream(ZSTD_seekable_CStream* zcs, ZSTD_outBuffer* output) { - if (!zcs->writingSeekTable && zcs->frameDSize) { + if (!zcs->writingSeekTable) { const size_t endFrame = ZSTD_seekable_endFrame(zcs, output); if (ZSTD_isError(endFrame)) return endFrame; /* return an accurate size hint */ diff --git a/priv/zstd/contrib/seekable_format/zstdseek_decompress.c b/priv/zstd/contrib/seekable_format/zstdseek_decompress.c index 5eed024..fbb2d4f 100644 --- a/priv/zstd/contrib/seekable_format/zstdseek_decompress.c +++ b/priv/zstd/contrib/seekable_format/zstdseek_decompress.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-present, Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -23,13 +23,64 @@ # endif #endif +/* ************************************************************ +* Detect POSIX version +* PLATFORM_POSIX_VERSION = 0 for non-Unix e.g. Windows +* PLATFORM_POSIX_VERSION = 1 for Unix-like but non-POSIX +* PLATFORM_POSIX_VERSION > 1 is equal to found _POSIX_VERSION +* Value of PLATFORM_POSIX_VERSION can be forced on command line +***************************************************************/ +#ifndef PLATFORM_POSIX_VERSION + +# if (defined(__APPLE__) && defined(__MACH__)) || defined(__SVR4) || defined(_AIX) || defined(__hpux) /* POSIX.1-2001 (SUSv3) conformant */ \ + || defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) /* BSD distros */ + /* exception rule : force posix version to 200112L, + * note: it's better to use unistd.h's _POSIX_VERSION whenever possible */ +# define PLATFORM_POSIX_VERSION 200112L + +/* try to determine posix version through official unistd.h's _POSIX_VERSION (https://pubs.opengroup.org/onlinepubs/7908799/xsh/unistd.h.html). + * note : there is no simple way to know in advance if is present or not on target system, + * Posix specification mandates its presence and its content, but target system must respect this spec. + * It's necessary to _not_ #include whenever target OS is not unix-like + * otherwise it will block preprocessing stage. + * The following list of build macros tries to "guess" if target OS is likely unix-like, and therefore can #include + */ +# elif !defined(_WIN32) \ + && ( defined(__unix__) || defined(__unix) \ + || defined(__midipix__) || defined(__VMS) || defined(__HAIKU__) ) + +# if defined(__linux__) || defined(__linux) || defined(__CYGWIN__) +# ifndef _POSIX_C_SOURCE +# define _POSIX_C_SOURCE 200809L /* feature test macro : https://www.gnu.org/software/libc/manual/html_node/Feature-Test-Macros.html */ +# endif +# endif +# include /* declares _POSIX_VERSION */ +# if defined(_POSIX_VERSION) /* POSIX compliant */ +# define PLATFORM_POSIX_VERSION _POSIX_VERSION +# else +# define PLATFORM_POSIX_VERSION 1 +# endif + +# ifdef __UCLIBC__ +# ifndef __USE_MISC +# define __USE_MISC /* enable st_mtim on uclibc */ +# endif +# endif + +# else /* non-unix target platform (like Windows) */ +# define PLATFORM_POSIX_VERSION 0 +# endif + +#endif /* PLATFORM_POSIX_VERSION */ + + /* ************************************************************ * Avoid fseek()'s 2GiB barrier with MSVC, macOS, *BSD, MinGW ***************************************************************/ #if defined(_MSC_VER) && _MSC_VER >= 1400 # define LONG_SEEK _fseeki64 #elif !defined(__64BIT__) && (PLATFORM_POSIX_VERSION >= 200112L) /* No point defining Large file for 64 bit */ -# define LONG_SEEK fseeko +# define LONG_SEEK fseeko #elif defined(__MINGW32__) && !defined(__STRICT_ANSI__) && !defined(__NO_MINGW_LFS) && defined(__MSVCRT__) # define LONG_SEEK fseeko64 #elif defined(_WIN32) && !defined(__DJGPP__) @@ -466,9 +517,9 @@ size_t ZSTD_seekable_decompress(ZSTD_seekable* zs, void* dst, size_t len, unsign size_t forwardProgress; if (zs->decompressedOffset < offset) { /* dummy decompressions until we get to the target offset */ - outTmp = (ZSTD_outBuffer){zs->outBuff, MIN(SEEKABLE_BUFF_SIZE, offset - zs->decompressedOffset), 0}; + outTmp = (ZSTD_outBuffer){zs->outBuff, (size_t) (MIN(SEEKABLE_BUFF_SIZE, offset - zs->decompressedOffset)), 0}; } else { - outTmp = (ZSTD_outBuffer){dst, len, zs->decompressedOffset - offset}; + outTmp = (ZSTD_outBuffer){dst, len, (size_t) (zs->decompressedOffset - offset)}; } prevOutPos = outTmp.pos; diff --git a/priv/zstd/contrib/seqBench/Makefile b/priv/zstd/contrib/seqBench/Makefile new file mode 100644 index 0000000..e7f08a4 --- /dev/null +++ b/priv/zstd/contrib/seqBench/Makefile @@ -0,0 +1,58 @@ +# ################################################################ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under both the BSD-style license (found in the +# LICENSE file in the root directory of this source tree) and the GPLv2 (found +# in the COPYING file in the root directory of this source tree). +# ################################################################ + +PROGDIR = ../../programs +LIBDIR = ../../lib + +LIBZSTD = $(LIBDIR)/libzstd.a + +CPPFLAGS+= -I$(LIBDIR) -I$(LIBDIR)/common -I$(LIBDIR)/dictBuilder -I$(PROGDIR) + +CFLAGS ?= -O3 -g +CFLAGS += -std=gnu99 +DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \ + -Wstrict-aliasing=1 -Wswitch-enum \ + -Wstrict-prototypes -Wundef -Wpointer-arith \ + -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \ + -Wredundant-decls +CFLAGS += $(DEBUGFLAGS) $(MOREFLAGS) + + +default: seqBench + +all : seqBench + +seqBench: util.o timefn.o benchfn.o datagen.o xxhash.o seqBench.c $(LIBZSTD) + $(CC) $(CPPFLAGS) $(CFLAGS) $^ $(LDFLAGS) -o $@ + +.PHONY: $(LIBZSTD) +$(LIBZSTD): + $(MAKE) -C $(LIBDIR) libzstd.a CFLAGS="$(CFLAGS)" + +benchfn.o: $(PROGDIR)/benchfn.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + +timefn.o: $(PROGDIR)/timefn.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + +datagen.o: $(PROGDIR)/datagen.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + +util.o: $(PROGDIR)/util.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + + +xxhash.o : $(LIBDIR)/common/xxhash.c + $(CC) $(CPPFLAGS) $(CFLAGS) $^ -c + + +clean: + $(RM) *.o + $(MAKE) -C $(LIBDIR) clean > /dev/null + $(RM) seqBench diff --git a/priv/zstd/contrib/seqBench/seqBench.c b/priv/zstd/contrib/seqBench/seqBench.c new file mode 100644 index 0000000..7efebec --- /dev/null +++ b/priv/zstd/contrib/seqBench/seqBench.c @@ -0,0 +1,53 @@ +#define ZSTD_STATIC_LINKING_ONLY +#include +#include +#include +#include +#include + +int main(int argc, char *argv[]) { + ZSTD_CCtx* zc = ZSTD_createCCtx(); + + if (argc != 2) { + printf("Usage: seqBench \n"); // TODO provide the block delim option here + return 1; + } + + FILE *f = fopen(argv[1], "rb"); + fseek(f, 0, SEEK_END); + long inBufSize = ftell(f); + fseek(f, 0, SEEK_SET); + + char *inBuf = malloc(inBufSize + 1); + fread(inBuf, inBufSize, 1, f); + fclose(f); + + size_t seqsSize = ZSTD_sequenceBound(inBufSize); + ZSTD_Sequence *seqs = (ZSTD_Sequence*)malloc(seqsSize * sizeof(ZSTD_Sequence)); + char *outBuf = malloc(ZSTD_compressBound(inBufSize)); + + ZSTD_generateSequences(zc, seqs, seqsSize, inBuf, inBufSize); + ZSTD_CCtx_setParameter(zc, ZSTD_c_blockDelimiters, ZSTD_sf_explicitBlockDelimiters); + size_t outBufSize = ZSTD_compressSequences(zc, outBuf, inBufSize, seqs, seqsSize, inBuf, inBufSize); + if (ZSTD_isError(outBufSize)) { + printf("ERROR: %lu\n", outBufSize); + return 1; + } + + char *validationBuf = malloc(inBufSize); + ZSTD_decompress(validationBuf, inBufSize, outBuf, outBufSize); + + if (memcmp(inBuf, validationBuf, inBufSize) == 0) { + printf("Compression and decompression were successful!\n"); + } else { + printf("ERROR: input and validation buffers don't match!\n"); + for (int i = 0; i < inBufSize; i++) { + if (inBuf[i] != validationBuf[i]) { + printf("First bad index: %d\n", i); + break; + } + } + } + + return 0; +} diff --git a/priv/zstd/doc/README.md b/priv/zstd/doc/README.md index bb7a3e4..8f3babc 100644 --- a/priv/zstd/doc/README.md +++ b/priv/zstd/doc/README.md @@ -5,8 +5,9 @@ This directory contains material defining the Zstandard format, as well as detailed instructions to use `zstd` library. __`zstd_manual.html`__ : Documentation of `zstd.h` API, in html format. -Click on this link: [http://zstd.net/zstd_manual.html](http://zstd.net/zstd_manual.html) -to display documentation of latest release in readable format within a browser. +Unfortunately, Github doesn't display `html` files in parsed format, just as source code. +For a readable display of html API documentation of latest release, +use this link: [https://raw.githack.com/facebook/zstd/release/doc/zstd_manual.html](https://raw.githack.com/facebook/zstd/release/doc/zstd_manual.html) . __`zstd_compression_format.md`__ : This document defines the Zstandard compression format. Compliant decoders must adhere to this document, diff --git a/priv/zstd/doc/educational_decoder/Makefile b/priv/zstd/doc/educational_decoder/Makefile index a9c601e..f6deeb1 100644 --- a/priv/zstd/doc/educational_decoder/Makefile +++ b/priv/zstd/doc/educational_decoder/Makefile @@ -1,5 +1,5 @@ # ################################################################ -# Copyright (c) Yann Collet, Facebook, Inc. +# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/doc/educational_decoder/harness.c b/priv/zstd/doc/educational_decoder/harness.c index 935f60d..12c5a80 100644 --- a/priv/zstd/doc/educational_decoder/harness.c +++ b/priv/zstd/doc/educational_decoder/harness.c @@ -1,5 +1,5 @@ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/doc/educational_decoder/zstd_decompress.c b/priv/zstd/doc/educational_decoder/zstd_decompress.c index 62e6f0d..9ade765 100644 --- a/priv/zstd/doc/educational_decoder/zstd_decompress.c +++ b/priv/zstd/doc/educational_decoder/zstd_decompress.c @@ -1,5 +1,5 @@ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the @@ -1890,7 +1890,7 @@ static size_t HUF_decompress_4stream(const HUF_dtable *const dtable, /// Initializes a Huffman table using canonical Huffman codes /// For more explanation on canonical Huffman codes see -/// http://www.cs.uofs.edu/~mccloske/courses/cmps340/huff_canonical_dec2015.html +/// https://www.cs.scranton.edu/~mccloske/courses/cmps340/huff_canonical_dec2015.html /// Codes within a level are allocated in symbol order (i.e. smaller symbols get /// earlier codes) static void HUF_init_dtable(HUF_dtable *const table, const u8 *const bits, @@ -2145,7 +2145,7 @@ static void FSE_init_dtable(FSE_dtable *const dtable, // "All remaining symbols are sorted in their natural order. Starting from // symbol 0 and table position 0, each symbol gets attributed as many cells - // as its probability. Cell allocation is spreaded, not linear." + // as its probability. Cell allocation is spread, not linear." // Place the rest in the table const u16 step = (size >> 1) + (size >> 3) + 3; const u16 mask = size - 1; diff --git a/priv/zstd/doc/educational_decoder/zstd_decompress.h b/priv/zstd/doc/educational_decoder/zstd_decompress.h index d89c835..c13c813 100644 --- a/priv/zstd/doc/educational_decoder/zstd_decompress.h +++ b/priv/zstd/doc/educational_decoder/zstd_decompress.h @@ -1,5 +1,5 @@ /* - * Copyright (c) Facebook, Inc. + * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the diff --git a/priv/zstd/doc/images/zstd_logo86.png b/priv/zstd/doc/images/zstd_logo86.png index 216f228061c460c4e4725785486ebcff38804fbc..8abefe21b1e172b9683392057421984f64c8c9a4 100644 GIT binary patch literal 13069 zcmeHucQ~Bi*7oS4Bw7&NDA9%)z4sb52qJ>PVAL^2??jF05+s7?q9#g02!as3mk^@& zE;`>xekJF7&wJkQT-W#hcU*gB_Fn5=_g;JLwf5tghe#bQ6=DJ!0ssI&tfs1{i~iI@ zzqIkO(C?X^*9OoZcqlbZCA>vEI#MQ<80M1<0DxA{Ti*z!3-w?^Af2t@4lpK^7Xrov z^MqRi0G<;i&+k7q5ph?#gb>`oXjGD*tF-sRWoS5*vl?J!yH{A$MIjK}{{e4uQ4lcS zb9r{?cG;St`vAo`YT(WJQl>DU0Z%*a^6b`@$M(v(f502RdMbavsoc{8<#{&0<+YF! zU%!jfgY7xYjPRznXM>>=C*GD9C(7Dpqs|{6a`-c7ung>-yJcOtF+a`_X&%LCvoUVi zHwbvyESd4SY%8F-(JbGr@LYUd=B$paKK*@ct0zy&l1JtCt-uv$l(47nF@4ESQ)4f9 z&g|q|>K^8XEH~wi8`uw|S`VXdXvsTm*@tF)neN&<@fVlrzFYq| z$@$~iyiF(r5dNMylwSGW_Ya?MQYvWdw0yCQew?#XWi9mMU6;h0qOU{0rAp^;6bGO5u!ktDNW1Bo zxmePSeY6nv>{xs6j6BFVDv{#Ri#h8gNb1c5h0h4*2E9+G4qZEY{3%MnH#d!6$7jlw zk~J!y4-FM$#e?!+xONEDOzv?Q=yv2L8Dslr8*2)U>e@S2jrJGkK79bM5*qcUm2$_T zbLFf5G2+7*^89+K zMr=`xj$Zdu&v9Id@SPg*j?aL-Sreb$bnC#8jTryq>I^7Z0v0S?LC0@ApH>0c;v2sr zv6}im6v8DUwbLWCSviu*nN^)QT9jXG31%7m3X2(2TUND+oeW{QzEiEdB_gfrFdE|( z?pp?s>7{kJK085cIPNpl;H(C(7J9t(I_Fb<`>Q(cJTcw+H7CDg*4N(qb4LjtG*tb~ zwzJ=S#_?_9o~db?8u|6MC4ElKnO0%JJKem;8R+r$N zTsR+Dp5P+v@Zsf@vdH&i8~nCWe=p}!@;`(Vuw! zWytd5p1{PevP85=+I0BI{n4-QOC)qn0>`g^g`C85li!J&5ezt5+An0%ZxaiTD$%I2 zsT_3_WBwtp^LEm;`<^wuEzybpG)}o$5dzUItxnIvZEVo^9!xb1{>qtND`tYEul=a^ zLZoueiSbkm@v6jb4R_Y4ji~ZAHLCiaPH~3Xmt*AP>~N{K8`ZPKJ35$iP6s)!-4><4 zNLA%k*wtphNRviKjeMa9W^o!6lDdYFEqgPgXn<;UgdrWN7*W2_- z@8Ea+W{s}~8La%vj2D80wBVKaGT4lBNG`xm3kwPnN_y z-7V~1AZkm$CyV(rgUdj;`D4gUj^}sSc=&wp2cMTQbZ~J7b=p-lWvYKvekaRPj(qRp z{s`z0WXja|NM_gToLFZOC;Ab8L{Mg4&LqXPiLR4cD;7v>@>{pby9=|}9~7AMpV**C zz7tupTZi)Ij30`(m+B!K8mudT^>m?`RReD7bTeJTk3>p0=%L?DmF!p{{W4xpUIHbL z-_RLIe6eBU6eLrKVmNzcAMrs4?#ckur$ktTPdKM4s zCehXjIVt~GHo1?KT(@PNnuuM@*O!yoRr47~aul)lS2@2a&|}7W_N$FgYt{lR1gWv_ zC9tBBi9As`gu?*K5MsRDpjG(pruI$hM16v^=Iq!T@eS>?vQY$jf>>W7=7SNNBJGb^ z5y{oIh=jnT=Rx?};%9m{S~2iG>BL(BcvL@O(LqALO9~ERF2^JV4f_Zem6Ci-GyvIx zPR2(vR|TXL4SSwrUnP= zx{sz7g!eb$S-H8?7t25HPQTV_9m4@Ohc2htFe?%0eE1@_1rgA*g8T4)`vTCF5ABef z-sq7oObkk_SuJ!l+0tkUdm3ICDk86zW0wYKG`)4h^<*=_&gPCzgSQ*LR>Aul`nbfy zru6e8+RkQdqyRHdMgyjK`Y-uJItY8M6r2m820Jk5*7`AjJ2wf8X4#YTYZ3u{%ttxD zV5X|9uz{Q^Ks)9k21^2Hh>QRqGPBAwdRrtxwW+k-)Au!CTbXTGuPI>Lw~+UNX6sgv zoqFz;*Q=x57T($?aucgNN=I-+Q=fzb5D?#}m>M7HsQgU8 zyD{LoZpa;zz1d3M6U7Mau-mm71R-DKHT$jWRrC!fDkw-*FSwx|lh;WPUI27YN_x0V%|Hjt5rvj6I{+?whg6BUBo(-YIlr zUid!?{=qV)`jqD`Ru%)RQ&RaS0i9?k0U!nY^~7c(eaAu+UIKEnLL6BZTiUA%ZFZEs z;W=uMXfL(fiVsXdh_}mcW)IXAsxNh-`2UDu%waan-_5Lu502%gA za(lw$k(*JmFdf4IJQj9?ZVU*!ph~FS!Rvb-Wr0+1;v)0 zr+xv2eKL;LDSscgd`n?|#?Kt-_yF`MpE;^2V?pofe z5^n_DNg{0S$BHj70=r`K#8b^ceesz+l-&j-Yue2a_2#joriHo|9yG5YFy`!#fDOzByf%72ftjUv#>cZD`EH0AaTCuv0&d& z3htY{qa6n^Nd>V|3$30brgg$@O$o;j#-VS3Qqc`TL)?qO&2*;E#)(JcsMv4JJypnX zTz4|V9zWIYQ?(3H^gIFLuAD;^BC8fcdfJ*(Oz>yvb5seF#vJV^hO<6jn=7;d*hLU_ zF?TDeiL8I4W(Kh97TH#+*>0CJ$$wy}M``Fs?LR0hg<@eXo;*E&b&tO-GNt5|lV;^a zri4XKIO+r+=(wUU5C{6=p2xBX?0C0RYaQ~o`w)nX;jD% zVjXp`);u14sW)uBI>CBi-bEAnNFEoXzh9ZfEGdC>yS~y(HWadjVZ~T-;9h=X_yz&7 z;l1t! zn(cU9&7;jaSwnY4Y?oF&X?0B+W)O4orGZu%XG(X-wH}V(HN85zUaVV4iWwVNit5;P z?ru?dqV;Yu_Jo)NGm^Eh>V1$#^M##jitl{Y-Ir3RQmM-*mkXIHLs|QAp;Yp z-Lp9RCj{GTYHOZtm5=WazhXBUuCi7%9(*&yuvzS06;%gxa}JL4uJX6Sc4SSGV;26* zvUAixr<(~{v7zC#58_UGFx5OmBb}C@?9s@6JNKx|ec)+F$6!m=MBDe*_|;>INWJ>_ zaMe_~u9q*m5J9+=g26{X&51jStrvv-`rHTA+R4u0sr?cI?A@(}_f$wum>$p`ci-$( zP^;o7njxjl%MC&RY;rh}!RsPN?3>ZTbPw0ezx7WdhRE)BR@u45Qb(2o5qSqyC!1oA z5f2MoM&-%9U+SN$#Qj zYJ#fI1`hSOX%j2V7N;vr-`P{&U<-@}N6y)4=_wh^;$tSWn6iwu+h?W_>U5Oj&cbg` z06QKy(QN}fv>eFA#_*B%2E(3as{jKBYlu`aZSe-ojT)X-e}2}~C;Aps5EwrIp2sq- zcFodHvUQHj`s4tfDXNk@<1w&FxQ)GwyEbuF%sOZ?AIi|7adSw38*c&ClJEJTkO(2zs#0n*ylgf9cm!*UzlhsU-{(!z5<~S@$TFwt;`gRdN~er^*UUgj4EpKGSek ziw$0%bZ_QSdu=@4@^4#tu#ZSJB)VR)3W#xihS4RrN`lU2c=D z7(C0)a^NUYgx!sS^NHe<`lc4{Vh$*7KVe_A3k;*6{qe;At4SAILBt9irrXx>GAElO zwUMseP}D`zp1(Zu9brziK~YQ!qI2hn&20EZSqoR(<5BO%%BLxaROOLv`+@*RO-kDr zoW3ux5{*9&p1HT5t4S7SO38TDVa2vd|mb6j9#GR64jb;NeA zqokraISqsZ`=CsG!M2PIvqU6|!C#*9gQ~ji48CiG+JuoVy)rwU~f)bk`%gUSl9Zj8Fm+{;p*}iy z%A>GIXBKT0_2;*B#Ve_z&fbMk7ruE@8JNFn5mjAP@7QASIMZAzDB(~!;u`UtNIS6O z=Ll{@`!r8eJq5Qhv`ZJc<8T31wzh z^-D>M>k!;%EP;sUa`B1C4JtdoYf$Wo`Oh;{!7XGYHb8LSewP7-;BEf;p6li57y;Nx z`riCUEQ9qwj&x+=guekQ@BnqmHGE@P;6}aE{mkIL3uNSJon`=PMTe!M`& z_Ij4y+!%1({j*#q0j7vbfPEREamR<)g0&g2qO=JdDAoGPT|+l+aZp)F-^|)6o;$_L zt`~Sawp%&U&g+9h=KTerx(j_Vq_SS}i|1arWnQ(1bZt1H8|b}-0`Q1>yD(x&9+%pS z2Ix2dII?6;Gl-CV;Cl<)sUDr8bxhF8xKBw%%(9Sd&`FP!ZQs@uWOZ#x@@7YE^Q6f0 zun||x5#+=pHzzi5H8co6^3SXkE)`3$mB9!qyFK0OAGI#`u-F`#?Ps$Y69u_DdkVyC zV+c=IuDutk_KM@6&=o;tGZ&;63uUXT`v}Cr;=ML!ypEhXhgQrW9(w%L5h-L0z<>_i4HWgx(cST+wQa!_TF(LpR88|4 zQNNsLzJ$iyD!(pDeImsP@3R^vU`WhlX3_b8R!TAWI@|O3#j^U-ib$_=`1x zf_v00wb&?M3enP|5hY!aEl(1glrmHp740VS#g~KQPPqHQAJXHB;~>QQPTMzf_Xyq3Jn0}J}F1z!-|(l&SUYGtBa_K@Eo*8{uW z_O8)y#!F!t;nyW7T#YsZztZ8G8WB~BieEz2xy`{;<+D|eB}yly!;(cVJV4JhR`WKA9^#SQYKaKKLGKb_Ye?Y~ z;7hhSOQnGuF7Y-ulc8*%Pl=;bs;ezltzM?{=%)4HPW-iKMLGdXe@c6U^v~nkTJR@4 z#Q|ZHG7MAG=QD?MJbCGD@~1?)k%WP@aT&a;)EIqkl>%n*IS19N zr;h+okmPdGNDB>#!h7NmPgocqGTb2wg|u-1rEh2iYtL=6%J7s7|QLsB+g?1rsj_yi>nCJgI(OrmY&F7iK7?l_SU=D%Gq z(de{V=;>Fn5N?_T&HOy?r>1PgM=3M~RCij1s0jm>cStnau(I!;P)w_guePgiM}UvM zzS9WsDXY<4lI2Z0k|y6X?hcwQ5Iv&5rp#?Q%)}h1uy|jzhhX##>8^~|MphD}m{sbZM2J1}yW|n8= zElM{}cy*2|eUzVuZV@@`n%;?D&R6jsxr?O~6G^|}=zTqjK~+|HLwi&;SZQu$I$Ew7 zqw~38nwO19wsB??1hw;3JE)!V>D&$4hRZel3wXsD@Lu)-W3FUJwQjojATKw`LT1HlMBQ3s zRPPu`EOCk=)0=bi2u@VlV{b2u945bWpCeAjj#r}cCPSUq+X9y z9sAIO?MZkXKB@HVvj@rb@49|l7;dg1d)Nfmf^id2A77pn4nzuOj5tlxl$Z&Aqznjq z+K5Bje*YjQ31^l_s7meQyf6=axi0O1Vv&PnM^=$uO9?%FAAy4O&QPS{z#GU@%In^j z_1WaJRoh|`JHjUM`dshmFNX67$1x~gV_UiUaqSxQ19xRtgwk6gCf+VJ(*e%Q)LEhmTIp3Nv(GWPJwQz#{_66;Mf70BIM&#Yp< zfQ>la`3WdGlDtS zRL$9nX;$y$y)lsB)V!sOs_rSXc1$IL`pRaV&M9!V{C3#x=Ki3p$bNsBdD*1vBpN_4nKR;s%n(lfM3C3e7Jo_>tG%Su%lu5VY=v_i3O z)7@Sf7kIBzmUT=e3!@1l44)dD_m21qQM9{u(^O6dLnF_!)`RC6N`8!_n99BBP^*N| z;$-&$phdvRF%b$_^1eMA+`~N&s9n;K;B^`=pxH7%gKQu7rH)uvia%@{pWypK8OMjE zqn*dnJEo*O6XXoKw~EbhSB1e5S90oDeizfU>T>2o_-W0JsIwBnyBT#pTNhL=hI|TV z*Kf@O02q(p3JN-E3JOd*y4uQW8cOJgYA^iarBpx3(zTjBr~=+3m589|(n;ruh!VBC z8KuKKL{K80Vse6`Ns~Z9rP5S2GxOPExT*ENA)cWe$r-=}h15mGlU+WeFmCZgOvZW~ zGI(qO0_wZ;-&)e?2z9Em&C0XaabcU0Dq^2gD?e+|@YAA5+@>=-+}=6hZF~En_gu_7 z`9%jHLsqgA^$GZm>E*XdRL^M()F6ps)^!#V!mE1L?l!(4(t{IzE-B!Uqqo-rwaDsg zL(;x|l|`mas&eyQwr@ZiEOd|3rpg4??Xyc2hMf4qKMGflYXm-Jk`IjLy2J5FKHA_3 z1v%$Ltj|$6-A7g4R1-EZk|z`MX2KlaWythY+HuD<5l#v`aWjWgOKkj2%qMA%0gPYW z=dkf5u?&qn%?hyuGIBpjiJa~>0Jss6gbJ_0V4?a zi{IA^9lD$VatdG16C)7#(UcrMQ*%>O8};WtdE?VJw6KvWz_Q$Z0N}#tvw=HF0k`Tt0-4}eQb0o?O|T|H0cHzV z^+v+8Ub?kaza5pK~AnLR}{ZE6k)EGNH_uocXndB;)Gf_yP>3jK(wCe z4|25DQppo7{9}Tv@lSeJlog*EdLW2i4;sM74+e{Z!2BS7F}~mJ(OON-zpR~H|A->+f^A zxx%hQKmB&FvgSiO^>gOmZB*1Wb^fxs3Zo6&5%JUF3jI6M%JMH9!VT&06Juq`2XlZq zq8)KXGeiCXfAw_yFN=Rz&z0uCID&#(|2Oa}&3}QTz5D0ne}?y`>p!hxp~-pSXdXc!elY<-3u_SAN(=%L6c)4qiHZn7K-N$JE2yZbwWToB@;3@? zBph8hP>0{Wx}vf|Q;9-FAtFLnLLf1ikO-Pe#PZ4;D-gs=R2bb6;)e)8e^Om-o4A~g zniP$a?gz2H6S3!da3yFdG#e{@}1R-Eyev!XH z1~8;6x+bnTAz)s9$WM)xrMNPh5sJ=DxFggC#)ojS`Kh?tMR6UtC(OY}5zX%O%M>Ck z`pcic)%Sl_gN1&nuQpX29SZceK(7iB?blCMbR@(TkT5998L98=>>vfaS{KumLqF@6 zN%D`%6xVRJ{Au_T470i_?mx;~9%{q)b12F8ufYF>Nzc~V!|DHx=Wpm=EOJPchcnV% z8>ww!2eU-|Yo31w{)kOfTC8U%w_fnlN+LJ+96;6JjvI$NVWph%dU4LS$V`;0EkpZmFOwwS)#?4`@+7a$H%(3P-vc0db^eK~UFEHw8v@|~havwlFn=b> ze}ntY{*R>mpUi)U{jyeYMtGr1-WH|p;q))3|0lp-3>t7tba^@dE7yOA{0hr&7df=g zzhvl(C;D2<_t(YvkF>a|od3n2Kho`gaRxN?esCE&L@%tA0uT0{ z|7`0Ufc4Ek+TRki@Xxlk4*1np*Z%IH`{=f^`bS${`MZx6F0Wi(UhZ66Y+qdLo}cfW zpU<70&7PeNpPUSxoU|Puw;mt293QtG9km`Ey*oUrIXtR9JSzQhSn}hr;NVC8!H=AS zA8GpsDfMTd&gsY2f4NFvbAlu zwGH3ew%yvc-rRb)v1z=qd4FTmXk*i0V^e#5Lu-9QaeYI6eM4@2Lt$-QW^G+)bxmM( zjc0X@Yjurdb&YmqmEhMx(CfRKoUl#J3HUHyB8CZ-m4j?OL~z5#*3p%IbMPvR1iQeM1#m0MU;Qd(C2=H2_c z`i75fpF6vHdi(kZhDXLHzb`JWtZi;>@9Z5M9v`2ao}FKGzudG$Pl#cmuA*3=gD;!_ z0APuzDaz@4PE3wqy=7G=!I?4nTJ4B#`x9+dj!uihquiM&=%#76>*hshFLl)AZ4amIH?clmC=@q9( zE&+loB?_XWVTnCtkK959+3_D$r+bc)yiC)4@i1cc0>4z6Af7HCHA>oVa;J*n?o<|q z0t5OF$aCl93Z2qcyXh{9gaqewR?&U7q0giV6`*#Mbk&)AyAHmkCDK6xfi0U{=X z8KcExLFqW3RF4cMafa{zI8=$!1uMrdkWxDn4L&y`CVN$}9lnt-b^GISr~? z?}nkvdT>Y>J5Uy;Y`7`$>Bosj@*|{X6NsNa_+|0YL0fr!uTvJzI9N$4YzKRZ@>-L&(&G!5bp1L#}5`BR0I~q!a9D; z&XrAi+ioBOGRjJf-Jkh_iRM1ld`I|S- z?Ks&U4v*|6yNY>%wgp2IViaD#jwsB12^@5P7k|KMn1f+_*ip%34tiqLZuG1}P(PU2 zz6eB5RuDTpu$xC1y}>YT7T_;>&gpDu4VseE>a$@Le~>OO-W0!RNg%J{{S|N)3pEq delta 5771 zcmZvfbx_n{x5t<6knToM5SH!`mTu`#azP}PrIGlN(j_Tf0)imn(kva)UE+e2bf*Yh z-?(?)JNKS{o-^nB{me7-%$YOiiMJo-OQHp^K_E~@>CGw#gkRyPsHm%^sK}!04zqW3 zhJrw<(P`;82KuIS14sV+?Yd-9o(ix~PjuBUg}75)%BsQPWQr_NQ89z7xN%sNk91W$ zNg-XeAVDe!1p(W_Eha4_)vLxz5z1T>p8KP|)uAJY|H=oseAThi&VQ972TFaF0H~&7 z^MI20M#v&>`~+=n)-$hzg0Y0m*mLZ(S5`J0cToTqiEaI_uDxp7-XdgmmheZeDgCGT{B5uBi4k zSMEl~>&Jt_5;=CXI!Dl_xTsJ~JvxWY2S0`fsi|oWA*bykwV$zfJrjDwGW9Jpn<0W! zk9l!FSfjz-+*6!!-X{x z%^ysM+mz~D6LR4NEDGfeCr0J{bbdY)N;{@WM3R%c&RH4DX(u=HPON|=;$ADPL5w~5N8D)UodB^e-eu@r0WkyG zfINXX(O8(qD&01gPbWJIR@f>f4Hk2!IOcQTDEvuQLsfdg;1L$5D6D1$SzS^J>^_AP zT|tg;4qc|NOix3G3e;aS8<5=L=Z20c8W+fJqs@>IDZ&c*!RY!{oF?R=;lH|g7chv! z*;hH;1ar`Sbw?aOZUWG&Lr8i?kNvLc0&vp1zpM&ive<=(Y_n|P5}YU~B#Cq2Jz{rC zvMFKs^qekPw1l8JLN)o(h7fjSp=~CON1RYUGkNq+8$qxW_+zHJCSxh>baZpX&On?E z()N=H(P!E@yy5EsU=!Lh4O60VU-r+J$E?1@vILglD-m*Wa*-X;%)J0x65bM~Qz$=c0t*KP=@4Ed zt^=MJo?2z@em8$!r>8pRWx9gP#^(o5g_d`(;}*$(2)s+9CXbx)PDM0rYS%63Xh z@WW>fZw`JAfW?DdkAAl>S@{=Rj;ysz54R( z@{1XQnX(xjcTWidi3SE8*D8xNi~41)=J`{IABG>v&BxQ`)6LU?P1s-}Hz_3{WgCS! zz7s_Sh4zzWeH#4|9g<_?ETsi+m;h%p{pUh#+D*#A(W*9=xCMcUBS)*0oElpy`n=RZ@_Pz zHATFI#=zAANS9C{>klPa`=UD{5n2bDH>HN9Vk1^U{z4&(uJy>~h2{j02oI+%lu%iw zf2KpWSmu>Em^YLcpVy4{)%bKiwHR!#t=ezp*#Yk~FNuk4U6Co438e|bv>hRe*fKLV zU1_I#I4>QmbF)29W>axiIx=Cxa`5|*8w|b{K8B1yz8A)&PHlVR z)i1;t#S+ZpzYF=u_~iLl-6lmwtHcagC(Z+T>3KHeDS6E8Ic-M%2!D1|5ULL)BP*P* zny;Q;hHOE8LlPm2gS>+(|190r-iO?q-yPM8ELwQW_zQQkT6xJkuyBH?46#~&ux!!L{ECjjIq z?peEqY&?34Ro7M3>{L<%X4cAomnTl5CZUs4nSKJ5w$*(VefT!!>&I|CI09}P8$fkn zBHXm#3ZGnS)rd>q;BBrmaB*Ck5*psffTuGi4)m)GFvmO$;r*O(>nB^lo5d@(;6Kzp zsL+<)+LYKpsgRiBr%dZE;EFgy; zw7>Ye9zNfH)YrCF$(hc%$!V$4{LvVw`~7LHELuw~`>p7A7^)ttQK@P3(5SJ>3mOA$ zhGKjVZ4`kyxbma!g9hpt0plT;;j2TAgVg<0BTU1uhf^yE=={8T&}Qz_@ru$n@w@S} z%syw{E6eN1GWeG*)cbbJAgVtBK>iAHi>@c@^seb{@iw11N^;_YU@zmh8Di=E5)3B- z=cAXkSMCMz#YLnrv5RXkJqJ-wBsKG)%vxXxVzLcxNnZSEgIoM5Ao2Tj{@U`0i9mV| zp5xVa65bWPn{uAgH>!I!v^7Z;6#G@A0dZu|`6)s4S3^pN=^yXouVXtELPHLs`JB1W zq^00|SIZ`?<$)Ukr#B%9P{u^QUW=w5%{BGEY*zb*GXnTuZl7*T`Ydjq?(k?EYrllP zbZNE+s+zx7-)?xd0()Hk`~F!l@#WVCUi&V4*(*MmrQb)-8rxtmTSk$A2Vz_~2{)Tl?Zi*Hy!jRKLPNUXbbo_`dc!Ww)$l{PXxr%c2g#_%GKn4`wH$8|L4y zI*$^$rnnwM7z7-K@L67WtEqw7WN<+uRG_!~AdY57m^9_Y*Zh38<9y#GIh_a@5aJL7 zX~0;`Cy}28N^qWkc(+v|661zb6Ccm}@aTh@=-jj!#f-I$oR((WG$v>TyHo6>xA*!m zIvQ3!puz;2$P+mDR@OG^1uELmlbiO7<{!a%AOmnPr!Y$QTey)R5IVM_fr+PymZqdF z%!SX|4rT-8^L26iyXt~K(!P>^Qx~YGHH)u{b4mmwH@*m;Fdvvr7N3PBrIk?${qJ$l z$brf7H&U#bh5nT!G4lccaQr)@p}vmxlDdw*P-jz!qYKp4!x82xD*zS&OY{G~(*KK^ z{0Ay3{=d-wkp3G5SXiVz9qsKUb=)1HuAbJO|GZL~|Nqqf4M_95+u2J(pdPmFj&A=n z|3mm!^#w#(7GIkGf8IhCe-P&j-rrAvx*FuUfiK!Yu5lEjAVbEI*DHagL#0!D&a?=s)hT)S%_ZQeGD#i82v~%{)Xo*GN1NB$S zGE#wuIn7f3V$3-s`?<|!e~#wI#Znp`AShS&P$pjH`3ntu&0@UakNH7y=5Qe6Urp&uM1}>d70>iu0SNnJck-uU6U?{R4-o zU=iY9comjG=byX60^PFv>OY1>d->}yT+j5!qqx@k?WbSE_ahYF`BIm|Up)rWupig7 z*EyzKeO=;T8$2w%&+1%Uo1yJ_=Xa|%pZ@$~@J$)Scdo??6*(1YSe(2zX09A}JjNhA zcr%j|+N?3{Z%oR;^!+f7o6XTHZi5pmtxxvkR$G7v`XOU-G7eB%HMA?vHr%UE5^w-D<&P z9zn=WT|j?pZ*rA_%99u~u(xv$!xA`ie0XBHH7Kxh)x;7fSe8{DN0jur$VMMu8hMZ< z0XkmMFvSvz4KW+#kV8MhFS0=7H0N8D(`e3S-&~J&kw%~?C=OcuQvlTPMQBO+XT zBBWCZMx=W0A!6+jHFzZ$^)KwfmZ@DKcmCv$`hIYzjbyQmU_PE4Rm6Yf&u-1mx8{m;J@faP{30W22)q@A#9_@+rrzTAs;u zgYm@BZ78&ifD0_5HQD~i6sk{%Sb|lvPAn79nEJMf=ZX5o&ilbEwC&nEy%)I@Ltphd zyryC($IjBWhbs+0otW>oR)vE_b~}Bk_XAa+-9*0kr*9-_7#p>%+^c8+pV0orG&6#| z+9#M>{#kzvD~-QjiOb=b^u$O0<>IIr2%5^V?=L~XXhZefiW_tSV^xT{7kOK3+wZya z6mx660xv6jZO3P)aFsa~S35b1nLRB#=X3#VjaFgB;P6X8TrHo6-Ry`2R7uhp~23_8=Tbw6wntoM5f zM6rU$f8WB6ei1ge{*EEnF6z{5`q4pwIfDdX(NqQKhmIn9Fv~z&aUOZUt7eD7kG3cdiGBICFWxL#n>c-!}^>u+f(dY@uG_S&c( z#AaV`IHqIy9Xt!6e>AY$6EB>-z^UX}yFJc*0-A>>l2nU}ZU@n33?26L-82Me?V>-v z1n+Gn$9`L7dPLg=2W6yj*K}hPJVzQ*51=25EP=w}Zj}T~QP0df!U!Lnbob#-5#k!MbD8;6hPS;7hftL; zMsJ9I_U_!2wwK>gK_$VDj47%Sje2ZUL2qKVYkcbYkc`I1mVPrbYIF}|1K>?9zJ${H z5DS9`uCpibN_k?MWXKa!Hj-;EQJ|BH+1S4$@@XZf{^eH1EE;2rCZ1>*=``8+*zlK) ztYaX<6UCdHZnbO$t5b&cbHmDg|CY_R$%*7|bY3+EyA_|q2mA!HeD(^_|2oJXa(J(z zIBGTM*vF*Is>gz}1XTC3fu+2T)EN<1I%3%zg89c030*<(j_Ne_qorRt&sf54>b^hJjOKahBAhW9$4Hu-p@GCuBa8-8a&W^rcEG6!MHPM2|X06AEo@9 z_aU-pmlvc)M#2TL@wIRD3W8XJwK4|-;*q00i1QWPWEx_0HM!&*Kn=0~HI0X!mg`bV zpR=@#rvk3#c96KXz>In5j_`V+_53QfB+5ykLsbX&><=`Ft!tx{R-KHzyO|?gx^m3f z&Vr1@RV477_tksKMD-C!jKoV)j}~llOOo9AlpRu+N(kDwtZp<9myh~b_6}KvT>L8s z%YLm241J;B%rSiZflk+5NN`)<4PIt%s2X*KQ%ZYIIPeS8e~LDGzk~}Q@7n0P{IR@v z;?a}l-EKfZ&hd8~SCjgjW@V$ajh_Z)_NJ|-#E^N(h?Q)xnlrxsECwr3D>eY{1k=o* zn6>x5&Q`A!>o)D(zdpwo3i)R2;MBn$KUy5Qy_o-s=hFxsP-Kst5_igrwv-)ZtYu+T z={;6=IAS%mxD)e*f2=^Nj?m)5UzHZczx546H_!j>!tx8~!S~LlN)r-vW}sqsbYy1= za|PpVG0&MB5gWxV#e^67sqWdDv+-|Yr=TrpW___tkUTSDM|G!_b&tIQobktyPYwf$ z1?A_T@I2@N{Z;c#K^4b0qyo5PDt=zY2Dq9++YIeNKO@2iR!0I2We)lAvUyk}s93qI z1Vj2NTAH*hvZ|yD22+0oWmsV{;yF-pI#>xNAAOB*%N4B-nr%>-#j}MY7*2;bRj49s zZ4ugCB^sAd-%a)iW5dx;>(Qi>n||_@7}XmX!9MjWhGq*_8GfiLE9GZ#_=NWw*!NQJE+oqmax1L!)}XiH sF+_C_>2) & 3` `Regenerated_Size = (Literals_Section_Header[0]>>4) + (Literals_Section_Header[1]<<4) + (Literals_Section_Header[2]<<12)` Only Stream1 is present for these cases. -Note : it's allowed to represent a short value (for example `13`) +Note : it's allowed to represent a short value (for example `27`) using a long format, even if it's less efficient. __`Size_Format` for `Compressed_Literals_Block` and `Treeless_Literals_Block`__ : @@ -521,19 +521,33 @@ __`Size_Format` for `Compressed_Literals_Block` and `Treeless_Literals_Block`__ Both `Regenerated_Size` and `Compressed_Size` use 10 bits (0-1023). `Literals_Section_Header` uses 3 bytes. - `Size_Format` == 01 : 4 streams. - Both `Regenerated_Size` and `Compressed_Size` use 10 bits (0-1023). + Both `Regenerated_Size` and `Compressed_Size` use 10 bits (6-1023). `Literals_Section_Header` uses 3 bytes. - `Size_Format` == 10 : 4 streams. - Both `Regenerated_Size` and `Compressed_Size` use 14 bits (0-16383). + Both `Regenerated_Size` and `Compressed_Size` use 14 bits (6-16383). `Literals_Section_Header` uses 4 bytes. - `Size_Format` == 11 : 4 streams. - Both `Regenerated_Size` and `Compressed_Size` use 18 bits (0-262143). + Both `Regenerated_Size` and `Compressed_Size` use 18 bits (6-262143). `Literals_Section_Header` uses 5 bytes. Both `Compressed_Size` and `Regenerated_Size` fields follow __little-endian__ convention. Note: `Compressed_Size` __includes__ the size of the Huffman Tree description _when_ it is present. +4 streams is superior to 1 stream in decompression speed, +by exploiting instruction level parallelism. +But it's also more expensive, +costing on average ~7.3 bytes more than the 1 stream mode, mostly from the jump table. + +In general, use the 4 streams mode when there are more literals to decode, +to favor higher decompression speeds. +Beyond 1KB, the 4 streams mode is compulsory anyway. + +Note that a minimum of 6 bytes is required for the 4 streams mode. +That's a technical minimum, but it's not recommended to employ the 4 streams mode +for such a small quantity, that would be wasteful. +A more practical lower bound would be around ~256 bytes. + #### Raw Literals Block The data in Stream1 is `Regenerated_Size` bytes long, it contains the raw literals data to be used during [Sequence Execution]. @@ -945,14 +959,14 @@ sequences are applied to them: |:--------------:|:-----------------:|:------------------:|:------------------:|:------------------:|:-----------------------:| | | | 1 | 4 | 8 | starting values | | 1114 | 11 | 1111 | 1 | 4 | non-repeat | -| 1 | 22 | 1111 | 1 | 4 | repeat 1; no change | +| 1 | 22 | 1111 | 1 | 4 | repeat 1: no change | | 2225 | 22 | 2222 | 1111 | 1 | non-repeat | | 1114 | 111 | 1111 | 2222 | 1111 | non-repeat | | 3336 | 33 | 3333 | 1111 | 2222 | non-repeat | -| 2 | 22 | 1111 | 3333 | 2222 | repeat 2; swap 1 & 2 | -| 3 | 33 | 2222 | 1111 | 3333 | repeat 3; rotate 3 to 1 | -| 3 | 0 | 2221 | 2222 | 1111 | insert resolved offset | -| 1 | 0 | 2222 | 2221 | 3333 | repeat 2 | +| 2 | 22 | 1111 | 3333 | 2222 | repeat 2: swap 1 & 2 | +| 3 | 33 | 2222 | 1111 | 3333 | repeat 3: rotate 3 to 1 | +| 3 | 0 | 2221 | 2222 | 1111 | special case : insert `repeat1 - 1` | +| 1 | 0 | 2222 | 2221 | 1111 | == repeat 2 | Skippable Frames @@ -967,14 +981,14 @@ into a flow of concatenated frames. Skippable frames defined in this specification are compatible with [LZ4] ones. -[LZ4]:http://www.lz4.org +[LZ4]:https://lz4.github.io/lz4/ From a compliant decoder perspective, skippable frames need just be skipped, and their content ignored, resuming decoding after the skippable frame. It can be noted that a skippable frame can be used to watermark a stream of concatenated frames -embedding any kind of tracking information (even just an UUID). +embedding any kind of tracking information (even just a UUID). Users wary of such possibility should scan the stream of concatenated frames in an attempt to detect such frame for analysis or removal. @@ -1124,7 +1138,7 @@ These symbols define a full state reset, reading `Accuracy_Log` bits. Then, all remaining symbols, sorted in natural order, are allocated cells. Starting from symbol `0` (if it exists), and table position `0`, each symbol gets allocated as many cells as its probability. -Cell allocation is spreaded, not linear : +Cell allocation is spread, not linear : each successor position follows this rule : ``` @@ -1669,7 +1683,7 @@ or at least provide a meaningful error code explaining for which reason it canno Version changes --------------- -- 0.3.7 : clarifications for Repeat_Offsets +- 0.3.7 : clarifications for Repeat_Offsets, matching RFC8878 - 0.3.6 : clarifications for Dictionary_ID - 0.3.5 : clarifications for Block_Maximum_Size - 0.3.4 : clarifications for FSE decoding table diff --git a/priv/zstd/doc/zstd_manual.html b/priv/zstd/doc/zstd_manual.html index 010f10a..c33f0e2 100644 --- a/priv/zstd/doc/zstd_manual.html +++ b/priv/zstd/doc/zstd_manual.html @@ -1,10 +1,10 @@ -zstd 1.5.0 Manual +zstd 1.5.4 Manual -

zstd 1.5.0 Manual

+

zstd 1.5.4 Manual

Contents

    @@ -121,7 +121,9 @@

    zstd 1.5.0 Manual

    note 6 : This function replaces ZSTD_getDecompressedSize()


    -
    unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
+
    ZSTD_DEPRECATED("Replaced by ZSTD_getFrameContentSize")
+ZSTDLIB_API
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
 
      NOTE: This function is now obsolete, in favor of ZSTD_getFrameContentSize().
   Both functions work the same way, but ZSTD_getDecompressedSize() blends
   "empty", "unknown" and "error" results to the same return value (0),
@@ -137,8 +139,30 @@ 
    zstd 1.5.0 Manual
    
         or an error code if input is invalid 
 

    
 
-
    Helper functions
    #define ZSTD_COMPRESSBOUND(srcSize)   ((srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
-size_t      ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+
    Helper functions
    /* ZSTD_compressBound() :
+ * maximum compressed size in worst case single-pass scenario.
+ * When invoking `ZSTD_compress()` or any other one-pass compression function,
+ * it's recommended to provide @dstCapacity >= ZSTD_compressBound(srcSize)
+ * as it eliminates one potential failure scenario,
+ * aka not enough room in dst buffer to write the compressed frame.
+ * Note : ZSTD_compressBound() itself can fail, if @srcSize > ZSTD_MAX_INPUT_SIZE .
+ *        In which case, ZSTD_compressBound() will return an error code
+ *        which can be tested using ZSTD_isError().
+ *
+ * ZSTD_COMPRESSBOUND() :
+ * same as ZSTD_compressBound(), but as a macro.
+ * It can be used to produce constants, which can be useful for static allocation,
+ * for example to size a static array on stack.
+ * Will produce constant value 0 if srcSize too large.
+ */
+#define ZSTD_MAX_INPUT_SIZE ((sizeof(size_t)==8) ? 0xFF00FF00FF00FF00LLU : 0xFF00FF00U)
+#define ZSTD_COMPRESSBOUND(srcSize)   (((size_t)(srcSize) >= ZSTD_MAX_INPUT_SIZE) ? 0 : (srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
+size_t ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+/* ZSTD_isError() :
+ * Most ZSTD_* functions returning a size_t value can be tested for error,
+ * using ZSTD_isError().
+ * @return 1 if error, 0 otherwise
+ */
 unsigned    ZSTD_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
 const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
 int         ZSTD_minCLevel(void);               /*!< minimum negative compression level allowed, requires v1.4.0+ */
@@ -357,8 +381,11 @@ 
    Decompression context
      When decompressing many times,
      * ZSTD_c_stableOutBuffer
      * ZSTD_c_blockDelimiters
      * ZSTD_c_validateSequences
-     * ZSTD_c_splitBlocks
+     * ZSTD_c_useBlockSplitter
      * ZSTD_c_useRowMatchFinder
+     * ZSTD_c_prefetchCDictTables
+     * ZSTD_c_enableSeqProducerFallback
+     * ZSTD_c_maxBlockSize
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -377,7 +404,11 @@ 
    Decompression context
      When decompressing many times,
      ZSTD_c_experimentalParam12=1009,
      ZSTD_c_experimentalParam13=1010,
      ZSTD_c_experimentalParam14=1011,
-     ZSTD_c_experimentalParam15=1012
+     ZSTD_c_experimentalParam15=1012,
+     ZSTD_c_experimentalParam16=1013,
+     ZSTD_c_experimentalParam17=1014,
+     ZSTD_c_experimentalParam18=1015,
+     ZSTD_c_experimentalParam19=1016
 } ZSTD_cParameter;
 

    
 
    typedef struct {
@@ -440,7 +471,7 @@ 
    Decompression context
      When decompressing many times,
                   They will be used to compress next frame.
                   Resetting session never fails.
   - The parameters : changes all parameters back to "default".
-                  This removes any reference to any dictionary too.
+                  This also removes any reference to any dictionary or external sequence producer.
                   Parameters can only be changed between 2 sessions (i.e. no compression is currently ongoing)
                   otherwise the reset fails, and function returns an error value (which can be tested using ZSTD_isError())
   - Both : similar to resetting the session, followed by resetting parameters.
@@ -479,13 +510,15 @@ 
    Decompression context
      When decompressing many times,
      * ZSTD_d_stableOutBuffer
      * ZSTD_d_forceIgnoreChecksum
      * ZSTD_d_refMultipleDDicts
+     * ZSTD_d_disableHuffmanAssembly
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly
      */
      ZSTD_d_experimentalParam1=1000,
      ZSTD_d_experimentalParam2=1001,
      ZSTD_d_experimentalParam3=1002,
-     ZSTD_d_experimentalParam4=1003
+     ZSTD_d_experimentalParam4=1003,
+     ZSTD_d_experimentalParam5=1004
 
 } ZSTD_dParameter;
 

    
@@ -654,6 +687,9 @@ 
    Streaming compression functions
    typedef enum {
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
      ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
      ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+
+ Note that ZSTD_initCStream() clears any previously set dictionary. Use the new API
+ to compress with a dictionary.
  
 

    
 
@@ -689,6 +725,33 @@ 
    ZSTD_DStream management functions
    ZSTD_DStream* ZSTD_
 size_t ZSTD_freeDStream(ZSTD_DStream* zds);  /* accept NULL pointer */
 

    
 
    Streaming decompression functions

    
+
    size_t ZSTD_initDStream(ZSTD_DStream* zds);
+
     Initialize/reset DStream state for new decompression operation.
+ Call before new decompression operation using same DStream.
+
+ Note : This function is redundant with the advanced API and equivalent to:
+     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
+     ZSTD_DCtx_refDDict(zds, NULL);
+ 
+

    
+
+
    size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+
     Streaming decompression function.
+ Call repetitively to consume full input updating it as necessary.
+ Function will update both input and output `pos` fields exposing current state via these fields:
+ - `input.pos < input.size`, some input remaining and caller should provide remaining input
+   on the next call.
+ - `output.pos < output.size`, decoder finished and flushed all remaining buffers.
+ - `output.pos == output.size`, potentially uncflushed data present in the internal buffers,
+   call ZSTD_decompressStream() again to flush remaining data to output.
+ Note : with no additional input, amount of data flushed <= ZSTD_BLOCKSIZE_MAX.
+
+ @return : 0 when a frame is completely decoded and fully flushed,
+           or an error code, which can be tested using ZSTD_isError(),
+           or any other value > 0, which means there is some decoding or flushing to do to complete current frame.
+ 
+

    
+
 
    size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
 

    
 
    size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output buffer. Guarantee to successfully flush at least one complete block in all circumstances. */
@@ -794,7 +857,7 @@ 
    Streaming decompression functions

    
   If @return == 0, the dictID could not be decoded.
   This could for one of the following reasons :
   - The frame does not require a dictionary to be decoded (most common case).
-  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden information.
+  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden piece of information.
     Note : this use case also happens when using a non-conformant dictionary.
   - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
   - This is not a Zstandard frame.
@@ -803,7 +866,7 @@ 
    Streaming decompression functions

    
 
 
    Advanced dictionary and prefix API (Requires v1.4.0+)
      This API allows dictionaries to be used with ZSTD_compress2(),
- ZSTD_compressStream2(), and ZSTD_decompress(). Dictionaries are sticky, and
+ ZSTD_compressStream2(), and ZSTD_decompressDCtx(). Dictionaries are sticky, and
  only reset with the context is reset with ZSTD_reset_parameters or
  ZSTD_reset_session_and_parameters. Prefixes are single-use.
 
    
@@ -814,8 +877,9 @@ 
    Streaming decompression functions

    
  @result : 0, or an error code (which can be tested with ZSTD_isError()).
   Special: Loading a NULL (or 0-size) dictionary invalidates previous dictionary,
            meaning "return to no-dictionary mode".
-  Note 1 : Dictionary is sticky, it will be used for all future compressed frames.
-           To return to "no-dictionary" situation, load a NULL dictionary (or reset parameters).
+  Note 1 : Dictionary is sticky, it will be used for all future compressed frames,
+           until parameters are reset, a new dictionary is loaded, or the dictionary
+           is explicitly invalidated by loading a NULL dictionary.
   Note 2 : Loading a dictionary involves building tables.
            It's also a CPU consuming operation, with non-negligible impact on latency.
            Tables are dependent on compression parameters, and for this reason,
@@ -828,7 +892,7 @@ 
    Streaming decompression functions

    
 

    
 
 
    size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
-
      Reference a prepared dictionary, to be used for all next compressed frames.
+
      Reference a prepared dictionary, to be used for all future compressed frames.
   Note that compression parameters are enforced from within CDict,
   and supersede any compression parameter previously set within CCtx.
   The parameters ignored are labelled as "superseded-by-cdict" in the ZSTD_cParameter enum docs.
@@ -863,9 +927,9 @@ 
    Streaming decompression functions

    
 

    
 
 
    size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
-
      Create an internal DDict from dict buffer,
-  to be used to decompress next frames.
-  The dictionary remains valid for all future frames, until explicitly invalidated.
+
      Create an internal DDict from dict buffer, to be used to decompress all future frames.
+  The dictionary remains valid for all future frames, until explicitly invalidated, or
+  a new dictionary is loaded.
  @result : 0, or an error code (which can be tested with ZSTD_isError()).
   Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
             meaning "return to no-dictionary mode".
@@ -889,9 +953,10 @@ 
    Streaming decompression functions

    
   The memory for the table is allocated on the first call to refDDict, and can be
   freed with ZSTD_freeDCtx().
 
+  If called with ZSTD_d_refMultipleDDicts disabled (the default), only one dictionary
+  will be managed, and referencing a dictionary effectively "discards" any previous one.
+
  @result : 0, or an error code (which can be tested with ZSTD_isError()).
-  Note 1 : Currently, only one dictionary can be managed.
-           Referencing a new dictionary effectively "discards" any previous one.
   Special: referencing a NULL DDict means "return to no-dictionary mode".
   Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
  
@@ -1072,14 +1137,18 @@ 
    Streaming decompression functions

    
 } ZSTD_literalCompressionMode_e;
 

    
 
    typedef enum {
-  ZSTD_urm_auto = 0,                   /* Automatically determine whether or not we use row matchfinder */
-  ZSTD_urm_disableRowMatchFinder = 1,  /* Never use row matchfinder */
-  ZSTD_urm_enableRowMatchFinder = 2    /* Always use row matchfinder when applicable */
-} ZSTD_useRowMatchFinderMode_e;
+  /* Note: This enum controls features which are conditionally beneficial. Zstd typically will make a final
+   * decision on whether or not to enable the feature (ZSTD_ps_auto), but setting the switch to ZSTD_ps_enable
+   * or ZSTD_ps_disable allow for a force enable/disable the feature.
+   */
+  ZSTD_ps_auto = 0,         /* Let the library automatically determine whether the feature shall be enabled */
+  ZSTD_ps_enable = 1,       /* Force-enable the feature */
+  ZSTD_ps_disable = 2       /* Do not use the feature */
+} ZSTD_paramSwitch_e;
 

    
 
    Frame size functions
    
 
-
    unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
+
    ZSTDLIB_STATIC_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
 
      `src` should point to the start of a series of ZSTD encoded and/or skippable frames
   `srcSize` must be the _exact_ size of this series
        (i.e. there should be a frame boundary at `src + srcSize`)
@@ -1102,7 +1171,7 @@ 
    Streaming decompression functions

    
             however it does mean that all frame data must be present and valid. 
 

    
 
-
    unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize);
+
    ZSTDLIB_STATIC_API unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize);
 
      `src` should point to the start of a series of ZSTD encoded and/or skippable frames
   `srcSize` must be the _exact_ size of this series
        (i.e. there should be a frame boundary at `src + srcSize`)
@@ -1117,26 +1186,80 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
+
    ZSTDLIB_STATIC_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
 
      srcSize must be >= ZSTD_FRAMEHEADERSIZE_PREFIX.
  @return : size of the Frame Header,
            or an error code (if srcSize is too small) 
 

    
 
+
    ZSTDLIB_STATIC_API size_t ZSTD_decompressionMargin(const void* src, size_t srcSize);
+
     Zstd supports in-place decompression, where the input and output buffers overlap.
+ In this case, the output buffer must be at least (Margin + Output_Size) bytes large,
+ and the input buffer must be at the end of the output buffer.
+
+  _______________________ Output Buffer ________________________
+ |                                                              |
+ |                                        ____ Input Buffer ____|
+ |                                       |                      |
+ v                                       v                      v
+ |---------------------------------------|-----------|----------|
+ ^                                                   ^          ^
+ |___________________ Output_Size ___________________|_ Margin _|
+
+ NOTE: See also ZSTD_DECOMPRESSION_MARGIN().
+ NOTE: This applies only to single-pass decompression through ZSTD_decompress() or
+ ZSTD_decompressDCtx().
+ NOTE: This function supports multi-frame input.
+
+ @param src The compressed frame(s)
+ @param srcSize The size of the compressed frame(s)
+ @returns The decompression margin or an error that can be checked with ZSTD_isError().
+ 
+

    
+
+
    #define ZSTD_DECOMPRESSION_MARGIN(originalSize, blockSize) ((size_t)(                                              \
+        ZSTD_FRAMEHEADERSIZE_MAX                                                              /* Frame header */ + \
+        4                                                                                         /* checksum */ + \
+        ((originalSize) == 0 ? 0 : 3 * (((originalSize) + (blockSize) - 1) / blockSize)) /* 3 bytes per block */ + \
+        (blockSize)                                                                    /* One block of margin */   \
+    ))
+
     Similar to ZSTD_decompressionMargin(), but instead of computing the margin from
+ the compressed frame, compute it from the original size and the blockSizeLog.
+ See ZSTD_decompressionMargin() for details.
+
+ WARNING: This macro does not support multi-frame input, the input must be a single
+ zstd frame. If you need that support use the function, or implement it yourself.
+
+ @param originalSize The original uncompressed size of the data.
+ @param blockSize    The block size == MIN(windowSize, ZSTD_BLOCKSIZE_MAX).
+                     Unless you explicitly set the windowLog smaller than
+                     ZSTD_BLOCKSIZELOG_MAX you can just use ZSTD_BLOCKSIZE_MAX.
+ 
+

    
+
 
    typedef enum {
   ZSTD_sf_noBlockDelimiters = 0,         /* Representation of ZSTD_Sequence has no block delimiters, sequences only */
   ZSTD_sf_explicitBlockDelimiters = 1    /* Representation of ZSTD_Sequence contains explicit block delimiters */
 } ZSTD_sequenceFormat_e;
 

    
-
     Generate sequences using ZSTD_compress2, given a source buffer.
+
    ZSTDLIB_STATIC_API size_t ZSTD_sequenceBound(size_t srcSize);
+
     `srcSize` : size of the input buffer
+  @return : upper-bound for the number of sequences that can be generated
+            from a buffer of srcSize bytes
+
+  note : returns number of sequences - to get bytes, multiply by sizeof(ZSTD_Sequence).
+ 
+

    
+
+
     Generate sequences using ZSTD_compress2(), given a source buffer.
 
  Each block will end with a dummy sequence
  with offset == 0, matchLength == 0, and litLength == length of last literals.
  litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
  simply acts as a block delimiter.
 
- zc can be used to insert custom compression params.
- This function invokes ZSTD_compress2
+ @zc can be used to insert custom compression params.
+ This function invokes ZSTD_compress2().
 
  The output of this function can be fed into ZSTD_compressSequences() with CCtx
  setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
@@ -1144,9 +1267,9 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
+
    ZSTDLIB_STATIC_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
 
     Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
- by merging them into into the literals of the next sequence.
+ by merging them into the literals of the next sequence.
 
  As such, the final generated result has no explicit representation of block boundaries,
  and the final last literals segment is not represented in the sequences.
@@ -1157,10 +1280,13 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstSize,
-                      const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
-                      const void* src, size_t srcSize);
-
     Compress an array of ZSTD_Sequence, generated from the original source buffer, into dst.
+
    ZSTDLIB_STATIC_API size_t
+ZSTD_compressSequences( ZSTD_CCtx* cctx, void* dst, size_t dstSize,
+            const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+            const void* src, size_t srcSize);
+
     Compress an array of ZSTD_Sequence, associated with @src buffer, into dst.
+ @src contains the entire input (not just the literals).
+ If @srcSize > sum(sequence.length), the remaining bytes are considered all literals
  If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
  The entire source is compressed into a single frame.
 
@@ -1185,15 +1311,15 @@ 
    Streaming decompression functions

    
  Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
  Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
          and cannot emit an RLE block that disagrees with the repcode history
- @return : final compressed size or a ZSTD error.
+ @return : final compressed size, or a ZSTD error code.
  
 

    
 
-
    size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
+
    ZSTDLIB_STATIC_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
                                 const void* src, size_t srcSize, unsigned magicVariant);
 
     Generates a zstd skippable frame containing data given by src, and writes it to dst buffer.
 
- Skippable frames begin with a a 4-byte magic number. There are 16 possible choices of magic number,
+ Skippable frames begin with a 4-byte magic number. There are 16 possible choices of magic number,
  ranging from ZSTD_MAGIC_SKIPPABLE_START to ZSTD_MAGIC_SKIPPABLE_START+15.
  As such, the parameter magicVariant controls the exact skippable frame magic number variant used, so
  the magic number used will be ZSTD_MAGIC_SKIPPABLE_START + magicVariant.
@@ -1205,12 +1331,31 @@ 
    Streaming decompression functions

    
  
 

    
 
+
    size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
+                                const void* src, size_t srcSize);
+
     Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
+
+ The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
+ i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
+ in the magicVariant.
+
+ Returns an error if destination buffer is not large enough, or if the frame is not skippable.
+
+ @return : number of bytes written or a ZSTD error.
+ 
+

    
+
+
    unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size);
+
      Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
+ 
+

    
+
 
    Memory management
    
 
-
    size_t ZSTD_estimateCCtxSize(int compressionLevel);
-size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
-size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
-size_t ZSTD_estimateDCtxSize(void);
+
    ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDCtxSize(void);
 
      These functions make it possible to estimate memory usage
   of a future {D,C}Ctx, before its creation.
 
@@ -1230,16 +1375,19 @@ 
    Streaming decompression functions

    
   and ZSTD_estimateCCtxSize_usingCCtxParams(), which can be used in tandem with ZSTD_CCtxParams_setParameter().
   Both can be used to estimate memory using custom compression parameters and arbitrary srcSize limits.
 
-  Note 2 : only single-threaded compression is supported.
+  Note : only single-threaded compression is supported.
   ZSTD_estimateCCtxSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+
+  Note 2 : ZSTD_estimateCCtxSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+  Size estimates assume that no external sequence producer is registered.
  
 

    
 
-
    size_t ZSTD_estimateCStreamSize(int compressionLevel);
-size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
-size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
-size_t ZSTD_estimateDStreamSize(size_t windowSize);
-size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
+
    ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
 
      ZSTD_estimateCStreamSize() will provide a budget large enough for any compression level up to selected one.
   It will also consider src size to be arbitrarily "large", which is worst case.
   If srcSize is known to always be small, ZSTD_estimateCStreamSize_usingCParams() can provide a tighter estimation.
@@ -1251,20 +1399,25 @@ 
    Streaming decompression functions

    
   or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
   Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
          an internal ?Dict will be created, which additional size is not estimated here.
-         In this case, get total size by adding ZSTD_estimate?DictSize 
+         In this case, get total size by adding ZSTD_estimate?DictSize
+  Note 2 : only single-threaded compression is supported.
+  ZSTD_estimateCStreamSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+  Note 3 : ZSTD_estimateCStreamSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+  Size estimates assume that no external sequence producer is registered.
+ 
 

    
 
-
    size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
-size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
-size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
+
    ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
 
      ZSTD_estimateCDictSize() will bet that src size is relatively "small", and content is copied, like ZSTD_createCDict().
   ZSTD_estimateCDictSize_advanced() makes it possible to control compression parameters precisely, like ZSTD_createCDict_advanced().
   Note : dictionaries created by reference (`ZSTD_dlm_byRef`) are logically smaller.
  
 

    
 
-
    ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
-ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
+
    ZSTDLIB_STATIC_API ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
 
      Initialize an object using a pre-allocated fixed-size buffer.
   workspace: The memory area to emplace the object into.
              Provided pointer *must be 8-bytes aligned*.
@@ -1287,7 +1440,7 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
+
    ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
 

    
 
    typedef void* (*ZSTD_allocFunction) (void* opaque, size_t size);
 typedef void  (*ZSTD_freeFunction) (void* opaque, void* address);
@@ -1303,9 +1456,24 @@ 
    Streaming decompression functions

    
  
 

    
 
+
    typedef struct POOL_ctx_s ZSTD_threadPool;
+ZSTDLIB_STATIC_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
+ZSTDLIB_STATIC_API void ZSTD_freeThreadPool (ZSTD_threadPool* pool);  /* accept NULL pointer */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool);
+
      These prototypes make it possible to share a thread pool among multiple compression contexts.
+  This can limit resources for applications with multiple threads where each one uses
+  a threaded compression mode (via ZSTD_c_nbWorkers parameter).
+  ZSTD_createThreadPool creates a new thread pool with a given number of threads.
+  Note that the lifetime of such pool must exist while being used.
+  ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
+  to use an internal thread pool).
+  ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
+ 
+

    
+
 
    Advanced compression functions
    
 
-
    ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
+
    ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
 
      Create a digested dictionary for compression
   Dictionary content is just referenced, not duplicated.
   As a consequence, `dictBuffer` **must** outlive CDict,
@@ -1313,22 +1481,22 @@ 
    Streaming decompression functions

    
   note: equivalent to ZSTD_createCDict_advanced(), with dictLoadMethod==ZSTD_dlm_byRef 
 

    
 
-
    ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+
    ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
     @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
  `estimatedSrcSize` value is optional, select 0 if not known 
 

    
 
-
    ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+
    ZSTDLIB_STATIC_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
      same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
   All fields of `ZSTD_frameParameters` are set to default : contentSize=1, checksum=0, noDictID=0 
 

    
 
-
    size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
+
    ZSTDLIB_STATIC_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
 
      Ensure param values remain within authorized range.
  @return 0 on success, or an error code (can be checked with ZSTD_isError()) 
 

    
 
-
    ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
+
    ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
 
      optimize params for a given `srcSize` and `dictSize`.
  `srcSize` can be unknown, in which case use ZSTD_CONTENTSIZE_UNKNOWN.
  `dictSize` must be `0` when there is no dictionary.
@@ -1336,18 +1504,27 @@ 
    Streaming decompression functions

    
   This function never fails (wide contract) 
 

    
 
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams);
+
      Set all parameters provided within @cparams into the working @cctx.
+  Note : if modifying parameters during compression (MT mode only),
+         note that changes to the .windowLog parameter will be ignored.
+ @return 0 on success, or an error code (can be checked with ZSTD_isError()) 
+

    
+
 
    ZSTD_DEPRECATED("use ZSTD_compress2")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compress_advanced(ZSTD_CCtx* cctx,
-                              void* dst, size_t dstCapacity,
-                        const void* src, size_t srcSize,
-                        const void* dict,size_t dictSize,
-                              ZSTD_parameters params);
+                  void* dst, size_t dstCapacity,
+            const void* src, size_t srcSize,
+            const void* dict,size_t dictSize,
+                  ZSTD_parameters params);
 
      Note : this function is now DEPRECATED.
          It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_setParameter() and other parameter setters.
   This prototype will generate compilation warnings. 
 

    
 
 
    ZSTD_DEPRECATED("use ZSTD_compress2 with ZSTD_CCtx_loadDictionary")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
                                   void* dst, size_t dstCapacity,
                             const void* src, size_t srcSize,
@@ -1358,31 +1535,31 @@ 
    Streaming decompression functions

    
   This prototype will generate compilation warnings. 
 

    
 
-
    size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
 
      Same as ZSTD_CCtx_loadDictionary(), but dictionary content is referenced, instead of being copied into CCtx.
   It saves some memory, but also requires that `dict` outlives its usage within `cctx` 
 

    
 
-
    size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
 
      Same as ZSTD_CCtx_loadDictionary(), but gives finer control over
   how to load the dictionary (by copy ? by reference ?)
   and how to interpret it (automatic ? force raw mode ? full mode only ?) 
 

    
 
-
    size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
 
      Same as ZSTD_CCtx_refPrefix(), but gives finer control over
   how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) 
 

    
 
-
    size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value);
 
      Get the requested compression parameter value, selected by enum ZSTD_cParameter,
   and store it into int* value.
  @return : 0, or an error code (which can be tested with ZSTD_isError()).
  
 

    
 
-
    ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
-size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);  /* accept NULL pointer */
+
    ZSTDLIB_STATIC_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
+ZSTDLIB_STATIC_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);  /* accept NULL pointer */
 
      Quick howto :
   - ZSTD_createCCtxParams() : Create a ZSTD_CCtx_params structure
   - ZSTD_CCtxParams_setParameter() : Push parameters one by one into
@@ -1401,24 +1578,24 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
 
      Reset params to default values.
  
 

    
 
-
    size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
 
      Initializes the compression parameters of cctxParams according to
   compression level. All other parameters are reset to their default values.
  
 

    
 
-
    size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
 
      Initializes the compression and frame parameters of cctxParams according to
   params. All other parameters are reset to their default values.
  
 

    
 
-
    size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
 
      Similar to ZSTD_CCtx_setParameter.
   Set one compression parameter, selected by enum ZSTD_cParameter.
   Parameters must be applied to a ZSTD_CCtx using
@@ -1428,14 +1605,14 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params, ZSTD_cParameter param, int* value);
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params, ZSTD_cParameter param, int* value);
 
     Similar to ZSTD_CCtx_getParameter.
  Get the requested value of one compression parameter, selected by enum ZSTD_cParameter.
  @result : 0, or an error code (which can be tested with ZSTD_isError()).
  
 

    
 
-
    size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+
    ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
         ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
 
      Apply a set of ZSTD_CCtx_params to the compression context.
   This can be done even after compression is started,
@@ -1445,7 +1622,7 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_compressStream2_simpleArgs (
+
    ZSTDLIB_STATIC_API size_t ZSTD_compressStream2_simpleArgs (
                 ZSTD_CCtx* cctx,
                 void* dst, size_t dstCapacity, size_t* dstPos,
           const void* src, size_t srcSize, size_t* srcPos,
@@ -1459,40 +1636,40 @@ 
    Streaming decompression functions

    
 
 
    Advanced decompression functions
    
 
-
    unsigned ZSTD_isFrame(const void* buffer, size_t size);
+
    ZSTDLIB_STATIC_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
 
      Tells if the content of `buffer` starts with a valid Frame Identifier.
   Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
   Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
   Note 3 : Skippable Frame Identifiers are considered valid. 
 

    
 
-
    ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
+
    ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
 
      Create a digested dictionary, ready to start decompression operation without startup delay.
   Dictionary content is referenced, and therefore stays in dictBuffer.
   It is important that dictBuffer outlives DDict,
   it must remain read accessible throughout the lifetime of DDict 
 

    
 
-
    size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+
    ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
 
      Same as ZSTD_DCtx_loadDictionary(),
   but references `dict` content instead of copying it into `dctx`.
   This saves memory if `dict` remains around.,
   However, it's imperative that `dict` remains accessible (and unmodified) while being used, so it must outlive decompression. 
 

    
 
-
    size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
    ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
 
      Same as ZSTD_DCtx_loadDictionary(),
   but gives direct control over
   how to load the dictionary (by copy ? by reference ?)
   and how to interpret it (automatic ? force raw mode ? full mode only ?). 
 

    
 
-
    size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
    ZSTDLIB_STATIC_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
 
      Same as ZSTD_DCtx_refPrefix(), but gives finer control over
   how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) 
 

    
 
-
    size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
+
    ZSTDLIB_STATIC_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
 
      Refuses allocating internal buffers for frames requiring a window size larger than provided limit.
   This protects a decoder context from reserving too much memory for itself (potential attack scenario).
   This parameter is only useful in streaming mode, since no internal buffer is allocated in single-pass mode.
@@ -1501,7 +1678,7 @@ 
    Streaming decompression functions

    
  
 

    
 
-
    size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
+
    ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
 
      Get the requested decompression parameter value, selected by enum ZSTD_dParameter,
   and store it into int* value.
  @return : 0, or an error code (which can be tested with ZSTD_isError()).
@@ -1509,6 +1686,7 @@ 
    Streaming decompression functions

    
 

    
 
 
    ZSTD_DEPRECATED("use ZSTD_DCtx_setParameter() instead")
+ZSTDLIB_STATIC_API
 size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
 
      This function is REDUNDANT. Prefer ZSTD_DCtx_setParameter().
   Instruct the decoder context about what kind of data to decode next.
@@ -1517,7 +1695,7 @@ 
    Streaming decompression functions

    
  @return : 0, or an error code (which can be tested using ZSTD_isError()). 
 

    
 
-
    size_t ZSTD_decompressStream_simpleArgs (
+
    ZSTDLIB_STATIC_API size_t ZSTD_decompressStream_simpleArgs (
                 ZSTD_DCtx* dctx,
                 void* dst, size_t dstCapacity, size_t* dstPos,
           const void* src, size_t srcSize, size_t* srcPos);
@@ -1535,6 +1713,7 @@ 
    Streaming decompression functions

    
 
 
    Advanced Streaming compression functions

    
 
    ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
              int compressionLevel,
              unsigned long long pledgedSrcSize);
@@ -1552,6 +1731,7 @@ 
    Advanced Streaming compression functions

    
 
 
    ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
          const void* dict, size_t dictSize,
                int compressionLevel);
@@ -1569,6 +1749,7 @@ 
    Advanced Streaming compression functions

    
 
 
    ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
         const void* dict, size_t dictSize,
               ZSTD_parameters params,
@@ -1590,17 +1771,19 @@ 
    Advanced Streaming compression functions

    
 
 
    ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
 
     This function is DEPRECATED, and equivalent to:
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
      ZSTD_CCtx_refCDict(zcs, cdict);
- 
+
  note : cdict will just be referenced, and must outlive compression session
  This prototype will generate compilation warnings.
  
 

    
 
 
    ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
                    const ZSTD_CDict* cdict,
                          ZSTD_frameParameters fParams,
@@ -1622,6 +1805,7 @@ 
    Advanced Streaming compression functions

    
 
 
    ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
 
     This function is DEPRECATED, and is equivalent to:
      ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
@@ -1651,7 +1835,7 @@ 
    Advanced Streaming compression functions
    /* MT only : nb of workers actively compressing at probe time */
 } ZSTD_frameProgression;
 

    
-
    size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
+
    ZSTDLIB_STATIC_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
 
      Tell how many bytes are ready to be flushed immediately.
   Useful for multithreading scenarios (nbWorkers >= 1).
   Probe the oldest active job, defined as oldest job not yet entirely flushed,
@@ -1667,32 +1851,32 @@ 
    Advanced Streaming compression functions

    
 
 
    Advanced Streaming decompression functions

    
-
    size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
+
    ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_loadDictionary, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
 
    
      ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
      ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
 
  note: no dictionary will be used if dict == NULL or dictSize < 8
- Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
 

    
 
-
    size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
+
    ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_refDDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
 
    
      ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
      ZSTD_DCtx_refDDict(zds, ddict);
 
  note : ddict is referenced, it must outlive decompression session
- Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
 

    
 
-
    size_t ZSTD_resetDStream(ZSTD_DStream* zds);
+
    ZSTD_DEPRECATED("use ZSTD_DCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
 
    
      ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
 
  re-use decompression parameters from previous init; saves dictionary loading
- Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  
 

    
 
@@ -1710,7 +1894,6 @@ 
    Advanced Streaming decompression functions
    
 
   Start by initializing a context.
   Use ZSTD_compressBegin(), or ZSTD_compressBegin_usingDict() for dictionary compression.
-  It's also possible to duplicate a reference context which has already been initialized, using ZSTD_copyCCtx()
 
   Then, consume your input using ZSTD_compressContinue().
   There are some important considerations to keep in mind when using this advanced function :
@@ -1731,11 +1914,12 @@ 
    Advanced Streaming decompression functions
    
   `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress again.
 
    
 
-
    Buffer-less streaming compression functions
    size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
-size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
-size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
-size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+
    Buffer-less streaming compression functions
    ZSTDLIB_STATIC_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
 

    
+
    size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+

    
 
    size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
 

    
 
    Buffer-less streaming decompression (synchronous mode)
    @@ -1747,8 +1931,8 @@ 
    Buffer-less streaming compression functions
    size_t ZS
   Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
   Data fragment must be large enough to ensure successful decoding.
  `ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
-  @result : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
-           >0 : `srcSize` is too small, please provide at least @result bytes on next attempt.
+  result  : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
+           >0 : `srcSize` is too small, please provide at least result bytes on next attempt.
            errorCode, which can be tested using ZSTD_isError().
 
   It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
@@ -1767,7 +1951,7 @@ 
    Buffer-less streaming compression functions
    size_t ZS
 
   The most memory efficient way is to use a round buffer of sufficient size.
   Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
-  which can @return an error code if required value is too large for current system (in 32-bits mode).
+  which can return an error code if required value is too large for current system (in 32-bits mode).
   In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
   up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
   which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
@@ -1787,7 +1971,7 @@ 
    Buffer-less streaming compression functions
    size_t ZS
   ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
   ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
 
- @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
   It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
   It can also be an error code, which can be tested with ZSTD_isError().
 
@@ -1818,14 +2002,16 @@ 
    Buffer-less streaming decompression functions
    typedef
     unsigned headerSize;
     unsigned dictID;
     unsigned checksumFlag;
+    unsigned _reserved1;
+    unsigned _reserved2;
 } ZSTD_frameHeader;
 

    
-
    size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+
    ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
 /*! ZSTD_getFrameHeader_advanced() :
  *  same as ZSTD_getFrameHeader(),
  *  with added capability to select a format (like ZSTD_f_zstd1_magicless) */
-size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
-size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
+ZSTDLIB_STATIC_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
 
      decode Frame Header, or requires larger `srcSize`.
  @return : 0, `zfhPtr` is correctly filled,
           >0, `srcSize` is too small, value is wanted `srcSize` amount,
@@ -1845,7 +2031,6 @@ 
    Buffer-less streaming decompression functions
    typedef
     - It is necessary to init context before starting
       + compression : any ZSTD_compressBegin*() variant, including with dictionary
       + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
-      + copyCCtx() and copyDCtx() can be used too
     - Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
       + If input is larger than a block size, it's necessary to split input data into multiple blocks
       + For inputs larger than a single block, consider using regular ZSTD_compress() instead.
@@ -1861,10 +2046,36 @@ 
    Buffer-less streaming decompression functions
    typedef
         Use ZSTD_insertBlock() for such a case.
 

    
 
-
    Raw zstd block functions
    size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
-size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
+
    Raw zstd block functions
    ZSTDLIB_STATIC_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
+ZSTDLIB_STATIC_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
 

    
+
    ZSTDLIB_STATIC_API void
+ZSTD_registerSequenceProducer(
+  ZSTD_CCtx* cctx,
+  void* sequenceProducerState,
+  ZSTD_sequenceProducer_F* sequenceProducer
+);
+
     Instruct zstd to use a block-level external sequence producer function.
+
+ The sequenceProducerState must be initialized by the caller, and the caller is
+ responsible for managing its lifetime. This parameter is sticky across
+ compressions. It will remain set until the user explicitly resets compression
+ parameters.
+
+ Sequence producer registration is considered to be an "advanced parameter",
+ part of the "advanced API". This means it will only have an effect on compression
+ APIs which respect advanced parameters, such as compress2() and compressStream2().
+ Older compression APIs such as compressCCtx(), which predate the introduction of
+ "advanced parameters", will ignore any external sequence producer setting.
+
+ The sequence producer can be "cleared" by registering a NULL function pointer. This
+ removes all limitations described above in the "LIMITATIONS" section of the API docs.
+
+ The user is strongly encouraged to read the full API documentation (above) before
+ calling this function. 
+

    
+
 
 
diff --git a/priv/zstd/examples/Makefile b/priv/zstd/examples/Makefile
index 8d7361d..31f52d3 100644
--- a/priv/zstd/examples/Makefile
+++ b/priv/zstd/examples/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/common.h b/priv/zstd/examples/common.h
index 181f6b3..4873e87 100644
--- a/priv/zstd/examples/common.h
+++ b/priv/zstd/examples/common.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -21,6 +21,17 @@
 #include   // stat
 #include 
 
+
+/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
+#if defined(__GNUC__)
+#  define UNUSED_ATTR __attribute__((unused))
+#else
+#  define UNUSED_ATTR
+#endif
+
+#define HEADER_FUNCTION static UNUSED_ATTR
+
+
 /*
  * Define the returned error code from utility functions.
  */
@@ -57,7 +68,7 @@ typedef enum {
  * Check the zstd error code and die if an error occurred after printing a
  * message.
  */
-#define CHECK_ZSTD(fn, ...)                                      \
+#define CHECK_ZSTD(fn)                                           \
     do {                                                         \
         size_t const err = (fn);                                 \
         CHECK(!ZSTD_isError(err), "%s", ZSTD_getErrorName(err)); \
@@ -68,7 +79,7 @@ typedef enum {
  *
  * @return The size of a given file path.
  */
-static size_t fsize_orDie(const char *filename)
+HEADER_FUNCTION size_t fsize_orDie(const char *filename)
 {
     struct stat st;
     if (stat(filename, &st) != 0) {
@@ -96,7 +107,7 @@ static size_t fsize_orDie(const char *filename)
  * @return If successful this function will return a FILE pointer to an
  * opened file otherwise it sends an error to stderr and exits.
  */
-static FILE* fopen_orDie(const char *filename, const char *instruction)
+HEADER_FUNCTION FILE* fopen_orDie(const char *filename, const char *instruction)
 {
     FILE* const inFile = fopen(filename, instruction);
     if (inFile) return inFile;
@@ -108,7 +119,7 @@ static FILE* fopen_orDie(const char *filename, const char *instruction)
 /*! fclose_orDie() :
  * Close an opened file using given FILE pointer.
  */
-static void fclose_orDie(FILE* file)
+HEADER_FUNCTION void fclose_orDie(FILE* file)
 {
     if (!fclose(file)) { return; };
     /* error */
@@ -123,7 +134,7 @@ static void fclose_orDie(FILE* file)
  *
  * @return The number of bytes read.
  */
-static size_t fread_orDie(void* buffer, size_t sizeToRead, FILE* file)
+HEADER_FUNCTION size_t fread_orDie(void* buffer, size_t sizeToRead, FILE* file)
 {
     size_t const readSize = fread(buffer, 1, sizeToRead, file);
     if (readSize == sizeToRead) return readSize;   /* good */
@@ -143,7 +154,7 @@ static size_t fread_orDie(void* buffer, size_t sizeToRead, FILE* file)
  *
  * @return The number of bytes written.
  */
-static size_t fwrite_orDie(const void* buffer, size_t sizeToWrite, FILE* file)
+HEADER_FUNCTION size_t fwrite_orDie(const void* buffer, size_t sizeToWrite, FILE* file)
 {
     size_t const writtenSize = fwrite(buffer, 1, sizeToWrite, file);
     if (writtenSize == sizeToWrite) return sizeToWrite;   /* good */
@@ -159,7 +170,7 @@ static size_t fwrite_orDie(const void* buffer, size_t sizeToWrite, FILE* file)
  * cated memory.  If there is an error, this function will send that
  * error to stderr and exit.
  */
-static void* malloc_orDie(size_t size)
+HEADER_FUNCTION void* malloc_orDie(size_t size)
 {
     void* const buff = malloc(size);
     if (buff) return buff;
@@ -177,7 +188,7 @@ static void* malloc_orDie(size_t size)
  * @return If successful this function will load file into buffer and
  * return file size, otherwise it will printout an error to stderr and exit.
  */
-static size_t loadFile_orDie(const char* fileName, void* buffer, size_t bufferSize)
+HEADER_FUNCTION size_t loadFile_orDie(const char* fileName, void* buffer, size_t bufferSize)
 {
     size_t const fileSize = fsize_orDie(fileName);
     CHECK(fileSize <= bufferSize, "File too large!");
@@ -201,7 +212,8 @@ static size_t loadFile_orDie(const char* fileName, void* buffer, size_t bufferSi
  * @return If successful this function will return buffer and bufferSize(=fileSize),
  * otherwise it will printout an error to stderr and exit.
  */
-static void* mallocAndLoadFile_orDie(const char* fileName, size_t* bufferSize) {
+HEADER_FUNCTION void* mallocAndLoadFile_orDie(const char* fileName, size_t* bufferSize)
+{
     size_t const fileSize = fsize_orDie(fileName);
     *bufferSize = fileSize;
     void* const buffer = malloc_orDie(*bufferSize);
@@ -217,7 +229,7 @@ static void* mallocAndLoadFile_orDie(const char* fileName, size_t* bufferSize) {
  * Note: This function will send an error to stderr and exit if it
  * cannot write to a given file.
  */
-static void saveFile_orDie(const char* fileName, const void* buff, size_t buffSize)
+HEADER_FUNCTION void saveFile_orDie(const char* fileName, const void* buff, size_t buffSize)
 {
     FILE* const oFile = fopen_orDie(fileName, "wb");
     size_t const wSize = fwrite(buff, 1, buffSize, oFile);
diff --git a/priv/zstd/examples/dictionary_compression.c b/priv/zstd/examples/dictionary_compression.c
index 0eee650..83edc1c 100644
--- a/priv/zstd/examples/dictionary_compression.c
+++ b/priv/zstd/examples/dictionary_compression.c
@@ -1,12 +1,22 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
  * You may select, at your option, one of the above-listed licenses.
- */
+**/
+
+/* This example deals with Dictionary compression,
+ * its counterpart is `examples/dictionary_decompression.c` .
+ * These examples presume that a dictionary already exists.
+ * The main method to create a dictionary is `zstd --train`,
+ * look at the CLI documentation for details.
+ * Another possible method is to employ dictionary training API,
+ * published in `lib/zdict.h` .
+**/
+
 #include      // printf
 #include     // free
 #include     // memset, strcat
@@ -14,7 +24,7 @@
 #include "common.h"    // Helper functions, CHECK(), and CHECK_ZSTD()
 
 /* createDict() :
-   `dictFileName` is supposed to have been created using `zstd --train` */
+** `dictFileName` is supposed already created using `zstd --train` */
 static ZSTD_CDict* createCDict_orDie(const char* dictFileName, int cLevel)
 {
     size_t dictSize;
diff --git a/priv/zstd/examples/dictionary_decompression.c b/priv/zstd/examples/dictionary_decompression.c
index 107cfc1..e6c9999 100644
--- a/priv/zstd/examples/dictionary_decompression.c
+++ b/priv/zstd/examples/dictionary_decompression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/multiple_simple_compression.c b/priv/zstd/examples/multiple_simple_compression.c
index 5d2a28f..bf77ca1 100644
--- a/priv/zstd/examples/multiple_simple_compression.c
+++ b/priv/zstd/examples/multiple_simple_compression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/multiple_streaming_compression.c b/priv/zstd/examples/multiple_streaming_compression.c
index d4efc8e..b12ad03 100644
--- a/priv/zstd/examples/multiple_streaming_compression.c
+++ b/priv/zstd/examples/multiple_streaming_compression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/simple_compression.c b/priv/zstd/examples/simple_compression.c
index 27a65b1..7c88072 100644
--- a/priv/zstd/examples/simple_compression.c
+++ b/priv/zstd/examples/simple_compression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/simple_decompression.c b/priv/zstd/examples/simple_decompression.c
index 59c1fd4..f499156 100644
--- a/priv/zstd/examples/simple_decompression.c
+++ b/priv/zstd/examples/simple_decompression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/streaming_compression.c b/priv/zstd/examples/streaming_compression.c
index e20bcde..ed0a3a6 100644
--- a/priv/zstd/examples/streaming_compression.c
+++ b/priv/zstd/examples/streaming_compression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,9 +15,12 @@
 #include       // presumes zstd library is installed
 #include "common.h"    // Helper functions, CHECK(), and CHECK_ZSTD()
 
-
-static void compressFile_orDie(const char* fname, const char* outName, int cLevel)
+static void compressFile_orDie(const char* fname, const char* outName, int cLevel,
+                               int nbThreads)
 {
+    fprintf (stderr, "Starting compression of %s with level %d, using %d threads\n",
+             fname, cLevel, nbThreads);
+
     /* Open the input and output files. */
     FILE* const fin  = fopen_orDie(fname, "rb");
     FILE* const fout = fopen_orDie(outName, "wb");
@@ -39,7 +42,7 @@ static void compressFile_orDie(const char* fname, const char* outName, int cLeve
      */
     CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, cLevel) );
     CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1) );
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 4);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, nbThreads);
 
     /* This loop read from the input file, compresses that entire chunk,
      * and writes all output produced to the output file.
@@ -106,19 +109,32 @@ int main(int argc, const char** argv)
 {
     const char* const exeName = argv[0];
 
-    if (argc!=2) {
+    if (argc < 2) {
         printf("wrong arguments\n");
         printf("usage:\n");
-        printf("%s FILE\n", exeName);
+        printf("%s FILE [LEVEL] [THREADS]\n", exeName);
         return 1;
     }
 
+    int cLevel = 1;
+    int nbThreads = 4;
+
+    if (argc >= 3) {
+      cLevel = atoi (argv[2]);
+      CHECK(cLevel != 0, "can't parse LEVEL!");
+    }
+
+    if (argc >= 4) {
+      nbThreads = atoi (argv[3]);
+      CHECK(nbThreads != 0, "can't parse THREADS!");
+    }
+
     const char* const inFilename = argv[1];
 
     char* const outFilename = createOutFilename_orDie(inFilename);
-    compressFile_orDie(inFilename, outFilename, 1);
+    compressFile_orDie(inFilename, outFilename, cLevel, nbThreads);
 
     free(outFilename);   /* not strictly required, since program execution stops there,
-                          * but some static analyzer main complain otherwise */
+                          * but some static analyzer may complain otherwise */
     return 0;
 }
diff --git a/priv/zstd/examples/streaming_compression_thread_pool.c b/priv/zstd/examples/streaming_compression_thread_pool.c
index 5a6551b..a1a0241 100644
--- a/priv/zstd/examples/streaming_compression_thread_pool.c
+++ b/priv/zstd/examples/streaming_compression_thread_pool.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Martin Liska, SUSE, Facebook, Inc.
+ * Copyright (c) Martin Liska, SUSE, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -28,8 +28,10 @@ typedef struct compress_args
 
 static void *compressFile_orDie(void *data)
 {
+    const int nbThreads = 16;
+
     compress_args_t *args = (compress_args_t *)data;
-    fprintf (stderr, "Starting compression of %s with level %d\n", args->fname, args->cLevel);
+    fprintf (stderr, "Starting compression of %s with level %d, using %d threads\n", args->fname, args->cLevel, nbThreads);
     /* Open the input and output files. */
     FILE* const fin  = fopen_orDie(args->fname, "rb");
     FILE* const fout = fopen_orDie(args->outName, "wb");
@@ -56,9 +58,9 @@ static void *compressFile_orDie(void *data)
      */
     CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, args->cLevel) );
     CHECK_ZSTD( ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1) );
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 16);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, nbThreads);
 
-    /* This loop read from the input file, compresses that entire chunk,
+    /* This loop reads from the input file, compresses that entire chunk,
      * and writes all output produced to the output file.
      */
     size_t const toRead = buffInSize;
diff --git a/priv/zstd/examples/streaming_decompression.c b/priv/zstd/examples/streaming_decompression.c
index 6dc4c22..95fa112 100644
--- a/priv/zstd/examples/streaming_decompression.c
+++ b/priv/zstd/examples/streaming_decompression.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/examples/streaming_memory_usage.c b/priv/zstd/examples/streaming_memory_usage.c
index a5219ef..957acb6 100644
--- a/priv/zstd/examples/streaming_memory_usage.c
+++ b/priv/zstd/examples/streaming_memory_usage.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/Makefile b/priv/zstd/lib/Makefile
index 9109476..a4cf61a 100644
--- a/priv/zstd/lib/Makefile
+++ b/priv/zstd/lib/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -8,110 +8,13 @@
 # You may select, at your option, one of the above-listed licenses.
 # ################################################################
 
-# Note: by default, the static library is built single-threaded and dynamic library is built
-# multi-threaded. It is possible to force multi or single threaded builds by appending
-# -mt or -nomt to the build target (like lib-mt for multi-threaded, lib-nomt for single-threaded).
-.PHONY: default
-default: lib-release
-
-# define silent mode as default (verbose mode with V=1 or VERBOSE=1)
-$(V)$(VERBOSE).SILENT:
-
-# When cross-compiling from linux to windows,
-# one might need to specify TARGET_SYSTEM as "Windows."
-# Building from Fedora fails without it.
-# (but Ubuntu and Debian don't need to set anything)
-TARGET_SYSTEM ?= $(OS)
-
-# Version numbers
-LIBVER_MAJOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MAJOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < ./zstd.h`
-LIBVER_MINOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MINOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < ./zstd.h`
-LIBVER_PATCH_SCRIPT:=`sed -n '/define ZSTD_VERSION_RELEASE/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < ./zstd.h`
-LIBVER_SCRIPT:= $(LIBVER_MAJOR_SCRIPT).$(LIBVER_MINOR_SCRIPT).$(LIBVER_PATCH_SCRIPT)
-LIBVER_MAJOR := $(shell echo $(LIBVER_MAJOR_SCRIPT))
-LIBVER_MINOR := $(shell echo $(LIBVER_MINOR_SCRIPT))
-LIBVER_PATCH := $(shell echo $(LIBVER_PATCH_SCRIPT))
-LIBVER := $(shell echo $(LIBVER_SCRIPT))
-VERSION?= $(LIBVER)
-CCVER := $(shell $(CC) --version)
-
-# ZSTD_LIB_MINIFY is a helper variable that
-# configures a bunch of other variables to space-optimized defaults.
-ZSTD_LIB_MINIFY ?= 0
-ifneq ($(ZSTD_LIB_MINIFY), 0)
-  HAVE_CC_OZ ?= $(shell echo "" | $(CC) -Oz -x c -c - -o /dev/null 2> /dev/null && echo 1 || echo 0)
-  ZSTD_LEGACY_SUPPORT ?= 0
-  ZSTD_LIB_DEPRECATED ?= 0
-  HUF_FORCE_DECOMPRESS_X1 ?= 1
-  ZSTD_FORCE_DECOMPRESS_SHORT ?= 1
-  ZSTD_NO_INLINE ?= 1
-  ZSTD_STRIP_ERROR_STRINGS ?= 1
-ifneq ($(HAVE_CC_OZ), 0)
-    # Some compilers (clang) support an even more space-optimized setting.
-    CFLAGS += -Oz
-else
-    CFLAGS += -Os
-endif
-  CFLAGS += -fno-stack-protector -fomit-frame-pointer -fno-ident \
-            -DDYNAMIC_BMI2=0 -DNDEBUG
-else
-  CFLAGS += -O3
-endif
-
-DEBUGLEVEL ?= 0
-CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -DDEBUGLEVEL=$(DEBUGLEVEL)
-ifeq ($(TARGET_SYSTEM),Windows_NT)   # MinGW assumed
-  CPPFLAGS += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
-endif
-DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
-            -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
-            -Wstrict-prototypes -Wundef -Wpointer-arith \
-            -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
-            -Wredundant-decls -Wmissing-prototypes -Wc++-compat
-CFLAGS   += $(DEBUGFLAGS) $(MOREFLAGS)
-FLAGS     = $(CPPFLAGS) $(CFLAGS)
-
-CPPFLAGS_DYNLIB  = -DZSTD_MULTITHREAD # dynamic library build defaults to multi-threaded
-LDFLAGS_DYNLIB   = -pthread
-CPPFLAGS_STATLIB =                    # static library build defaults to single-threaded
-
-HAVE_COLORNEVER = $(shell echo a | grep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
-GREP_OPTIONS ?=
-ifeq ($HAVE_COLORNEVER, 1)
-  GREP_OPTIONS += --color=never
-endif
-GREP = grep $(GREP_OPTIONS)
-SED_ERE_OPT ?= -E
-
-ZSTDCOMMON_FILES := $(sort $(wildcard common/*.c))
-ZSTDCOMP_FILES := $(sort $(wildcard compress/*.c))
-ZSTDDECOMP_FILES := $(sort $(wildcard decompress/*.c))
-ZDICT_FILES := $(sort $(wildcard dictBuilder/*.c))
-ZDEPR_FILES := $(sort $(wildcard deprecated/*.c))
-ZSTD_FILES := $(ZSTDCOMMON_FILES)
-
-ifeq ($(findstring GCC,$(CCVER)),GCC)
-decompress/zstd_decompress_block.o : CFLAGS+=-fno-tree-vectorize
-endif
-
 # Modules
 ZSTD_LIB_COMPRESSION ?= 1
 ZSTD_LIB_DECOMPRESSION ?= 1
 ZSTD_LIB_DICTBUILDER ?= 1
 ZSTD_LIB_DEPRECATED ?= 0
 
-# Legacy support
-ZSTD_LEGACY_SUPPORT ?= 5
-ZSTD_LEGACY_MULTITHREADED_API ?= 0
-
-# Build size optimizations
-HUF_FORCE_DECOMPRESS_X1 ?= 0
-HUF_FORCE_DECOMPRESS_X2 ?= 0
-ZSTD_FORCE_DECOMPRESS_SHORT ?= 0
-ZSTD_FORCE_DECOMPRESS_LONG ?= 0
-ZSTD_NO_INLINE ?= 0
-ZSTD_STRIP_ERROR_STRINGS ?= 0
-
+# Input variables for libzstd.mk
 ifeq ($(ZSTD_LIB_COMPRESSION), 0)
   ZSTD_LIB_DICTBUILDER = 0
   ZSTD_LIB_DEPRECATED = 0
@@ -122,86 +25,46 @@ ifeq ($(ZSTD_LIB_DECOMPRESSION), 0)
   ZSTD_LIB_DEPRECATED = 0
 endif
 
+include libzstd.mk
+
+ZSTD_FILES := $(ZSTD_COMMON_FILES) $(ZSTD_LEGACY_FILES)
+
 ifneq ($(ZSTD_LIB_COMPRESSION), 0)
-  ZSTD_FILES += $(ZSTDCOMP_FILES)
+  ZSTD_FILES += $(ZSTD_COMPRESS_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DECOMPRESSION), 0)
-  ZSTD_FILES += $(ZSTDDECOMP_FILES)
+  ZSTD_FILES += $(ZSTD_DECOMPRESS_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DEPRECATED), 0)
-  ZSTD_FILES += $(ZDEPR_FILES)
+  ZSTD_FILES += $(ZSTD_DEPRECATED_FILES)
 endif
 
 ifneq ($(ZSTD_LIB_DICTBUILDER), 0)
-  ZSTD_FILES += $(ZDICT_FILES)
-endif
-
-ifneq ($(HUF_FORCE_DECOMPRESS_X1), 0)
-  CFLAGS += -DHUF_FORCE_DECOMPRESS_X1
-endif
-
-ifneq ($(HUF_FORCE_DECOMPRESS_X2), 0)
-  CFLAGS += -DHUF_FORCE_DECOMPRESS_X2
-endif
-
-ifneq ($(ZSTD_FORCE_DECOMPRESS_SHORT), 0)
-  CFLAGS += -DZSTD_FORCE_DECOMPRESS_SHORT
-endif
-
-ifneq ($(ZSTD_FORCE_DECOMPRESS_LONG), 0)
-  CFLAGS += -DZSTD_FORCE_DECOMPRESS_LONG
-endif
-
-ifneq ($(ZSTD_NO_INLINE), 0)
-  CFLAGS += -DZSTD_NO_INLINE
-endif
-
-ifneq ($(ZSTD_STRIP_ERROR_STRINGS), 0)
-  CFLAGS += -DZSTD_STRIP_ERROR_STRINGS
+  ZSTD_FILES += $(ZSTD_DICTBUILDER_FILES)
 endif
 
-ifneq ($(ZSTD_LEGACY_MULTITHREADED_API), 0)
-  CFLAGS += -DZSTD_LEGACY_MULTITHREADED_API
-endif
-
-ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
-ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
-  ZSTD_FILES += $(shell ls legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
-endif
-endif
-CPPFLAGS  += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
-
 ZSTD_LOCAL_SRC := $(notdir $(ZSTD_FILES))
-ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_SRC:.c=.o)
+ZSTD_LOCAL_OBJ0 := $(ZSTD_LOCAL_SRC:.c=.o)
+ZSTD_LOCAL_OBJ := $(ZSTD_LOCAL_OBJ0:.S=.o)
 
-ZSTD_SUBDIR := common compress decompress dictBuilder legacy deprecated
-vpath %.c $(ZSTD_SUBDIR)
+VERSION := $(ZSTD_VERSION)
 
-UNAME := $(shell uname)
+# Note: by default, the static library is built single-threaded and dynamic library is built
+# multi-threaded. It is possible to force multi or single threaded builds by appending
+# -mt or -nomt to the build target (like lib-mt for multi-threaded, lib-nomt for single-threaded).
+.PHONY: default
+default: lib-release
+
+CPPFLAGS_DYNLIB  += -DZSTD_MULTITHREAD # dynamic library build defaults to multi-threaded
+LDFLAGS_DYNLIB   += -pthread
+CPPFLAGS_STATLIB +=                    # static library build defaults to single-threaded
 
-ifndef BUILD_DIR
-ifeq ($(UNAME), Darwin)
-  ifeq ($(shell md5 < /dev/null > /dev/null; echo $$?), 0)
-    HASH ?= md5
-  endif
-else ifeq ($(UNAME), FreeBSD)
-  HASH ?= gmd5sum
-else ifeq ($(UNAME), NetBSD)
-  HASH ?= md5 -n
-else ifeq ($(UNAME), OpenBSD)
-  HASH ?= md5
-endif
-HASH ?= md5sum
 
-HASH_DIR = conf_$(shell echo $(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $(ZSTD_FILES) | $(HASH) | cut -f 1 -d " " )
-HAVE_HASH :=$(shell echo 1 | $(HASH) > /dev/null && echo 1 || echo 0)
-ifeq ($(HAVE_HASH),0)
-  $(info warning : could not find HASH ($(HASH)), needed to differentiate builds using different flags)
-  BUILD_DIR := obj/generic_noconf
+ifeq ($(findstring GCC,$(CCVER)),GCC)
+decompress/zstd_decompress_block.o : CFLAGS+=-fno-tree-vectorize
 endif
-endif # BUILD_DIR
 
 
 # macOS linker doesn't support -soname, and use different extension
@@ -212,19 +75,16 @@ ifeq ($(UNAME), Darwin)
   SHARED_EXT_VER = $(LIBVER).$(SHARED_EXT)
   SONAME_FLAGS = -install_name $(LIBDIR)/libzstd.$(SHARED_EXT_MAJOR) -compatibility_version $(LIBVER_MAJOR) -current_version $(LIBVER)
 else
-  SONAME_FLAGS = -Wl,-soname=libzstd.$(SHARED_EXT).$(LIBVER_MAJOR)
+  ifeq ($(UNAME), AIX)
+    SONAME_FLAGS =
+  else
+    SONAME_FLAGS = -Wl,-soname=libzstd.$(SHARED_EXT).$(LIBVER_MAJOR)
+  endif
   SHARED_EXT = so
   SHARED_EXT_MAJOR = $(SHARED_EXT).$(LIBVER_MAJOR)
   SHARED_EXT_VER = $(SHARED_EXT).$(LIBVER)
 endif
 
-SET_CACHE_DIRECTORY = \
-   +$(MAKE) --no-print-directory $@ \
-    BUILD_DIR=obj/$(HASH_DIR) \
-    CPPFLAGS="$(CPPFLAGS)" \
-    CFLAGS="$(CFLAGS)" \
-    LDFLAGS="$(LDFLAGS)"
-
 
 .PHONY: all
 all: lib
@@ -233,6 +93,13 @@ all: lib
 .PHONY: libzstd.a  # must be run every time
 libzstd.a: CPPFLAGS += $(CPPFLAGS_STATLIB)
 
+SET_CACHE_DIRECTORY = \
+   +$(MAKE) --no-print-directory $@ \
+    BUILD_DIR=obj/$(HASH_DIR) \
+    CPPFLAGS="$(CPPFLAGS)" \
+    CFLAGS="$(CFLAGS)" \
+    LDFLAGS="$(LDFLAGS)"
+
 ifndef BUILD_DIR
 # determine BUILD_DIR from compilation flags
 
@@ -343,6 +210,14 @@ $(ZSTD_STATLIB_DIR)/%.o : %.c $(ZSTD_STATLIB_DIR)/%.d | $(ZSTD_STATLIB_DIR)
 	@echo CC $@
 	$(COMPILE.c) $(DEPFLAGS) $(ZSTD_STATLIB_DIR)/$*.d $(OUTPUT_OPTION) $<
 
+$(ZSTD_DYNLIB_DIR)/%.o : %.S | $(ZSTD_DYNLIB_DIR)
+	@echo AS $@
+	$(COMPILE.S) $(OUTPUT_OPTION) $<
+
+$(ZSTD_STATLIB_DIR)/%.o : %.S | $(ZSTD_STATLIB_DIR)
+	@echo AS $@
+	$(COMPILE.S) $(OUTPUT_OPTION) $<
+
 MKDIR ?= mkdir
 $(BUILD_DIR) $(ZSTD_DYNLIB_DIR) $(ZSTD_STATLIB_DIR):
 	$(MKDIR) -p $@
@@ -374,9 +249,9 @@ clean:
 #-----------------------------------------------------------------------------
 # make install is validated only for below listed environments
 #-----------------------------------------------------------------------------
-ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
+ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku AIX))
 
-all: libzstd.pc
+lib: libzstd.pc
 
 HAS_EXPLICIT_EXEC_PREFIX := $(if $(or $(EXEC_PREFIX),$(exec_prefix)),1,)
 
@@ -421,16 +296,16 @@ INSTALL_PROGRAM ?= $(INSTALL)
 INSTALL_DATA    ?= $(INSTALL) -m 644
 
 
-libzstd.pc:
 libzstd.pc: libzstd.pc.in
 	@echo creating pkgconfig
-	@sed $(SED_ERE_OPT) \
+	@sed \
 	        -e 's|@PREFIX@|$(PREFIX)|' \
 	        -e 's|@EXEC_PREFIX@|$(PCEXEC_PREFIX)|' \
-          -e 's|@INCLUDEDIR@|$(PCINCPREFIX)$(PCINCDIR)|' \
-          -e 's|@LIBDIR@|$(PCLIBPREFIX)$(PCLIBDIR)|' \
-          -e 's|@VERSION@|$(VERSION)|' \
-          $< >$@
+	        -e 's|@INCLUDEDIR@|$(PCINCPREFIX)$(PCINCDIR)|' \
+	        -e 's|@LIBDIR@|$(PCLIBPREFIX)$(PCLIBDIR)|' \
+	        -e 's|@VERSION@|$(VERSION)|' \
+	        -e 's|@LIBS_PRIVATE@|$(LDFLAGS_DYNLIB)|' \
+	        $< >$@
 
 .PHONY: install
 install: install-pc install-static install-shared install-includes
diff --git a/priv/zstd/lib/README.md b/priv/zstd/lib/README.md
index f781ac5..c3b5d18 100644
--- a/priv/zstd/lib/README.md
+++ b/priv/zstd/lib/README.md
@@ -91,7 +91,7 @@ The file structure is designed to make this selection manually achievable for an
         `ZSTD_LIB_COMPRESSION, ZSTD_LIB_DECOMPRESSION`, `ZSTD_LIB_DICTBUILDER`,
         and `ZSTD_LIB_DEPRECATED` as `0` to forgo compilation of the
         corresponding features. This will also disable compilation of all
-        dependencies (eg. `ZSTD_LIB_COMPRESSION=0` will also disable
+        dependencies (e.g. `ZSTD_LIB_COMPRESSION=0` will also disable
         dictBuilder).
 
 - There are a number of options that can help minimize the binary size of
@@ -125,7 +125,7 @@ The file structure is designed to make this selection manually achievable for an
   `ZSTD_getErrorName` (implied by `ZSTD_LIB_MINIFY`).
 
   Finally, when integrating into your application, make sure you're doing link-
-  time optimation and unused symbol garbage collection (via some combination of,
+  time optimization and unused symbol garbage collection (via some combination of,
   e.g., `-flto`, `-ffat-lto-objects`, `-fuse-linker-plugin`,
   `-ffunction-sections`, `-fdata-sections`, `-fmerge-all-constants`,
   `-Wl,--gc-sections`, `-Wl,-z,norelro`, and an archiver that understands
@@ -155,6 +155,19 @@ The file structure is designed to make this selection manually achievable for an
 - The build macro `ZSTD_NO_INTRINSICS` can be defined to disable all explicit intrinsics.
   Compiler builtins are still used.
 
+- The build macro `ZSTD_DECODER_INTERNAL_BUFFER` can be set to control
+  the amount of extra memory used during decompression to store literals.
+  This defaults to 64kB.  Reducing this value reduces the memory footprint of
+  `ZSTD_DCtx` decompression contexts,
+  but might also result in a small decompression speed cost.
+
+- The C compiler macros `ZSTDLIB_VISIBLE`, `ZSTDERRORLIB_VISIBLE` and `ZDICTLIB_VISIBLE`
+  can be overridden to control the visibility of zstd's API. Additionally,
+  `ZSTDLIB_STATIC_API` and `ZDICTLIB_STATIC_API` can be overridden to control the visibility
+  of zstd's static API. Specifically, it can be set to `ZSTDLIB_HIDDEN` to hide the symbols
+  from the shared library. These macros default to `ZSTDLIB_VISIBILITY`,
+  `ZSTDERRORLIB_VSIBILITY`, and `ZDICTLIB_VISIBILITY` if unset, for backwards compatibility
+  with the old macro names.
 
 #### Windows : using MinGW+MSYS to create DLL
 
diff --git a/priv/zstd/lib/common/bits.h b/priv/zstd/lib/common/bits.h
new file mode 100644
index 0000000..7939f3d
--- /dev/null
+++ b/priv/zstd/lib/common/bits.h
@@ -0,0 +1,175 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_BITS_H
+#define ZSTD_BITS_H
+
+#include "mem.h"
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros32_fallback(U32 val)
+{
+    assert(val != 0);
+    {
+        static const int DeBruijnBytePos[32] = {0, 1, 28, 2, 29, 14, 24, 3,
+                                                30, 22, 20, 15, 25, 17, 4, 8,
+                                                31, 27, 13, 23, 21, 19, 16, 7,
+                                                26, 12, 18, 6, 11, 5, 10, 9};
+        return DeBruijnBytePos[((U32) ((val & -(S32) val) * 0x077CB531U)) >> 27];
+    }
+}
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros32(U32 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER)
+#       if STATIC_BMI2 == 1
+            return _tzcnt_u32(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanForward(&r, val);
+                return (unsigned)r;
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4)
+        return (unsigned)__builtin_ctz(val);
+#   else
+        return ZSTD_countTrailingZeros32_fallback(val);
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros32_fallback(U32 val) {
+    assert(val != 0);
+    {
+        static const U32 DeBruijnClz[32] = {0, 9, 1, 10, 13, 21, 2, 29,
+                                            11, 14, 16, 18, 22, 25, 3, 30,
+                                            8, 12, 20, 28, 15, 17, 24, 7,
+                                            19, 27, 23, 6, 26, 5, 4, 31};
+        val |= val >> 1;
+        val |= val >> 2;
+        val |= val >> 4;
+        val |= val >> 8;
+        val |= val >> 16;
+        return 31 - DeBruijnClz[(val * 0x07C4ACDDU) >> 27];
+    }
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros32(U32 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER)
+#       if STATIC_BMI2 == 1
+            return _lzcnt_u32(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanReverse(&r, val);
+                return (unsigned)(31 - r);
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4)
+        return (unsigned)__builtin_clz(val);
+#   else
+        return ZSTD_countLeadingZeros32_fallback(val);
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countTrailingZeros64(U64 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER) && defined(_WIN64)
+#       if STATIC_BMI2 == 1
+            return _tzcnt_u64(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanForward64(&r, val);
+                return (unsigned)r;
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4) && defined(__LP64__)
+        return (unsigned)__builtin_ctzll(val);
+#   else
+        {
+            U32 mostSignificantWord = (U32)(val >> 32);
+            U32 leastSignificantWord = (U32)val;
+            if (leastSignificantWord == 0) {
+                return 32 + ZSTD_countTrailingZeros32(mostSignificantWord);
+            } else {
+                return ZSTD_countTrailingZeros32(leastSignificantWord);
+            }
+        }
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_countLeadingZeros64(U64 val)
+{
+    assert(val != 0);
+#   if defined(_MSC_VER) && defined(_WIN64)
+#       if STATIC_BMI2 == 1
+            return _lzcnt_u64(val);
+#       else
+            if (val != 0) {
+                unsigned long r;
+                _BitScanReverse64(&r, val);
+                return (unsigned)(63 - r);
+            } else {
+                /* Should not reach this code path */
+                __assume(0);
+            }
+#       endif
+#   elif defined(__GNUC__) && (__GNUC__ >= 4)
+        return (unsigned)(__builtin_clzll(val));
+#   else
+        {
+            U32 mostSignificantWord = (U32)(val >> 32);
+            U32 leastSignificantWord = (U32)val;
+            if (mostSignificantWord == 0) {
+                return 32 + ZSTD_countLeadingZeros32(leastSignificantWord);
+            } else {
+                return ZSTD_countLeadingZeros32(mostSignificantWord);
+            }
+        }
+#   endif
+}
+
+MEM_STATIC unsigned ZSTD_NbCommonBytes(size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+            return ZSTD_countTrailingZeros64((U64)val) >> 3;
+        } else {
+            return ZSTD_countTrailingZeros32((U32)val) >> 3;
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+            return ZSTD_countLeadingZeros64((U64)val) >> 3;
+        } else {
+            return ZSTD_countLeadingZeros32((U32)val) >> 3;
+        }
+    }
+}
+
+MEM_STATIC unsigned ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
+{
+    assert(val != 0);
+    return 31 - ZSTD_countLeadingZeros32(val);
+}
+
+#endif /* ZSTD_BITS_H */
diff --git a/priv/zstd/lib/common/bitstream.h b/priv/zstd/lib/common/bitstream.h
index 2e5a933..db1b4cf 100644
--- a/priv/zstd/lib/common/bitstream.h
+++ b/priv/zstd/lib/common/bitstream.h
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * bitstream
  * Part of FSE library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -30,14 +30,15 @@ extern "C" {
 #include "compiler.h"       /* UNLIKELY() */
 #include "debug.h"          /* assert(), DEBUGLOG(), RAWLOG() */
 #include "error_private.h"  /* error codes and messages */
+#include "bits.h"           /* ZSTD_highbit32 */
 
 
 /*=========================================
 *  Target specific
 =========================================*/
 #ifndef ZSTD_NO_INTRINSICS
-#  if defined(__BMI__) && defined(__GNUC__)
-#    include    /* support for bextr (experimental) */
+#  if (defined(__BMI__) || defined(__BMI2__)) && defined(__GNUC__)
+#    include    /* support for bextr (experimental)/bzhi */
 #  elif defined(__ICCARM__)
 #    include 
 #  endif
@@ -132,42 +133,6 @@ MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC);
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
 /* faster, but works only if nbBits >= 1 */
 
-
-
-/*-**************************************************************
-*  Internal functions
-****************************************************************/
-MEM_STATIC unsigned BIT_highbit32 (U32 val)
-{
-    assert(val != 0);
-    {
-#   if defined(_MSC_VER)   /* Visual */
-#       if STATIC_BMI2 == 1
-		return _lzcnt_u32(val) ^ 31;
-#       else
-		unsigned long r = 0;
-		return _BitScanReverse(&r, val) ? (unsigned)r : 0;
-#       endif
-#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
-        return __builtin_clz (val) ^ 31;
-#   elif defined(__ICCARM__)    /* IAR Intrinsic */
-        return 31 - __CLZ(val);
-#   else   /* Software version */
-        static const unsigned DeBruijnClz[32] = { 0,  9,  1, 10, 13, 21,  2, 29,
-                                                 11, 14, 16, 18, 22, 25,  3, 30,
-                                                  8, 12, 20, 28, 15, 17, 24,  7,
-                                                 19, 27, 23,  6, 26,  5,  4, 31 };
-        U32 v = val;
-        v |= v >> 1;
-        v |= v >> 2;
-        v |= v >> 4;
-        v |= v >> 8;
-        v |= v >> 16;
-        return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
-#   endif
-    }
-}
-
 /*=====    Local Constants   =====*/
 static const unsigned BIT_mask[] = {
     0,          1,         3,         7,         0xF,       0x1F,
@@ -197,6 +162,16 @@ MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
     return 0;
 }
 
+MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
+{
+#if defined(STATIC_BMI2) && STATIC_BMI2 == 1 && !defined(ZSTD_NO_INTRINSICS)
+    return  _bzhi_u64(bitContainer, nbBits);
+#else
+    assert(nbBits < BIT_MASK_SIZE);
+    return bitContainer & BIT_mask[nbBits];
+#endif
+}
+
 /*! BIT_addBits() :
  *  can add up to 31 bits into `bitC`.
  *  Note : does not check for register overflow ! */
@@ -206,7 +181,7 @@ MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
     DEBUG_STATIC_ASSERT(BIT_MASK_SIZE == 32);
     assert(nbBits < BIT_MASK_SIZE);
     assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
-    bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
+    bitC->bitContainer |= BIT_getLowerBits(value, nbBits) << bitC->bitPos;
     bitC->bitPos += nbBits;
 }
 
@@ -285,7 +260,7 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
         bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
         bitD->bitContainer = MEM_readLEST(bitD->ptr);
         { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
-          bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
+          bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
           if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
     } else {
         bitD->ptr   = bitD->start;
@@ -293,27 +268,27 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
         switch(srcSize)
         {
         case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
-                /* fall-through */
+                ZSTD_FALLTHROUGH;
 
         case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
-                /* fall-through */
+                ZSTD_FALLTHROUGH;
 
         case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
-                /* fall-through */
+                ZSTD_FALLTHROUGH;
 
         case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
-                /* fall-through */
+                ZSTD_FALLTHROUGH;
 
         case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
-                /* fall-through */
+                ZSTD_FALLTHROUGH;
 
         case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
-                /* fall-through */
+                ZSTD_FALLTHROUGH;
 
         default: break;
         }
         {   BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
-            bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
+            bitD->bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
             if (lastByte == 0) return ERROR(corruption_detected);  /* endMark not present */
         }
         bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
@@ -332,16 +307,15 @@ MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getMiddleBits(size_t bitContainer, U32 c
     U32 const regMask = sizeof(bitContainer)*8 - 1;
     /* if start > regMask, bitstream is corrupted, and result is undefined */
     assert(nbBits < BIT_MASK_SIZE);
-    return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
-}
-
-MEM_STATIC FORCE_INLINE_ATTR size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
-{
-#if defined(STATIC_BMI2) && STATIC_BMI2 == 1
-	return  _bzhi_u64(bitContainer, nbBits);
+    /* x86 transform & ((1 << nbBits) - 1) to bzhi instruction, it is better
+     * than accessing memory. When bmi2 instruction is not present, we consider
+     * such cpus old (pre-Haswell, 2013) and their performance is not of that
+     * importance.
+     */
+#if defined(__x86_64__) || defined(_M_X86)
+    return (bitContainer >> (start & regMask)) & ((((U64)1) << nbBits) - 1);
 #else
-    assert(nbBits < BIT_MASK_SIZE);
-    return bitContainer & BIT_mask[nbBits];
+    return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
 #endif
 }
 
@@ -391,7 +365,7 @@ MEM_STATIC FORCE_INLINE_ATTR size_t BIT_readBits(BIT_DStream_t* bitD, unsigned n
 }
 
 /*! BIT_readBitsFast() :
- *  unsafe version; only works only if nbBits >= 1 */
+ *  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBitsFast(bitD, nbBits);
diff --git a/priv/zstd/lib/common/compiler.h b/priv/zstd/lib/common/compiler.h
index a951d0a..d4f2f28 100644
--- a/priv/zstd/lib/common/compiler.h
+++ b/priv/zstd/lib/common/compiler.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,6 +11,8 @@
 #ifndef ZSTD_COMPILER_H
 #define ZSTD_COMPILER_H
 
+#include "portability_macros.h"
+
 /*-*******************************************************
 *  Compiler specifics
 *********************************************************/
@@ -40,7 +42,7 @@
 
 /**
   On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
-  This explictly marks such functions as __cdecl so that the code will still compile
+  This explicitly marks such functions as __cdecl so that the code will still compile
   if a CC other than __cdecl has been made the default.
 */
 #if  defined(_MSC_VER)
@@ -92,29 +94,17 @@
 
 
 /* target attribute */
-#ifndef __has_attribute
-  #define __has_attribute(x) 0  /* Compatibility with non-clang compilers. */
-#endif
 #if defined(__GNUC__) || defined(__ICCARM__)
 #  define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
 #else
 #  define TARGET_ATTRIBUTE(target)
 #endif
 
-/* Enable runtime BMI2 dispatch based on the CPU.
- * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
+/* Target attribute for BMI2 dynamic dispatch.
+ * Enable lzcnt, bmi, and bmi2.
+ * We test for bmi1 & bmi2. lzcnt is included in bmi1.
  */
-#ifndef DYNAMIC_BMI2
-  #if ((defined(__clang__) && __has_attribute(__target__)) \
-      || (defined(__GNUC__) \
-          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
-      && (defined(__x86_64__) || defined(_M_X86)) \
-      && !defined(__BMI2__)
-  #  define DYNAMIC_BMI2 1
-  #else
-  #  define DYNAMIC_BMI2 0
-  #endif
-#endif
+#define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2")
 
 /* prefetch
  * can be disabled, by declaring NO_PREFETCH build macro */
@@ -150,8 +140,9 @@
 }
 
 /* vectorization
- * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax */
-#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__)
+ * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax,
+ * and some compilers, like Intel ICC and MCST LCC, do not support it at all. */
+#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__)
 #  if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
 #    define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
 #  else
@@ -174,6 +165,12 @@
 #define UNLIKELY(x) (x)
 #endif
 
+#if __has_builtin(__builtin_unreachable) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)))
+#  define ZSTD_UNREACHABLE { assert(0), __builtin_unreachable(); }
+#else
+#  define ZSTD_UNREACHABLE { assert(0); }
+#endif
+
 /* disable warnings */
 #ifdef _MSC_VER    /* Visual Studio */
 #  include                     /* For Visual 2005 */
@@ -190,6 +187,8 @@
 #    ifdef __AVX2__  //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2
 #       define STATIC_BMI2 1
 #    endif
+#  elif defined(__BMI2__) && defined(__x86_64__) && defined(__GNUC__)
+#    define STATIC_BMI2 1
 #  endif
 #endif
 
@@ -197,26 +196,103 @@
     #define STATIC_BMI2 0
 #endif
 
-/* compat. with non-clang compilers */
-#ifndef __has_builtin
-#  define __has_builtin(x) 0
+/* compile time determination of SIMD support */
+#if !defined(ZSTD_NO_INTRINSICS)
+#  if defined(__SSE2__) || defined(_M_AMD64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2))
+#    define ZSTD_ARCH_X86_SSE2
+#  endif
+#  if defined(__ARM_NEON) || defined(_M_ARM64)
+#    define ZSTD_ARCH_ARM_NEON
+#  endif
+#
+#  if defined(ZSTD_ARCH_X86_SSE2)
+#    include 
+#  elif defined(ZSTD_ARCH_ARM_NEON)
+#    include 
+#  endif
 #endif
 
-/* compat. with non-clang compilers */
-#ifndef __has_feature
-#  define __has_feature(x) 0
+/* C-language Attributes are added in C23. */
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
+# define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+#else
+# define ZSTD_HAS_C_ATTRIBUTE(x) 0
 #endif
 
-/* detects whether we are being compiled under msan */
-#ifndef ZSTD_MEMORY_SANITIZER
-#  if __has_feature(memory_sanitizer)
-#    define ZSTD_MEMORY_SANITIZER 1
-#  else
-#    define ZSTD_MEMORY_SANITIZER 0
-#  endif
+/* Only use C++ attributes in C++. Some compilers report support for C++
+ * attributes when compiling with C.
+ */
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+# define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+# define ZSTD_HAS_CPP_ATTRIBUTE(x) 0
 #endif
 
-#if ZSTD_MEMORY_SANITIZER
+/* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute.
+ * - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough
+ * - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+ * - Else: __attribute__((__fallthrough__))
+ */
+#ifndef ZSTD_FALLTHROUGH
+# if ZSTD_HAS_C_ATTRIBUTE(fallthrough)
+#  define ZSTD_FALLTHROUGH [[fallthrough]]
+# elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough)
+#  define ZSTD_FALLTHROUGH [[fallthrough]]
+# elif __has_attribute(__fallthrough__)
+/* Leading semicolon is to satisfy gcc-11 with -pedantic. Without the semicolon
+ * gcc complains about: a label can only be part of a statement and a declaration is not a statement.
+ */
+#  define ZSTD_FALLTHROUGH ; __attribute__((__fallthrough__))
+# else
+#  define ZSTD_FALLTHROUGH
+# endif
+#endif
+
+/*-**************************************************************
+*  Alignment check
+*****************************************************************/
+
+/* this test was initially positioned in mem.h,
+ * but this file is removed (or replaced) for linux kernel
+ * so it's now hosted in compiler.h,
+ * which remains valid for both user & kernel spaces.
+ */
+
+#ifndef ZSTD_ALIGNOF
+# if defined(__GNUC__) || defined(_MSC_VER)
+/* covers gcc, clang & MSVC */
+/* note : this section must come first, before C11,
+ * due to a limitation in the kernel source generator */
+#  define ZSTD_ALIGNOF(T) __alignof(T)
+
+# elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)
+/* C11 support */
+#  include 
+#  define ZSTD_ALIGNOF(T) alignof(T)
+
+# else
+/* No known support for alignof() - imperfect backup */
+#  define ZSTD_ALIGNOF(T) (sizeof(void*) < sizeof(T) ? sizeof(void*) : sizeof(T))
+
+# endif
+#endif /* ZSTD_ALIGNOF */
+
+/*-**************************************************************
+*  Sanitizer
+*****************************************************************/
+
+/* Issue #3240 reports an ASAN failure on an llvm-mingw build. Out of an
+ * abundance of caution, disable our custom poisoning on mingw. */
+#ifdef __MINGW32__
+#ifndef ZSTD_ASAN_DONT_POISON_WORKSPACE
+#define ZSTD_ASAN_DONT_POISON_WORKSPACE 1
+#endif
+#ifndef ZSTD_MSAN_DONT_POISON_WORKSPACE
+#define ZSTD_MSAN_DONT_POISON_WORKSPACE 1
+#endif
+#endif
+
+#if ZSTD_MEMORY_SANITIZER && !defined(ZSTD_MSAN_DONT_POISON_WORKSPACE)
 /* Not all platforms that support msan provide sanitizers/msan_interface.h.
  * We therefore declare the functions we need ourselves, rather than trying to
  * include the header file... */
@@ -237,18 +313,7 @@ void __msan_poison(const volatile void *a, size_t size);
 intptr_t __msan_test_shadow(const volatile void *x, size_t size);
 #endif
 
-/* detects whether we are being compiled under asan */
-#ifndef ZSTD_ADDRESS_SANITIZER
-#  if __has_feature(address_sanitizer)
-#    define ZSTD_ADDRESS_SANITIZER 1
-#  elif defined(__SANITIZE_ADDRESS__)
-#    define ZSTD_ADDRESS_SANITIZER 1
-#  else
-#    define ZSTD_ADDRESS_SANITIZER 0
-#  endif
-#endif
-
-#if ZSTD_ADDRESS_SANITIZER
+#if ZSTD_ADDRESS_SANITIZER && !defined(ZSTD_ASAN_DONT_POISON_WORKSPACE)
 /* Not all platforms that support asan provide sanitizers/asan_interface.h.
  * We therefore declare the functions we need ourselves, rather than trying to
  * include the header file... */
diff --git a/priv/zstd/lib/common/cpu.h b/priv/zstd/lib/common/cpu.h
index 8acd33b..8bc34a3 100644
--- a/priv/zstd/lib/common/cpu.h
+++ b/priv/zstd/lib/common/cpu.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/common/debug.c b/priv/zstd/lib/common/debug.c
index bb863c9..ebf7bfc 100644
--- a/priv/zstd/lib/common/debug.c
+++ b/priv/zstd/lib/common/debug.c
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * debug
  * Part of FSE library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
diff --git a/priv/zstd/lib/common/debug.h b/priv/zstd/lib/common/debug.h
index 3b2a320..0e9817e 100644
--- a/priv/zstd/lib/common/debug.h
+++ b/priv/zstd/lib/common/debug.h
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * debug
  * Part of FSE library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
diff --git a/priv/zstd/lib/common/entropy_common.c b/priv/zstd/lib/common/entropy_common.c
index 41cd695..e2173af 100644
--- a/priv/zstd/lib/common/entropy_common.c
+++ b/priv/zstd/lib/common/entropy_common.c
@@ -1,6 +1,6 @@
 /* ******************************************************************
  * Common functions of New Generation Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -19,8 +19,8 @@
 #include "error_private.h"       /* ERR_*, ERROR */
 #define FSE_STATIC_LINKING_ONLY  /* FSE_MIN_TABLELOG */
 #include "fse.h"
-#define HUF_STATIC_LINKING_ONLY  /* HUF_TABLELOG_ABSOLUTEMAX */
 #include "huf.h"
+#include "bits.h"                /* ZSDT_highbit32, ZSTD_countTrailingZeros32 */
 
 
 /*===   Version   ===*/
@@ -38,28 +38,6 @@ const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
 /*-**************************************************************
 *  FSE NCount encoding-decoding
 ****************************************************************/
-static U32 FSE_ctz(U32 val)
-{
-    assert(val != 0);
-    {
-#   if defined(_MSC_VER)   /* Visual */
-        unsigned long r=0;
-        return _BitScanForward(&r, val) ? (unsigned)r : 0;
-#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
-        return __builtin_ctz(val);
-#   elif defined(__ICCARM__)    /* IAR Intrinsic */
-        return __CTZ(val);
-#   else   /* Software version */
-        U32 count = 0;
-        while ((val & 1) == 0) {
-            val >>= 1;
-            ++count;
-        }
-        return count;
-#   endif
-    }
-}
-
 FORCE_INLINE_TEMPLATE
 size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
                            const void* headerBuffer, size_t hbSize)
@@ -107,7 +85,7 @@ size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigne
              * repeat.
              * Avoid UB by setting the high bit to 1.
              */
-            int repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
+            int repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
             while (repeats >= 12) {
                 charnum += 3 * 12;
                 if (LIKELY(ip <= iend-7)) {
@@ -118,7 +96,7 @@ size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigne
                     ip = iend - 4;
                 }
                 bitStream = MEM_readLE32(ip) >> bitCount;
-                repeats = FSE_ctz(~bitStream | 0x80000000) >> 1;
+                repeats = ZSTD_countTrailingZeros32(~bitStream | 0x80000000) >> 1;
             }
             charnum += 3 * repeats;
             bitStream >>= 2 * repeats;
@@ -183,7 +161,7 @@ size_t FSE_readNCount_body(short* normalizedCounter, unsigned* maxSVPtr, unsigne
                  * know that threshold > 1.
                  */
                 if (remaining <= 1) break;
-                nbBits = BIT_highbit32(remaining) + 1;
+                nbBits = ZSTD_highbit32(remaining) + 1;
                 threshold = 1 << (nbBits - 1);
             }
             if (charnum >= maxSV1) break;
@@ -217,7 +195,7 @@ static size_t FSE_readNCount_body_default(
 }
 
 #if DYNAMIC_BMI2
-TARGET_ATTRIBUTE("bmi2") static size_t FSE_readNCount_body_bmi2(
+BMI2_TARGET_ATTRIBUTE static size_t FSE_readNCount_body_bmi2(
         short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
         const void* headerBuffer, size_t hbSize)
 {
@@ -258,7 +236,7 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      const void* src, size_t srcSize)
 {
     U32 wksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
-    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* bmi2 */ 0);
+    return HUF_readStats_wksp(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, wksp, sizeof(wksp), /* flags */ 0);
 }
 
 FORCE_INLINE_TEMPLATE size_t
@@ -299,21 +277,21 @@ HUF_readStats_body(BYTE* huffWeight, size_t hwSize, U32* rankStats,
     ZSTD_memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
     weightTotal = 0;
     {   U32 n; for (n=0; n= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
+            if (huffWeight[n] > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
             rankStats[huffWeight[n]]++;
             weightTotal += (1 << huffWeight[n]) >> 1;
     }   }
     if (weightTotal == 0) return ERROR(corruption_detected);
 
     /* get last non-null symbol weight (implied, total must be 2^n) */
-    {   U32 const tableLog = BIT_highbit32(weightTotal) + 1;
+    {   U32 const tableLog = ZSTD_highbit32(weightTotal) + 1;
         if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
         *tableLogPtr = tableLog;
         /* determine last weight */
         {   U32 const total = 1 << tableLog;
             U32 const rest = total - weightTotal;
-            U32 const verif = 1 << BIT_highbit32(rest);
-            U32 const lastWeight = BIT_highbit32(rest) + 1;
+            U32 const verif = 1 << ZSTD_highbit32(rest);
+            U32 const lastWeight = ZSTD_highbit32(rest) + 1;
             if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
             huffWeight[oSize] = (BYTE)lastWeight;
             rankStats[lastWeight]++;
@@ -337,7 +315,7 @@ static size_t HUF_readStats_body_default(BYTE* huffWeight, size_t hwSize, U32* r
 }
 
 #if DYNAMIC_BMI2
-static TARGET_ATTRIBUTE("bmi2") size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+static BMI2_TARGET_ATTRIBUTE size_t HUF_readStats_body_bmi2(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize,
                      void* workSpace, size_t wkspSize)
@@ -350,13 +328,13 @@ size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                      U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize,
                      void* workSpace, size_t wkspSize,
-                     int bmi2)
+                     int flags)
 {
 #if DYNAMIC_BMI2
-    if (bmi2) {
+    if (flags & HUF_flags_bmi2) {
         return HUF_readStats_body_bmi2(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
     }
 #endif
-    (void)bmi2;
+    (void)flags;
     return HUF_readStats_body_default(huffWeight, hwSize, rankStats, nbSymbolsPtr, tableLogPtr, src, srcSize, workSpace, wkspSize);
 }
diff --git a/priv/zstd/lib/common/error_private.c b/priv/zstd/lib/common/error_private.c
index 6d1135f..075fc5e 100644
--- a/priv/zstd/lib/common/error_private.c
+++ b/priv/zstd/lib/common/error_private.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -27,9 +27,11 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(version_unsupported): return "Version not supported";
     case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
     case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
-    case PREFIX(corruption_detected): return "Corrupted block detected";
+    case PREFIX(corruption_detected): return "Data corruption detected";
     case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(literals_headerWrong): return "Header of Literals' block doesn't respect format specification";
     case PREFIX(parameter_unsupported): return "Unsupported parameter";
+    case PREFIX(parameter_combination_unsupported): return "Unsupported combination of parameters";
     case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
     case PREFIX(init_missing): return "Context should be init first";
     case PREFIX(memory_allocation): return "Allocation error : not enough memory";
@@ -38,17 +40,22 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
     case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
     case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
+    case PREFIX(stabilityCondition_notRespected): return "pledged buffer stability condition is not respected";
     case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
     case PREFIX(dictionary_wrong): return "Dictionary mismatch";
     case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
     case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
     case PREFIX(srcSize_wrong): return "Src size is incorrect";
     case PREFIX(dstBuffer_null): return "Operation on NULL destination buffer";
+    case PREFIX(noForwardProgress_destFull): return "Operation made no progress over multiple calls, due to output buffer being full";
+    case PREFIX(noForwardProgress_inputEmpty): return "Operation made no progress over multiple calls, due to input being empty";
         /* following error codes are not stable and may be removed or changed in a future version */
     case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
     case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
     case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
     case PREFIX(srcBuffer_wrong): return "Source buffer is wrong";
+    case PREFIX(sequenceProducer_failed): return "Block-level external sequence producer returned an error code";
+    case PREFIX(externalSequences_invalid): return "External sequences are not valid";
     case PREFIX(maxCode):
     default: return notErrorCode;
     }
diff --git a/priv/zstd/lib/common/error_private.h b/priv/zstd/lib/common/error_private.h
index 6d8b9f7..325daad 100644
--- a/priv/zstd/lib/common/error_private.h
+++ b/priv/zstd/lib/common/error_private.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -22,6 +22,8 @@ extern "C" {
 *  Dependencies
 ******************************************/
 #include "../zstd_errors.h"  /* enum list */
+#include "compiler.h"
+#include "debug.h"
 #include "zstd_deps.h"       /* size_t */
 
 
@@ -73,6 +75,83 @@ ERR_STATIC const char* ERR_getErrorName(size_t code)
     return ERR_getErrorString(ERR_getErrorCode(code));
 }
 
+/**
+ * Ignore: this is an internal helper.
+ *
+ * This is a helper function to help force C99-correctness during compilation.
+ * Under strict compilation modes, variadic macro arguments can't be empty.
+ * However, variadic function arguments can be. Using a function therefore lets
+ * us statically check that at least one (string) argument was passed,
+ * independent of the compilation flags.
+ */
+static INLINE_KEYWORD UNUSED_ATTR
+void _force_has_format_string(const char *format, ...) {
+  (void)format;
+}
+
+/**
+ * Ignore: this is an internal helper.
+ *
+ * We want to force this function invocation to be syntactically correct, but
+ * we don't want to force runtime evaluation of its arguments.
+ */
+#define _FORCE_HAS_FORMAT_STRING(...) \
+  if (0) { \
+    _force_has_format_string(__VA_ARGS__); \
+  }
+
+#define ERR_QUOTE(str) #str
+
+/**
+ * Return the specified error if the condition evaluates to true.
+ *
+ * In debug modes, prints additional information.
+ * In order to do that (particularly, printing the conditional that failed),
+ * this can't just wrap RETURN_ERROR().
+ */
+#define RETURN_ERROR_IF(cond, err, ...) \
+  if (cond) { \
+    RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
+           __FILE__, __LINE__, ERR_QUOTE(cond), ERR_QUOTE(ERROR(err))); \
+    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+    RAWLOG(3, ": " __VA_ARGS__); \
+    RAWLOG(3, "\n"); \
+    return ERROR(err); \
+  }
+
+/**
+ * Unconditionally return the specified error.
+ *
+ * In debug modes, prints additional information.
+ */
+#define RETURN_ERROR(err, ...) \
+  do { \
+    RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
+           __FILE__, __LINE__, ERR_QUOTE(ERROR(err))); \
+    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+    RAWLOG(3, ": " __VA_ARGS__); \
+    RAWLOG(3, "\n"); \
+    return ERROR(err); \
+  } while(0);
+
+/**
+ * If the provided expression evaluates to an error code, returns that error code.
+ *
+ * In debug modes, prints additional information.
+ */
+#define FORWARD_IF_ERROR(err, ...) \
+  do { \
+    size_t const err_code = (err); \
+    if (ERR_isError(err_code)) { \
+      RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
+             __FILE__, __LINE__, ERR_QUOTE(err), ERR_getErrorName(err_code)); \
+      _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+      RAWLOG(3, ": " __VA_ARGS__); \
+      RAWLOG(3, "\n"); \
+      return err_code; \
+    } \
+  } while(0);
+
 #if defined (__cplusplus)
 }
 #endif
diff --git a/priv/zstd/lib/common/fse.h b/priv/zstd/lib/common/fse.h
index 19dd4fe..02a1f0b 100644
--- a/priv/zstd/lib/common/fse.h
+++ b/priv/zstd/lib/common/fse.h
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * FSE : Finite State Entropy codec
  * Public Prototypes declaration
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -53,34 +53,6 @@ extern "C" {
 FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
 
 
-/*-****************************************
-*  FSE simple functions
-******************************************/
-/*! FSE_compress() :
-    Compress content of buffer 'src', of size 'srcSize', into destination buffer 'dst'.
-    'dst' buffer must be already allocated. Compression runs faster is dstCapacity >= FSE_compressBound(srcSize).
-    @return : size of compressed data (<= dstCapacity).
-    Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
-                     if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression instead.
-                     if FSE_isError(return), compression failed (more details using FSE_getErrorName())
-*/
-FSE_PUBLIC_API size_t FSE_compress(void* dst, size_t dstCapacity,
-                             const void* src, size_t srcSize);
-
-/*! FSE_decompress():
-    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
-    into already allocated destination buffer 'dst', of size 'dstCapacity'.
-    @return : size of regenerated data (<= maxDstSize),
-              or an error code, which can be tested using FSE_isError() .
-
-    ** Important ** : FSE_decompress() does not decompress non-compressible nor RLE data !!!
-    Why ? : making this distinction requires a header.
-    Header management is intentionally delegated to the user layer, which can better manage special cases.
-*/
-FSE_PUBLIC_API size_t FSE_decompress(void* dst,  size_t dstCapacity,
-                               const void* cSrc, size_t cSrcSize);
-
-
 /*-*****************************************
 *  Tool functions
 ******************************************/
@@ -91,20 +63,6 @@ FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return
 FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
 
 
-/*-*****************************************
-*  FSE advanced functions
-******************************************/
-/*! FSE_compress2() :
-    Same as FSE_compress(), but allows the selection of 'maxSymbolValue' and 'tableLog'
-    Both parameters can be defined as '0' to mean : use default value
-    @return : size of compressed data
-    Special values : if return == 0, srcData is not compressible => Nothing is stored within cSrc !!!
-                     if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression.
-                     if FSE_isError(return), it's an error code.
-*/
-FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
-
-
 /*-*****************************************
 *  FSE detailed API
 ******************************************/
@@ -164,8 +122,6 @@ FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
 /*! Constructor and Destructor of FSE_CTable.
     Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
 typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
-FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog);
-FSE_PUBLIC_API void        FSE_freeCTable (FSE_CTable* ct);
 
 /*! FSE_buildCTable():
     Builds `ct`, which must be already allocated, using FSE_createCTable().
@@ -241,23 +197,7 @@ FSE_PUBLIC_API size_t FSE_readNCount_bmi2(short* normalizedCounter,
                            unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
                            const void* rBuffer, size_t rBuffSize, int bmi2);
 
-/*! Constructor and Destructor of FSE_DTable.
-    Note that its size depends on 'tableLog' */
 typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
-FSE_PUBLIC_API FSE_DTable* FSE_createDTable(unsigned tableLog);
-FSE_PUBLIC_API void        FSE_freeDTable(FSE_DTable* dt);
-
-/*! FSE_buildDTable():
-    Builds 'dt', which must be already allocated, using FSE_createDTable().
-    return : 0, or an errorCode, which can be tested using FSE_isError() */
-FSE_PUBLIC_API size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
-
-/*! FSE_decompress_usingDTable():
-    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
-    into `dst` which must be already allocated.
-    @return : size of regenerated data (necessarily <= `dstCapacity`),
-              or an errorCode, which can be tested using FSE_isError() */
-FSE_PUBLIC_API size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
 
 /*!
 Tutorial :
@@ -320,24 +260,15 @@ If there is an error, the function will return an error code, which can be teste
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
 /**< same as FSE_optimalTableLog(), which used `minus==2` */
 
-/* FSE_compress_wksp() :
- * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
- * FSE_COMPRESS_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
- */
-#define FSE_COMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
-size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
-
-size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
-/**< build a fake FSE_CTable, designed for a flat distribution, where each symbol uses nbBits */
-
 size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
 /**< build a fake FSE_CTable, designed to compress always the same symbolValue */
 
 /* FSE_buildCTable_wksp() :
  * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
  * `wkspSize` must be >= `FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog)` of `unsigned`.
+ * See FSE_buildCTable_wksp() for breakdown of workspace usage.
  */
-#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (maxSymbolValue + 2 + (1ull << (tableLog - 2)))
+#define FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog) (((maxSymbolValue + 2) + (1ull << (tableLog)))/2 + sizeof(U64)/sizeof(U32) /* additional 8 bytes for potential table overwrite */)
 #define FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) (sizeof(unsigned) * FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(maxSymbolValue, tableLog))
 size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 
@@ -346,19 +277,11 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
 FSE_PUBLIC_API size_t FSE_buildDTable_wksp(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 /**< Same as FSE_buildDTable(), using an externally allocated `workspace` produced with `FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxSymbolValue)` */
 
-size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
-/**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
-
-size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
-/**< build a fake FSE_DTable, designed to always generate the same symbolValue */
-
-#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
+#define FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) (FSE_DTABLE_SIZE_U32(maxTableLog) + 1 + FSE_BUILD_DTABLE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) + (FSE_MAX_SYMBOL_VALUE + 1) / 2 + 1)
 #define FSE_DECOMPRESS_WKSP_SIZE(maxTableLog, maxSymbolValue) (FSE_DECOMPRESS_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(unsigned))
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize);
-/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)` */
-
 size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize, int bmi2);
-/**< Same as FSE_decompress_wksp() but with dynamic BMI2 support. Pass 1 if your CPU supports BMI2 or 0 if it doesn't. */
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DECOMPRESS_WKSP_SIZE_U32(maxLog, maxSymbolValue)`.
+ * Set bmi2 to 1 if your CPU supports BMI2 or 0 if it doesn't */
 
 typedef enum {
    FSE_repeat_none,  /**< Cannot use the previous table */
@@ -554,7 +477,7 @@ MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePt
 
 /* FSE_getMaxNbBits() :
  * Approximate maximum cost of a symbol, in bits.
- * Fractional get rounded up (i.e : a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
+ * Fractional get rounded up (i.e. a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
  * note 1 : assume symbolValue is valid (<= maxSymbolValue)
  * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
 MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
diff --git a/priv/zstd/lib/common/fse_decompress.c b/priv/zstd/lib/common/fse_decompress.c
index f4ff58f..1e1c9f9 100644
--- a/priv/zstd/lib/common/fse_decompress.c
+++ b/priv/zstd/lib/common/fse_decompress.c
@@ -1,6 +1,6 @@
 /* ******************************************************************
  * FSE : Finite State Entropy decoder
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -24,6 +24,7 @@
 #include "error_private.h"
 #define ZSTD_DEPS_NEED_MALLOC
 #include "zstd_deps.h"
+#include "bits.h"       /* ZSTD_highbit32 */
 
 
 /* **************************************************************
@@ -55,19 +56,6 @@
 #define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
 #define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
 
-
-/* Function templates */
-FSE_DTable* FSE_createDTable (unsigned tableLog)
-{
-    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
-    return (FSE_DTable*)ZSTD_malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
-}
-
-void FSE_freeDTable (FSE_DTable* dt)
-{
-    ZSTD_free(dt);
-}
-
 static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
 {
     void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
@@ -127,10 +115,10 @@ static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCo
             }
         }
         /* Now we spread those positions across the table.
-         * The benefit of doing it in two stages is that we avoid the the
+         * The benefit of doing it in two stages is that we avoid the
          * variable size inner loop, which caused lots of branch misses.
          * Now we can run through all the positions without any branch misses.
-         * We unroll the loop twice, since that is what emperically worked best.
+         * We unroll the loop twice, since that is what empirically worked best.
          */
         {
             size_t position = 0;
@@ -166,7 +154,7 @@ static size_t FSE_buildDTable_internal(FSE_DTable* dt, const short* normalizedCo
         for (u=0; utableLog = 0;
-    DTableH->fastMode = 0;
-
-    cell->newState = 0;
-    cell->symbol = symbolValue;
-    cell->nbBits = 0;
-
-    return 0;
-}
-
-
-size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
-{
-    void* ptr = dt;
-    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
-    void* dPtr = dt + 1;
-    FSE_decode_t* const dinfo = (FSE_decode_t*)dPtr;
-    const unsigned tableSize = 1 << nbBits;
-    const unsigned tableMask = tableSize - 1;
-    const unsigned maxSV1 = tableMask+1;
-    unsigned s;
-
-    /* Sanity checks */
-    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
-
-    /* Build Decoding Table */
-    DTableH->tableLog = (U16)nbBits;
-    DTableH->fastMode = 1;
-    for (s=0; sfastMode;
-
-    /* select fast mode (static) */
-    if (fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
-    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
-}
-
-
-size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
-{
-    return FSE_decompress_wksp_bmi2(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, /* bmi2 */ 0);
-}
-
 typedef struct {
     short ncount[FSE_MAX_SYMBOL_VALUE + 1];
     FSE_DTable dtable[1]; /* Dynamically sized */
@@ -342,7 +267,8 @@ FORCE_INLINE_TEMPLATE size_t FSE_decompress_wksp_body(
     }
 
     if (FSE_DECOMPRESS_WKSP_SIZE(tableLog, maxSymbolValue) > wkspSize) return ERROR(tableLog_tooLarge);
-    workSpace = wksp->dtable + FSE_DTABLE_SIZE_U32(tableLog);
+    assert(sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog) <= wkspSize);
+    workSpace = (BYTE*)workSpace + sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
     wkspSize -= sizeof(*wksp) + FSE_DTABLE_SIZE(tableLog);
 
     CHECK_F( FSE_buildDTable_internal(wksp->dtable, wksp->ncount, maxSymbolValue, tableLog, workSpace, wkspSize) );
@@ -365,7 +291,7 @@ static size_t FSE_decompress_wksp_body_default(void* dst, size_t dstCapacity, co
 }
 
 #if DYNAMIC_BMI2
-TARGET_ATTRIBUTE("bmi2") static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
+BMI2_TARGET_ATTRIBUTE static size_t FSE_decompress_wksp_body_bmi2(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, unsigned maxLog, void* workSpace, size_t wkspSize)
 {
     return FSE_decompress_wksp_body(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize, 1);
 }
@@ -382,22 +308,4 @@ size_t FSE_decompress_wksp_bmi2(void* dst, size_t dstCapacity, const void* cSrc,
     return FSE_decompress_wksp_body_default(dst, dstCapacity, cSrc, cSrcSize, maxLog, workSpace, wkspSize);
 }
 
-
-typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
-
-#ifndef ZSTD_NO_UNUSED_FUNCTIONS
-size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog) {
-    U32 wksp[FSE_BUILD_DTABLE_WKSP_SIZE_U32(FSE_TABLELOG_ABSOLUTE_MAX, FSE_MAX_SYMBOL_VALUE)];
-    return FSE_buildDTable_wksp(dt, normalizedCounter, maxSymbolValue, tableLog, wksp, sizeof(wksp));
-}
-
-size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
-{
-    /* Static analyzer seems unable to understand this table will be properly initialized later */
-    U32 wksp[FSE_DECOMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
-    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, FSE_MAX_TABLELOG, wksp, sizeof(wksp));
-}
-#endif
-
-
 #endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/priv/zstd/lib/common/huf.h b/priv/zstd/lib/common/huf.h
index 3d47ced..73d1ee5 100644
--- a/priv/zstd/lib/common/huf.h
+++ b/priv/zstd/lib/common/huf.h
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * huff0 huffman codec,
  * part of Finite State Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  * You can contact the author at :
  * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -21,106 +21,29 @@ extern "C" {
 
 /* *** Dependencies *** */
 #include "zstd_deps.h"    /* size_t */
-
-
-/* *** library symbols visibility *** */
-/* Note : when linking with -fvisibility=hidden on gcc, or by default on Visual,
- *        HUF symbols remain "private" (internal symbols for library only).
- *        Set macro FSE_DLL_EXPORT to 1 if you want HUF symbols visible on DLL interface */
-#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
-#  define HUF_PUBLIC_API __attribute__ ((visibility ("default")))
-#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
-#  define HUF_PUBLIC_API __declspec(dllexport)
-#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
-#  define HUF_PUBLIC_API __declspec(dllimport)  /* not required, just to generate faster code (saves a function pointer load from IAT and an indirect jump) */
-#else
-#  define HUF_PUBLIC_API
-#endif
-
-
-/* ========================== */
-/* ***  simple functions  *** */
-/* ========================== */
-
-/** HUF_compress() :
- *  Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
- * 'dst' buffer must be already allocated.
- *  Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
- * `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
- * @return : size of compressed data (<= `dstCapacity`).
- *  Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
- *                   if HUF_isError(return), compression failed (more details using HUF_getErrorName())
- */
-HUF_PUBLIC_API size_t HUF_compress(void* dst, size_t dstCapacity,
-                             const void* src, size_t srcSize);
-
-/** HUF_decompress() :
- *  Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
- *  into already allocated buffer 'dst', of minimum size 'dstSize'.
- * `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
- *  Note : in contrast with FSE, HUF_decompress can regenerate
- *         RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
- *         because it knows size to regenerate (originalSize).
- * @return : size of regenerated data (== originalSize),
- *           or an error code, which can be tested using HUF_isError()
- */
-HUF_PUBLIC_API size_t HUF_decompress(void* dst,  size_t originalSize,
-                               const void* cSrc, size_t cSrcSize);
+#include "mem.h"          /* U32 */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
 
 
 /* ***   Tool functions *** */
-#define HUF_BLOCKSIZE_MAX (128 * 1024)                  /**< maximum input size for a single block compressed with HUF_compress */
-HUF_PUBLIC_API size_t HUF_compressBound(size_t size);   /**< maximum compressed size (worst case) */
+#define HUF_BLOCKSIZE_MAX (128 * 1024)   /**< maximum input size for a single block compressed with HUF_compress */
+size_t HUF_compressBound(size_t size);   /**< maximum compressed size (worst case) */
 
 /* Error Management */
-HUF_PUBLIC_API unsigned    HUF_isError(size_t code);       /**< tells if a return value is an error code */
-HUF_PUBLIC_API const char* HUF_getErrorName(size_t code);  /**< provides error code string (useful for debugging) */
-
-
-/* ***   Advanced function   *** */
-
-/** HUF_compress2() :
- *  Same as HUF_compress(), but offers control over `maxSymbolValue` and `tableLog`.
- * `maxSymbolValue` must be <= HUF_SYMBOLVALUE_MAX .
- * `tableLog` must be `<= HUF_TABLELOG_MAX` . */
-HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,
-                               const void* src, size_t srcSize,
-                               unsigned maxSymbolValue, unsigned tableLog);
-
-/** HUF_compress4X_wksp() :
- *  Same as HUF_compress2(), but uses externally allocated `workSpace`.
- * `workspace` must have minimum alignment of 4, and be at least as large as HUF_WORKSPACE_SIZE */
-#define HUF_WORKSPACE_SIZE ((6 << 10) + 256)
-#define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
-HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
-                                     const void* src, size_t srcSize,
-                                     unsigned maxSymbolValue, unsigned tableLog,
-                                     void* workSpace, size_t wkspSize);
-
-#endif   /* HUF_H_298734234 */
+unsigned    HUF_isError(size_t code);       /**< tells if a return value is an error code */
+const char* HUF_getErrorName(size_t code);  /**< provides error code string (useful for debugging) */
 
-/* ******************************************************************
- *  WARNING !!
- *  The following section contains advanced and experimental definitions
- *  which shall never be used in the context of a dynamic library,
- *  because they are not guaranteed to remain stable in the future.
- *  Only consider them in association with static linking.
- * *****************************************************************/
-#if defined(HUF_STATIC_LINKING_ONLY) && !defined(HUF_H_HUF_STATIC_LINKING_ONLY)
-#define HUF_H_HUF_STATIC_LINKING_ONLY
-
-/* *** Dependencies *** */
-#include "mem.h"   /* U32 */
-#define FSE_STATIC_LINKING_ONLY
-#include "fse.h"
 
+#define HUF_WORKSPACE_SIZE ((8 << 10) + 512 /* sorting scratch space */)
+#define HUF_WORKSPACE_SIZE_U64 (HUF_WORKSPACE_SIZE / sizeof(U64))
 
 /* *** Constants *** */
-#define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_TABLELOG_ABSOLUTEMAX */
 #define HUF_TABLELOG_DEFAULT  11      /* default tableLog value when none specified */
 #define HUF_SYMBOLVALUE_MAX  255
 
-#define HUF_TABLELOG_ABSOLUTEMAX  15  /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_TABLELOG_ABSOLUTEMAX  12  /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
 #if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
 #  error "HUF_TABLELOG_MAX is too large !"
 #endif
@@ -136,15 +59,11 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
 
 /* static allocation of HUF's Compression Table */
 /* this is a private definition, just exposed for allocation and strict aliasing purpose. never EVER access its members directly */
-struct HUF_CElt_s {
-  U16  val;
-  BYTE nbBits;
-};   /* typedef'd to HUF_CElt */
-typedef struct HUF_CElt_s HUF_CElt;   /* consider it an incomplete type */
-#define HUF_CTABLE_SIZE_U32(maxSymbolValue)   ((maxSymbolValue)+1)   /* Use tables of U32, for proper alignment */
-#define HUF_CTABLE_SIZE(maxSymbolValue)       (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
+typedef size_t HUF_CElt;   /* consider it an incomplete type */
+#define HUF_CTABLE_SIZE_ST(maxSymbolValue)   ((maxSymbolValue)+2)   /* Use tables of size_t, for proper alignment */
+#define HUF_CTABLE_SIZE(maxSymbolValue)       (HUF_CTABLE_SIZE_ST(maxSymbolValue) * sizeof(size_t))
 #define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
-    HUF_CElt name[HUF_CTABLE_SIZE_U32(maxSymbolValue)] /* no final ; */
+    HUF_CElt name[HUF_CTABLE_SIZE_ST(maxSymbolValue)] /* no final ; */
 
 /* static allocation of HUF's DTable */
 typedef U32 HUF_DTable;
@@ -158,25 +77,49 @@ typedef U32 HUF_DTable;
 /* ****************************************
 *  Advanced decompression functions
 ******************************************/
-size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-#endif
 
-size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
-size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
-size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize); /**< considers RLE and uncompressed as errors */
-size_t HUF_decompress4X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
-#endif
+/**
+ * Huffman flags bitset.
+ * For all flags, 0 is the default value.
+ */
+typedef enum {
+    /**
+     * If compiled with DYNAMIC_BMI2: Set flag only if the CPU supports BMI2 at runtime.
+     * Otherwise: Ignored.
+     */
+    HUF_flags_bmi2 = (1 << 0),
+    /**
+     * If set: Test possible table depths to find the one that produces the smallest header + encoded size.
+     * If unset: Use heuristic to find the table depth.
+     */
+    HUF_flags_optimalDepth = (1 << 1),
+    /**
+     * If set: If the previous table can encode the input, always reuse the previous table.
+     * If unset: If the previous table can encode the input, reuse the previous table if it results in a smaller output.
+     */
+    HUF_flags_preferRepeat = (1 << 2),
+    /**
+     * If set: Sample the input and check if the sample is uncompressible, if it is then don't attempt to compress.
+     * If unset: Always histogram the entire input.
+     */
+    HUF_flags_suspectUncompressible = (1 << 3),
+    /**
+     * If set: Don't use assembly implementations
+     * If unset: Allow using assembly implementations
+     */
+    HUF_flags_disableAsm = (1 << 4),
+    /**
+     * If set: Don't use the fast decoding loop, always use the fallback decoding loop.
+     * If unset: Use the fast decoding loop when possible.
+     */
+    HUF_flags_disableFast = (1 << 5)
+} HUF_flags_e;
 
 
 /* ****************************************
  *  HUF detailed API
  * ****************************************/
+#define HUF_OPTIMAL_DEPTH_THRESHOLD ZSTD_btultra
 
 /*! HUF_compress() does the following:
  *  1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
@@ -189,11 +132,12 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
  *  For example, it's possible to compress several blocks using the same 'CTable',
  *  or to save and regenerate 'CTable' using external methods.
  */
-unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
-size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);   /* @return : maxNbBits; CTable and count can overlap. In which case, CTable will overwrite count content */
-size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
+unsigned HUF_minTableLog(unsigned symbolCardinality);
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue);
+unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, void* workSpace,
+ size_t wkspSize, HUF_CElt* table, const unsigned* count, int flags); /* table is used as scratch space for building and testing tables, not a return value */
 size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog, void* workspace, size_t workspaceSize);
-size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
 size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
 int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
 
@@ -202,22 +146,24 @@ typedef enum {
    HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
    HUF_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
  } HUF_repeat;
+
 /** HUF_compress4X_repeat() :
  *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
  *  If it uses hufTable it does not modify hufTable or repeat.
  *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
- *  If preferRepeat then the old table will always be used if valid. */
+ *  If preferRepeat then the old table will always be used if valid.
+ *  If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
 size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
                        const void* src, size_t srcSize,
                        unsigned maxSymbolValue, unsigned tableLog,
                        void* workSpace, size_t wkspSize,    /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
-                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
 
 /** HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
  * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
  */
-#define HUF_CTABLE_WORKSPACE_SIZE_U32 (2*HUF_SYMBOLVALUE_MAX +1 +1)
+#define HUF_CTABLE_WORKSPACE_SIZE_U32 ((4 * (HUF_SYMBOLVALUE_MAX + 1)) + 192)
 #define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
 size_t HUF_buildCTable_wksp (HUF_CElt* tree,
                        const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
@@ -243,17 +189,16 @@ size_t HUF_readStats_wksp(BYTE* huffWeight, size_t hwSize,
                           U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
                           const void* src, size_t srcSize,
                           void* workspace, size_t wkspSize,
-                          int bmi2);
+                          int flags);
 
 /** HUF_readCTable() :
  *  Loading a CTable saved with HUF_writeCTable() */
 size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
 
-/** HUF_getNbBits() :
+/** HUF_getNbBitsFromCTable() :
  *  Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
- *  Note 1 : is not inlined, as HUF_CElt definition is private
- *  Note 2 : const void* used, so that it can provide a statically allocated table as argument (which uses type U32) */
-U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue);
+ *  Note 1 : is not inlined, as HUF_CElt definition is private */
+U32 HUF_getNbBitsFromCTable(const HUF_CElt* symbolTable, U32 symbolValue);
 
 /*
  * HUF_decompress() does the following:
@@ -282,80 +227,46 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
 #define HUF_DECOMPRESS_WORKSPACE_SIZE ((2 << 10) + (1 << 9))
 #define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
 
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_readDTableX1 (HUF_DTable* DTable, const void* src, size_t srcSize);
-size_t HUF_readDTableX1_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
-#endif
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize);
-size_t HUF_readDTableX2_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
-#endif
-
-size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress4X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#endif
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#endif
-
 
 /* ====================== */
 /* single stream variants */
 /* ====================== */
 
-size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
-size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
-size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags);
 /** HUF_compress1X_repeat() :
  *  Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
  *  If it uses hufTable it does not modify hufTable or repeat.
  *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
- *  If preferRepeat then the old table will always be used if valid. */
+ *  If preferRepeat then the old table will always be used if valid.
+ *  If suspectUncompressible then some sampling checks will be run to potentially skip huffman coding */
 size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
                        const void* src, size_t srcSize,
                        unsigned maxSymbolValue, unsigned tableLog,
                        void* workSpace, size_t wkspSize,   /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
-                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
-
-size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
-#endif
-
-size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
-size_t HUF_decompress1X_DCtx_wksp (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
-#endif
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
-#endif
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int flags);
 
-size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /**< automatic selection of sing or double symbol decoder, based on DTable */
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-#endif
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
 #ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);   /**< double-symbols decoder */
 #endif
 
 /* BMI2 variants.
  * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
  */
-size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
 #ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
 #endif
-size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
-size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags);
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags);
 #ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
+#endif
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags);
 #endif
 
-#endif /* HUF_STATIC_LINKING_ONLY */
+#endif   /* HUF_H_298734234 */
 
 #if defined (__cplusplus)
 }
diff --git a/priv/zstd/lib/common/mem.h b/priv/zstd/lib/common/mem.h
index 9f3b81a..98dd47a 100644
--- a/priv/zstd/lib/common/mem.h
+++ b/priv/zstd/lib/common/mem.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -51,6 +51,8 @@ extern "C" {
 #    include  /* intptr_t */
 #  endif
   typedef   uint8_t BYTE;
+  typedef   uint8_t U8;
+  typedef    int8_t S8;
   typedef  uint16_t U16;
   typedef   int16_t S16;
   typedef  uint32_t U32;
@@ -63,6 +65,8 @@ extern "C" {
 #  error "this implementation requires char to be exactly 8-bit type"
 #endif
   typedef unsigned char      BYTE;
+  typedef unsigned char      U8;
+  typedef   signed char      S8;
 #if USHRT_MAX != 65535
 #  error "this implementation requires short to be exactly 16-bit type"
 #endif
@@ -129,21 +133,15 @@ MEM_STATIC size_t MEM_swapST(size_t in);
 /*-**************************************************************
 *  Memory I/O Implementation
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS :
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+/* MEM_FORCE_MEMORY_ACCESS : For accessing unaligned memory:
+ * Method 0 : always use `memcpy()`. Safe and portable.
+ * Method 1 : Use compiler extension to set unaligned access.
  * Method 2 : direct access. This method is portable but violate C standard.
  *            It can generate buggy code on targets depending on alignment.
- *            In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
+ * Default  : method 1 if supported, else method 0
  */
 #ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__INTEL_COMPILER) || defined(__GNUC__) || defined(__ICCARM__)
+#  ifdef __GNUC__
 #    define MEM_FORCE_MEMORY_ACCESS 1
 #  endif
 #endif
@@ -153,8 +151,22 @@ MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
 
 MEM_STATIC unsigned MEM_isLittleEndian(void)
 {
+#if defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+    return 1;
+#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+    return 0;
+#elif defined(__clang__) && __LITTLE_ENDIAN__
+    return 1;
+#elif defined(__clang__) && __BIG_ENDIAN__
+    return 0;
+#elif defined(_MSC_VER) && (_M_AMD64 || _M_IX86)
+    return 1;
+#elif defined(__DMC__) && defined(_M_IX86)
+    return 1;
+#else
     const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
     return one.c[0];
+#endif
 }
 
 #if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
@@ -172,30 +184,19 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
 
 #elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
 
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-#if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
-    __pragma( pack(push, 1) )
-    typedef struct { U16 v; } unalign16;
-    typedef struct { U32 v; } unalign32;
-    typedef struct { U64 v; } unalign64;
-    typedef struct { size_t v; } unalignArch;
-    __pragma( pack(pop) )
-#else
-    typedef struct { U16 v; } __attribute__((packed)) unalign16;
-    typedef struct { U32 v; } __attribute__((packed)) unalign32;
-    typedef struct { U64 v; } __attribute__((packed)) unalign64;
-    typedef struct { size_t v; } __attribute__((packed)) unalignArch;
-#endif
+typedef __attribute__((aligned(1))) U16 unalign16;
+typedef __attribute__((aligned(1))) U32 unalign32;
+typedef __attribute__((aligned(1))) U64 unalign64;
+typedef __attribute__((aligned(1))) size_t unalignArch;
 
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; }
-MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; }
+MEM_STATIC U16 MEM_read16(const void* ptr) { return *(const unalign16*)ptr; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return *(const unalign32*)ptr; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return *(const unalign64*)ptr; }
+MEM_STATIC size_t MEM_readST(const void* ptr) { return *(const unalignArch*)ptr; }
 
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; }
-MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; }
-MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; }
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(unalign16*)memPtr = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(unalign32*)memPtr = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(unalign64*)memPtr = value; }
 
 #else
 
@@ -239,6 +240,14 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value)
 
 #endif /* MEM_FORCE_MEMORY_ACCESS */
 
+MEM_STATIC U32 MEM_swap32_fallback(U32 in)
+{
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+}
+
 MEM_STATIC U32 MEM_swap32(U32 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
@@ -247,22 +256,13 @@ MEM_STATIC U32 MEM_swap32(U32 in)
   || (defined(__clang__) && __has_builtin(__builtin_bswap32))
     return __builtin_bswap32(in);
 #else
-    return  ((in << 24) & 0xff000000 ) |
-            ((in <<  8) & 0x00ff0000 ) |
-            ((in >>  8) & 0x0000ff00 ) |
-            ((in >> 24) & 0x000000ff );
+    return MEM_swap32_fallback(in);
 #endif
 }
 
-MEM_STATIC U64 MEM_swap64(U64 in)
+MEM_STATIC U64 MEM_swap64_fallback(U64 in)
 {
-#if defined(_MSC_VER)     /* Visual Studio */
-    return _byteswap_uint64(in);
-#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
-  || (defined(__clang__) && __has_builtin(__builtin_bswap64))
-    return __builtin_bswap64(in);
-#else
-    return  ((in << 56) & 0xff00000000000000ULL) |
+     return  ((in << 56) & 0xff00000000000000ULL) |
             ((in << 40) & 0x00ff000000000000ULL) |
             ((in << 24) & 0x0000ff0000000000ULL) |
             ((in << 8)  & 0x000000ff00000000ULL) |
@@ -270,6 +270,17 @@ MEM_STATIC U64 MEM_swap64(U64 in)
             ((in >> 24) & 0x0000000000ff0000ULL) |
             ((in >> 40) & 0x000000000000ff00ULL) |
             ((in >> 56) & 0x00000000000000ffULL);
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap64))
+    return __builtin_bswap64(in);
+#else
+    return MEM_swap64_fallback(in);
 #endif
 }
 
diff --git a/priv/zstd/lib/common/pool.c b/priv/zstd/lib/common/pool.c
index ea70b8b..f3d9d08 100644
--- a/priv/zstd/lib/common/pool.c
+++ b/priv/zstd/lib/common/pool.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -12,7 +12,7 @@
 /* ======   Dependencies   ======= */
 #include "zstd_deps.h" /* size_t */
 #include "debug.h"     /* assert */
-#include "zstd_internal.h"  /* ZSTD_customMalloc, ZSTD_customFree */
+#include "zstd_internal.h"  /* ZSTD_customCalloc, ZSTD_customFree */
 #include "pool.h"
 
 /* ======   Compiler specifics   ====== */
@@ -86,7 +86,7 @@ static void* POOL_thread(void* opaque) {
         {   POOL_job const job = ctx->queue[ctx->queueHead];
             ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
             ctx->numThreadsBusy++;
-            ctx->queueEmpty = ctx->queueHead == ctx->queueTail;
+            ctx->queueEmpty = (ctx->queueHead == ctx->queueTail);
             /* Unlock the mutex, signal a pusher, and run the job */
             ZSTD_pthread_cond_signal(&ctx->queuePushCond);
             ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
@@ -96,15 +96,14 @@ static void* POOL_thread(void* opaque) {
             /* If the intended queue size was 0, signal after finishing job */
             ZSTD_pthread_mutex_lock(&ctx->queueMutex);
             ctx->numThreadsBusy--;
-            if (ctx->queueSize == 1) {
-                ZSTD_pthread_cond_signal(&ctx->queuePushCond);
-            }
+            ZSTD_pthread_cond_signal(&ctx->queuePushCond);
             ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
         }
     }  /* for (;;) */
     assert(0);  /* Unreachable */
 }
 
+/* ZSTD_createThreadPool() : public access point */
 POOL_ctx* ZSTD_createThreadPool(size_t numThreads) {
     return POOL_create (numThreads, 0);
 }
@@ -114,7 +113,8 @@ POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
 }
 
 POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
-                               ZSTD_customMem customMem) {
+                               ZSTD_customMem customMem)
+{
     POOL_ctx* ctx;
     /* Check parameters */
     if (!numThreads) { return NULL; }
@@ -126,7 +126,7 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
      * empty and full queues.
      */
     ctx->queueSize = queueSize + 1;
-    ctx->queue = (POOL_job*)ZSTD_customMalloc(ctx->queueSize * sizeof(POOL_job), customMem);
+    ctx->queue = (POOL_job*)ZSTD_customCalloc(ctx->queueSize * sizeof(POOL_job), customMem);
     ctx->queueHead = 0;
     ctx->queueTail = 0;
     ctx->numThreadsBusy = 0;
@@ -140,7 +140,7 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
     }
     ctx->shutdown = 0;
     /* Allocate space for the thread handles */
-    ctx->threads = (ZSTD_pthread_t*)ZSTD_customMalloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
+    ctx->threads = (ZSTD_pthread_t*)ZSTD_customCalloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
     ctx->threadCapacity = 0;
     ctx->customMem = customMem;
     /* Check for errors */
@@ -173,7 +173,7 @@ static void POOL_join(POOL_ctx* ctx) {
     /* Join all of the threads */
     {   size_t i;
         for (i = 0; i < ctx->threadCapacity; ++i) {
-            ZSTD_pthread_join(ctx->threads[i], NULL);  /* note : could fail */
+            ZSTD_pthread_join(ctx->threads[i]);  /* note : could fail */
     }   }
 }
 
@@ -188,11 +188,22 @@ void POOL_free(POOL_ctx *ctx) {
     ZSTD_customFree(ctx, ctx->customMem);
 }
 
+/*! POOL_joinJobs() :
+ *  Waits for all queued jobs to finish executing.
+ */
+void POOL_joinJobs(POOL_ctx* ctx) {
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    while(!ctx->queueEmpty || ctx->numThreadsBusy > 0) {
+        ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
+    }
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+}
+
 void ZSTD_freeThreadPool (ZSTD_threadPool* pool) {
   POOL_free (pool);
 }
 
-size_t POOL_sizeof(POOL_ctx *ctx) {
+size_t POOL_sizeof(const POOL_ctx* ctx) {
     if (ctx==NULL) return 0;  /* supports sizeof NULL */
     return sizeof(*ctx)
         + ctx->queueSize * sizeof(POOL_job)
@@ -209,7 +220,7 @@ static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
         return 0;
     }
     /* numThreads > threadCapacity */
-    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_customMalloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
+    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_customCalloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
         if (!threadPool) return 1;
         /* replace existing thread pool */
         ZSTD_memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
@@ -257,9 +268,12 @@ static int isQueueFull(POOL_ctx const* ctx) {
 }
 
 
-static void POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
+static void
+POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
 {
-    POOL_job const job = {function, opaque};
+    POOL_job job;
+    job.function = function;
+    job.opaque = opaque;
     assert(ctx != NULL);
     if (ctx->shutdown) return;
 
@@ -313,7 +327,9 @@ POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
     return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
 }
 
-POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
+POOL_ctx*
+POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem)
+{
     (void)numThreads;
     (void)queueSize;
     (void)customMem;
@@ -325,6 +341,11 @@ void POOL_free(POOL_ctx* ctx) {
     (void)ctx;
 }
 
+void POOL_joinJobs(POOL_ctx* ctx){
+    assert(!ctx || ctx == &g_poolCtx);
+    (void)ctx;
+}
+
 int POOL_resize(POOL_ctx* ctx, size_t numThreads) {
     (void)ctx; (void)numThreads;
     return 0;
@@ -341,7 +362,7 @@ int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
     return 1;
 }
 
-size_t POOL_sizeof(POOL_ctx* ctx) {
+size_t POOL_sizeof(const POOL_ctx* ctx) {
     if (ctx==NULL) return 0;  /* supports sizeof NULL */
     assert(ctx == &g_poolCtx);
     return sizeof(*ctx);
diff --git a/priv/zstd/lib/common/pool.h b/priv/zstd/lib/common/pool.h
index e18aa07..eb22ff5 100644
--- a/priv/zstd/lib/common/pool.h
+++ b/priv/zstd/lib/common/pool.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -38,6 +38,12 @@ POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
  */
 void POOL_free(POOL_ctx* ctx);
 
+
+/*! POOL_joinJobs() :
+ *  Waits for all queued jobs to finish executing.
+ */
+void POOL_joinJobs(POOL_ctx* ctx);
+
 /*! POOL_resize() :
  *  Expands or shrinks pool's number of threads.
  *  This is more efficient than releasing + creating a new context,
@@ -53,7 +59,7 @@ int POOL_resize(POOL_ctx* ctx, size_t numThreads);
  * @return threadpool memory usage
  *  note : compatible with NULL (returns 0 in this case)
  */
-size_t POOL_sizeof(POOL_ctx* ctx);
+size_t POOL_sizeof(const POOL_ctx* ctx);
 
 /*! POOL_function :
  *  The function type that can be added to a thread pool.
@@ -70,7 +76,7 @@ void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
 
 
 /*! POOL_tryAdd() :
- *  Add the job `function(opaque)` to thread pool _if_ a worker is available.
+ *  Add the job `function(opaque)` to thread pool _if_ a queue slot is available.
  *  Returns immediately even if not (does not block).
  * @return : 1 if successful, 0 if not.
  */
diff --git a/priv/zstd/lib/common/portability_macros.h b/priv/zstd/lib/common/portability_macros.h
new file mode 100644
index 0000000..8fd6ea8
--- /dev/null
+++ b/priv/zstd/lib/common/portability_macros.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_PORTABILITY_MACROS_H
+#define ZSTD_PORTABILITY_MACROS_H
+
+/**
+ * This header file contains macro definitions to support portability.
+ * This header is shared between C and ASM code, so it MUST only
+ * contain macro definitions. It MUST not contain any C code.
+ *
+ * This header ONLY defines macros to detect platforms/feature support.
+ *
+ */
+
+
+/* compat. with non-clang compilers */
+#ifndef __has_attribute
+  #define __has_attribute(x) 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_builtin
+#  define __has_builtin(x) 0
+#endif
+
+/* compat. with non-clang compilers */
+#ifndef __has_feature
+#  define __has_feature(x) 0
+#endif
+
+/* detects whether we are being compiled under msan */
+#ifndef ZSTD_MEMORY_SANITIZER
+#  if __has_feature(memory_sanitizer)
+#    define ZSTD_MEMORY_SANITIZER 1
+#  else
+#    define ZSTD_MEMORY_SANITIZER 0
+#  endif
+#endif
+
+/* detects whether we are being compiled under asan */
+#ifndef ZSTD_ADDRESS_SANITIZER
+#  if __has_feature(address_sanitizer)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  elif defined(__SANITIZE_ADDRESS__)
+#    define ZSTD_ADDRESS_SANITIZER 1
+#  else
+#    define ZSTD_ADDRESS_SANITIZER 0
+#  endif
+#endif
+
+/* detects whether we are being compiled under dfsan */
+#ifndef ZSTD_DATAFLOW_SANITIZER
+# if __has_feature(dataflow_sanitizer)
+#  define ZSTD_DATAFLOW_SANITIZER 1
+# else
+#  define ZSTD_DATAFLOW_SANITIZER 0
+# endif
+#endif
+
+/* Mark the internal assembly functions as hidden  */
+#ifdef __ELF__
+# define ZSTD_HIDE_ASM_FUNCTION(func) .hidden func
+#else
+# define ZSTD_HIDE_ASM_FUNCTION(func)
+#endif
+
+/* Enable runtime BMI2 dispatch based on the CPU.
+ * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
+ */
+#ifndef DYNAMIC_BMI2
+  #if ((defined(__clang__) && __has_attribute(__target__)) \
+      || (defined(__GNUC__) \
+          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
+      && (defined(__x86_64__) || defined(_M_X64)) \
+      && !defined(__BMI2__)
+  #  define DYNAMIC_BMI2 1
+  #else
+  #  define DYNAMIC_BMI2 0
+  #endif
+#endif
+
+/**
+ * Only enable assembly for GNUC compatible compilers,
+ * because other platforms may not support GAS assembly syntax.
+ *
+ * Only enable assembly for Linux / MacOS, other platforms may
+ * work, but they haven't been tested. This could likely be
+ * extended to BSD systems.
+ *
+ * Disable assembly when MSAN is enabled, because MSAN requires
+ * 100% of code to be instrumented to work.
+ */
+#if defined(__GNUC__)
+#  if defined(__linux__) || defined(__linux) || defined(__APPLE__)
+#    if ZSTD_MEMORY_SANITIZER
+#      define ZSTD_ASM_SUPPORTED 0
+#    elif ZSTD_DATAFLOW_SANITIZER
+#      define ZSTD_ASM_SUPPORTED 0
+#    else
+#      define ZSTD_ASM_SUPPORTED 1
+#    endif
+#  else
+#    define ZSTD_ASM_SUPPORTED 0
+#  endif
+#else
+#  define ZSTD_ASM_SUPPORTED 0
+#endif
+
+/**
+ * Determines whether we should enable assembly for x86-64
+ * with BMI2.
+ *
+ * Enable if all of the following conditions hold:
+ * - ASM hasn't been explicitly disabled by defining ZSTD_DISABLE_ASM
+ * - Assembly is supported
+ * - We are compiling for x86-64 and either:
+ *   - DYNAMIC_BMI2 is enabled
+ *   - BMI2 is supported at compile time
+ */
+#if !defined(ZSTD_DISABLE_ASM) &&                                 \
+    ZSTD_ASM_SUPPORTED &&                                         \
+    defined(__x86_64__) &&                                        \
+    (DYNAMIC_BMI2 || defined(__BMI2__))
+# define ZSTD_ENABLE_ASM_X86_64_BMI2 1
+#else
+# define ZSTD_ENABLE_ASM_X86_64_BMI2 0
+#endif
+
+/*
+ * For x86 ELF targets, add .note.gnu.property section for Intel CET in
+ * assembly sources when CET is enabled.
+ *
+ * Additionally, any function that may be called indirectly must begin
+ * with ZSTD_CET_ENDBRANCH.
+ */
+#if defined(__ELF__) && (defined(__x86_64__) || defined(__i386__)) \
+    && defined(__has_include)
+# if __has_include()
+#  include 
+#  define ZSTD_CET_ENDBRANCH _CET_ENDBR
+# endif
+#endif
+
+#ifndef ZSTD_CET_ENDBRANCH
+# define ZSTD_CET_ENDBRANCH
+#endif
+
+#endif /* ZSTD_PORTABILITY_MACROS_H */
diff --git a/priv/zstd/lib/common/threading.c b/priv/zstd/lib/common/threading.c
index 92cf57c..f234110 100644
--- a/priv/zstd/lib/common/threading.c
+++ b/priv/zstd/lib/common/threading.c
@@ -23,8 +23,7 @@ int g_ZSTD_threading_useless_symbol;
 #if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
 
 /**
- * Windows minimalist Pthread Wrapper, based on :
- * http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ * Windows minimalist Pthread Wrapper
  */
 
 
@@ -35,37 +34,92 @@ int g_ZSTD_threading_useless_symbol;
 
 /* ===  Implementation  === */
 
+typedef struct {
+    void* (*start_routine)(void*);
+    void* arg;
+    int initialized;
+    ZSTD_pthread_cond_t initialized_cond;
+    ZSTD_pthread_mutex_t initialized_mutex;
+} ZSTD_thread_params_t;
+
 static unsigned __stdcall worker(void *arg)
 {
-    ZSTD_pthread_t* const thread = (ZSTD_pthread_t*) arg;
-    thread->arg = thread->start_routine(thread->arg);
+    void* (*start_routine)(void*);
+    void* thread_arg;
+
+    /* Inialized thread_arg and start_routine and signal main thread that we don't need it
+     * to wait any longer.
+     */
+    {
+        ZSTD_thread_params_t*  thread_param = (ZSTD_thread_params_t*)arg;
+        thread_arg = thread_param->arg;
+        start_routine = thread_param->start_routine;
+
+        /* Signal main thread that we are running and do not depend on its memory anymore */
+        ZSTD_pthread_mutex_lock(&thread_param->initialized_mutex);
+        thread_param->initialized = 1;
+        ZSTD_pthread_cond_signal(&thread_param->initialized_cond);
+        ZSTD_pthread_mutex_unlock(&thread_param->initialized_mutex);
+    }
+
+    start_routine(thread_arg);
+
     return 0;
 }
 
 int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
             void* (*start_routine) (void*), void* arg)
 {
+    ZSTD_thread_params_t thread_param;
     (void)unused;
-    thread->arg = arg;
-    thread->start_routine = start_routine;
-    thread->handle = (HANDLE) _beginthreadex(NULL, 0, worker, thread, 0, NULL);
 
-    if (!thread->handle)
+    thread_param.start_routine = start_routine;
+    thread_param.arg = arg;
+    thread_param.initialized = 0;
+    *thread = NULL;
+
+    /* Setup thread initialization synchronization */
+    if(ZSTD_pthread_cond_init(&thread_param.initialized_cond, NULL)) {
+        /* Should never happen on Windows */
+        return -1;
+    }
+    if(ZSTD_pthread_mutex_init(&thread_param.initialized_mutex, NULL)) {
+        /* Should never happen on Windows */
+        ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+        return -1;
+    }
+
+    /* Spawn thread */
+    *thread = (HANDLE)_beginthreadex(NULL, 0, worker, &thread_param, 0, NULL);
+    if (!thread) {
+        ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
+        ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
         return errno;
-    else
-        return 0;
+    }
+
+    /* Wait for thread to be initialized */
+    ZSTD_pthread_mutex_lock(&thread_param.initialized_mutex);
+    while(!thread_param.initialized) {
+        ZSTD_pthread_cond_wait(&thread_param.initialized_cond, &thread_param.initialized_mutex);
+    }
+    ZSTD_pthread_mutex_unlock(&thread_param.initialized_mutex);
+    ZSTD_pthread_mutex_destroy(&thread_param.initialized_mutex);
+    ZSTD_pthread_cond_destroy(&thread_param.initialized_cond);
+
+    return 0;
 }
 
-int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
+int ZSTD_pthread_join(ZSTD_pthread_t thread)
 {
     DWORD result;
 
-    if (!thread.handle) return 0;
+    if (!thread) return 0;
+
+    result = WaitForSingleObject(thread, INFINITE);
+    CloseHandle(thread);
 
-    result = WaitForSingleObject(thread.handle, INFINITE);
     switch (result) {
     case WAIT_OBJECT_0:
-        if (value_ptr) *value_ptr = thread.arg;
         return 0;
     case WAIT_ABANDONED:
         return EINVAL;
diff --git a/priv/zstd/lib/common/threading.h b/priv/zstd/lib/common/threading.h
index fd0060d..fb5c1c8 100644
--- a/priv/zstd/lib/common/threading.h
+++ b/priv/zstd/lib/common/threading.h
@@ -23,8 +23,7 @@ extern "C" {
 #if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
 
 /**
- * Windows minimalist Pthread Wrapper, based on :
- * http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ * Windows minimalist Pthread Wrapper
  */
 #ifdef WINVER
 #  undef WINVER
@@ -62,16 +61,12 @@ extern "C" {
 #define ZSTD_pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
 
 /* ZSTD_pthread_create() and ZSTD_pthread_join() */
-typedef struct {
-    HANDLE handle;
-    void* (*start_routine)(void*);
-    void* arg;
-} ZSTD_pthread_t;
+typedef HANDLE ZSTD_pthread_t;
 
 int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
                    void* (*start_routine) (void*), void* arg);
 
-int ZSTD_pthread_join(ZSTD_pthread_t thread, void** value_ptr);
+int ZSTD_pthread_join(ZSTD_pthread_t thread);
 
 /**
  * add here more wrappers as required
@@ -99,7 +94,7 @@ int ZSTD_pthread_join(ZSTD_pthread_t thread, void** value_ptr);
 
 #define ZSTD_pthread_t                  pthread_t
 #define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
-#define ZSTD_pthread_join(a, b)         pthread_join((a),(b))
+#define ZSTD_pthread_join(a)         pthread_join((a),NULL)
 
 #else /* DEBUGLEVEL >= 1 */
 
@@ -124,7 +119,7 @@ int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond);
 
 #define ZSTD_pthread_t                  pthread_t
 #define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
-#define ZSTD_pthread_join(a, b)         pthread_join((a),(b))
+#define ZSTD_pthread_join(a)         pthread_join((a),NULL)
 
 #endif
 
diff --git a/priv/zstd/lib/common/xxhash.c b/priv/zstd/lib/common/xxhash.c
index 926b336..fd237c9 100644
--- a/priv/zstd/lib/common/xxhash.c
+++ b/priv/zstd/lib/common/xxhash.c
@@ -1,11 +1,11 @@
 /*
  *  xxHash - Fast Hash algorithm
- *  Copyright (c) Yann Collet, Facebook, Inc.
+ *  Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
- *  - xxHash homepage: http://www.xxhash.com
+ *  - xxHash homepage: https://cyan4973.github.io/xxHash/
  *  - xxHash source repository : https://github.com/Cyan4973/xxHash
- * 
+ *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
@@ -13,812 +13,12 @@
 */
 
 
-/* *************************************
-*  Tuning parameters
-***************************************/
-/*!XXH_FORCE_MEMORY_ACCESS :
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method doesn't depend on compiler but violate C standard.
- *            It can generate buggy code on targets which do not support unaligned memory accesses.
- *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://stackoverflow.com/a/32095106/646947 for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) )) || \
-  defined(__ICCARM__)
-#    define XXH_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
-
-/*!XXH_ACCEPT_NULL_INPUT_POINTER :
- * If the input pointer is a null pointer, xxHash default behavior is to trigger a memory access error, since it is a bad pointer.
- * When this option is enabled, xxHash output for null input pointers will be the same as a null-length input.
- * By default, this option is disabled. To enable it, uncomment below define :
- */
-/* #define XXH_ACCEPT_NULL_INPUT_POINTER 1 */
-
-/*!XXH_FORCE_NATIVE_FORMAT :
- * By default, xxHash library provides endian-independent Hash values, based on little-endian convention.
- * Results are therefore identical for little-endian and big-endian CPU.
- * This comes at a performance cost for big-endian CPU, since some swapping is required to emulate little-endian format.
- * Should endian-independence be of no importance for your application, you may set the #define below to 1,
- * to improve speed for Big-endian CPU.
- * This option has no impact on Little_Endian CPU.
- */
-#ifndef XXH_FORCE_NATIVE_FORMAT   /* can be defined externally */
-#  define XXH_FORCE_NATIVE_FORMAT 0
-#endif
 
-/*!XXH_FORCE_ALIGN_CHECK :
- * This is a minor performance trick, only useful with lots of very small keys.
- * It means : check for aligned/unaligned input.
- * The check costs one initial branch per hash; set to 0 when the input data
- * is guaranteed to be aligned.
+/*
+ * xxhash.c instantiates functions defined in xxhash.h
  */
-#ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */
-#  if defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64)
-#    define XXH_FORCE_ALIGN_CHECK 0
-#  else
-#    define XXH_FORCE_ALIGN_CHECK 1
-#  endif
-#endif
-
 
-/* *************************************
-*  Includes & Memory related functions
-***************************************/
-/* Modify the local functions below should you wish to use some other memory routines */
-/* for ZSTD_malloc(), ZSTD_free() */
-#define ZSTD_DEPS_NEED_MALLOC
-#include "zstd_deps.h"  /* size_t, ZSTD_malloc, ZSTD_free, ZSTD_memcpy */
-static void* XXH_malloc(size_t s) { return ZSTD_malloc(s); }
-static void  XXH_free  (void* p)  { ZSTD_free(p); }
-static void* XXH_memcpy(void* dest, const void* src, size_t size) { return ZSTD_memcpy(dest,src,size); }
+#define XXH_STATIC_LINKING_ONLY   /* access advanced declarations */
+#define XXH_IMPLEMENTATION   /* access definitions */
 
-#ifndef XXH_STATIC_LINKING_ONLY
-#  define XXH_STATIC_LINKING_ONLY
-#endif
 #include "xxhash.h"
-
-
-/* *************************************
-*  Compiler Specific Options
-***************************************/
-#include "compiler.h"
-
-
-/* *************************************
-*  Basic Types
-***************************************/
-#include "mem.h"  /* BYTE, U32, U64, size_t */
-
-#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
-
-/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
-static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; }
-static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign;
-
-static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-static U64 XXH_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-#else
-
-/* portable and safe solution. Generally efficient.
- * see : http://stackoverflow.com/a/32095106/646947
- */
-
-static U32 XXH_read32(const void* memPtr)
-{
-    U32 val;
-    ZSTD_memcpy(&val, memPtr, sizeof(val));
-    return val;
-}
-
-static U64 XXH_read64(const void* memPtr)
-{
-    U64 val;
-    ZSTD_memcpy(&val, memPtr, sizeof(val));
-    return val;
-}
-
-#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
-
-
-/* ****************************************
-*  Compiler-specific Functions and Macros
-******************************************/
-#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-
-/* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */
-#if defined(_MSC_VER)
-#  define XXH_rotl32(x,r) _rotl(x,r)
-#  define XXH_rotl64(x,r) _rotl64(x,r)
-#else
-#if defined(__ICCARM__)
-#  include 
-#  define XXH_rotl32(x,r) __ROR(x,(32 - r))
-#else
-#  define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r)))
-#endif
-#  define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r)))
-#endif
-
-#if defined(_MSC_VER)     /* Visual Studio */
-#  define XXH_swap32 _byteswap_ulong
-#  define XXH_swap64 _byteswap_uint64
-#elif GCC_VERSION >= 403
-#  define XXH_swap32 __builtin_bswap32
-#  define XXH_swap64 __builtin_bswap64
-#else
-static U32 XXH_swap32 (U32 x)
-{
-    return  ((x << 24) & 0xff000000 ) |
-            ((x <<  8) & 0x00ff0000 ) |
-            ((x >>  8) & 0x0000ff00 ) |
-            ((x >> 24) & 0x000000ff );
-}
-static U64 XXH_swap64 (U64 x)
-{
-    return  ((x << 56) & 0xff00000000000000ULL) |
-            ((x << 40) & 0x00ff000000000000ULL) |
-            ((x << 24) & 0x0000ff0000000000ULL) |
-            ((x << 8)  & 0x000000ff00000000ULL) |
-            ((x >> 8)  & 0x00000000ff000000ULL) |
-            ((x >> 24) & 0x0000000000ff0000ULL) |
-            ((x >> 40) & 0x000000000000ff00ULL) |
-            ((x >> 56) & 0x00000000000000ffULL);
-}
-#endif
-
-
-/* *************************************
-*  Architecture Macros
-***************************************/
-typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;
-
-/* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */
-#ifndef XXH_CPU_LITTLE_ENDIAN
-    static const int g_one = 1;
-#   define XXH_CPU_LITTLE_ENDIAN   (*(const char*)(&g_one))
-#endif
-
-
-/* ***************************
-*  Memory reads
-*****************************/
-typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;
-
-FORCE_INLINE_TEMPLATE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
-{
-    if (align==XXH_unaligned)
-        return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
-    else
-        return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr);
-}
-
-FORCE_INLINE_TEMPLATE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
-{
-    return XXH_readLE32_align(ptr, endian, XXH_unaligned);
-}
-
-static U32 XXH_readBE32(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
-}
-
-FORCE_INLINE_TEMPLATE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
-{
-    if (align==XXH_unaligned)
-        return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
-    else
-        return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
-}
-
-FORCE_INLINE_TEMPLATE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
-{
-    return XXH_readLE64_align(ptr, endian, XXH_unaligned);
-}
-
-static U64 XXH_readBE64(const void* ptr)
-{
-    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
-}
-
-
-/* *************************************
-*  Macros
-***************************************/
-#define XXH_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(int)(!!(c)) }; }    /* use only *after* variable declarations */
-
-
-/* *************************************
-*  Constants
-***************************************/
-static const U32 PRIME32_1 = 2654435761U;
-static const U32 PRIME32_2 = 2246822519U;
-static const U32 PRIME32_3 = 3266489917U;
-static const U32 PRIME32_4 =  668265263U;
-static const U32 PRIME32_5 =  374761393U;
-
-static const U64 PRIME64_1 = 11400714785074694791ULL;
-static const U64 PRIME64_2 = 14029467366897019727ULL;
-static const U64 PRIME64_3 =  1609587929392839161ULL;
-static const U64 PRIME64_4 =  9650029242287828579ULL;
-static const U64 PRIME64_5 =  2870177450012600261ULL;
-
-XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
-
-
-/* **************************
-*  Utils
-****************************/
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
-{
-    ZSTD_memcpy(dstState, srcState, sizeof(*dstState));
-}
-
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
-{
-    ZSTD_memcpy(dstState, srcState, sizeof(*dstState));
-}
-
-
-/* ***************************
-*  Simple Hash Functions
-*****************************/
-
-static U32 XXH32_round(U32 seed, U32 input)
-{
-    seed += input * PRIME32_2;
-    seed  = XXH_rotl32(seed, 13);
-    seed *= PRIME32_1;
-    return seed;
-}
-
-FORCE_INLINE_TEMPLATE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
-{
-    const BYTE* p = (const BYTE*)input;
-    const BYTE* bEnd = p + len;
-    U32 h32;
-#define XXH_get32bits(p) XXH_readLE32_align(p, endian, align)
-
-#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (p==NULL) {
-        len=0;
-        bEnd=p=(const BYTE*)(size_t)16;
-    }
-#endif
-
-    if (len>=16) {
-        const BYTE* const limit = bEnd - 16;
-        U32 v1 = seed + PRIME32_1 + PRIME32_2;
-        U32 v2 = seed + PRIME32_2;
-        U32 v3 = seed + 0;
-        U32 v4 = seed - PRIME32_1;
-
-        do {
-            v1 = XXH32_round(v1, XXH_get32bits(p)); p+=4;
-            v2 = XXH32_round(v2, XXH_get32bits(p)); p+=4;
-            v3 = XXH32_round(v3, XXH_get32bits(p)); p+=4;
-            v4 = XXH32_round(v4, XXH_get32bits(p)); p+=4;
-        } while (p<=limit);
-
-        h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
-    } else {
-        h32  = seed + PRIME32_5;
-    }
-
-    h32 += (U32) len;
-
-    while (p+4<=bEnd) {
-        h32 += XXH_get32bits(p) * PRIME32_3;
-        h32  = XXH_rotl32(h32, 17) * PRIME32_4 ;
-        p+=4;
-    }
-
-    while (p> 15;
-    h32 *= PRIME32_2;
-    h32 ^= h32 >> 13;
-    h32 *= PRIME32_3;
-    h32 ^= h32 >> 16;
-
-    return h32;
-}
-
-
-XXH_PUBLIC_API unsigned int XXH32 (const void* input, size_t len, unsigned int seed)
-{
-#if 0
-    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
-    XXH32_CREATESTATE_STATIC(state);
-    XXH32_reset(state, seed);
-    XXH32_update(state, input, len);
-    return XXH32_digest(state);
-#else
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if (XXH_FORCE_ALIGN_CHECK) {
-        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
-            if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-                return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
-            else
-                return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
-    }   }
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
-    else
-        return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
-#endif
-}
-
-
-static U64 XXH64_round(U64 acc, U64 input)
-{
-    acc += input * PRIME64_2;
-    acc  = XXH_rotl64(acc, 31);
-    acc *= PRIME64_1;
-    return acc;
-}
-
-static U64 XXH64_mergeRound(U64 acc, U64 val)
-{
-    val  = XXH64_round(0, val);
-    acc ^= val;
-    acc  = acc * PRIME64_1 + PRIME64_4;
-    return acc;
-}
-
-FORCE_INLINE_TEMPLATE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
-{
-    const BYTE* p = (const BYTE*)input;
-    const BYTE* const bEnd = p + len;
-    U64 h64;
-#define XXH_get64bits(p) XXH_readLE64_align(p, endian, align)
-
-#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (p==NULL) {
-        len=0;
-        bEnd=p=(const BYTE*)(size_t)32;
-    }
-#endif
-
-    if (len>=32) {
-        const BYTE* const limit = bEnd - 32;
-        U64 v1 = seed + PRIME64_1 + PRIME64_2;
-        U64 v2 = seed + PRIME64_2;
-        U64 v3 = seed + 0;
-        U64 v4 = seed - PRIME64_1;
-
-        do {
-            v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8;
-            v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8;
-            v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8;
-            v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8;
-        } while (p<=limit);
-
-        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
-        h64 = XXH64_mergeRound(h64, v1);
-        h64 = XXH64_mergeRound(h64, v2);
-        h64 = XXH64_mergeRound(h64, v3);
-        h64 = XXH64_mergeRound(h64, v4);
-
-    } else {
-        h64  = seed + PRIME64_5;
-    }
-
-    h64 += (U64) len;
-
-    while (p+8<=bEnd) {
-        U64 const k1 = XXH64_round(0, XXH_get64bits(p));
-        h64 ^= k1;
-        h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
-        p+=8;
-    }
-
-    if (p+4<=bEnd) {
-        h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
-        h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
-        p+=4;
-    }
-
-    while (p> 33;
-    h64 *= PRIME64_2;
-    h64 ^= h64 >> 29;
-    h64 *= PRIME64_3;
-    h64 ^= h64 >> 32;
-
-    return h64;
-}
-
-
-XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed)
-{
-#if 0
-    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
-    XXH64_CREATESTATE_STATIC(state);
-    XXH64_reset(state, seed);
-    XXH64_update(state, input, len);
-    return XXH64_digest(state);
-#else
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if (XXH_FORCE_ALIGN_CHECK) {
-        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
-            if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-                return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
-            else
-                return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
-    }   }
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
-    else
-        return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
-#endif
-}
-
-
-/* **************************************************
-*  Advanced Hash Functions
-****************************************************/
-
-XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
-{
-    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
-}
-XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
-{
-    XXH_free(statePtr);
-    return XXH_OK;
-}
-
-XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
-{
-    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
-}
-XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
-{
-    XXH_free(statePtr);
-    return XXH_OK;
-}
-
-
-/*** Hash feed ***/
-
-XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
-{
-    XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
-    ZSTD_memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
-    state.v1 = seed + PRIME32_1 + PRIME32_2;
-    state.v2 = seed + PRIME32_2;
-    state.v3 = seed + 0;
-    state.v4 = seed - PRIME32_1;
-    ZSTD_memcpy(statePtr, &state, sizeof(state));
-    return XXH_OK;
-}
-
-
-XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
-{
-    XXH64_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
-    ZSTD_memset(&state, 0, sizeof(state)-8);   /* do not write into reserved, for future removal */
-    state.v1 = seed + PRIME64_1 + PRIME64_2;
-    state.v2 = seed + PRIME64_2;
-    state.v3 = seed + 0;
-    state.v4 = seed - PRIME64_1;
-    ZSTD_memcpy(statePtr, &state, sizeof(state));
-    return XXH_OK;
-}
-
-
-FORCE_INLINE_TEMPLATE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
-{
-    const BYTE* p = (const BYTE*)input;
-    const BYTE* const bEnd = p + len;
-
-#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (input==NULL) return XXH_ERROR;
-#endif
-
-    state->total_len_32 += (unsigned)len;
-    state->large_len |= (len>=16) | (state->total_len_32>=16);
-
-    if (state->memsize + len < 16)  {   /* fill in tmp buffer */
-        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
-        state->memsize += (unsigned)len;
-        return XXH_OK;
-    }
-
-    if (state->memsize) {   /* some data left from previous update */
-        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
-        {   const U32* p32 = state->mem32;
-            state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
-            state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
-            state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
-            state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++;
-        }
-        p += 16-state->memsize;
-        state->memsize = 0;
-    }
-
-    if (p <= bEnd-16) {
-        const BYTE* const limit = bEnd - 16;
-        U32 v1 = state->v1;
-        U32 v2 = state->v2;
-        U32 v3 = state->v3;
-        U32 v4 = state->v4;
-
-        do {
-            v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
-            v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
-            v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
-            v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
-        } while (p<=limit);
-
-        state->v1 = v1;
-        state->v2 = v2;
-        state->v3 = v3;
-        state->v4 = v4;
-    }
-
-    if (p < bEnd) {
-        XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
-        state->memsize = (unsigned)(bEnd-p);
-    }
-
-    return XXH_OK;
-}
-
-XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
-{
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
-    else
-        return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
-}
-
-
-
-FORCE_INLINE_TEMPLATE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
-{
-    const BYTE * p = (const BYTE*)state->mem32;
-    const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
-    U32 h32;
-
-    if (state->large_len) {
-        h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
-    } else {
-        h32 = state->v3 /* == seed */ + PRIME32_5;
-    }
-
-    h32 += state->total_len_32;
-
-    while (p+4<=bEnd) {
-        h32 += XXH_readLE32(p, endian) * PRIME32_3;
-        h32  = XXH_rotl32(h32, 17) * PRIME32_4;
-        p+=4;
-    }
-
-    while (p> 15;
-    h32 *= PRIME32_2;
-    h32 ^= h32 >> 13;
-    h32 *= PRIME32_3;
-    h32 ^= h32 >> 16;
-
-    return h32;
-}
-
-
-XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
-{
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH32_digest_endian(state_in, XXH_littleEndian);
-    else
-        return XXH32_digest_endian(state_in, XXH_bigEndian);
-}
-
-
-
-/* **** XXH64 **** */
-
-FORCE_INLINE_TEMPLATE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
-{
-    const BYTE* p = (const BYTE*)input;
-    const BYTE* const bEnd = p + len;
-
-#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
-    if (input==NULL) return XXH_ERROR;
-#endif
-
-    state->total_len += len;
-
-    if (state->memsize + len < 32) {  /* fill in tmp buffer */
-        if (input != NULL) {
-            XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len);
-        }
-        state->memsize += (U32)len;
-        return XXH_OK;
-    }
-
-    if (state->memsize) {   /* tmp buffer is full */
-        XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize);
-        state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian));
-        state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian));
-        state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian));
-        state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian));
-        p += 32-state->memsize;
-        state->memsize = 0;
-    }
-
-    if (p+32 <= bEnd) {
-        const BYTE* const limit = bEnd - 32;
-        U64 v1 = state->v1;
-        U64 v2 = state->v2;
-        U64 v3 = state->v3;
-        U64 v4 = state->v4;
-
-        do {
-            v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8;
-            v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8;
-            v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8;
-            v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8;
-        } while (p<=limit);
-
-        state->v1 = v1;
-        state->v2 = v2;
-        state->v3 = v3;
-        state->v4 = v4;
-    }
-
-    if (p < bEnd) {
-        XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
-        state->memsize = (unsigned)(bEnd-p);
-    }
-
-    return XXH_OK;
-}
-
-XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len)
-{
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH64_update_endian(state_in, input, len, XXH_littleEndian);
-    else
-        return XXH64_update_endian(state_in, input, len, XXH_bigEndian);
-}
-
-
-
-FORCE_INLINE_TEMPLATE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
-{
-    const BYTE * p = (const BYTE*)state->mem64;
-    const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize;
-    U64 h64;
-
-    if (state->total_len >= 32) {
-        U64 const v1 = state->v1;
-        U64 const v2 = state->v2;
-        U64 const v3 = state->v3;
-        U64 const v4 = state->v4;
-
-        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
-        h64 = XXH64_mergeRound(h64, v1);
-        h64 = XXH64_mergeRound(h64, v2);
-        h64 = XXH64_mergeRound(h64, v3);
-        h64 = XXH64_mergeRound(h64, v4);
-    } else {
-        h64  = state->v3 + PRIME64_5;
-    }
-
-    h64 += (U64) state->total_len;
-
-    while (p+8<=bEnd) {
-        U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian));
-        h64 ^= k1;
-        h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
-        p+=8;
-    }
-
-    if (p+4<=bEnd) {
-        h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1;
-        h64  = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
-        p+=4;
-    }
-
-    while (p> 33;
-    h64 *= PRIME64_2;
-    h64 ^= h64 >> 29;
-    h64 *= PRIME64_3;
-    h64 ^= h64 >> 32;
-
-    return h64;
-}
-
-
-XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in)
-{
-    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
-
-    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
-        return XXH64_digest_endian(state_in, XXH_littleEndian);
-    else
-        return XXH64_digest_endian(state_in, XXH_bigEndian);
-}
-
-
-/* **************************
-*  Canonical representation
-****************************/
-
-/*! Default XXH result types are basic unsigned 32 and 64 bits.
-*   The canonical representation follows human-readable write convention, aka big-endian (large digits first).
-*   These functions allow transformation of hash result into and from its canonical format.
-*   This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
-*/
-
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
-{
-    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
-    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
-    ZSTD_memcpy(dst, &hash, sizeof(*dst));
-}
-
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
-{
-    XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
-    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
-    ZSTD_memcpy(dst, &hash, sizeof(*dst));
-}
-
-XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
-{
-    return XXH_readBE32(src);
-}
-
-XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
-{
-    return XXH_readBE64(src);
-}
diff --git a/priv/zstd/lib/common/xxhash.h b/priv/zstd/lib/common/xxhash.h
index 16c1f16..b8b7329 100644
--- a/priv/zstd/lib/common/xxhash.h
+++ b/priv/zstd/lib/common/xxhash.h
@@ -1,20 +1,36 @@
 /*
- * xxHash - Extremely Fast Hash algorithm
- * Header File
- * Copyright (c) Yann Collet, Facebook, Inc.
+ *  xxHash - Fast Hash algorithm
+ *  Copyright (c) Meta Platforms, Inc. and affiliates.
+ *
+ *  You can contact the author at :
+ *  - xxHash homepage: https://cyan4973.github.io/xxHash/
+ *  - xxHash source repository : https://github.com/Cyan4973/xxHash
  *
- * You can contact the author at :
- * - xxHash source repository : https://github.com/Cyan4973/xxHash
- * 
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
  * You may select, at your option, one of the above-listed licenses.
 */
 
-/* Notice extracted from xxHash homepage :
 
-xxHash is an extremely fast Hash algorithm, running at RAM speed limits.
+#ifndef XXH_NO_XXH3
+# define XXH_NO_XXH3
+#endif
+
+#ifndef XXH_NAMESPACE
+# define XXH_NAMESPACE ZSTD_
+#endif
+
+/*!
+ * @mainpage xxHash
+ *
+ * @file xxhash.h
+ * xxHash prototypes and implementation
+ */
+/* TODO: update */
+/* Notice extracted from xxHash homepage:
+
+xxHash is an extremely fast hash algorithm, running at RAM speed limits.
 It also successfully passes all tests from the SMHasher suite.
 
 Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)
@@ -22,7 +38,7 @@ Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo
 Name            Speed       Q.Score   Author
 xxHash          5.4 GB/s     10
 CrapWow         3.2 GB/s      2       Andrew
-MumurHash 3a    2.7 GB/s     10       Austin Appleby
+MurmurHash 3a   2.7 GB/s     10       Austin Appleby
 SpookyHash      2.0 GB/s     10       Bob Jenkins
 SBox            1.4 GB/s      9       Bret Mulvey
 Lookup3         1.2 GB/s      9       Bob Jenkins
@@ -37,8 +53,13 @@ Q.Score is a measure of quality of the hash function.
 It depends on successfully passing SMHasher test set.
 10 is a perfect score.
 
-A 64-bits version, named XXH64, is available since r35.
-It offers much better speed, but for 64-bits applications only.
+Note: SMHasher's CRC32 implementation is not the fastest one.
+Other speed-oriented implementations can be faster,
+especially in combination with PCLMUL instruction:
+https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735
+
+A 64-bit version, named XXH64, is available since r35.
+It offers much better speed, but for 64-bit applications only.
 Name     Speed on 64 bits    Speed on 32 bits
 XXH64       13.8 GB/s            1.9 GB/s
 XXH32        6.8 GB/s            6.0 GB/s
@@ -48,33 +69,34 @@ XXH32        6.8 GB/s            6.0 GB/s
 extern "C" {
 #endif
 
-#ifndef XXHASH_H_5627135585666179
-#define XXHASH_H_5627135585666179 1
-
-
-/* ****************************
-*  Definitions
-******************************/
-#include "zstd_deps.h"
-typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
-
-
 /* ****************************
-*  API modifier
-******************************/
-/** XXH_PRIVATE_API
-*   This is useful if you want to include xxhash functions in `static` mode
-*   in order to inline them, and remove their symbol from the public list.
-*   Methodology :
-*     #define XXH_PRIVATE_API
-*     #include "xxhash.h"
-*   `xxhash.c` is automatically included.
-*   It's not useful to compile and link it as a separate module anymore.
-*/
-#ifdef XXH_PRIVATE_API
-#  ifndef XXH_STATIC_LINKING_ONLY
-#    define XXH_STATIC_LINKING_ONLY
-#  endif
+ *  INLINE mode
+ ******************************/
+/*!
+ * XXH_INLINE_ALL (and XXH_PRIVATE_API)
+ * Use these build macros to inline xxhash into the target unit.
+ * Inlining improves performance on small inputs, especially when the length is
+ * expressed as a compile-time constant:
+ *
+ *      https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
+ *
+ * It also keeps xxHash symbols private to the unit, so they are not exported.
+ *
+ * Usage:
+ *     #define XXH_INLINE_ALL
+ *     #include "xxhash.h"
+ *
+ * Do not compile and link xxhash.o as a separate object, as it is not useful.
+ */
+#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
+    && !defined(XXH_INLINE_ALL_31684351384)
+   /* this section should be traversed only once */
+#  define XXH_INLINE_ALL_31684351384
+   /* give access to the advanced API, required to compile implementations */
+#  undef XXH_STATIC_LINKING_ONLY   /* avoid macro redef */
+#  define XXH_STATIC_LINKING_ONLY
+   /* make all functions private */
+#  undef XXH_PUBLIC_API
 #  if defined(__GNUC__)
 #    define XXH_PUBLIC_API static __inline __attribute__((unused))
 #  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
@@ -82,45 +104,205 @@ typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
 #  elif defined(_MSC_VER)
 #    define XXH_PUBLIC_API static __inline
 #  else
-#    define XXH_PUBLIC_API static   /* this version may generate warnings for unused static functions; disable the relevant warning */
+     /* note: this version may generate warnings for unused static functions */
+#    define XXH_PUBLIC_API static
 #  endif
-#else
-#  define XXH_PUBLIC_API   /* do nothing */
-#endif /* XXH_PRIVATE_API */
 
-/*!XXH_NAMESPACE, aka Namespace Emulation :
+   /*
+    * This part deals with the special case where a unit wants to inline xxHash,
+    * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
+    * such as part of some previously included *.h header file.
+    * Without further action, the new include would just be ignored,
+    * and functions would effectively _not_ be inlined (silent failure).
+    * The following macros solve this situation by prefixing all inlined names,
+    * avoiding naming collision with previous inclusions.
+    */
+   /* Before that, we unconditionally #undef all symbols,
+    * in case they were already defined with XXH_NAMESPACE.
+    * They will then be redefined for XXH_INLINE_ALL
+    */
+#  undef XXH_versionNumber
+    /* XXH32 */
+#  undef XXH32
+#  undef XXH32_createState
+#  undef XXH32_freeState
+#  undef XXH32_reset
+#  undef XXH32_update
+#  undef XXH32_digest
+#  undef XXH32_copyState
+#  undef XXH32_canonicalFromHash
+#  undef XXH32_hashFromCanonical
+    /* XXH64 */
+#  undef XXH64
+#  undef XXH64_createState
+#  undef XXH64_freeState
+#  undef XXH64_reset
+#  undef XXH64_update
+#  undef XXH64_digest
+#  undef XXH64_copyState
+#  undef XXH64_canonicalFromHash
+#  undef XXH64_hashFromCanonical
+    /* XXH3_64bits */
+#  undef XXH3_64bits
+#  undef XXH3_64bits_withSecret
+#  undef XXH3_64bits_withSeed
+#  undef XXH3_64bits_withSecretandSeed
+#  undef XXH3_createState
+#  undef XXH3_freeState
+#  undef XXH3_copyState
+#  undef XXH3_64bits_reset
+#  undef XXH3_64bits_reset_withSeed
+#  undef XXH3_64bits_reset_withSecret
+#  undef XXH3_64bits_update
+#  undef XXH3_64bits_digest
+#  undef XXH3_generateSecret
+    /* XXH3_128bits */
+#  undef XXH128
+#  undef XXH3_128bits
+#  undef XXH3_128bits_withSeed
+#  undef XXH3_128bits_withSecret
+#  undef XXH3_128bits_reset
+#  undef XXH3_128bits_reset_withSeed
+#  undef XXH3_128bits_reset_withSecret
+#  undef XXH3_128bits_reset_withSecretandSeed
+#  undef XXH3_128bits_update
+#  undef XXH3_128bits_digest
+#  undef XXH128_isEqual
+#  undef XXH128_cmp
+#  undef XXH128_canonicalFromHash
+#  undef XXH128_hashFromCanonical
+    /* Finally, free the namespace itself */
+#  undef XXH_NAMESPACE
 
-If you want to include _and expose_ xxHash functions from within your own library,
-but also want to avoid symbol collisions with another library which also includes xxHash,
+    /* employ the namespace for XXH_INLINE_ALL */
+#  define XXH_NAMESPACE XXH_INLINE_
+   /*
+    * Some identifiers (enums, type names) are not symbols,
+    * but they must nonetheless be renamed to avoid redeclaration.
+    * Alternative solution: do not redeclare them.
+    * However, this requires some #ifdefs, and has a more dispersed impact.
+    * Meanwhile, renaming can be achieved in a single place.
+    */
+#  define XXH_IPREF(Id)   XXH_NAMESPACE ## Id
+#  define XXH_OK XXH_IPREF(XXH_OK)
+#  define XXH_ERROR XXH_IPREF(XXH_ERROR)
+#  define XXH_errorcode XXH_IPREF(XXH_errorcode)
+#  define XXH32_canonical_t  XXH_IPREF(XXH32_canonical_t)
+#  define XXH64_canonical_t  XXH_IPREF(XXH64_canonical_t)
+#  define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
+#  define XXH32_state_s XXH_IPREF(XXH32_state_s)
+#  define XXH32_state_t XXH_IPREF(XXH32_state_t)
+#  define XXH64_state_s XXH_IPREF(XXH64_state_s)
+#  define XXH64_state_t XXH_IPREF(XXH64_state_t)
+#  define XXH3_state_s  XXH_IPREF(XXH3_state_s)
+#  define XXH3_state_t  XXH_IPREF(XXH3_state_t)
+#  define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
+   /* Ensure the header is parsed again, even if it was previously included */
+#  undef XXHASH_H_5627135585666179
+#  undef XXHASH_H_STATIC_13879238742
+#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
 
-you can use XXH_NAMESPACE, to automatically prefix any public symbol from xxhash library
-with the value of XXH_NAMESPACE (so avoid to keep it NULL and avoid numeric values).
 
-Note that no change is required within the calling program as long as it includes `xxhash.h` :
-regular symbol name will be automatically translated by this header.
-*/
+
+/* ****************************************************************
+ *  Stable API
+ *****************************************************************/
+#ifndef XXHASH_H_5627135585666179
+#define XXHASH_H_5627135585666179 1
+
+
+/*!
+ * @defgroup public Public API
+ * Contains details on the public xxHash functions.
+ * @{
+ */
+/* specific declaration modes for Windows */
+#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+#    ifdef XXH_EXPORT
+#      define XXH_PUBLIC_API __declspec(dllexport)
+#    elif XXH_IMPORT
+#      define XXH_PUBLIC_API __declspec(dllimport)
+#    endif
+#  else
+#    define XXH_PUBLIC_API   /* do nothing */
+#  endif
+#endif
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Emulate a namespace by transparently prefixing all symbols.
+ *
+ * If you want to include _and expose_ xxHash functions from within your own
+ * library, but also want to avoid symbol collisions with other libraries which
+ * may also include xxHash, you can use XXH_NAMESPACE to automatically prefix
+ * any public symbol from xxhash library with the value of XXH_NAMESPACE
+ * (therefore, avoid empty or numeric values).
+ *
+ * Note that no change is required within the calling program as long as it
+ * includes `xxhash.h`: Regular symbol names will be automatically translated
+ * by this header.
+ */
+#  define XXH_NAMESPACE /* YOUR NAME HERE */
+#  undef XXH_NAMESPACE
+#endif
+
 #ifdef XXH_NAMESPACE
 #  define XXH_CAT(A,B) A##B
 #  define XXH_NAME2(A,B) XXH_CAT(A,B)
-#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
-#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
 #  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+/* XXH32 */
+#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
 #  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
-#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
 #  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
-#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
 #  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
-#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
 #  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
-#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
 #  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
-#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
 #  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
-#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
 #  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
-#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
 #  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+/* XXH64 */
+#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
 #  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+/* XXH3_64bits */
+#  define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
+#  define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
+#  define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
+#  define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
+#  define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
+#  define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
+#  define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
+#  define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
+#  define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
+#  define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
+#  define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
+#  define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
+#  define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
+#  define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
+#  define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
+/* XXH3_128bits */
+#  define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
+#  define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
+#  define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
+#  define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
+#  define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
+#  define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
+#  define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
+#  define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
+#  define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
+#  define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
+#  define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
+#  define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
+#  define XXH128_cmp     XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
+#  define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
+#  define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
 #endif
 
 
@@ -128,156 +310,5375 @@ regular symbol name will be automatically translated by this header.
 *  Version
 ***************************************/
 #define XXH_VERSION_MAJOR    0
-#define XXH_VERSION_MINOR    6
-#define XXH_VERSION_RELEASE  2
+#define XXH_VERSION_MINOR    8
+#define XXH_VERSION_RELEASE  1
 #define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
+
+/*!
+ * @brief Obtains the xxHash version.
+ *
+ * This is mostly useful when xxHash is compiled as a shared library,
+ * since the returned value comes from the library, as opposed to header file.
+ *
+ * @return `XXH_VERSION_NUMBER` of the invoked library.
+ */
 XXH_PUBLIC_API unsigned XXH_versionNumber (void);
 
 
 /* ****************************
-*  Simple Hash Functions
+*  Common basic types
 ******************************/
-typedef unsigned int       XXH32_hash_t;
-typedef unsigned long long XXH64_hash_t;
-
-XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, unsigned int seed);
-XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, unsigned long long seed);
-
-/*!
-XXH32() :
-    Calculate the 32-bits hash of sequence "length" bytes stored at memory address "input".
-    The memory between input & input+length must be valid (allocated and read-accessible).
-    "seed" can be used to alter the result predictably.
-    Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s
-XXH64() :
-    Calculate the 64-bits hash of sequence of length "len" stored at memory address "input".
-    "seed" can be used to alter the result predictably.
-    This function runs 2x faster on 64-bits systems, but slower on 32-bits systems (see benchmark).
-*/
+#include    /* size_t */
+typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
 
 
-/* ****************************
-*  Streaming Hash Functions
-******************************/
-typedef struct XXH32_state_s XXH32_state_t;   /* incomplete type */
-typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+/*-**********************************************************************
+*  32-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* Don't show  include */
+/*!
+ * @brief An unsigned 32-bit integer.
+ *
+ * Not necessarily defined to `uint32_t` but functionally equivalent.
+ */
+typedef uint32_t XXH32_hash_t;
 
-/*! State allocation, compatible with dynamic libraries */
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#   include 
+    typedef uint32_t XXH32_hash_t;
 
-XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
-XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
+#else
+#   include 
+#   if UINT_MAX == 0xFFFFFFFFUL
+      typedef unsigned int XXH32_hash_t;
+#   else
+#     if ULONG_MAX == 0xFFFFFFFFUL
+        typedef unsigned long XXH32_hash_t;
+#     else
+#       error "unsupported platform: need a 32-bit type"
+#     endif
+#   endif
+#endif
 
-XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
-XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+/*!
+ * @}
+ *
+ * @defgroup xxh32_family XXH32 family
+ * @ingroup public
+ * Contains functions used in the classic 32-bit xxHash algorithm.
+ *
+ * @note
+ *   XXH32 is useful for older platforms, with no or poor 64-bit performance.
+ *   Note that @ref xxh3_family provides competitive speed
+ *   for both 32-bit and 64-bit systems, and offers true 64/128 bit hash results.
+ *
+ * @see @ref xxh64_family, @ref xxh3_family : Other xxHash families
+ * @see @ref xxh32_impl for implementation details
+ * @{
+ */
+
+/*!
+ * @brief Calculates the 32-bit hash of @p input using xxHash32.
+ *
+ * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
+ *
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed The 32-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 32-bit hash value.
+ *
+ * @see
+ *    XXH64(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
+ *    Direct equivalents for the other variants of xxHash.
+ * @see
+ *    XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
+
+/*!
+ * Streaming functions generate the xxHash value from an incremental input.
+ * This method is slower than single-call functions, due to state management.
+ * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
+ *
+ * An XXH state must first be allocated using `XXH*_createState()`.
+ *
+ * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
+ *
+ * Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
+ *
+ * The function returns an error code, with 0 meaning OK, and any other value
+ * meaning there is an error.
+ *
+ * Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
+ * This function returns the nn-bits hash as an int or long long.
+ *
+ * It's still possible to continue inserting input into the hash state after a
+ * digest, and generate new hash values later on by invoking `XXH*_digest()`.
+ *
+ * When done, release the state using `XXH*_freeState()`.
+ *
+ * Example code for incrementally hashing a file:
+ * @code{.c}
+ *    #include 
+ *    #include 
+ *    #define BUFFER_SIZE 256
+ *
+ *    // Note: XXH64 and XXH3 use the same interface.
+ *    XXH32_hash_t
+ *    hashFile(FILE* stream)
+ *    {
+ *        XXH32_state_t* state;
+ *        unsigned char buf[BUFFER_SIZE];
+ *        size_t amt;
+ *        XXH32_hash_t hash;
+ *
+ *        state = XXH32_createState();       // Create a state
+ *        assert(state != NULL);             // Error check here
+ *        XXH32_reset(state, 0xbaad5eed);    // Reset state with our seed
+ *        while ((amt = fread(buf, 1, sizeof(buf), stream)) != 0) {
+ *            XXH32_update(state, buf, amt); // Hash the file in chunks
+ *        }
+ *        hash = XXH32_digest(state);        // Finalize the hash
+ *        XXH32_freeState(state);            // Clean up
+ *        return hash;
+ *    }
+ * @endcode
+ */
+
+/*!
+ * @typedef struct XXH32_state_s XXH32_state_t
+ * @brief The opaque state struct for the XXH32 streaming API.
+ *
+ * @see XXH32_state_s for details.
+ */
+typedef struct XXH32_state_s XXH32_state_t;
 
+/*!
+ * @brief Allocates an @ref XXH32_state_t.
+ *
+ * Must be freed with XXH32_freeState().
+ * @return An allocated XXH32_state_t on success, `NULL` on failure.
+ */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
+/*!
+ * @brief Frees an @ref XXH32_state_t.
+ *
+ * Must be allocated with XXH32_createState().
+ * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
+ * @return XXH_OK.
+ */
+XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
+/*!
+ * @brief Copies one @ref XXH32_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
+ */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);
 
-/* hash streaming */
+/*!
+ * @brief Resets an @ref XXH32_state_t to begin a new hash.
+ *
+ * This function resets and seeds a state. Call it before @ref XXH32_update().
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 32-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t seed);
 
-XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, unsigned int seed);
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH32_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
 XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated hash value from an @ref XXH32_state_t.
+ *
+ * @note
+ *   Calling XXH32_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ *  @p statePtr must not be `NULL`.
+ *
+ * @return The calculated xxHash32 value from that state.
+ */
 XXH_PUBLIC_API XXH32_hash_t  XXH32_digest (const XXH32_state_t* statePtr);
 
-XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, unsigned long long seed);
-XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
-XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
+/*******   Canonical representation   *******/
 
 /*
-These functions generate the xxHash of an input provided in multiple segments.
-Note that, for small input, they are slower than single-call functions, due to state management.
-For small input, prefer `XXH32()` and `XXH64()` .
+ * The default return values from XXH functions are unsigned 32 and 64 bit
+ * integers.
+ * This the simplest and fastest format for further post-processing.
+ *
+ * However, this leaves open the question of what is the order on the byte level,
+ * since little and big endian conventions will store the same number differently.
+ *
+ * The canonical representation settles this issue by mandating big-endian
+ * convention, the same convention as human-readable numbers (large digits first).
+ *
+ * When writing hash values to storage, sending them over a network, or printing
+ * them, it's highly recommended to use the canonical representation to ensure
+ * portability across a wider range of systems, present and future.
+ *
+ * The following functions allow transformation of hash values to and from
+ * canonical format.
+ */
+
+/*!
+ * @brief Canonical (big endian) representation of @ref XXH32_hash_t.
+ */
+typedef struct {
+    unsigned char digest[4]; /*!< Hash bytes, big endian */
+} XXH32_canonical_t;
 
-XXH state must first be allocated, using XXH*_createState() .
+/*!
+ * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t.
+ *
+ * @param dst The @ref XXH32_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH32_hash_t to be converted.
+ *
+ * @pre
+ *   @p dst must not be `NULL`.
+ */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
+
+/*!
+ * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
+ *
+ * @param src The @ref XXH32_canonical_t to convert.
+ *
+ * @pre
+ *   @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
 
-Start a new hash by initializing state with a seed, using XXH*_reset().
 
-Then, feed the hash state by calling XXH*_update() as many times as necessary.
-Obviously, input must be allocated and read accessible.
-The function returns an error code, with 0 meaning OK, and any other value meaning there is an error.
+#ifdef __has_attribute
+# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
+#else
+# define XXH_HAS_ATTRIBUTE(x) 0
+#endif
 
-Finally, a hash value can be produced anytime, by using XXH*_digest().
-This function returns the nn-bits hash as an int or long long.
+/* C-language Attributes are added in C23. */
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
+# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+#else
+# define XXH_HAS_C_ATTRIBUTE(x) 0
+#endif
 
-It's still possible to continue inserting input into the hash state after a digest,
-and generate some new hashes later on, by calling again XXH*_digest().
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+# define XXH_HAS_CPP_ATTRIBUTE(x) 0
+#endif
 
-When done, free XXH state space if it was allocated dynamically.
+/*
+Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
+introduced in CPP17 and C23.
+CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+C23   : https://en.cppreference.com/w/c/language/attributes/fallthrough
 */
+#if XXH_HAS_C_ATTRIBUTE(x)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_CPP_ATTRIBUTE(x)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
+# define XXH_FALLTHROUGH __attribute__ ((fallthrough))
+#else
+# define XXH_FALLTHROUGH
+#endif
 
+/*!
+ * @}
+ * @ingroup public
+ * @{
+ */
 
-/* **************************
-*  Utils
-****************************/
-#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* ! C99 */
-#  define restrict   /* disable restrict */
+#ifndef XXH_NO_LONG_LONG
+/*-**********************************************************************
+*  64-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* don't include  */
+/*!
+ * @brief An unsigned 64-bit integer.
+ *
+ * Not necessarily defined to `uint64_t` but functionally equivalent.
+ */
+typedef uint64_t XXH64_hash_t;
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#  include 
+   typedef uint64_t XXH64_hash_t;
+#else
+#  include 
+#  if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
+     /* LP64 ABI says uint64_t is unsigned long */
+     typedef unsigned long XXH64_hash_t;
+#  else
+     /* the following type must have a width of 64-bit */
+     typedef unsigned long long XXH64_hash_t;
+#  endif
 #endif
 
-XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dst_state, const XXH32_state_t* restrict src_state);
-XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dst_state, const XXH64_state_t* restrict src_state);
+/*!
+ * @}
+ *
+ * @defgroup xxh64_family XXH64 family
+ * @ingroup public
+ * @{
+ * Contains functions used in the classic 64-bit xxHash algorithm.
+ *
+ * @note
+ *   XXH3 provides competitive speed for both 32-bit and 64-bit systems,
+ *   and offers true 64/128 bit hash results.
+ *   It provides better speed for systems with vector processing capabilities.
+ */
 
 
-/* **************************
-*  Canonical representation
-****************************/
-/* Default result type for XXH functions are primitive unsigned 32 and 64 bits.
-*  The canonical representation uses human-readable write convention, aka big-endian (large digits first).
-*  These functions allow transformation of hash result into and from its canonical format.
-*  This way, hash values can be written into a file / memory, and remain comparable on different systems and programs.
-*/
-typedef struct { unsigned char digest[4]; } XXH32_canonical_t;
-typedef struct { unsigned char digest[8]; } XXH64_canonical_t;
+/*!
+ * @brief Calculates the 64-bit hash of @p input using xxHash64.
+ *
+ * This function usually runs faster on 64-bit systems, but slower on 32-bit
+ * systems (see benchmark).
+ *
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed The 64-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 64-bit hash.
+ *
+ * @see
+ *    XXH32(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
+ *    Direct equivalents for the other variants of xxHash.
+ * @see
+ *    XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH64(const void* input, size_t length, XXH64_hash_t seed);
 
-XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
-XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
+/*******   Streaming   *******/
+/*!
+ * @brief The opaque state struct for the XXH64 streaming API.
+ *
+ * @see XXH64_state_s for details.
+ */
+typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
+XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dst_state, const XXH64_state_t* src_state);
 
-XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
+
+/*******   Canonical representation   *******/
+typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
 XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
 
+#ifndef XXH_NO_XXH3
+/*!
+ * @}
+ * ************************************************************************
+ * @defgroup xxh3_family XXH3 family
+ * @ingroup public
+ * @{
+ *
+ * XXH3 is a more recent hash algorithm featuring:
+ *  - Improved speed for both small and large inputs
+ *  - True 64-bit and 128-bit outputs
+ *  - SIMD acceleration
+ *  - Improved 32-bit viability
+ *
+ * Speed analysis methodology is explained here:
+ *
+ *    https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html
+ *
+ * Compared to XXH64, expect XXH3 to run approximately
+ * ~2x faster on large inputs and >3x faster on small ones,
+ * exact differences vary depending on platform.
+ *
+ * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
+ * but does not require it.
+ * Any 32-bit and 64-bit targets that can run XXH32 smoothly
+ * can run XXH3 at competitive speeds, even without vector support.
+ * Further details are explained in the implementation.
+ *
+ * Optimized implementations are provided for AVX512, AVX2, SSE2, NEON, POWER8,
+ * ZVector and scalar targets. This can be controlled via the XXH_VECTOR macro.
+ *
+ * XXH3 implementation is portable:
+ * it has a generic C90 formulation that can be compiled on any platform,
+ * all implementations generage exactly the same hash value on all platforms.
+ * Starting from v0.8.0, it's also labelled "stable", meaning that
+ * any future version will also generate the same hash value.
+ *
+ * XXH3 offers 2 variants, _64bits and _128bits.
+ *
+ * When only 64 bits are needed, prefer invoking the _64bits variant, as it
+ * reduces the amount of mixing, resulting in faster speed on small inputs.
+ * It's also generally simpler to manipulate a scalar return type than a struct.
+ *
+ * The API supports one-shot hashing, streaming mode, and custom secrets.
+ */
+
+/*-**********************************************************************
+*  XXH3 64-bit variant
+************************************************************************/
+
+/* XXH3_64bits():
+ * default 64-bit variant, using default secret and default seed of 0.
+ * It's the fastest variant. */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* data, size_t len);
+
+/*
+ * XXH3_64bits_withSeed():
+ * This variant generates a custom secret on the fly
+ * based on default secret altered using the `seed` value.
+ * While this operation is decently fast, note that it's not completely free.
+ * Note: seed==0 produces the same results as XXH3_64bits().
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+
+/*!
+ * The bare minimum size for a custom secret.
+ *
+ * @see
+ *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
+ *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
+ */
+#define XXH3_SECRET_SIZE_MIN 136
+
+/*
+ * XXH3_64bits_withSecret():
+ * It's possible to provide any blob of bytes as a "secret" to generate the hash.
+ * This makes it more difficult for an external actor to prepare an intentional collision.
+ * The main condition is that secretSize *must* be large enough (>= XXH3_SECRET_SIZE_MIN).
+ * However, the quality of the secret impacts the dispersion of the hash algorithm.
+ * Therefore, the secret _must_ look like a bunch of random bytes.
+ * Avoid "trivial" or structured data such as repeated sequences or a text document.
+ * Whenever in doubt about the "randomness" of the blob of bytes,
+ * consider employing "XXH3_generateSecret()" instead (see below).
+ * It will generate a proper high entropy secret derived from the blob of bytes.
+ * Another advantage of using XXH3_generateSecret() is that
+ * it guarantees that all bits within the initial blob of bytes
+ * will impact every bit of the output.
+ * This is not necessarily the case when using the blob of bytes directly
+ * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+
+
+/*******   Streaming   *******/
+/*
+ * Streaming requires state maintenance.
+ * This operation costs memory and CPU.
+ * As a consequence, streaming is slower than one-shot hashing.
+ * For better performance, prefer one-shot functions whenever applicable.
+ */
+
+/*!
+ * @brief The state struct for the XXH3 streaming API.
+ *
+ * @see XXH3_state_s for details.
+ */
+typedef struct XXH3_state_s XXH3_state_t;
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void);
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
+XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state);
+
+/*
+ * XXH3_64bits_reset():
+ * Initialize with default parameters.
+ * digest will be equivalent to `XXH3_64bits()`.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t* statePtr);
+/*
+ * XXH3_64bits_reset_withSeed():
+ * Generate a custom secret from `seed`, and store it into `statePtr`.
+ * digest will be equivalent to `XXH3_64bits_withSeed()`.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
+/*
+ * XXH3_64bits_reset_withSecret():
+ * `secret` is referenced, it _must outlive_ the hash streaming session.
+ * Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
+ * and the quality of produced hash values depends on secret's entropy
+ * (secret's content should look like a bunch of random bytes).
+ * When in doubt about the randomness of a candidate `secret`,
+ * consider employing `XXH3_generateSecret()` instead (see below).
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_digest (const XXH3_state_t* statePtr);
+
+/* note : canonical representation of XXH3 is the same as XXH64
+ * since they both produce XXH64_hash_t values */
+
+
+/*-**********************************************************************
+*  XXH3 128-bit variant
+************************************************************************/
+
+/*!
+ * @brief The return value from 128-bit hashes.
+ *
+ * Stored in little endian order, although the fields themselves are in native
+ * endianness.
+ */
+typedef struct {
+    XXH64_hash_t low64;   /*!< `value & 0xFFFFFFFFFFFFFFFF` */
+    XXH64_hash_t high64;  /*!< `value >> 64` */
+} XXH128_hash_t;
+
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* data, size_t len);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+
+/*******   Streaming   *******/
+/*
+ * Streaming requires state maintenance.
+ * This operation costs memory and CPU.
+ * As a consequence, streaming is slower than one-shot hashing.
+ * For better performance, prefer one-shot functions whenever applicable.
+ *
+ * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
+ * Use already declared XXH3_createState() and XXH3_freeState().
+ *
+ * All reset and streaming functions have same meaning as their 64-bit counterpart.
+ */
+
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr);
+
+/* Following helper functions make it possible to compare XXH128_hast_t values.
+ * Since XXH128_hash_t is a structure, this capability is not offered by the language.
+ * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */
+
+/*!
+ * XXH128_isEqual():
+ * Return: 1 if `h1` and `h2` are equal, 0 if they are not.
+ */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
+
+/*!
+ * XXH128_cmp():
+ *
+ * This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
+ *
+ * return: >0 if *h128_1  > *h128_2
+ *         =0 if *h128_1 == *h128_2
+ *         <0 if *h128_1  < *h128_2
+ */
+XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2);
+
+
+/*******   Canonical representation   *******/
+typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
+XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash);
+XXH_PUBLIC_API XXH128_hash_t XXH128_hashFromCanonical(const XXH128_canonical_t* src);
+
+
+#endif  /* !XXH_NO_XXH3 */
+#endif  /* XXH_NO_LONG_LONG */
+
+/*!
+ * @}
+ */
 #endif /* XXHASH_H_5627135585666179 */
 
 
 
-/* ================================================================================================
-   This section contains definitions which are not guaranteed to remain stable.
-   They may change in future versions, becoming incompatible with a different version of the library.
-   They shall only be used with static linking.
-   Never use these definitions in association with dynamic linking !
-=================================================================================================== */
-#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXH_STATIC_H_3543687687345)
-#define XXH_STATIC_H_3543687687345
-
-/* These definitions are only meant to allow allocation of XXH state
-   statically, on stack, or in a struct for example.
-   Do not use members directly. */
-
-   struct XXH32_state_s {
-       unsigned total_len_32;
-       unsigned large_len;
-       unsigned v1;
-       unsigned v2;
-       unsigned v3;
-       unsigned v4;
-       unsigned mem32[4];   /* buffer defined as U32 for alignment */
-       unsigned memsize;
-       unsigned reserved;   /* never read nor write, will be removed in a future version */
-   };   /* typedef'd to XXH32_state_t */
-
-   struct XXH64_state_s {
-       unsigned long long total_len;
-       unsigned long long v1;
-       unsigned long long v2;
-       unsigned long long v3;
-       unsigned long long v4;
-       unsigned long long mem64[4];   /* buffer defined as U64 for alignment */
-       unsigned memsize;
-       unsigned reserved[2];          /* never read nor write, will be removed in a future version */
-   };   /* typedef'd to XXH64_state_t */
-
-
-#  ifdef XXH_PRIVATE_API
-#    include "xxhash.c"   /* include xxhash functions as `static`, for inlining */
-#  endif
+#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
+#define XXHASH_H_STATIC_13879238742
+/* ****************************************************************************
+ * This section contains declarations which are not guaranteed to remain stable.
+ * They may change in future versions, becoming incompatible with a different
+ * version of the library.
+ * These declarations should only be used with static linking.
+ * Never use them in association with dynamic linking!
+ ***************************************************************************** */
+
+/*
+ * These definitions are only present to allow static allocation
+ * of XXH states, on stack or in a struct, for example.
+ * Never **ever** access their members directly.
+ */
+
+/*!
+ * @internal
+ * @brief Structure for XXH32 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
+ * an opaque type. This allows fields to safely be changed.
+ *
+ * Typedef'd to @ref XXH32_state_t.
+ * Do not access the members of this struct directly.
+ * @see XXH64_state_s, XXH3_state_s
+ */
+struct XXH32_state_s {
+   XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
+   XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
+   XXH32_hash_t v[4];         /*!< Accumulator lanes */
+   XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem32 */
+   XXH32_hash_t reserved;     /*!< Reserved field. Do not read nor write to it. */
+};   /* typedef'd to XXH32_state_t */
+
 
-#endif /* XXH_STATIC_LINKING_ONLY && XXH_STATIC_H_3543687687345 */
+#ifndef XXH_NO_LONG_LONG  /* defined when there is no 64-bit support */
+
+/*!
+ * @internal
+ * @brief Structure for XXH64 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
+ * an opaque type. This allows fields to safely be changed.
+ *
+ * Typedef'd to @ref XXH64_state_t.
+ * Do not access the members of this struct directly.
+ * @see XXH32_state_s, XXH3_state_s
+ */
+struct XXH64_state_s {
+   XXH64_hash_t total_len;    /*!< Total length hashed. This is always 64-bit. */
+   XXH64_hash_t v[4];         /*!< Accumulator lanes */
+   XXH64_hash_t mem64[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem64 */
+   XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
+   XXH64_hash_t reserved64;   /*!< Reserved field. Do not read or write to it. */
+};   /* typedef'd to XXH64_state_t */
+
+
+#ifndef XXH_NO_XXH3
+
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
+#  include 
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
+/* In C++ alignas() is a keyword */
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__GNUC__)
+#  define XXH_ALIGN(n)      __attribute__ ((aligned(n)))
+#elif defined(_MSC_VER)
+#  define XXH_ALIGN(n)      __declspec(align(n))
+#else
+#  define XXH_ALIGN(n)   /* disabled */
+#endif
+
+/* Old GCC versions only accept the attribute after the type in structures. */
+#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L))   /* C11+ */ \
+    && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
+    && defined(__GNUC__)
+#   define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
+#else
+#   define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
+#endif
+
+/*!
+ * @brief The size of the internal XXH3 buffer.
+ *
+ * This is the optimal update size for incremental hashing.
+ *
+ * @see XXH3_64b_update(), XXH3_128b_update().
+ */
+#define XXH3_INTERNALBUFFER_SIZE 256
+
+/*!
+ * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
+ *
+ * This is the size used in @ref XXH3_kSecret and the seeded functions.
+ *
+ * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
+ */
+#define XXH3_SECRET_DEFAULT_SIZE 192
+
+/*!
+ * @internal
+ * @brief Structure for XXH3 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined.
+ * Otherwise it is an opaque type.
+ * Never use this definition in combination with dynamic library.
+ * This allows fields to safely be changed in the future.
+ *
+ * @note ** This structure has a strict alignment requirement of 64 bytes!! **
+ * Do not allocate this with `malloc()` or `new`,
+ * it will not be sufficiently aligned.
+ * Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation.
+ *
+ * Typedef'd to @ref XXH3_state_t.
+ * Do never access the members of this struct directly.
+ *
+ * @see XXH3_INITSTATE() for stack initialization.
+ * @see XXH3_createState(), XXH3_freeState().
+ * @see XXH32_state_s, XXH64_state_s
+ */
+struct XXH3_state_s {
+   XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
+       /*!< The 8 accumulators. Similar to `vN` in @ref XXH32_state_s::v1 and @ref XXH64_state_s */
+   XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
+       /*!< Used to store a custom secret generated from a seed. */
+   XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
+       /*!< The internal buffer. @see XXH32_state_s::mem32 */
+   XXH32_hash_t bufferedSize;
+       /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
+   XXH32_hash_t useSeed;
+       /*!< Reserved field. Needed for padding on 64-bit. */
+   size_t nbStripesSoFar;
+       /*!< Number or stripes processed. */
+   XXH64_hash_t totalLen;
+       /*!< Total length hashed. 64-bit even on 32-bit targets. */
+   size_t nbStripesPerBlock;
+       /*!< Number of stripes per block. */
+   size_t secretLimit;
+       /*!< Size of @ref customSecret or @ref extSecret */
+   XXH64_hash_t seed;
+       /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
+   XXH64_hash_t reserved64;
+       /*!< Reserved field. */
+   const unsigned char* extSecret;
+       /*!< Reference to an external secret for the _withSecret variants, NULL
+        *   for other variants. */
+   /* note: there may be some padding at the end due to alignment on 64 bytes */
+}; /* typedef'd to XXH3_state_t */
+
+#undef XXH_ALIGN_MEMBER
+
+/*!
+ * @brief Initializes a stack-allocated `XXH3_state_s`.
+ *
+ * When the @ref XXH3_state_t structure is merely emplaced on stack,
+ * it should be initialized with XXH3_INITSTATE() or a memset()
+ * in case its first reset uses XXH3_NNbits_reset_withSeed().
+ * This init can be omitted if the first reset uses default or _withSecret mode.
+ * This operation isn't necessary when the state is created with XXH3_createState().
+ * Note that this doesn't prepare the state for a streaming operation,
+ * it's still necessary to use XXH3_NNbits_reset*() afterwards.
+ */
+#define XXH3_INITSTATE(XXH3_state_ptr)   { (XXH3_state_ptr)->seed = 0; }
+
+
+/* XXH128() :
+ * simple alias to pre-selected XXH3_128bits variant
+ */
+XXH_PUBLIC_API XXH128_hash_t XXH128(const void* data, size_t len, XXH64_hash_t seed);
+
+
+/* ===   Experimental API   === */
+/* Symbols defined below must be considered tied to a specific library version. */
+
+/*
+ * XXH3_generateSecret():
+ *
+ * Derive a high-entropy secret from any user-defined content, named customSeed.
+ * The generated secret can be used in combination with `*_withSecret()` functions.
+ * The `_withSecret()` variants are useful to provide a higher level of protection than 64-bit seed,
+ * as it becomes much more difficult for an external actor to guess how to impact the calculation logic.
+ *
+ * The function accepts as input a custom seed of any length and any content,
+ * and derives from it a high-entropy secret of length @secretSize
+ * into an already allocated buffer @secretBuffer.
+ * @secretSize must be >= XXH3_SECRET_SIZE_MIN
+ *
+ * The generated secret can then be used with any `*_withSecret()` variant.
+ * Functions `XXH3_128bits_withSecret()`, `XXH3_64bits_withSecret()`,
+ * `XXH3_128bits_reset_withSecret()` and `XXH3_64bits_reset_withSecret()`
+ * are part of this list. They all accept a `secret` parameter
+ * which must be large enough for implementation reasons (>= XXH3_SECRET_SIZE_MIN)
+ * _and_ feature very high entropy (consist of random-looking bytes).
+ * These conditions can be a high bar to meet, so
+ * XXH3_generateSecret() can be employed to ensure proper quality.
+ *
+ * customSeed can be anything. It can have any size, even small ones,
+ * and its content can be anything, even "poor entropy" sources such as a bunch of zeroes.
+ * The resulting `secret` will nonetheless provide all required qualities.
+ *
+ * When customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize);
+
+
+/*
+ * XXH3_generateSecret_fromSeed():
+ *
+ * Generate the same secret as the _withSeed() variants.
+ *
+ * The resulting secret has a length of XXH3_SECRET_DEFAULT_SIZE (necessarily).
+ * @secretBuffer must be already allocated, of size at least XXH3_SECRET_DEFAULT_SIZE bytes.
+ *
+ * The generated secret can be used in combination with
+ *`*_withSecret()` and `_withSecretandSeed()` variants.
+ * This generator is notably useful in combination with `_withSecretandSeed()`,
+ * as a way to emulate a faster `_withSeed()` variant.
+ */
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed);
+
+/*
+ * *_withSecretandSeed() :
+ * These variants generate hash values using either
+ * @seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
+ * or @secret for "large" keys (>= XXH3_MIDSIZE_MAX).
+ *
+ * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
+ * `_withSeed()` has to generate the secret on the fly for "large" keys.
+ * It's fast, but can be perceptible for "not so large" keys (< 1 KB).
+ * `_withSecret()` has to generate the masks on the fly for "small" keys,
+ * which requires more instructions than _withSeed() variants.
+ * Therefore, _withSecretandSeed variant combines the best of both worlds.
+ *
+ * When @secret has been generated by XXH3_generateSecret_fromSeed(),
+ * this variant produces *exactly* the same results as `_withSeed()` variant,
+ * hence offering only a pure speed benefit on "large" input,
+ * by skipping the need to regenerate the secret for every large input.
+ *
+ * Another usage scenario is to hash the secret to a 64-bit hash value,
+ * for example with XXH3_64bits(), which then becomes the seed,
+ * and then employ both the seed and the secret in _withSecretandSeed().
+ * On top of speed, an added benefit is that each bit in the secret
+ * has a 50% chance to swap each bit in the output,
+ * via its impact to the seed.
+ * This is not guaranteed when using the secret directly in "small data" scenarios,
+ * because only portions of the secret are employed for small data.
+ */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(const void* data, size_t len,
+                              const void* secret, size_t secretSize,
+                              XXH64_hash_t seed);
+
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(const void* data, size_t len,
+                               const void* secret, size_t secretSize,
+                               XXH64_hash_t seed64);
+
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
+                                    const void* secret, size_t secretSize,
+                                    XXH64_hash_t seed64);
+
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
+                                     const void* secret, size_t secretSize,
+                                     XXH64_hash_t seed64);
+
+
+#endif  /* XXH_NO_XXH3 */
+#endif  /* XXH_NO_LONG_LONG */
+#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
+#  define XXH_IMPLEMENTATION
+#endif
+
+#endif  /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */
+
+
+/* ======================================================================== */
+/* ======================================================================== */
+/* ======================================================================== */
+
+
+/*-**********************************************************************
+ * xxHash implementation
+ *-**********************************************************************
+ * xxHash's implementation used to be hosted inside xxhash.c.
+ *
+ * However, inlining requires implementation to be visible to the compiler,
+ * hence be included alongside the header.
+ * Previously, implementation was hosted inside xxhash.c,
+ * which was then #included when inlining was activated.
+ * This construction created issues with a few build and install systems,
+ * as it required xxhash.c to be stored in /include directory.
+ *
+ * xxHash implementation is now directly integrated within xxhash.h.
+ * As a consequence, xxhash.c is no longer needed in /include.
+ *
+ * xxhash.c is still available and is still useful.
+ * In a "normal" setup, when xxhash is not inlined,
+ * xxhash.h only exposes the prototypes and public symbols,
+ * while xxhash.c can be built into an object file xxhash.o
+ * which can then be linked into the final binary.
+ ************************************************************************/
+
+#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
+   || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
+#  define XXH_IMPLEM_13a8737387
+
+/* *************************************
+*  Tuning parameters
+***************************************/
+
+/*!
+ * @defgroup tuning Tuning parameters
+ * @{
+ *
+ * Various macros to control xxHash's behavior.
+ */
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Define this to disable 64-bit code.
+ *
+ * Useful if only using the @ref xxh32_family and you have a strict C90 compiler.
+ */
+#  define XXH_NO_LONG_LONG
+#  undef XXH_NO_LONG_LONG /* don't actually */
+/*!
+ * @brief Controls how unaligned memory is accessed.
+ *
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is
+ * safe and portable.
+ *
+ * Unfortunately, on some target/compiler combinations, the generated assembly
+ * is sub-optimal.
+ *
+ * The below switch allow selection of a different access method
+ * in the search for improved performance.
+ *
+ * @par Possible options:
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
+ *   @par
+ *     Use `memcpy()`. Safe and portable. Note that most modern compilers will
+ *     eliminate the function call and treat it as an unaligned access.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((packed))`
+ *   @par
+ *     Depends on compiler extensions and is therefore not portable.
+ *     This method is safe _if_ your compiler supports it,
+ *     and *generally* as fast or faster than `memcpy`.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
+ *  @par
+ *     Casts directly and dereferences. This method doesn't depend on the
+ *     compiler, but it violates the C standard as it directly dereferences an
+ *     unaligned pointer. It can generate buggy code on targets which do not
+ *     support unaligned memory accesses, but in some circumstances, it's the
+ *     only known way to get the most performance.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
+ *  @par
+ *     Also portable. This can generate the best code on old compilers which don't
+ *     inline small `memcpy()` calls, and it might also be faster on big-endian
+ *     systems which lack a native byteswap instruction. However, some compilers
+ *     will emit literal byteshifts even if the target supports unaligned access.
+ *  .
+ *
+ * @warning
+ *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
+ *   care, as what works on one compiler/platform/optimization level may cause
+ *   another to read garbage data or even crash.
+ *
+ * See https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
+ *
+ * Prefer these methods in priority order (0 > 3 > 1 > 2)
+ */
+#  define XXH_FORCE_MEMORY_ACCESS 0
+
+/*!
+ * @def XXH_FORCE_ALIGN_CHECK
+ * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
+ * and XXH64() only).
+ *
+ * This is an important performance trick for architectures without decent
+ * unaligned memory access performance.
+ *
+ * It checks for input alignment, and when conditions are met, uses a "fast
+ * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
+ * faster_ read speed.
+ *
+ * The check costs one initial branch per hash, which is generally negligible,
+ * but not zero.
+ *
+ * Moreover, it's not useful to generate an additional code path if memory
+ * access uses the same instruction for both aligned and unaligned
+ * addresses (e.g. x86 and aarch64).
+ *
+ * In these cases, the alignment check can be removed by setting this macro to 0.
+ * Then the code will always use unaligned memory access.
+ * Align check is automatically disabled on x86, x64 & arm64,
+ * which are platforms known to offer good unaligned memory accesses performance.
+ *
+ * This option does not affect XXH3 (only XXH32 and XXH64).
+ */
+#  define XXH_FORCE_ALIGN_CHECK 0
+
+/*!
+ * @def XXH_NO_INLINE_HINTS
+ * @brief When non-zero, sets all functions to `static`.
+ *
+ * By default, xxHash tries to force the compiler to inline almost all internal
+ * functions.
+ *
+ * This can usually improve performance due to reduced jumping and improved
+ * constant folding, but significantly increases the size of the binary which
+ * might not be favorable.
+ *
+ * Additionally, sometimes the forced inlining can be detrimental to performance,
+ * depending on the architecture.
+ *
+ * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
+ * compiler full control on whether to inline or not.
+ *
+ * When not optimizing (-O0), optimizing for size (-Os, -Oz), or using
+ * -fno-inline with GCC or Clang, this will automatically be defined.
+ */
+#  define XXH_NO_INLINE_HINTS 0
+
+/*!
+ * @def XXH32_ENDJMP
+ * @brief Whether to use a jump for `XXH32_finalize`.
+ *
+ * For performance, `XXH32_finalize` uses multiple branches in the finalizer.
+ * This is generally preferable for performance,
+ * but depending on exact architecture, a jmp may be preferable.
+ *
+ * This setting is only possibly making a difference for very small inputs.
+ */
+#  define XXH32_ENDJMP 0
+
+/*!
+ * @internal
+ * @brief Redefines old internal names.
+ *
+ * For compatibility with code that uses xxHash's internals before the names
+ * were changed to improve namespacing. There is no other reason to use this.
+ */
+#  define XXH_OLD_NAMES
+#  undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
+#endif /* XXH_DOXYGEN */
+/*!
+ * @}
+ */
+
+#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+   /* prefer __packed__ structures (method 1) for gcc on armv7+ and mips */
+#  if !defined(__clang__) && \
+( \
+    (defined(__INTEL_COMPILER) && !defined(_WIN32)) || \
+    ( \
+        defined(__GNUC__) && ( \
+            (defined(__ARM_ARCH) && __ARM_ARCH >= 7) || \
+            ( \
+                defined(__mips__) && \
+                (__mips <= 5 || __mips_isa_rev < 6) && \
+                (!defined(__mips16) || defined(__mips_mips16e2)) \
+            ) \
+        ) \
+    ) \
+)
+#    define XXH_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+#ifndef XXH_FORCE_ALIGN_CHECK  /* can be defined externally */
+#  if defined(__i386)  || defined(__x86_64__) || defined(__aarch64__) \
+   || defined(_M_IX86) || defined(_M_X64)     || defined(_M_ARM64) /* visual */
+#    define XXH_FORCE_ALIGN_CHECK 0
+#  else
+#    define XXH_FORCE_ALIGN_CHECK 1
+#  endif
+#endif
+
+#ifndef XXH_NO_INLINE_HINTS
+#  if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ \
+   || defined(__NO_INLINE__)     /* -O0, -fno-inline */
+#    define XXH_NO_INLINE_HINTS 1
+#  else
+#    define XXH_NO_INLINE_HINTS 0
+#  endif
+#endif
+
+#ifndef XXH32_ENDJMP
+/* generally preferable for performance */
+#  define XXH32_ENDJMP 0
+#endif
+
+/*!
+ * @defgroup impl Implementation
+ * @{
+ */
+
+
+/* *************************************
+*  Includes & Memory related functions
+***************************************/
+/* Modify the local functions below should you wish to use some other memory routines */
+/* for ZSTD_malloc(), ZSTD_free() */
+#define ZSTD_DEPS_NEED_MALLOC
+#include "zstd_deps.h"  /* size_t, ZSTD_malloc, ZSTD_free, ZSTD_memcpy */
+static void* XXH_malloc(size_t s) { return ZSTD_malloc(s); }
+static void  XXH_free  (void* p)  { ZSTD_free(p); }
+static void* XXH_memcpy(void* dest, const void* src, size_t size) { return ZSTD_memcpy(dest,src,size); }
+
+
+/* *************************************
+*  Compiler Specific Options
+***************************************/
+#ifdef _MSC_VER /* Visual Studio warning fix */
+#  pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
+#endif
+
+#if XXH_NO_INLINE_HINTS  /* disable inlining hints */
+#  if defined(__GNUC__) || defined(__clang__)
+#    define XXH_FORCE_INLINE static __attribute__((unused))
+#  else
+#    define XXH_FORCE_INLINE static
+#  endif
+#  define XXH_NO_INLINE static
+/* enable inlining hints */
+#elif defined(__GNUC__) || defined(__clang__)
+#  define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
+#  define XXH_NO_INLINE static __attribute__((noinline))
+#elif defined(_MSC_VER)  /* Visual Studio */
+#  define XXH_FORCE_INLINE static __forceinline
+#  define XXH_NO_INLINE static __declspec(noinline)
+#elif defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
+#  define XXH_FORCE_INLINE static inline
+#  define XXH_NO_INLINE static
+#else
+#  define XXH_FORCE_INLINE static
+#  define XXH_NO_INLINE static
+#endif
+
+
+
+/* *************************************
+*  Debug
+***************************************/
+/*!
+ * @ingroup tuning
+ * @def XXH_DEBUGLEVEL
+ * @brief Sets the debugging level.
+ *
+ * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
+ * compiler's command line options. The value must be a number.
+ */
+#ifndef XXH_DEBUGLEVEL
+#  ifdef DEBUGLEVEL /* backwards compat */
+#    define XXH_DEBUGLEVEL DEBUGLEVEL
+#  else
+#    define XXH_DEBUGLEVEL 0
+#  endif
+#endif
+
+#if (XXH_DEBUGLEVEL>=1)
+#  include    /* note: can still be disabled with NDEBUG */
+#  define XXH_ASSERT(c)   assert(c)
+#else
+#  define XXH_ASSERT(c)   ((void)0)
+#endif
+
+/* note: use after variable declarations */
+#ifndef XXH_STATIC_ASSERT
+#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)    /* C11 */
+#    include 
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
+#  elif defined(__cplusplus) && (__cplusplus >= 201103L)            /* C++11 */
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
+#  else
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
+#  endif
+#  define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
+#endif
+
+/*!
+ * @internal
+ * @def XXH_COMPILER_GUARD(var)
+ * @brief Used to prevent unwanted optimizations for @p var.
+ *
+ * It uses an empty GCC inline assembly statement with a register constraint
+ * which forces @p var into a general purpose register (e.g. eax, ebx, ecx
+ * on x86) and marks it as modified.
+ *
+ * This is used in a few places to avoid unwanted autovectorization (e.g.
+ * XXH32_round()). All vectorization we want is explicit via intrinsics,
+ * and _usually_ isn't wanted elsewhere.
+ *
+ * We also use it to prevent unwanted constant folding for AArch64 in
+ * XXH3_initCustomSecret_scalar().
+ */
+#if defined(__GNUC__) || defined(__clang__)
+#  define XXH_COMPILER_GUARD(var) __asm__ __volatile__("" : "+r" (var))
+#else
+#  define XXH_COMPILER_GUARD(var) ((void)0)
+#endif
+
+/* *************************************
+*  Basic Types
+***************************************/
+#if !defined (__VMS) \
+ && (defined (__cplusplus) \
+ || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include 
+  typedef uint8_t xxh_u8;
+#else
+  typedef unsigned char xxh_u8;
+#endif
+typedef XXH32_hash_t xxh_u32;
+
+#ifdef XXH_OLD_NAMES
+#  define BYTE xxh_u8
+#  define U8   xxh_u8
+#  define U32  xxh_u32
+#endif
+
+/* ***   Memory access   *** */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_read32(const void* ptr)
+ * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit native endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit little endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readBE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit big endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit big endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align)
+ * @brief Like @ref XXH_readLE32(), but has an option for aligned reads.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ * Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is
+ * always @ref XXH_alignment::XXH_unaligned.
+ *
+ * @param ptr The pointer to read from.
+ * @param align Whether @p ptr is aligned.
+ * @pre
+ *   If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte
+ *   aligned.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
+ */
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE32 and XXH_readBE32.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/*
+ * Force direct memory access. Only works on CPU which support unaligned memory
+ * access in hardware.
+ */
+static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/*
+ * __pack instructions are safer but compiler specific, hence potentially
+ * problematic for some compilers.
+ *
+ * Currently only defined for GCC and ICC.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
+#endif
+static xxh_u32 XXH_read32(const void* ptr)
+{
+    typedef union { xxh_u32 u32; } __attribute__((packed)) xxh_unalign;
+    return ((const xxh_unalign*)ptr)->u32;
+}
+
+#else
+
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u32 XXH_read32(const void* memPtr)
+{
+    xxh_u32 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+
+/* ***   Endianness   *** */
+
+/*!
+ * @ingroup tuning
+ * @def XXH_CPU_LITTLE_ENDIAN
+ * @brief Whether the target is little endian.
+ *
+ * Defined to 1 if the target is little endian, or 0 if it is big endian.
+ * It can be defined externally, for example on the compiler command line.
+ *
+ * If it is not defined,
+ * a runtime check (which is usually constant folded) is used instead.
+ *
+ * @note
+ *   This is not necessarily defined to an integer constant.
+ *
+ * @see XXH_isLittleEndian() for the runtime check.
+ */
+#ifndef XXH_CPU_LITTLE_ENDIAN
+/*
+ * Try to detect endianness automatically, to avoid the nonstandard behavior
+ * in `XXH_isLittleEndian()`
+ */
+#  if defined(_WIN32) /* Windows is always little endian */ \
+     || defined(__LITTLE_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 1
+#  elif defined(__BIG_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 0
+#  else
+/*!
+ * @internal
+ * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
+ *
+ * Most compilers will constant fold this.
+ */
+static int XXH_isLittleEndian(void)
+{
+    /*
+     * Portable and well-defined behavior.
+     * Don't use static: it is detrimental to performance.
+     */
+    const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
+    return one.c[0];
+}
+#   define XXH_CPU_LITTLE_ENDIAN   XXH_isLittleEndian()
+#  endif
+#endif
+
+
+
+
+/* ****************************************
+*  Compiler-specific Functions and Macros
+******************************************/
+#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
+#ifdef __has_builtin
+#  define XXH_HAS_BUILTIN(x) __has_builtin(x)
+#else
+#  define XXH_HAS_BUILTIN(x) 0
+#endif
+
+/*!
+ * @internal
+ * @def XXH_rotl32(x,r)
+ * @brief 32-bit rotate left.
+ *
+ * @param x The 32-bit integer to be rotated.
+ * @param r The number of bits to rotate.
+ * @pre
+ *   @p r > 0 && @p r < 32
+ * @note
+ *   @p x and @p r may be evaluated multiple times.
+ * @return The rotated result.
+ */
+#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
+                               && XXH_HAS_BUILTIN(__builtin_rotateleft64)
+#  define XXH_rotl32 __builtin_rotateleft32
+#  define XXH_rotl64 __builtin_rotateleft64
+/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
+#elif defined(_MSC_VER)
+#  define XXH_rotl32(x,r) _rotl(x,r)
+#  define XXH_rotl64(x,r) _rotl64(x,r)
+#else
+#  define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
+#  define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
+#endif
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_swap32(xxh_u32 x)
+ * @brief A 32-bit byteswap.
+ *
+ * @param x The 32-bit integer to byteswap.
+ * @return @p x, byteswapped.
+ */
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap32 _byteswap_ulong
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap32 __builtin_bswap32
+#else
+static xxh_u32 XXH_swap32 (xxh_u32 x)
+{
+    return  ((x << 24) & 0xff000000 ) |
+            ((x <<  8) & 0x00ff0000 ) |
+            ((x >>  8) & 0x0000ff00 ) |
+            ((x >> 24) & 0x000000ff );
+}
+#endif
+
+
+/* ***************************
+*  Memory reads
+*****************************/
+
+/*!
+ * @internal
+ * @brief Enum to indicate whether a pointer is aligned.
+ */
+typedef enum {
+    XXH_aligned,  /*!< Aligned */
+    XXH_unaligned /*!< Possibly unaligned */
+} XXH_alignment;
+
+/*
+ * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
+ *
+ * This is ideal for older compilers which don't inline memcpy.
+ */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u32)bytePtr[1] << 8)
+         | ((xxh_u32)bytePtr[2] << 16)
+         | ((xxh_u32)bytePtr[3] << 24);
+}
+
+XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[3]
+         | ((xxh_u32)bytePtr[2] << 8)
+         | ((xxh_u32)bytePtr[1] << 16)
+         | ((xxh_u32)bytePtr[0] << 24);
+}
+
+#else
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
+}
+
+static xxh_u32 XXH_readBE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
+}
+#endif
+
+XXH_FORCE_INLINE xxh_u32
+XXH_readLE32_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned) {
+        return XXH_readLE32(ptr);
+    } else {
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
+    }
+}
+
+
+/* *************************************
+*  Misc
+***************************************/
+/*! @ingroup public */
+XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
+
+
+/* *******************************************************************
+*  32-bit hash functions
+*********************************************************************/
+/*!
+ * @}
+ * @defgroup xxh32_impl XXH32 implementation
+ * @ingroup impl
+ * @{
+ */
+ /* #define instead of static const, to be used as initializers */
+#define XXH_PRIME32_1  0x9E3779B1U  /*!< 0b10011110001101110111100110110001 */
+#define XXH_PRIME32_2  0x85EBCA77U  /*!< 0b10000101111010111100101001110111 */
+#define XXH_PRIME32_3  0xC2B2AE3DU  /*!< 0b11000010101100101010111000111101 */
+#define XXH_PRIME32_4  0x27D4EB2FU  /*!< 0b00100111110101001110101100101111 */
+#define XXH_PRIME32_5  0x165667B1U  /*!< 0b00010110010101100110011110110001 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME32_1 XXH_PRIME32_1
+#  define PRIME32_2 XXH_PRIME32_2
+#  define PRIME32_3 XXH_PRIME32_3
+#  define PRIME32_4 XXH_PRIME32_4
+#  define PRIME32_5 XXH_PRIME32_5
+#endif
+
+/*!
+ * @internal
+ * @brief Normal stripe processing routine.
+ *
+ * This shuffles the bits so that any bit from @p input impacts several bits in
+ * @p acc.
+ *
+ * @param acc The accumulator lane.
+ * @param input The stripe of input to mix.
+ * @return The mixed accumulator lane.
+ */
+static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
+{
+    acc += input * XXH_PRIME32_2;
+    acc  = XXH_rotl32(acc, 13);
+    acc *= XXH_PRIME32_1;
+#if (defined(__SSE4_1__) || defined(__aarch64__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
+    /*
+     * UGLY HACK:
+     * A compiler fence is the only thing that prevents GCC and Clang from
+     * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
+     * reason) without globally disabling SSE4.1.
+     *
+     * The reason we want to avoid vectorization is because despite working on
+     * 4 integers at a time, there are multiple factors slowing XXH32 down on
+     * SSE4:
+     * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
+     *   newer chips!) making it slightly slower to multiply four integers at
+     *   once compared to four integers independently. Even when pmulld was
+     *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
+     *   just to multiply unless doing a long operation.
+     *
+     * - Four instructions are required to rotate,
+     *      movqda tmp,  v // not required with VEX encoding
+     *      pslld  tmp, 13 // tmp <<= 13
+     *      psrld  v,   19 // x >>= 19
+     *      por    v,  tmp // x |= tmp
+     *   compared to one for scalar:
+     *      roll   v, 13    // reliably fast across the board
+     *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
+     *
+     * - Instruction level parallelism is actually more beneficial here because
+     *   the SIMD actually serializes this operation: While v1 is rotating, v2
+     *   can load data, while v3 can multiply. SSE forces them to operate
+     *   together.
+     *
+     * This is also enabled on AArch64, as Clang autovectorizes it incorrectly
+     * and it is pointless writing a NEON implementation that is basically the
+     * same speed as scalar for XXH32.
+     */
+    XXH_COMPILER_GUARD(acc);
+#endif
+    return acc;
+}
+
+/*!
+ * @internal
+ * @brief Mixes all bits to finalize the hash.
+ *
+ * The final mix ensures that all input bits have a chance to impact any bit in
+ * the output digest, resulting in an unbiased distribution.
+ *
+ * @param h32 The hash to avalanche.
+ * @return The avalanched hash.
+ */
+static xxh_u32 XXH32_avalanche(xxh_u32 h32)
+{
+    h32 ^= h32 >> 15;
+    h32 *= XXH_PRIME32_2;
+    h32 ^= h32 >> 13;
+    h32 *= XXH_PRIME32_3;
+    h32 ^= h32 >> 16;
+    return(h32);
+}
+
+#define XXH_get32bits(p) XXH_readLE32_align(p, align)
+
+/*!
+ * @internal
+ * @brief Processes the last 0-15 bytes of @p ptr.
+ *
+ * There may be up to 15 bytes remaining to consume from the input.
+ * This final stage will digest them to ensure that all input bytes are present
+ * in the final mix.
+ *
+ * @param h32 The hash to finalize.
+ * @param ptr The pointer to the remaining input.
+ * @param len The remaining length, modulo 16.
+ * @param align Whether @p ptr is aligned.
+ * @return The finalized hash.
+ */
+static xxh_u32
+XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+#define XXH_PROCESS1 do {                           \
+    h32 += (*ptr++) * XXH_PRIME32_5;                \
+    h32 = XXH_rotl32(h32, 11) * XXH_PRIME32_1;      \
+} while (0)
+
+#define XXH_PROCESS4 do {                           \
+    h32 += XXH_get32bits(ptr) * XXH_PRIME32_3;      \
+    ptr += 4;                                   \
+    h32  = XXH_rotl32(h32, 17) * XXH_PRIME32_4;     \
+} while (0)
+
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+
+    /* Compact rerolled version; generally faster */
+    if (!XXH32_ENDJMP) {
+        len &= 15;
+        while (len >= 4) {
+            XXH_PROCESS4;
+            len -= 4;
+        }
+        while (len > 0) {
+            XXH_PROCESS1;
+            --len;
+        }
+        return XXH32_avalanche(h32);
+    } else {
+         switch(len&15) /* or switch(bEnd - p) */ {
+           case 12:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 8:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 4:       XXH_PROCESS4;
+                         return XXH32_avalanche(h32);
+
+           case 13:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 9:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 5:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(h32);
+
+           case 14:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 10:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 6:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(h32);
+
+           case 15:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 11:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 7:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 3:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 2:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 1:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 0:       return XXH32_avalanche(h32);
+        }
+        XXH_ASSERT(0);
+        return h32;   /* reaching this point is deemed impossible */
+    }
+}
+
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1 XXH_PROCESS1
+#  define PROCESS4 XXH_PROCESS4
+#else
+#  undef XXH_PROCESS1
+#  undef XXH_PROCESS4
+#endif
+
+/*!
+ * @internal
+ * @brief The implementation for @ref XXH32().
+ *
+ * @param input , len , seed Directly passed from @ref XXH32().
+ * @param align Whether @p input is aligned.
+ * @return The calculated hash.
+ */
+XXH_FORCE_INLINE xxh_u32
+XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
+{
+    xxh_u32 h32;
+
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=16) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 15;
+        xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+        xxh_u32 v2 = seed + XXH_PRIME32_2;
+        xxh_u32 v3 = seed + 0;
+        xxh_u32 v4 = seed - XXH_PRIME32_1;
+
+        do {
+            v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
+            v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
+            v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
+            v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
+        } while (input < limit);
+
+        h32 = XXH_rotl32(v1, 1)  + XXH_rotl32(v2, 7)
+            + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
+    } else {
+        h32  = seed + XXH_PRIME32_5;
+    }
+
+    h32 += (xxh_u32)len;
+
+    return XXH32_finalize(h32, input, len&15, align);
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH32_state_t state;
+    XXH32_reset(&state, seed);
+    XXH32_update(&state, (const xxh_u8*)input, len);
+    return XXH32_digest(&state);
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
+            return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
+
+    return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+#endif
+}
+
+
+
+/*******   Hash streaming   *******/
+/*!
+ * @ingroup xxh32_family
+ */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
+{
+    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
+}
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
+{
+    XXH_ASSERT(statePtr != NULL);
+    memset(statePtr, 0, sizeof(*statePtr));
+    statePtr->v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+    statePtr->v[1] = seed + XXH_PRIME32_2;
+    statePtr->v[2] = seed + 0;
+    statePtr->v[3] = seed - XXH_PRIME32_1;
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH32_update(XXH32_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
+
+        state->total_len_32 += (XXH32_hash_t)len;
+        state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));
+
+        if (state->memsize + len < 16)  {   /* fill in tmp buffer */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
+            state->memsize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
+
+        if (state->memsize) {   /* some data left from previous update */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
+            {   const xxh_u32* p32 = state->mem32;
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
+            }
+            p += 16-state->memsize;
+            state->memsize = 0;
+        }
+
+        if (p <= bEnd-16) {
+            const xxh_u8* const limit = bEnd - 16;
+
+            do {
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
+            } while (p<=limit);
+
+        }
+
+        if (p < bEnd) {
+            XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
+    }
+
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
+{
+    xxh_u32 h32;
+
+    if (state->large_len) {
+        h32 = XXH_rotl32(state->v[0], 1)
+            + XXH_rotl32(state->v[1], 7)
+            + XXH_rotl32(state->v[2], 12)
+            + XXH_rotl32(state->v[3], 18);
+    } else {
+        h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
+    }
+
+    h32 += state->total_len_32;
+
+    return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
+}
+
+
+/*******   Canonical representation   *******/
+
+/*!
+ * @ingroup xxh32_family
+ * The default return values from XXH functions are unsigned 32 and 64 bit
+ * integers.
+ *
+ * The canonical representation uses big endian convention, the same convention
+ * as human-readable numbers (large digits first).
+ *
+ * This way, hash values can be written into a file or buffer, remaining
+ * comparable across different systems.
+ *
+ * The following functions allow transformation of hash values to and from their
+ * canonical format.
+ */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
+{
+    /* XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t)); */
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
+}
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
+{
+    return XXH_readBE32(src);
+}
+
+
+#ifndef XXH_NO_LONG_LONG
+
+/* *******************************************************************
+*  64-bit hash functions
+*********************************************************************/
+/*!
+ * @}
+ * @ingroup impl
+ * @{
+ */
+/*******   Memory access   *******/
+
+typedef XXH64_hash_t xxh_u64;
+
+#ifdef XXH_OLD_NAMES
+#  define U64 xxh_u64
+#endif
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE64 and XXH_readBE64.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    return *(const xxh_u64*) memPtr;
+}
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/*
+ * __pack instructions are safer, but compiler specific, hence potentially
+ * problematic for some compilers.
+ *
+ * Currently only defined for GCC and ICC.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
+#endif
+static xxh_u64 XXH_read64(const void* ptr)
+{
+    typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) xxh_unalign64;
+    return ((const xxh_unalign64*)ptr)->u64;
+}
+
+#else
+
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: https://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    xxh_u64 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap64 _byteswap_uint64
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap64 __builtin_bswap64
+#else
+static xxh_u64 XXH_swap64(xxh_u64 x)
+{
+    return  ((x << 56) & 0xff00000000000000ULL) |
+            ((x << 40) & 0x00ff000000000000ULL) |
+            ((x << 24) & 0x0000ff0000000000ULL) |
+            ((x << 8)  & 0x000000ff00000000ULL) |
+            ((x >> 8)  & 0x00000000ff000000ULL) |
+            ((x >> 24) & 0x0000000000ff0000ULL) |
+            ((x >> 40) & 0x000000000000ff00ULL) |
+            ((x >> 56) & 0x00000000000000ffULL);
+}
+#endif
+
+
+/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u64)bytePtr[1] << 8)
+         | ((xxh_u64)bytePtr[2] << 16)
+         | ((xxh_u64)bytePtr[3] << 24)
+         | ((xxh_u64)bytePtr[4] << 32)
+         | ((xxh_u64)bytePtr[5] << 40)
+         | ((xxh_u64)bytePtr[6] << 48)
+         | ((xxh_u64)bytePtr[7] << 56);
+}
+
+XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[7]
+         | ((xxh_u64)bytePtr[6] << 8)
+         | ((xxh_u64)bytePtr[5] << 16)
+         | ((xxh_u64)bytePtr[4] << 24)
+         | ((xxh_u64)bytePtr[3] << 32)
+         | ((xxh_u64)bytePtr[2] << 40)
+         | ((xxh_u64)bytePtr[1] << 48)
+         | ((xxh_u64)bytePtr[0] << 56);
+}
+
+#else
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
+}
+
+static xxh_u64 XXH_readBE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+}
+#endif
+
+XXH_FORCE_INLINE xxh_u64
+XXH_readLE64_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return XXH_readLE64(ptr);
+    else
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
+}
+
+
+/*******   xxh64   *******/
+/*!
+ * @}
+ * @defgroup xxh64_impl XXH64 implementation
+ * @ingroup impl
+ * @{
+ */
+/* #define rather that static const, to be used as initializers */
+#define XXH_PRIME64_1  0x9E3779B185EBCA87ULL  /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
+#define XXH_PRIME64_2  0xC2B2AE3D27D4EB4FULL  /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
+#define XXH_PRIME64_3  0x165667B19E3779F9ULL  /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
+#define XXH_PRIME64_4  0x85EBCA77C2B2AE63ULL  /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
+#define XXH_PRIME64_5  0x27D4EB2F165667C5ULL  /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME64_1 XXH_PRIME64_1
+#  define PRIME64_2 XXH_PRIME64_2
+#  define PRIME64_3 XXH_PRIME64_3
+#  define PRIME64_4 XXH_PRIME64_4
+#  define PRIME64_5 XXH_PRIME64_5
+#endif
+
+static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
+{
+    acc += input * XXH_PRIME64_2;
+    acc  = XXH_rotl64(acc, 31);
+    acc *= XXH_PRIME64_1;
+    return acc;
+}
+
+static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
+{
+    val  = XXH64_round(0, val);
+    acc ^= val;
+    acc  = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
+    return acc;
+}
+
+static xxh_u64 XXH64_avalanche(xxh_u64 h64)
+{
+    h64 ^= h64 >> 33;
+    h64 *= XXH_PRIME64_2;
+    h64 ^= h64 >> 29;
+    h64 *= XXH_PRIME64_3;
+    h64 ^= h64 >> 32;
+    return h64;
+}
+
+
+#define XXH_get64bits(p) XXH_readLE64_align(p, align)
+
+static xxh_u64
+XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+    len &= 31;
+    while (len >= 8) {
+        xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
+        ptr += 8;
+        h64 ^= k1;
+        h64  = XXH_rotl64(h64,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
+        len -= 8;
+    }
+    if (len >= 4) {
+        h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
+        ptr += 4;
+        h64 = XXH_rotl64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
+        len -= 4;
+    }
+    while (len > 0) {
+        h64 ^= (*ptr++) * XXH_PRIME64_5;
+        h64 = XXH_rotl64(h64, 11) * XXH_PRIME64_1;
+        --len;
+    }
+    return  XXH64_avalanche(h64);
+}
+
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1_64 XXH_PROCESS1_64
+#  define PROCESS4_64 XXH_PROCESS4_64
+#  define PROCESS8_64 XXH_PROCESS8_64
+#else
+#  undef XXH_PROCESS1_64
+#  undef XXH_PROCESS4_64
+#  undef XXH_PROCESS8_64
+#endif
+
+XXH_FORCE_INLINE xxh_u64
+XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
+{
+    xxh_u64 h64;
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=32) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 31;
+        xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+        xxh_u64 v2 = seed + XXH_PRIME64_2;
+        xxh_u64 v3 = seed + 0;
+        xxh_u64 v4 = seed - XXH_PRIME64_1;
+
+        do {
+            v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
+            v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
+            v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
+            v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
+        } while (inputv[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+    statePtr->v[1] = seed + XXH_PRIME64_2;
+    statePtr->v[2] = seed + 0;
+    statePtr->v[3] = seed - XXH_PRIME64_1;
+    return XXH_OK;
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH64_update (XXH64_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
+
+        state->total_len += len;
+
+        if (state->memsize + len < 32) {  /* fill in tmp buffer */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
+            state->memsize += (xxh_u32)len;
+            return XXH_OK;
+        }
+
+        if (state->memsize) {   /* tmp buffer is full */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
+            state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
+            state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
+            state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
+            state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
+            p += 32 - state->memsize;
+            state->memsize = 0;
+        }
+
+        if (p+32 <= bEnd) {
+            const xxh_u8* const limit = bEnd - 32;
+
+            do {
+                state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
+                state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
+                state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
+                state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
+            } while (p<=limit);
+
+        }
+
+        if (p < bEnd) {
+            XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
+    }
+
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t* state)
+{
+    xxh_u64 h64;
+
+    if (state->total_len >= 32) {
+        h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
+        h64 = XXH64_mergeRound(h64, state->v[0]);
+        h64 = XXH64_mergeRound(h64, state->v[1]);
+        h64 = XXH64_mergeRound(h64, state->v[2]);
+        h64 = XXH64_mergeRound(h64, state->v[3]);
+    } else {
+        h64  = state->v[2] /*seed*/ + XXH_PRIME64_5;
+    }
+
+    h64 += (xxh_u64) state->total_len;
+
+    return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
+}
+
+
+/******* Canonical representation   *******/
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
+{
+    /* XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t)); */
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
+{
+    return XXH_readBE64(src);
+}
+
+#ifndef XXH_NO_XXH3
+
+/* *********************************************************************
+*  XXH3
+*  New generation hash designed for speed on small keys and vectorization
+************************************************************************ */
+/*!
+ * @}
+ * @defgroup xxh3_impl XXH3 implementation
+ * @ingroup impl
+ * @{
+ */
+
+/* ===   Compiler specifics   === */
+
+#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
+#  define XXH_RESTRICT /* disable */
+#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* >= C99 */
+#  define XXH_RESTRICT   restrict
+#else
+/* Note: it might be useful to define __restrict or __restrict__ for some C++ compilers */
+#  define XXH_RESTRICT   /* disable */
+#endif
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3))  \
+  || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
+  || defined(__clang__)
+#    define XXH_likely(x) __builtin_expect(x, 1)
+#    define XXH_unlikely(x) __builtin_expect(x, 0)
+#else
+#    define XXH_likely(x) (x)
+#    define XXH_unlikely(x) (x)
+#endif
+
+#if defined(__GNUC__) || defined(__clang__)
+#  if defined(__ARM_NEON__) || defined(__ARM_NEON) \
+   || defined(__aarch64__)  || defined(_M_ARM) \
+   || defined(_M_ARM64)     || defined(_M_ARM64EC)
+#    define inline __inline__  /* circumvent a clang bug */
+#    include 
+#    undef inline
+#  elif defined(__AVX2__)
+#    include 
+#  elif defined(__SSE2__)
+#    include 
+#  endif
+#endif
+
+#if defined(_MSC_VER)
+#  include 
+#endif
+
+/*
+ * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
+ * remaining a true 64-bit/128-bit hash function.
+ *
+ * This is done by prioritizing a subset of 64-bit operations that can be
+ * emulated without too many steps on the average 32-bit machine.
+ *
+ * For example, these two lines seem similar, and run equally fast on 64-bit:
+ *
+ *   xxh_u64 x;
+ *   x ^= (x >> 47); // good
+ *   x ^= (x >> 13); // bad
+ *
+ * However, to a 32-bit machine, there is a major difference.
+ *
+ * x ^= (x >> 47) looks like this:
+ *
+ *   x.lo ^= (x.hi >> (47 - 32));
+ *
+ * while x ^= (x >> 13) looks like this:
+ *
+ *   // note: funnel shifts are not usually cheap.
+ *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
+ *   x.hi ^= (x.hi >> 13);
+ *
+ * The first one is significantly faster than the second, simply because the
+ * shift is larger than 32. This means:
+ *  - All the bits we need are in the upper 32 bits, so we can ignore the lower
+ *    32 bits in the shift.
+ *  - The shift result will always fit in the lower 32 bits, and therefore,
+ *    we can ignore the upper 32 bits in the xor.
+ *
+ * Thanks to this optimization, XXH3 only requires these features to be efficient:
+ *
+ *  - Usable unaligned access
+ *  - A 32-bit or 64-bit ALU
+ *      - If 32-bit, a decent ADC instruction
+ *  - A 32 or 64-bit multiply with a 64-bit result
+ *  - For the 128-bit variant, a decent byteswap helps short inputs.
+ *
+ * The first two are already required by XXH32, and almost all 32-bit and 64-bit
+ * platforms which can run XXH32 can run XXH3 efficiently.
+ *
+ * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
+ * notable exception.
+ *
+ * First of all, Thumb-1 lacks support for the UMULL instruction which
+ * performs the important long multiply. This means numerous __aeabi_lmul
+ * calls.
+ *
+ * Second of all, the 8 functional registers are just not enough.
+ * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
+ * Lo registers, and this shuffling results in thousands more MOVs than A32.
+ *
+ * A32 and T32 don't have this limitation. They can access all 14 registers,
+ * do a 32->64 multiply with UMULL, and the flexible operand allowing free
+ * shifts is helpful, too.
+ *
+ * Therefore, we do a quick sanity check.
+ *
+ * If compiling Thumb-1 for a target which supports ARM instructions, we will
+ * emit a warning, as it is not a "sane" platform to compile for.
+ *
+ * Usually, if this happens, it is because of an accident and you probably need
+ * to specify -march, as you likely meant to compile for a newer architecture.
+ *
+ * Credit: large sections of the vectorial and asm source code paths
+ *         have been contributed by @easyaspi314
+ */
+#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
+#   warning "XXH3 is highly inefficient without ARM or Thumb-2."
+#endif
+
+/* ==========================================
+ * Vectorization detection
+ * ========================================== */
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @ingroup tuning
+ * @brief Overrides the vectorization implementation chosen for XXH3.
+ *
+ * Can be defined to 0 to disable SIMD or any of the values mentioned in
+ * @ref XXH_VECTOR_TYPE.
+ *
+ * If this is not defined, it uses predefined macros to determine the best
+ * implementation.
+ */
+#  define XXH_VECTOR XXH_SCALAR
+/*!
+ * @ingroup tuning
+ * @brief Possible values for @ref XXH_VECTOR.
+ *
+ * Note that these are actually implemented as macros.
+ *
+ * If this is not defined, it is detected automatically.
+ * @ref XXH_X86DISPATCH overrides this.
+ */
+enum XXH_VECTOR_TYPE /* fake enum */ {
+    XXH_SCALAR = 0,  /*!< Portable scalar version */
+    XXH_SSE2   = 1,  /*!<
+                      * SSE2 for Pentium 4, Opteron, all x86_64.
+                      *
+                      * @note SSE2 is also guaranteed on Windows 10, macOS, and
+                      * Android x86.
+                      */
+    XXH_AVX2   = 2,  /*!< AVX2 for Haswell and Bulldozer */
+    XXH_AVX512 = 3,  /*!< AVX512 for Skylake and Icelake */
+    XXH_NEON   = 4,  /*!< NEON for most ARMv7-A and all AArch64 */
+    XXH_VSX    = 5,  /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+};
+/*!
+ * @ingroup tuning
+ * @brief Selects the minimum alignment for XXH3's accumulators.
+ *
+ * When using SIMD, this should match the alignment required for said vector
+ * type, so, for example, 32 for AVX2.
+ *
+ * Default: Auto detected.
+ */
+#  define XXH_ACC_ALIGN 8
+#endif
+
+/* Actual definition */
+#ifndef XXH_DOXYGEN
+#  define XXH_SCALAR 0
+#  define XXH_SSE2   1
+#  define XXH_AVX2   2
+#  define XXH_AVX512 3
+#  define XXH_NEON   4
+#  define XXH_VSX    5
+#endif
+
+#ifndef XXH_VECTOR    /* can be defined on command line */
+#  if ( \
+        defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
+     || defined(_M_ARM) || defined(_M_ARM64) || defined(_M_ARM64EC) /* msvc */ \
+   ) && ( \
+        defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
+    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
+   )
+#    define XXH_VECTOR XXH_NEON
+#  elif defined(__AVX512F__)
+#    define XXH_VECTOR XXH_AVX512
+#  elif defined(__AVX2__)
+#    define XXH_VECTOR XXH_AVX2
+#  elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
+#    define XXH_VECTOR XXH_SSE2
+#  elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
+     || (defined(__s390x__) && defined(__VEC__)) \
+     && defined(__GNUC__) /* TODO: IBM XL */
+#    define XXH_VECTOR XXH_VSX
+#  else
+#    define XXH_VECTOR XXH_SCALAR
+#  endif
+#endif
+
+/*
+ * Controls the alignment of the accumulator,
+ * for compatibility with aligned vector loads, which are usually faster.
+ */
+#ifndef XXH_ACC_ALIGN
+#  if defined(XXH_X86DISPATCH)
+#     define XXH_ACC_ALIGN 64  /* for compatibility with avx512 */
+#  elif XXH_VECTOR == XXH_SCALAR  /* scalar */
+#     define XXH_ACC_ALIGN 8
+#  elif XXH_VECTOR == XXH_SSE2  /* sse2 */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX2  /* avx2 */
+#     define XXH_ACC_ALIGN 32
+#  elif XXH_VECTOR == XXH_NEON  /* neon */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_VSX   /* vsx */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX512  /* avx512 */
+#     define XXH_ACC_ALIGN 64
+#  endif
+#endif
+
+#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
+    || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
+#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
+#else
+#  define XXH_SEC_ALIGN 8
+#endif
+
+/*
+ * UGLY HACK:
+ * GCC usually generates the best code with -O3 for xxHash.
+ *
+ * However, when targeting AVX2, it is overzealous in its unrolling resulting
+ * in code roughly 3/4 the speed of Clang.
+ *
+ * There are other issues, such as GCC splitting _mm256_loadu_si256 into
+ * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
+ * only applies to Sandy and Ivy Bridge... which don't even support AVX2.
+ *
+ * That is why when compiling the AVX2 version, it is recommended to use either
+ *   -O2 -mavx2 -march=haswell
+ * or
+ *   -O2 -mavx2 -mno-avx256-split-unaligned-load
+ * for decent performance, or to use Clang instead.
+ *
+ * Fortunately, we can control the first one with a pragma that forces GCC into
+ * -O2, but the other one we can't control without "failed to inline always
+ * inline function due to target mismatch" warnings.
+ */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+#  pragma GCC push_options
+#  pragma GCC optimize("-O2")
+#endif
+
+
+#if XXH_VECTOR == XXH_NEON
+/*
+ * NEON's setup for vmlal_u32 is a little more complicated than it is on
+ * SSE2, AVX2, and VSX.
+ *
+ * While PMULUDQ and VMULEUW both perform a mask, VMLAL.U32 performs an upcast.
+ *
+ * To do the same operation, the 128-bit 'Q' register needs to be split into
+ * two 64-bit 'D' registers, performing this operation::
+ *
+ *   [                a                 |                 b                ]
+ *            |              '---------. .--------'                |
+ *            |                         x                          |
+ *            |              .---------' '--------.                |
+ *   [ a & 0xFFFFFFFF | b & 0xFFFFFFFF ],[    a >> 32     |     b >> 32    ]
+ *
+ * Due to significant changes in aarch64, the fastest method for aarch64 is
+ * completely different than the fastest method for ARMv7-A.
+ *
+ * ARMv7-A treats D registers as unions overlaying Q registers, so modifying
+ * D11 will modify the high half of Q5. This is similar to how modifying AH
+ * will only affect bits 8-15 of AX on x86.
+ *
+ * VZIP takes two registers, and puts even lanes in one register and odd lanes
+ * in the other.
+ *
+ * On ARMv7-A, this strangely modifies both parameters in place instead of
+ * taking the usual 3-operand form.
+ *
+ * Therefore, if we want to do this, we can simply use a D-form VZIP.32 on the
+ * lower and upper halves of the Q register to end up with the high and low
+ * halves where we want - all in one instruction.
+ *
+ *   vzip.32   d10, d11       @ d10 = { d10[0], d11[0] }; d11 = { d10[1], d11[1] }
+ *
+ * Unfortunately we need inline assembly for this: Instructions modifying two
+ * registers at once is not possible in GCC or Clang's IR, and they have to
+ * create a copy.
+ *
+ * aarch64 requires a different approach.
+ *
+ * In order to make it easier to write a decent compiler for aarch64, many
+ * quirks were removed, such as conditional execution.
+ *
+ * NEON was also affected by this.
+ *
+ * aarch64 cannot access the high bits of a Q-form register, and writes to a
+ * D-form register zero the high bits, similar to how writes to W-form scalar
+ * registers (or DWORD registers on x86_64) work.
+ *
+ * The formerly free vget_high intrinsics now require a vext (with a few
+ * exceptions)
+ *
+ * Additionally, VZIP was replaced by ZIP1 and ZIP2, which are the equivalent
+ * of PUNPCKL* and PUNPCKH* in SSE, respectively, in order to only modify one
+ * operand.
+ *
+ * The equivalent of the VZIP.32 on the lower and upper halves would be this
+ * mess:
+ *
+ *   ext     v2.4s, v0.4s, v0.4s, #2 // v2 = { v0[2], v0[3], v0[0], v0[1] }
+ *   zip1    v1.2s, v0.2s, v2.2s     // v1 = { v0[0], v2[0] }
+ *   zip2    v0.2s, v0.2s, v1.2s     // v0 = { v0[1], v2[1] }
+ *
+ * Instead, we use a literal downcast, vmovn_u64 (XTN), and vshrn_n_u64 (SHRN):
+ *
+ *   shrn    v1.2s, v0.2d, #32  // v1 = (uint32x2_t)(v0 >> 32);
+ *   xtn     v0.2s, v0.2d       // v0 = (uint32x2_t)(v0 & 0xFFFFFFFF);
+ *
+ * This is available on ARMv7-A, but is less efficient than a single VZIP.32.
+ */
+
+/*!
+ * Function-like macro:
+ * void XXH_SPLIT_IN_PLACE(uint64x2_t &in, uint32x2_t &outLo, uint32x2_t &outHi)
+ * {
+ *     outLo = (uint32x2_t)(in & 0xFFFFFFFF);
+ *     outHi = (uint32x2_t)(in >> 32);
+ *     in = UNDEFINED;
+ * }
+ */
+# if !defined(XXH_NO_VZIP_HACK) /* define to disable */ \
+   && (defined(__GNUC__) || defined(__clang__)) \
+   && (defined(__arm__) || defined(__thumb__) || defined(_M_ARM))
+#  define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                              \
+    do {                                                                                    \
+      /* Undocumented GCC/Clang operand modifier: %e0 = lower D half, %f0 = upper D half */ \
+      /* https://github.com/gcc-mirror/gcc/blob/38cf91e5/gcc/config/arm/arm.c#L22486 */     \
+      /* https://github.com/llvm-mirror/llvm/blob/2c4ca683/lib/Target/ARM/ARMAsmPrinter.cpp#L399 */ \
+      __asm__("vzip.32  %e0, %f0" : "+w" (in));                                             \
+      (outLo) = vget_low_u32 (vreinterpretq_u32_u64(in));                                   \
+      (outHi) = vget_high_u32(vreinterpretq_u32_u64(in));                                   \
+   } while (0)
+# else
+#  define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                            \
+    do {                                                                                  \
+      (outLo) = vmovn_u64    (in);                                                        \
+      (outHi) = vshrn_n_u64  ((in), 32);                                                  \
+    } while (0)
+# endif
+
+/*!
+ * @ingroup tuning
+ * @brief Controls the NEON to scalar ratio for XXH3
+ *
+ * On AArch64 when not optimizing for size, XXH3 will run 6 lanes using NEON and
+ * 2 lanes on scalar by default.
+ *
+ * This can be set to 2, 4, 6, or 8. ARMv7 will default to all 8 NEON lanes, as the
+ * emulated 64-bit arithmetic is too slow.
+ *
+ * Modern ARM CPUs are _very_ sensitive to how their pipelines are used.
+ *
+ * For example, the Cortex-A73 can dispatch 3 micro-ops per cycle, but it can't
+ * have more than 2 NEON (F0/F1) micro-ops. If you are only using NEON instructions,
+ * you are only using 2/3 of the CPU bandwidth.
+ *
+ * This is even more noticeable on the more advanced cores like the A76 which
+ * can dispatch 8 micro-ops per cycle, but still only 2 NEON micro-ops at once.
+ *
+ * Therefore, @ref XXH3_NEON_LANES lanes will be processed using NEON, and the
+ * remaining lanes will use scalar instructions. This improves the bandwidth
+ * and also gives the integer pipelines something to do besides twiddling loop
+ * counters and pointers.
+ *
+ * This change benefits CPUs with large micro-op buffers without negatively affecting
+ * other CPUs:
+ *
+ *  | Chipset               | Dispatch type       | NEON only | 6:2 hybrid | Diff. |
+ *  |:----------------------|:--------------------|----------:|-----------:|------:|
+ *  | Snapdragon 730 (A76)  | 2 NEON/8 micro-ops  |  8.8 GB/s |  10.1 GB/s |  ~16% |
+ *  | Snapdragon 835 (A73)  | 2 NEON/3 micro-ops  |  5.1 GB/s |   5.3 GB/s |   ~5% |
+ *  | Marvell PXA1928 (A53) | In-order dual-issue |  1.9 GB/s |   1.9 GB/s |    0% |
+ *
+ * It also seems to fix some bad codegen on GCC, making it almost as fast as clang.
+ *
+ * @see XXH3_accumulate_512_neon()
+ */
+# ifndef XXH3_NEON_LANES
+#  if (defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) || defined(_M_ARM64EC)) \
+   && !defined(__OPTIMIZE_SIZE__)
+#   define XXH3_NEON_LANES 6
+#  else
+#   define XXH3_NEON_LANES XXH_ACC_NB
+#  endif
+# endif
+#endif  /* XXH_VECTOR == XXH_NEON */
+
+/*
+ * VSX and Z Vector helpers.
+ *
+ * This is very messy, and any pull requests to clean this up are welcome.
+ *
+ * There are a lot of problems with supporting VSX and s390x, due to
+ * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
+ */
+#if XXH_VECTOR == XXH_VSX
+#  if defined(__s390x__)
+#    include 
+#  else
+/* gcc's altivec.h can have the unwanted consequence to unconditionally
+ * #define bool, vector, and pixel keywords,
+ * with bad consequences for programs already using these keywords for other purposes.
+ * The paragraph defining these macros is skipped when __APPLE_ALTIVEC__ is defined.
+ * __APPLE_ALTIVEC__ is _generally_ defined automatically by the compiler,
+ * but it seems that, in some cases, it isn't.
+ * Force the build macro to be defined, so that keywords are not altered.
+ */
+#    if defined(__GNUC__) && !defined(__APPLE_ALTIVEC__)
+#      define __APPLE_ALTIVEC__
+#    endif
+#    include 
+#  endif
+
+typedef __vector unsigned long long xxh_u64x2;
+typedef __vector unsigned char xxh_u8x16;
+typedef __vector unsigned xxh_u32x4;
+
+# ifndef XXH_VSX_BE
+#  if defined(__BIG_ENDIAN__) \
+  || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_VSX_BE 1
+#  elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
+#    warning "-maltivec=be is not recommended. Please use native endianness."
+#    define XXH_VSX_BE 1
+#  else
+#    define XXH_VSX_BE 0
+#  endif
+# endif /* !defined(XXH_VSX_BE) */
+
+# if XXH_VSX_BE
+#  if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
+#    define XXH_vec_revb vec_revb
+#  else
+/*!
+ * A polyfill for POWER9's vec_revb().
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
+{
+    xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
+                                  0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
+    return vec_perm(val, val, vByteSwap);
+}
+#  endif
+# endif /* XXH_VSX_BE */
+
+/*!
+ * Performs an unaligned vector load and byte swaps it on big endian.
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
+{
+    xxh_u64x2 ret;
+    XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
+# if XXH_VSX_BE
+    ret = XXH_vec_revb(ret);
+# endif
+    return ret;
+}
+
+/*
+ * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
+ *
+ * These intrinsics weren't added until GCC 8, despite existing for a while,
+ * and they are endian dependent. Also, their meaning swap depending on version.
+ * */
+# if defined(__s390x__)
+ /* s390x is always big endian, no issue on this platform */
+#  define XXH_vec_mulo vec_mulo
+#  define XXH_vec_mule vec_mule
+# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
+/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
+#  define XXH_vec_mulo __builtin_altivec_vmulouw
+#  define XXH_vec_mule __builtin_altivec_vmuleuw
+# else
+/* gcc needs inline assembly */
+/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+# endif /* XXH_vec_mulo, XXH_vec_mule */
+#endif /* XXH_VECTOR == XXH_VSX */
+
+
+/* prefetch
+ * can be disabled, by declaring XXH_NO_PREFETCH build macro */
+#if defined(XXH_NO_PREFETCH)
+#  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
+#else
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() not defined outside of x86/x64 */
+#    include    /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define XXH_PREFETCH(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+#  else
+#    define XXH_PREFETCH(ptr) (void)(ptr)  /* disabled */
+#  endif
+#endif  /* XXH_NO_PREFETCH */
+
+
+/* ==========================================
+ * XXH3 default settings
+ * ========================================== */
+
+#define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
+
+#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
+#  error "default keyset is not large enough"
+#endif
+
+/*! Pseudorandom secret taken directly from FARSH. */
+XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
+    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
+    0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
+    0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
+    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
+    0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
+    0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
+    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
+    0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
+    0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
+    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
+    0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
+    0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
+};
+
+
+#ifdef XXH_OLD_NAMES
+#  define kSecret XXH3_kSecret
+#endif
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Calculates a 32-bit to 64-bit long multiply.
+ *
+ * Implemented as a macro.
+ *
+ * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
+ * need to (but it shouldn't need to anyways, it is about 7 instructions to do
+ * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
+ * use that instead of the normal method.
+ *
+ * If you are compiling for platforms like Thumb-1 and don't have a better option,
+ * you may also want to write your own long multiply routine here.
+ *
+ * @param x, y Numbers to be multiplied
+ * @return 64-bit product of the low 32 bits of @p x and @p y.
+ */
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64(xxh_u64 x, xxh_u64 y)
+{
+   return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
+}
+#elif defined(_MSC_VER) && defined(_M_IX86)
+#    define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
+#else
+/*
+ * Downcast + upcast is usually better than masking on older compilers like
+ * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
+ *
+ * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
+ * and perform a full 64x64 multiply -- entirely redundant on 32-bit.
+ */
+#    define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
+#endif
+
+/*!
+ * @brief Calculates a 64->128-bit long multiply.
+ *
+ * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar
+ * version.
+ *
+ * @param lhs , rhs The 64-bit integers to be multiplied
+ * @return The 128-bit result represented in an @ref XXH128_hash_t.
+ */
+static XXH128_hash_t
+XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
+{
+    /*
+     * GCC/Clang __uint128_t method.
+     *
+     * On most 64-bit targets, GCC and Clang define a __uint128_t type.
+     * This is usually the best way as it usually uses a native long 64-bit
+     * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
+     *
+     * Usually.
+     *
+     * Despite being a 32-bit platform, Clang (and emscripten) define this type
+     * despite not having the arithmetic for it. This results in a laggy
+     * compiler builtin call which calculates a full 128-bit multiply.
+     * In that case it is best to use the portable one.
+     * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
+     */
+#if (defined(__GNUC__) || defined(__clang__)) && !defined(__wasm__) \
+    && defined(__SIZEOF_INT128__) \
+    || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+
+    __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
+    XXH128_hash_t r128;
+    r128.low64  = (xxh_u64)(product);
+    r128.high64 = (xxh_u64)(product >> 64);
+    return r128;
+
+    /*
+     * MSVC for x64's _umul128 method.
+     *
+     * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
+     *
+     * This compiles to single operand MUL on x64.
+     */
+#elif (defined(_M_X64) || defined(_M_IA64)) && !defined(_M_ARM64EC)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(_umul128)
+#endif
+    xxh_u64 product_high;
+    xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
+    XXH128_hash_t r128;
+    r128.low64  = product_low;
+    r128.high64 = product_high;
+    return r128;
+
+    /*
+     * MSVC for ARM64's __umulh method.
+     *
+     * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
+     */
+#elif defined(_M_ARM64) || defined(_M_ARM64EC)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(__umulh)
+#endif
+    XXH128_hash_t r128;
+    r128.low64  = lhs * rhs;
+    r128.high64 = __umulh(lhs, rhs);
+    return r128;
+
+#else
+    /*
+     * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
+     *
+     * This is a fast and simple grade school multiply, which is shown below
+     * with base 10 arithmetic instead of base 0x100000000.
+     *
+     *           9 3 // D2 lhs = 93
+     *         x 7 5 // D2 rhs = 75
+     *     ----------
+     *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
+     *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
+     *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
+     *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
+     *     ---------
+     *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
+     *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
+     *     ---------
+     *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
+     *
+     * The reasons for adding the products like this are:
+     *  1. It avoids manual carry tracking. Just like how
+     *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
+     *     This avoids a lot of complexity.
+     *
+     *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
+     *     instruction available in ARM's Digital Signal Processing extension
+     *     in 32-bit ARMv6 and later, which is shown below:
+     *
+     *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
+     *         {
+     *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
+     *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
+     *             *RdHi = (xxh_u32)(product >> 32);
+     *         }
+     *
+     *     This instruction was designed for efficient long multiplication, and
+     *     allows this to be calculated in only 4 instructions at speeds
+     *     comparable to some 64-bit ALUs.
+     *
+     *  3. It isn't terrible on other platforms. Usually this will be a couple
+     *     of 32-bit ADD/ADCs.
+     */
+
+    /* First calculate all of the cross products. */
+    xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
+    xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32,        rhs & 0xFFFFFFFF);
+    xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
+    xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32,        rhs >> 32);
+
+    /* Now add the products together. These will never overflow. */
+    xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
+    xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32)        + hi_hi;
+    xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
+
+    XXH128_hash_t r128;
+    r128.low64  = lower;
+    r128.high64 = upper;
+    return r128;
+#endif
+}
+
+/*!
+ * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
+ *
+ * The reason for the separate function is to prevent passing too many structs
+ * around by value. This will hopefully inline the multiply, but we don't force it.
+ *
+ * @param lhs , rhs The 64-bit integers to multiply
+ * @return The low 64 bits of the product XOR'd by the high 64 bits.
+ * @see XXH_mult64to128()
+ */
+static xxh_u64
+XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
+{
+    XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
+    return product.low64 ^ product.high64;
+}
+
+/*! Seems to produce slightly better code on GCC for some reason. */
+XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
+{
+    XXH_ASSERT(0 <= shift && shift < 64);
+    return v64 ^ (v64 >> shift);
+}
+
+/*
+ * This is a fast avalanche stage,
+ * suitable when input bits are already partially mixed
+ */
+static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
+{
+    h64 = XXH_xorshift64(h64, 37);
+    h64 *= 0x165667919E3779F9ULL;
+    h64 = XXH_xorshift64(h64, 32);
+    return h64;
+}
+
+/*
+ * This is a stronger avalanche,
+ * inspired by Pelle Evensen's rrmxmx
+ * preferable when input has not been previously mixed
+ */
+static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
+{
+    /* this mix is inspired by Pelle Evensen's rrmxmx */
+    h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
+    h64 *= 0x9FB21C651E98DF25ULL;
+    h64 ^= (h64 >> 35) + len ;
+    h64 *= 0x9FB21C651E98DF25ULL;
+    return XXH_xorshift64(h64, 28);
+}
+
+
+/* ==========================================
+ * Short keys
+ * ==========================================
+ * One of the shortcomings of XXH32 and XXH64 was that their performance was
+ * sub-optimal on short lengths. It used an iterative algorithm which strongly
+ * favored lengths that were a multiple of 4 or 8.
+ *
+ * Instead of iterating over individual inputs, we use a set of single shot
+ * functions which piece together a range of lengths and operate in constant time.
+ *
+ * Additionally, the number of multiplies has been significantly reduced. This
+ * reduces latency, especially when emulating 64-bit multiplies on 32-bit.
+ *
+ * Depending on the platform, this may or may not be faster than XXH32, but it
+ * is almost guaranteed to be faster than XXH64.
+ */
+
+/*
+ * At very short lengths, there isn't enough input to fully hide secrets, or use
+ * the entire secret.
+ *
+ * There is also only a limited amount of mixing we can do before significantly
+ * impacting performance.
+ *
+ * Therefore, we use different sections of the secret and always mix two secret
+ * samples with an XOR. This should have no effect on performance on the
+ * seedless or withSeed variants because everything _should_ be constant folded
+ * by modern compilers.
+ *
+ * The XOR mixing hides individual parts of the secret and increases entropy.
+ *
+ * This adds an extra layer of strength for custom secrets.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combined = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combined = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combined = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8  const c1 = input[0];
+        xxh_u8  const c2 = input[len >> 1];
+        xxh_u8  const c3 = input[len - 1];
+        xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2  << 24)
+                               | ((xxh_u32)c3 <<  0) | ((xxh_u32)len << 8);
+        xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
+        return XXH64_avalanche(keyed);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input1 = XXH_readLE32(input);
+        xxh_u32 const input2 = XXH_readLE32(input + len - 4);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
+        xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
+        xxh_u64 const keyed = input64 ^ bitflip;
+        return XXH3_rrmxmx(keyed, len);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
+        xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
+        xxh_u64 const input_lo = XXH_readLE64(input)           ^ bitflip1;
+        xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
+        xxh_u64 const acc = len
+                          + XXH_swap64(input_lo) + input_hi
+                          + XXH3_mul128_fold64(input_lo, input_hi);
+        return XXH3_avalanche(acc);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (XXH_likely(len >  8)) return XXH3_len_9to16_64b(input, len, secret, seed);
+        if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
+        return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
+    }
+}
+
+/*
+ * DISCLAIMER: There are known *seed-dependent* multicollisions here due to
+ * multiplication by zero, affecting hashes of lengths 17 to 240.
+ *
+ * However, they are very unlikely.
+ *
+ * Keep this in mind when using the unseeded XXH3_64bits() variant: As with all
+ * unseeded non-cryptographic hashes, it does not attempt to defend itself
+ * against specially crafted inputs, only random inputs.
+ *
+ * Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes
+ * cancelling out the secret is taken an arbitrary number of times (addressed
+ * in XXH3_accumulate_512), this collision is very unlikely with random inputs
+ * and/or proper seeding:
+ *
+ * This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a
+ * function that is only called up to 16 times per hash with up to 240 bytes of
+ * input.
+ *
+ * This is not too bad for a non-cryptographic hash function, especially with
+ * only 64 bit outputs.
+ *
+ * The 128-bit variant (which trades some speed for strength) is NOT affected
+ * by this, although it is always a good idea to use a proper seed if you care
+ * about strength.
+ */
+XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
+                                     const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
+{
+#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__i386__) && defined(__SSE2__)  /* x86 + SSE2 */ \
+  && !defined(XXH_ENABLE_AUTOVECTORIZE)      /* Define to disable like XXH32 hack */
+    /*
+     * UGLY HACK:
+     * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
+     * slower code.
+     *
+     * By forcing seed64 into a register, we disrupt the cost model and
+     * cause it to scalarize. See `XXH32_round()`
+     *
+     * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
+     * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
+     * GCC 9.2, despite both emitting scalar code.
+     *
+     * GCC generates much better scalar code than Clang for the rest of XXH3,
+     * which is why finding a more optimal codepath is an interest.
+     */
+    XXH_COMPILER_GUARD(seed64);
+#endif
+    {   xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64 const input_hi = XXH_readLE64(input+8);
+        return XXH3_mul128_fold64(
+            input_lo ^ (XXH_readLE64(secret)   + seed64),
+            input_hi ^ (XXH_readLE64(secret+8) - seed64)
+        );
+    }
+}
+
+/* For mid range keys, XXH3 uses a Mum-hash variant. */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                     const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                     XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc += XXH3_mix16B(input+48, secret+96, seed);
+                    acc += XXH3_mix16B(input+len-64, secret+112, seed);
+                }
+                acc += XXH3_mix16B(input+32, secret+64, seed);
+                acc += XXH3_mix16B(input+len-48, secret+80, seed);
+            }
+            acc += XXH3_mix16B(input+16, secret+32, seed);
+            acc += XXH3_mix16B(input+len-32, secret+48, seed);
+        }
+        acc += XXH3_mix16B(input+0, secret+0, seed);
+        acc += XXH3_mix16B(input+len-16, secret+16, seed);
+
+        return XXH3_avalanche(acc);
+    }
+}
+
+#define XXH3_MIDSIZE_MAX 240
+
+XXH_NO_INLINE XXH64_hash_t
+XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    #define XXH3_MIDSIZE_STARTOFFSET 3
+    #define XXH3_MIDSIZE_LASTOFFSET  17
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        int const nbRounds = (int)len / 16;
+        int i;
+        for (i=0; i<8; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
+        }
+        acc = XXH3_avalanche(acc);
+        XXH_ASSERT(nbRounds >= 8);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
+         * In everywhere else, it uses scalar code.
+         *
+         * For 64->128-bit multiplies, even if the NEON was 100% optimal, it
+         * would still be slower than UMAAL (see XXH_mult64to128).
+         *
+         * Unfortunately, Clang doesn't handle the long multiplies properly and
+         * converts them to the nonexistent "vmulq_u64" intrinsic, which is then
+         * scalarized into an ugly mess of VMOV.32 instructions.
+         *
+         * This mess is difficult to avoid without turning autovectorization
+         * off completely, but they are usually relatively minor and/or not
+         * worth it to fix.
+         *
+         * This loop is the easiest to fix, as unlike XXH32, this pragma
+         * _actually works_ because it is a loop vectorization instead of an
+         * SLP vectorization.
+         */
+        #pragma clang loop vectorize(disable)
+#endif
+        for (i=8 ; i < nbRounds; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
+        }
+        /* last bytes */
+        acc += XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
+        return XXH3_avalanche(acc);
+    }
+}
+
+
+/* =======     Long Keys     ======= */
+
+#define XXH_STRIPE_LEN 64
+#define XXH_SECRET_CONSUME_RATE 8   /* nb of secret bytes consumed at each accumulation */
+#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
+
+#ifdef XXH_OLD_NAMES
+#  define STRIPE_LEN XXH_STRIPE_LEN
+#  define ACC_NB XXH_ACC_NB
+#endif
+
+XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
+{
+    if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
+    XXH_memcpy(dst, &v64, sizeof(v64));
+}
+
+/* Several intrinsic functions below are supposed to accept __int64 as argument,
+ * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
+ * However, several environments do not define __int64 type,
+ * requiring a workaround.
+ */
+#if !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+    typedef int64_t xxh_i64;
+#else
+    /* the following type must have a width of 64-bit */
+    typedef long long xxh_i64;
+#endif
+
+
+/*
+ * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
+ *
+ * It is a hardened version of UMAC, based off of FARSH's implementation.
+ *
+ * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
+ * implementations, and it is ridiculously fast.
+ *
+ * We harden it by mixing the original input to the accumulators as well as the product.
+ *
+ * This means that in the (relatively likely) case of a multiply by zero, the
+ * original input is preserved.
+ *
+ * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
+ * cross-pollination, as otherwise the upper and lower halves would be
+ * essentially independent.
+ *
+ * This doesn't matter on 64-bit hashes since they all get merged together in
+ * the end, so we skip the extra step.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
+ */
+
+#if (XXH_VECTOR == XXH_AVX512) \
+     || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
+
+#ifndef XXH_TARGET_AVX512
+# define XXH_TARGET_AVX512  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    __m512i* const xacc = (__m512i *) acc;
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+
+    {
+        /* data_vec    = input[0]; */
+        __m512i const data_vec    = _mm512_loadu_si512   (input);
+        /* key_vec     = secret[0]; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        /* data_key    = data_vec ^ key_vec; */
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+        /* data_key_lo = data_key >> 32; */
+        __m512i const data_key_lo = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+        /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+        __m512i const product     = _mm512_mul_epu32     (data_key, data_key_lo);
+        /* xacc[0] += swap(data_vec); */
+        __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
+        __m512i const sum       = _mm512_add_epi64(*xacc, data_swap);
+        /* xacc[0] += product; */
+        *xacc = _mm512_add_epi64(product, sum);
+    }
+}
+
+/*
+ * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
+ *
+ * Multiplication isn't perfect, as explained by Google in HighwayHash:
+ *
+ *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
+ *  // varying degrees. In descending order of goodness, bytes
+ *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
+ *  // As expected, the upper and lower bytes are much worse.
+ *
+ * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
+ *
+ * Since our algorithm uses a pseudorandom secret to add some variance into the
+ * mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
+ *
+ * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
+ * extraction.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
+ */
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+    {   __m512i* const xacc = (__m512i*) acc;
+        const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
+
+        /* xacc[0] ^= (xacc[0] >> 47) */
+        __m512i const acc_vec     = *xacc;
+        __m512i const shifted     = _mm512_srli_epi64    (acc_vec, 47);
+        __m512i const data_vec    = _mm512_xor_si512     (acc_vec, shifted);
+        /* xacc[0] ^= secret; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+
+        /* xacc[0] *= XXH_PRIME32_1; */
+        __m512i const data_key_hi = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+        __m512i const prod_lo     = _mm512_mul_epu32     (data_key, prime32);
+        __m512i const prod_hi     = _mm512_mul_epu32     (data_key_hi, prime32);
+        *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
+    XXH_ASSERT(((size_t)customSecret & 63) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
+        __m512i const seed = _mm512_mask_set1_epi64(_mm512_set1_epi64((xxh_i64)seed64), 0xAA, (xxh_i64)(0U - seed64));
+
+        const __m512i* const src  = (const __m512i*) ((const void*) XXH3_kSecret);
+              __m512i* const dest = (      __m512i*) customSecret;
+        int i;
+        XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 63) == 0);
+        for (i=0; i < nbRounds; ++i) {
+            /* GCC has a bug, _mm512_stream_load_si512 accepts 'void*', not 'void const*',
+             * this will warn "discards 'const' qualifier". */
+            union {
+                const __m512i* cp;
+                void* p;
+            } remote_const_void;
+            remote_const_void.cp = src + i;
+            dest[i] = _mm512_add_epi64(_mm512_stream_load_si512(remote_const_void.p), seed);
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_AVX2) \
+    || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
+
+#ifndef XXH_TARGET_AVX2
+# define XXH_TARGET_AVX2  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc    =       (__m256i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason. */
+        const         __m256i* const xinput  = (const __m256i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m256i const data_vec    = _mm256_loadu_si256    (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m256i const data_key_lo = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m256i const product     = _mm256_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
+            __m256i const sum       = _mm256_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm256_add_epi64(product, sum);
+    }   }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc = (__m256i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+        const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m256i const acc_vec     = xacc[i];
+            __m256i const shifted     = _mm256_srli_epi64    (acc_vec, 47);
+            __m256i const data_vec    = _mm256_xor_si256     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m256i const data_key_hi = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m256i const prod_lo     = _mm256_mul_epu32     (data_key, prime32);
+            __m256i const prod_hi     = _mm256_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
+    (void)(&XXH_writeLE64);
+    XXH_PREFETCH(customSecret);
+    {   __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);
+
+        const __m256i* const src  = (const __m256i*) ((const void*) XXH3_kSecret);
+              __m256i*       dest = (      __m256i*) customSecret;
+
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dest);
+#       endif
+        XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 31) == 0);
+
+        /* GCC -O2 need unroll loop manually */
+        dest[0] = _mm256_add_epi64(_mm256_stream_load_si256(src+0), seed);
+        dest[1] = _mm256_add_epi64(_mm256_stream_load_si256(src+1), seed);
+        dest[2] = _mm256_add_epi64(_mm256_stream_load_si256(src+2), seed);
+        dest[3] = _mm256_add_epi64(_mm256_stream_load_si256(src+3), seed);
+        dest[4] = _mm256_add_epi64(_mm256_stream_load_si256(src+4), seed);
+        dest[5] = _mm256_add_epi64(_mm256_stream_load_si256(src+5), seed);
+    }
+}
+
+#endif
+
+/* x86dispatch always generates SSE2 */
+#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
+
+#ifndef XXH_TARGET_SSE2
+# define XXH_TARGET_SSE2  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* SSE2 is just a half-scale version of the AVX2 version. */
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc    =       (__m128i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xinput  = (const __m128i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m128i const data_vec    = _mm_loadu_si128   (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m128i const product     = _mm_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
+            __m128i const sum       = _mm_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm_add_epi64(product, sum);
+    }   }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc = (__m128i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+        const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m128i const acc_vec     = xacc[i];
+            __m128i const shifted     = _mm_srli_epi64    (acc_vec, 47);
+            __m128i const data_vec    = _mm_xor_si128     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m128i const prod_lo     = _mm_mul_epu32     (data_key, prime32);
+            __m128i const prod_hi     = _mm_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
+
+#       if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
+        /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
+        XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
+        __m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
+#       else
+        __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
+#       endif
+        int i;
+
+        const void* const src16 = XXH3_kSecret;
+        __m128i* dst16 = (__m128i*) customSecret;
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dst16);
+#       endif
+        XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dst16 & 15) == 0);
+
+        for (i=0; i < nbRounds; ++i) {
+            dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_NEON)
+
+/* forward declarations for the scalar routines */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc, void const* XXH_RESTRICT input,
+                 void const* XXH_RESTRICT secret, size_t lane);
+
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+                         void const* XXH_RESTRICT secret, size_t lane);
+
+/*!
+ * @internal
+ * @brief The bulk processing loop for NEON.
+ *
+ * The NEON code path is actually partially scalar when running on AArch64. This
+ * is to optimize the pipelining and can have up to 15% speedup depending on the
+ * CPU, and it also mitigates some GCC codegen issues.
+ *
+ * @see XXH3_NEON_LANES for configuring this and details about this optimization.
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    XXH_STATIC_ASSERT(XXH3_NEON_LANES > 0 && XXH3_NEON_LANES <= XXH_ACC_NB && XXH3_NEON_LANES % 2 == 0);
+    {
+        uint64x2_t* const xacc = (uint64x2_t *) acc;
+        /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
+        uint8_t const* const xinput = (const uint8_t *) input;
+        uint8_t const* const xsecret  = (const uint8_t *) secret;
+
+        size_t i;
+        /* NEON for the first few lanes (these loops are normally interleaved) */
+        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
+            /* data_vec = xinput[i]; */
+            uint8x16_t data_vec    = vld1q_u8(xinput  + (i * 16));
+            /* key_vec  = xsecret[i];  */
+            uint8x16_t key_vec     = vld1q_u8(xsecret + (i * 16));
+            uint64x2_t data_key;
+            uint32x2_t data_key_lo, data_key_hi;
+            /* xacc[i] += swap(data_vec); */
+            uint64x2_t const data64  = vreinterpretq_u64_u8(data_vec);
+            uint64x2_t const swapped = vextq_u64(data64, data64, 1);
+            xacc[i] = vaddq_u64 (xacc[i], swapped);
+            /* data_key = data_vec ^ key_vec; */
+            data_key = vreinterpretq_u64_u8(veorq_u8(data_vec, key_vec));
+            /* data_key_lo = (uint32x2_t) (data_key & 0xFFFFFFFF);
+             * data_key_hi = (uint32x2_t) (data_key >> 32);
+             * data_key = UNDEFINED; */
+            XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
+            /* xacc[i] += (uint64x2_t) data_key_lo * (uint64x2_t) data_key_hi; */
+            xacc[i] = vmlal_u32 (xacc[i], data_key_lo, data_key_hi);
+
+        }
+        /* Scalar for the remainder. This may be a zero iteration loop. */
+        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+            XXH3_scalarRound(acc, input, secret, i);
+        }
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {   uint64x2_t* xacc       = (uint64x2_t*) acc;
+        uint8_t const* xsecret = (uint8_t const*) secret;
+        uint32x2_t prime       = vdup_n_u32 (XXH_PRIME32_1);
+
+        size_t i;
+        /* NEON for the first few lanes (these loops are normally interleaved) */
+        for (i=0; i < XXH3_NEON_LANES / 2; i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            uint64x2_t acc_vec  = xacc[i];
+            uint64x2_t shifted  = vshrq_n_u64 (acc_vec, 47);
+            uint64x2_t data_vec = veorq_u64   (acc_vec, shifted);
+
+            /* xacc[i] ^= xsecret[i]; */
+            uint8x16_t key_vec  = vld1q_u8    (xsecret + (i * 16));
+            uint64x2_t data_key = veorq_u64   (data_vec, vreinterpretq_u64_u8(key_vec));
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            uint32x2_t data_key_lo, data_key_hi;
+            /* data_key_lo = (uint32x2_t) (xacc[i] & 0xFFFFFFFF);
+             * data_key_hi = (uint32x2_t) (xacc[i] >> 32);
+             * xacc[i] = UNDEFINED; */
+            XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
+            {   /*
+                 * prod_hi = (data_key >> 32) * XXH_PRIME32_1;
+                 *
+                 * Avoid vmul_u32 + vshll_n_u32 since Clang 6 and 7 will
+                 * incorrectly "optimize" this:
+                 *   tmp     = vmul_u32(vmovn_u64(a), vmovn_u64(b));
+                 *   shifted = vshll_n_u32(tmp, 32);
+                 * to this:
+                 *   tmp     = "vmulq_u64"(a, b); // no such thing!
+                 *   shifted = vshlq_n_u64(tmp, 32);
+                 *
+                 * However, unlike SSE, Clang lacks a 64-bit multiply routine
+                 * for NEON, and it scalarizes two 64-bit multiplies instead.
+                 *
+                 * vmull_u32 has the same timing as vmul_u32, and it avoids
+                 * this bug completely.
+                 * See https://bugs.llvm.org/show_bug.cgi?id=39967
+                 */
+                uint64x2_t prod_hi = vmull_u32 (data_key_hi, prime);
+                /* xacc[i] = prod_hi << 32; */
+                xacc[i] = vshlq_n_u64(prod_hi, 32);
+                /* xacc[i] += (prod_hi & 0xFFFFFFFF) * XXH_PRIME32_1; */
+                xacc[i] = vmlal_u32(xacc[i], data_key_lo, prime);
+            }
+        }
+        /* Scalar for the remainder. This may be a zero iteration loop. */
+        for (i = XXH3_NEON_LANES; i < XXH_ACC_NB; i++) {
+            XXH3_scalarScrambleRound(acc, secret, i);
+        }
+    }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_VSX)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_vsx(  void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* presumed aligned */
+    unsigned int* const xacc = (unsigned int*) acc;
+    xxh_u64x2 const* const xinput   = (xxh_u64x2 const*) input;   /* no alignment restriction */
+    xxh_u64x2 const* const xsecret  = (xxh_u64x2 const*) secret;    /* no alignment restriction */
+    xxh_u64x2 const v32 = { 32, 32 };
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+        /* data_vec = xinput[i]; */
+        xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + i);
+        /* key_vec = xsecret[i]; */
+        xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + i);
+        xxh_u64x2 const data_key = data_vec ^ key_vec;
+        /* shuffled = (data_key << 32) | (data_key >> 32); */
+        xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
+        /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
+        xxh_u64x2 const product  = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
+        /* acc_vec = xacc[i]; */
+        xxh_u64x2 acc_vec        = (xxh_u64x2)vec_xl(0, xacc + 4 * i);
+        acc_vec += product;
+
+        /* swap high and low halves */
+#ifdef __s390x__
+        acc_vec += vec_permi(data_vec, data_vec, 2);
+#else
+        acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
+#endif
+        /* xacc[i] = acc_vec; */
+        vec_xst((xxh_u32x4)acc_vec, 0, xacc + 4 * i);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {         xxh_u64x2* const xacc    =       (xxh_u64x2*) acc;
+        const xxh_u64x2* const xsecret = (const xxh_u64x2*) secret;
+        /* constants */
+        xxh_u64x2 const v32  = { 32, 32 };
+        xxh_u64x2 const v47 = { 47, 47 };
+        xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
+        size_t i;
+        for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            xxh_u64x2 const acc_vec  = xacc[i];
+            xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
+
+            /* xacc[i] ^= xsecret[i]; */
+            xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + i);
+            xxh_u64x2 const data_key = data_vec ^ key_vec;
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF);  */
+            xxh_u64x2 const prod_even  = XXH_vec_mule((xxh_u32x4)data_key, prime);
+            /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
+            xxh_u64x2 const prod_odd  = XXH_vec_mulo((xxh_u32x4)data_key, prime);
+            xacc[i] = prod_odd + (prod_even << v32);
+    }   }
+}
+
+#endif
+
+/* scalar variants - universal */
+
+/*!
+ * @internal
+ * @brief Scalar round for @ref XXH3_accumulate_512_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
+XXH_FORCE_INLINE void
+XXH3_scalarRound(void* XXH_RESTRICT acc,
+                 void const* XXH_RESTRICT input,
+                 void const* XXH_RESTRICT secret,
+                 size_t lane)
+{
+    xxh_u64* xacc = (xxh_u64*) acc;
+    xxh_u8 const* xinput  = (xxh_u8 const*) input;
+    xxh_u8 const* xsecret = (xxh_u8 const*) secret;
+    XXH_ASSERT(lane < XXH_ACC_NB);
+    XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
+    {
+        xxh_u64 const data_val = XXH_readLE64(xinput + lane * 8);
+        xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + lane * 8);
+        xacc[lane ^ 1] += data_val; /* swap adjacent lanes */
+        xacc[lane] += XXH_mult32to64(data_key & 0xFFFFFFFF, data_key >> 32);
+    }
+}
+
+/*!
+ * @internal
+ * @brief Processes a 64 byte block of data using the scalar path.
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    size_t i;
+    for (i=0; i < XXH_ACC_NB; i++) {
+        XXH3_scalarRound(acc, input, secret, i);
+    }
+}
+
+/*!
+ * @internal
+ * @brief Scalar scramble step for @ref XXH3_scrambleAcc_scalar().
+ *
+ * This is extracted to its own function because the NEON path uses a combination
+ * of NEON and scalar.
+ */
+XXH_FORCE_INLINE void
+XXH3_scalarScrambleRound(void* XXH_RESTRICT acc,
+                         void const* XXH_RESTRICT secret,
+                         size_t lane)
+{
+    xxh_u64* const xacc = (xxh_u64*) acc;   /* presumed aligned */
+    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
+    XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
+    XXH_ASSERT(lane < XXH_ACC_NB);
+    {
+        xxh_u64 const key64 = XXH_readLE64(xsecret + lane * 8);
+        xxh_u64 acc64 = xacc[lane];
+        acc64 = XXH_xorshift64(acc64, 47);
+        acc64 ^= key64;
+        acc64 *= XXH_PRIME32_1;
+        xacc[lane] = acc64;
+    }
+}
+
+/*!
+ * @internal
+ * @brief Scrambles the accumulators after a large chunk has been read
+ */
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    size_t i;
+    for (i=0; i < XXH_ACC_NB; i++) {
+        XXH3_scalarScrambleRound(acc, secret, i);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    /*
+     * We need a separate pointer for the hack below,
+     * which requires a non-const pointer.
+     * Any decent compiler will optimize this out otherwise.
+     */
+    const xxh_u8* kSecretPtr = XXH3_kSecret;
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+
+#if defined(__clang__) && defined(__aarch64__)
+    /*
+     * UGLY HACK:
+     * Clang generates a bunch of MOV/MOVK pairs for aarch64, and they are
+     * placed sequentially, in order, at the top of the unrolled loop.
+     *
+     * While MOVK is great for generating constants (2 cycles for a 64-bit
+     * constant compared to 4 cycles for LDR), it fights for bandwidth with
+     * the arithmetic instructions.
+     *
+     *   I   L   S
+     * MOVK
+     * MOVK
+     * MOVK
+     * MOVK
+     * ADD
+     * SUB      STR
+     *          STR
+     * By forcing loads from memory (as the asm line causes Clang to assume
+     * that XXH3_kSecretPtr has been changed), the pipelines are used more
+     * efficiently:
+     *   I   L   S
+     *      LDR
+     *  ADD LDR
+     *  SUB     STR
+     *          STR
+     *
+     * See XXH3_NEON_LANES for details on the pipsline.
+     *
+     * XXH3_64bits_withSeed, len == 256, Snapdragon 835
+     *   without hack: 2654.4 MB/s
+     *   with hack:    3202.9 MB/s
+     */
+    XXH_COMPILER_GUARD(kSecretPtr);
+#endif
+    /*
+     * Note: in debug mode, this overrides the asm optimization
+     * and Clang will emit MOVK chains again.
+     */
+    XXH_ASSERT(kSecretPtr == XXH3_kSecret);
+
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
+        int i;
+        for (i=0; i < nbRounds; i++) {
+            /*
+             * The asm hack causes Clang to assume that kSecretPtr aliases with
+             * customSecret, and on aarch64, this prevented LDP from merging two
+             * loads together for free. Putting the loads together before the stores
+             * properly generates LDP.
+             */
+            xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i)     + seed64;
+            xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i,     lo);
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
+    }   }
+}
+
+
+typedef void (*XXH3_f_accumulate_512)(void* XXH_RESTRICT, const void*, const void*);
+typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
+typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
+
+
+#if (XXH_VECTOR == XXH_AVX512)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx512
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
+
+#elif (XXH_VECTOR == XXH_AVX2)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
+
+#elif (XXH_VECTOR == XXH_SSE2)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_sse2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
+
+#elif (XXH_VECTOR == XXH_NEON)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_neon
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_neon
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#elif (XXH_VECTOR == XXH_VSX)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_vsx
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#else /* scalar */
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#endif
+
+
+
+#ifndef XXH_PREFETCH_DIST
+#  ifdef __clang__
+#    define XXH_PREFETCH_DIST 320
+#  else
+#    if (XXH_VECTOR == XXH_AVX512)
+#      define XXH_PREFETCH_DIST 512
+#    else
+#      define XXH_PREFETCH_DIST 384
+#    endif
+#  endif  /* __clang__ */
+#endif  /* XXH_PREFETCH_DIST */
+
+/*
+ * XXH3_accumulate()
+ * Loops over XXH3_accumulate_512().
+ * Assumption: nbStripes will not overflow the secret size
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate(     xxh_u64* XXH_RESTRICT acc,
+                const xxh_u8* XXH_RESTRICT input,
+                const xxh_u8* XXH_RESTRICT secret,
+                      size_t nbStripes,
+                      XXH3_f_accumulate_512 f_acc512)
+{
+    size_t n;
+    for (n = 0; n < nbStripes; n++ ) {
+        const xxh_u8* const in = input + n*XXH_STRIPE_LEN;
+        XXH_PREFETCH(in + XXH_PREFETCH_DIST);
+        f_acc512(acc,
+                 in,
+                 secret + n*XXH_SECRET_CONSUME_RATE);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
+                      const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate_512 f_acc512,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
+    size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
+    size_t const nb_blocks = (len - 1) / block_len;
+
+    size_t n;
+
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+
+    for (n = 0; n < nb_blocks; n++) {
+        XXH3_accumulate(acc, input + n*block_len, secret, nbStripesPerBlock, f_acc512);
+        f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
+    }
+
+    /* last partial block */
+    XXH_ASSERT(len > XXH_STRIPE_LEN);
+    {   size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
+        XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
+        XXH3_accumulate(acc, input + nb_blocks*block_len, secret, nbStripes, f_acc512);
+
+        /* last stripe */
+        {   const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
+#define XXH_SECRET_LASTACC_START 7  /* not aligned on 8, last secret is different from acc & scrambler */
+            f_acc512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
+    }   }
+}
+
+XXH_FORCE_INLINE xxh_u64
+XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
+{
+    return XXH3_mul128_fold64(
+               acc[0] ^ XXH_readLE64(secret),
+               acc[1] ^ XXH_readLE64(secret+8) );
+}
+
+static XXH64_hash_t
+XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
+{
+    xxh_u64 result64 = start;
+    size_t i = 0;
+
+    for (i = 0; i < 4; i++) {
+        result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
+         * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
+         * XXH3_64bits, len == 256, Snapdragon 835:
+         *   without hack: 2063.7 MB/s
+         *   with hack:    2560.7 MB/s
+         */
+        XXH_COMPILER_GUARD(result64);
+#endif
+    }
+
+    return XXH3_avalanche(result64);
+}
+
+#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
+                        XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
+                           const void* XXH_RESTRICT secret, size_t secretSize,
+                           XXH3_f_accumulate_512 f_acc512,
+                           XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc512, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    /* do not align on 8, so that the secret is different from the accumulator */
+#define XXH_SECRET_MERGEACCS_START 11
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
+}
+
+/*
+ * It's important for performance to transmit secret's size (when it's static)
+ * so that the compiler can properly optimize the vectorized loop.
+ * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                             XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * It's preferable for performance that XXH3_hashLong is not inlined,
+ * as it results in a smaller function for small data, easier to the instruction cache.
+ * Note that inside this no_inline function, we do inline the internal loop,
+ * and provide a statically defined secret size to allow optimization of vector loop.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
+                          XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * XXH3_hashLong_64b_withSeed():
+ * Generate a custom key based on alteration of default XXH3_kSecret with the seed,
+ * and then use this key for long mode hashing.
+ *
+ * This operation is decently fast but nonetheless costs a little bit of time.
+ * Try to avoid it whenever possible (typically when seed==0).
+ *
+ * It's important for performance that XXH3_hashLong is not inlined. Not sure
+ * why (uop cache maybe?), but the difference is large and easily measurable.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
+                                    XXH64_hash_t seed,
+                                    XXH3_f_accumulate_512 f_acc512,
+                                    XXH3_f_scrambleAcc f_scramble,
+                                    XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed == 0)
+        return XXH3_hashLong_64b_internal(input, len,
+                                          XXH3_kSecret, sizeof(XXH3_kSecret),
+                                          f_acc512, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed);
+        return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
+                                          f_acc512, f_scramble);
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed(const void* input, size_t len,
+                           XXH64_hash_t seed, const xxh_u8* secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
+                XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
+
+
+typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
+                                          XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
+                     XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                     XXH3_hashLong64_f f_hashLong)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secretLen` condition is not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     * Also, note that function signature doesn't offer room to return an error.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
+}
+
+
+/* ===   Public entry point   === */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* input, size_t len)
+{
+    return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_internal(input, len, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
+}
+
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_64b_withSecret(input, len, seed, (const xxh_u8*)secret, secretSize);
+}
+
+
+/* ===   XXH3 streaming   === */
+
+/*
+ * Malloc's a pointer that is always aligned to align.
+ *
+ * This must be freed with `XXH_alignedFree()`.
+ *
+ * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
+ * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
+ * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
+ *
+ * This underalignment previously caused a rather obvious crash which went
+ * completely unnoticed due to XXH3_createState() not actually being tested.
+ * Credit to RedSpah for noticing this bug.
+ *
+ * The alignment is done manually: Functions like posix_memalign or _mm_malloc
+ * are avoided: To maintain portability, we would have to write a fallback
+ * like this anyways, and besides, testing for the existence of library
+ * functions without relying on external build tools is impossible.
+ *
+ * The method is simple: Overallocate, manually align, and store the offset
+ * to the original behind the returned pointer.
+ *
+ * Align must be a power of 2 and 8 <= align <= 128.
+ */
+static void* XXH_alignedMalloc(size_t s, size_t align)
+{
+    XXH_ASSERT(align <= 128 && align >= 8); /* range check */
+    XXH_ASSERT((align & (align-1)) == 0);   /* power of 2 */
+    XXH_ASSERT(s != 0 && s < (s + align));  /* empty/overflow */
+    {   /* Overallocate to make room for manual realignment and an offset byte */
+        xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
+        if (base != NULL) {
+            /*
+             * Get the offset needed to align this pointer.
+             *
+             * Even if the returned pointer is aligned, there will always be
+             * at least one byte to store the offset to the original pointer.
+             */
+            size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
+            /* Add the offset for the now-aligned pointer */
+            xxh_u8* ptr = base + offset;
+
+            XXH_ASSERT((size_t)ptr % align == 0);
+
+            /* Store the offset immediately before the returned pointer. */
+            ptr[-1] = (xxh_u8)offset;
+            return ptr;
+        }
+        return NULL;
+    }
+}
+/*
+ * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
+ * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
+ */
+static void XXH_alignedFree(void* p)
+{
+    if (p != NULL) {
+        xxh_u8* ptr = (xxh_u8*)p;
+        /* Get the offset byte we added in XXH_malloc. */
+        xxh_u8 offset = ptr[-1];
+        /* Free the original malloc'd pointer */
+        xxh_u8* base = ptr - offset;
+        XXH_free(base);
+    }
+}
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
+{
+    XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
+    if (state==NULL) return NULL;
+    XXH3_INITSTATE(state);
+    return state;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
+{
+    XXH_alignedFree(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state)
+{
+    XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
+}
+
+static void
+XXH3_reset_internal(XXH3_state_t* statePtr,
+                    XXH64_hash_t seed,
+                    const void* secret, size_t secretSize)
+{
+    size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
+    size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
+    XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
+    XXH_ASSERT(statePtr != NULL);
+    /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
+    memset((char*)statePtr + initStart, 0, initLength);
+    statePtr->acc[0] = XXH_PRIME32_3;
+    statePtr->acc[1] = XXH_PRIME64_1;
+    statePtr->acc[2] = XXH_PRIME64_2;
+    statePtr->acc[3] = XXH_PRIME64_3;
+    statePtr->acc[4] = XXH_PRIME64_4;
+    statePtr->acc[5] = XXH_PRIME32_2;
+    statePtr->acc[6] = XXH_PRIME64_5;
+    statePtr->acc[7] = XXH_PRIME32_1;
+    statePtr->seed = seed;
+    statePtr->useSeed = (seed != 0);
+    statePtr->extSecret = (const unsigned char*)secret;
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+    statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
+    statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset(XXH3_state_t* statePtr)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, secret, secretSize);
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (seed==0) return XXH3_64bits_reset(statePtr);
+    if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
+        XXH3_initCustomSecret(statePtr->customSecret, seed);
+    XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed64)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, seed64, secret, secretSize);
+    statePtr->useSeed = 1; /* always, even if seed64==0 */
+    return XXH_OK;
+}
+
+/* Note : when XXH3_consumeStripes() is invoked,
+ * there must be a guarantee that at least one more byte must be consumed from input
+ * so that the function can blindly consume all stripes using the "normal" secret segment */
+XXH_FORCE_INLINE void
+XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
+                    size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
+                    const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
+                    const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
+                    XXH3_f_accumulate_512 f_acc512,
+                    XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ASSERT(nbStripes <= nbStripesPerBlock);  /* can handle max 1 scramble per invocation */
+    XXH_ASSERT(*nbStripesSoFarPtr < nbStripesPerBlock);
+    if (nbStripesPerBlock - *nbStripesSoFarPtr <= nbStripes) {
+        /* need a scrambling operation */
+        size_t const nbStripesToEndofBlock = nbStripesPerBlock - *nbStripesSoFarPtr;
+        size_t const nbStripesAfterBlock = nbStripes - nbStripesToEndofBlock;
+        XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripesToEndofBlock, f_acc512);
+        f_scramble(acc, secret + secretLimit);
+        XXH3_accumulate(acc, input + nbStripesToEndofBlock * XXH_STRIPE_LEN, secret, nbStripesAfterBlock, f_acc512);
+        *nbStripesSoFarPtr = nbStripesAfterBlock;
+    } else {
+        XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripes, f_acc512);
+        *nbStripesSoFarPtr += nbStripes;
+    }
+}
+
+#ifndef XXH3_STREAM_USE_STACK
+# ifndef __clang__ /* clang doesn't need additional stack space */
+#   define XXH3_STREAM_USE_STACK 1
+# endif
+#endif
+/*
+ * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
+ */
+XXH_FORCE_INLINE XXH_errorcode
+XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
+            const xxh_u8* XXH_RESTRICT input, size_t len,
+            XXH3_f_accumulate_512 f_acc512,
+            XXH3_f_scrambleAcc f_scramble)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    XXH_ASSERT(state != NULL);
+    {   const xxh_u8* const bEnd = input + len;
+        const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* For some reason, gcc and MSVC seem to suffer greatly
+         * when operating accumulators directly into state.
+         * Operating into stack space seems to enable proper optimization.
+         * clang, on the other hand, doesn't seem to need this trick */
+        XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; memcpy(acc, state->acc, sizeof(acc));
+#else
+        xxh_u64* XXH_RESTRICT const acc = state->acc;
+#endif
+        state->totalLen += len;
+        XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
+
+        /* small input : just fill in tmp buffer */
+        if (state->bufferedSize + len <= XXH3_INTERNALBUFFER_SIZE) {
+            XXH_memcpy(state->buffer + state->bufferedSize, input, len);
+            state->bufferedSize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
+
+        /* total input is now > XXH3_INTERNALBUFFER_SIZE */
+        #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
+        XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0);   /* clean multiple */
+
+        /*
+         * Internal buffer is partially filled (always, except at beginning)
+         * Complete it, then consume it.
+         */
+        if (state->bufferedSize) {
+            size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
+            XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
+            input += loadSize;
+            XXH3_consumeStripes(acc,
+                               &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                state->buffer, XXH3_INTERNALBUFFER_STRIPES,
+                                secret, state->secretLimit,
+                                f_acc512, f_scramble);
+            state->bufferedSize = 0;
+        }
+        XXH_ASSERT(input < bEnd);
+
+        /* large input to consume : ingest per full block */
+        if ((size_t)(bEnd - input) > state->nbStripesPerBlock * XXH_STRIPE_LEN) {
+            size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
+            XXH_ASSERT(state->nbStripesPerBlock >= state->nbStripesSoFar);
+            /* join to current block's end */
+            {   size_t const nbStripesToEnd = state->nbStripesPerBlock - state->nbStripesSoFar;
+                XXH_ASSERT(nbStripesToEnd <= nbStripes);
+                XXH3_accumulate(acc, input, secret + state->nbStripesSoFar * XXH_SECRET_CONSUME_RATE, nbStripesToEnd, f_acc512);
+                f_scramble(acc, secret + state->secretLimit);
+                state->nbStripesSoFar = 0;
+                input += nbStripesToEnd * XXH_STRIPE_LEN;
+                nbStripes -= nbStripesToEnd;
+            }
+            /* consume per entire blocks */
+            while(nbStripes >= state->nbStripesPerBlock) {
+                XXH3_accumulate(acc, input, secret, state->nbStripesPerBlock, f_acc512);
+                f_scramble(acc, secret + state->secretLimit);
+                input += state->nbStripesPerBlock * XXH_STRIPE_LEN;
+                nbStripes -= state->nbStripesPerBlock;
+            }
+            /* consume last partial block */
+            XXH3_accumulate(acc, input, secret, nbStripes, f_acc512);
+            input += nbStripes * XXH_STRIPE_LEN;
+            XXH_ASSERT(input < bEnd);  /* at least some bytes left */
+            state->nbStripesSoFar = nbStripes;
+            /* buffer predecessor of last partial stripe */
+            XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+            XXH_ASSERT(bEnd - input <= XXH_STRIPE_LEN);
+        } else {
+            /* content to consume <= block size */
+            /* Consume input by a multiple of internal buffer size */
+            if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
+                const xxh_u8* const limit = bEnd - XXH3_INTERNALBUFFER_SIZE;
+                do {
+                    XXH3_consumeStripes(acc,
+                                       &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                        input, XXH3_INTERNALBUFFER_STRIPES,
+                                        secret, state->secretLimit,
+                                        f_acc512, f_scramble);
+                    input += XXH3_INTERNALBUFFER_SIZE;
+                } while (inputbuffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+            }
+        }
+
+        /* Some remaining input (always) : buffer it */
+        XXH_ASSERT(input < bEnd);
+        XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
+        XXH_ASSERT(state->bufferedSize == 0);
+        XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
+        state->bufferedSize = (XXH32_hash_t)(bEnd-input);
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* save stack accumulators into state */
+        memcpy(state->acc, acc, sizeof(acc));
+#endif
+    }
+
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_update(XXH3_state_t* state, const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+
+XXH_FORCE_INLINE void
+XXH3_digest_long (XXH64_hash_t* acc,
+                  const XXH3_state_t* state,
+                  const unsigned char* secret)
+{
+    /*
+     * Digest on a local copy. This way, the state remains unaltered, and it can
+     * continue ingesting more input afterwards.
+     */
+    XXH_memcpy(acc, state->acc, sizeof(state->acc));
+    if (state->bufferedSize >= XXH_STRIPE_LEN) {
+        size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
+        size_t nbStripesSoFar = state->nbStripesSoFar;
+        XXH3_consumeStripes(acc,
+                           &nbStripesSoFar, state->nbStripesPerBlock,
+                            state->buffer, nbStripes,
+                            secret, state->secretLimit,
+                            XXH3_accumulate_512, XXH3_scrambleAcc);
+        /* last stripe */
+        XXH3_accumulate_512(acc,
+                            state->buffer + state->bufferedSize - XXH_STRIPE_LEN,
+                            secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+    } else {  /* bufferedSize < XXH_STRIPE_LEN */
+        xxh_u8 lastStripe[XXH_STRIPE_LEN];
+        size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
+        XXH_ASSERT(state->bufferedSize > 0);  /* there is always some input buffered */
+        XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
+        XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
+        XXH3_accumulate_512(acc,
+                            lastStripe,
+                            secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+    }
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        return XXH3_mergeAccs(acc,
+                              secret + XXH_SECRET_MERGEACCS_START,
+                              (xxh_u64)state->totalLen * XXH_PRIME64_1);
+    }
+    /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
+    if (state->useSeed)
+        return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                  secret, state->secretLimit + XXH_STRIPE_LEN);
+}
+
+
+
+/* ==========================================
+ * XXH3 128 bits (a.k.a XXH128)
+ * ==========================================
+ * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
+ * even without counting the significantly larger output size.
+ *
+ * For example, extra steps are taken to avoid the seed-dependent collisions
+ * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
+ *
+ * This strength naturally comes at the cost of some speed, especially on short
+ * lengths. Note that longer hashes are about as fast as the 64-bit version
+ * due to it using only a slight modification of the 64-bit loop.
+ *
+ * XXH128 is also more oriented towards 64-bit machines. It is still extremely
+ * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
+ */
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    /* A doubled version of 1to3_64b with different constants. */
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8 const c1 = input[0];
+        xxh_u8 const c2 = input[len >> 1];
+        xxh_u8 const c3 = input[len - 1];
+        xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
+                                | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
+        xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
+        xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
+        xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
+        xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
+        XXH128_hash_t h128;
+        h128.low64  = XXH64_avalanche(keyed_lo);
+        h128.high64 = XXH64_avalanche(keyed_hi);
+        return h128;
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input_lo = XXH_readLE32(input);
+        xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
+        xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
+        xxh_u64 const keyed = input_64 ^ bitflip;
+
+        /* Shift len to the left to ensure it is even, this avoids even multiplies. */
+        XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
+
+        m128.high64 += (m128.low64 << 1);
+        m128.low64  ^= (m128.high64 >> 3);
+
+        m128.low64   = XXH_xorshift64(m128.low64, 35);
+        m128.low64  *= 0x9FB21C651E98DF25ULL;
+        m128.low64   = XXH_xorshift64(m128.low64, 28);
+        m128.high64  = XXH3_avalanche(m128.high64);
+        return m128;
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
+        xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
+        xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64       input_hi = XXH_readLE64(input + len - 8);
+        XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
+        /*
+         * Put len in the middle of m128 to ensure that the length gets mixed to
+         * both the low and high bits in the 128x64 multiply below.
+         */
+        m128.low64 += (xxh_u64)(len - 1) << 54;
+        input_hi   ^= bitfliph;
+        /*
+         * Add the high 32 bits of input_hi to the high 32 bits of m128, then
+         * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
+         * the high 64 bits of m128.
+         *
+         * The best approach to this operation is different on 32-bit and 64-bit.
+         */
+        if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
+            /*
+             * 32-bit optimized version, which is more readable.
+             *
+             * On 32-bit, it removes an ADC and delays a dependency between the two
+             * halves of m128.high64, but it generates an extra mask on 64-bit.
+             */
+            m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
+        } else {
+            /*
+             * 64-bit optimized (albeit more confusing) version.
+             *
+             * Uses some properties of addition and multiplication to remove the mask:
+             *
+             * Let:
+             *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
+             *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
+             *    c = XXH_PRIME32_2
+             *
+             *    a + (b * c)
+             * Inverse Property: x + y - x == y
+             *    a + (b * (1 + c - 1))
+             * Distributive Property: x * (y + z) == (x * y) + (x * z)
+             *    a + (b * 1) + (b * (c - 1))
+             * Identity Property: x * 1 == x
+             *    a + b + (b * (c - 1))
+             *
+             * Substitute a, b, and c:
+             *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             *
+             * Since input_hi.hi + input_hi.lo == input_hi, we get this:
+             *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             */
+            m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
+        }
+        /* m128 ^= XXH_swap64(m128 >> 64); */
+        m128.low64  ^= XXH_swap64(m128.high64);
+
+        {   /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
+            XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
+            h128.high64 += m128.high64 * XXH_PRIME64_2;
+
+            h128.low64   = XXH3_avalanche(h128.low64);
+            h128.high64  = XXH3_avalanche(h128.high64);
+            return h128;
+    }   }
+}
+
+/*
+ * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
+        if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
+        {   XXH128_hash_t h128;
+            xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
+            xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
+            h128.low64 = XXH64_avalanche(seed ^ bitflipl);
+            h128.high64 = XXH64_avalanche( seed ^ bitfliph);
+            return h128;
+    }   }
+}
+
+/*
+ * A bit slower than XXH3_mix16B, but handles multiply by zero better.
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
+              const xxh_u8* secret, XXH64_hash_t seed)
+{
+    acc.low64  += XXH3_mix16B (input_1, secret+0, seed);
+    acc.low64  ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
+    acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
+    acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
+    return acc;
+}
+
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   XXH128_hash_t acc;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
+                }
+                acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
+            }
+            acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
+        }
+        acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
+
+XXH_NO_INLINE XXH128_hash_t
+XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                       const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                       XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    {   XXH128_hash_t acc;
+        int const nbRounds = (int)len / 32;
+        int i;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        for (i=0; i<4; i++) {
+            acc = XXH128_mix32B(acc,
+                                input  + (32 * i),
+                                input  + (32 * i) + 16,
+                                secret + (32 * i),
+                                seed);
+        }
+        acc.low64 = XXH3_avalanche(acc.low64);
+        acc.high64 = XXH3_avalanche(acc.high64);
+        XXH_ASSERT(nbRounds >= 4);
+        for (i=4 ; i < nbRounds; i++) {
+            acc = XXH128_mix32B(acc,
+                                input + (32 * i),
+                                input + (32 * i) + 16,
+                                secret + XXH3_MIDSIZE_STARTOFFSET + (32 * (i - 4)),
+                                seed);
+        }
+        /* last bytes */
+        acc = XXH128_mix32B(acc,
+                            input + len - 16,
+                            input + len - 32,
+                            secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
+                            0ULL - seed);
+
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
+                            const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate_512 f_acc512,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc512, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    {   XXH128_hash_t h128;
+        h128.low64  = XXH3_mergeAccs(acc,
+                                     secret + XXH_SECRET_MERGEACCS_START,
+                                     (xxh_u64)len * XXH_PRIME64_1);
+        h128.high64 = XXH3_mergeAccs(acc,
+                                     secret + secretSize
+                                            - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                     ~((xxh_u64)len * XXH_PRIME64_2));
+        return h128;
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
+                           XXH64_hash_t seed64,
+                           const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
+                                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * It's important for performance to pass @secretLen (when it's static)
+ * to the compiler, so that it can properly optimize the vectorized loop.
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                              XXH64_hash_t seed64,
+                              const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
+                                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
+                                XXH64_hash_t seed64,
+                                XXH3_f_accumulate_512 f_acc512,
+                                XXH3_f_scrambleAcc f_scramble,
+                                XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed64 == 0)
+        return XXH3_hashLong_128b_internal(input, len,
+                                           XXH3_kSecret, sizeof(XXH3_kSecret),
+                                           f_acc512, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed64);
+        return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
+                                           f_acc512, f_scramble);
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed(const void* input, size_t len,
+                            XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
+                XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
+
+typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
+                                            XXH64_hash_t, const void* XXH_RESTRICT, size_t);
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_128bits_internal(const void* input, size_t len,
+                      XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                      XXH3_hashLong128_f f_hl128)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secret` conditions are not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hl128(input, len, seed64, secret, secretLen);
+}
+
+
+/* ===   Public XXH128 API   === */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* input, size_t len)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_default);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 (const xxh_u8*)secret, secretSize,
+                                 XXH3_hashLong_128b_withSecret);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_internal(input, len, seed,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_withSeed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_withSeed(input, len, seed);
+}
+
+
+/* ===   XXH3 128-bit streaming   === */
+
+/*
+ * All initialization and update functions are identical to 64-bit streaming variant.
+ * The only difference is the finalization routine.
+ */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset(XXH3_state_t* statePtr)
+{
+    return XXH3_64bits_reset(statePtr);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSeed(statePtr, seed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_update(XXH3_state_t* state, const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+        {   XXH128_hash_t h128;
+            h128.low64  = XXH3_mergeAccs(acc,
+                                         secret + XXH_SECRET_MERGEACCS_START,
+                                         (xxh_u64)state->totalLen * XXH_PRIME64_1);
+            h128.high64 = XXH3_mergeAccs(acc,
+                                         secret + state->secretLimit + XXH_STRIPE_LEN
+                                                - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
+            return h128;
+        }
+    }
+    /* len <= XXH3_MIDSIZE_MAX : short code */
+    if (state->seed)
+        return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                   secret, state->secretLimit + XXH_STRIPE_LEN);
+}
+
+/* 128-bit utility functions */
+
+#include    /* memcmp, memcpy */
+
+/* return : 1 is equal, 0 if different */
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
+{
+    /* note : XXH128_hash_t is compact, it has no padding byte */
+    return !(memcmp(&h1, &h2, sizeof(h1)));
+}
+
+/* This prototype is compatible with stdlib's qsort().
+ * return : >0 if *h128_1  > *h128_2
+ *          <0 if *h128_1  < *h128_2
+ *          =0 if *h128_1 == *h128_2  */
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2)
+{
+    XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
+    XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
+    int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
+    /* note : bets that, in most cases, hash values are different */
+    if (hcmp) return hcmp;
+    return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
+}
+
+
+/*======   Canonical representation   ======*/
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) {
+        hash.high64 = XXH_swap64(hash.high64);
+        hash.low64  = XXH_swap64(hash.low64);
+    }
+    XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
+    XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128_hashFromCanonical(const XXH128_canonical_t* src)
+{
+    XXH128_hash_t h;
+    h.high64 = XXH_readBE64(src);
+    h.low64  = XXH_readBE64(src->digest + 8);
+    return h;
+}
+
+
+
+/* ==========================================
+ * Secret generators
+ * ==========================================
+ */
+#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
+
+XXH_FORCE_INLINE void XXH3_combine16(void* dst, XXH128_hash_t h128)
+{
+    XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
+    XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize)
+{
+#if (XXH_DEBUGLEVEL >= 1)
+    XXH_ASSERT(secretBuffer != NULL);
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+#else
+    /* production mode, assert() are disabled */
+    if (secretBuffer == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+#endif
+
+    if (customSeedSize == 0) {
+        customSeed = XXH3_kSecret;
+        customSeedSize = XXH_SECRET_DEFAULT_SIZE;
+    }
+#if (XXH_DEBUGLEVEL >= 1)
+    XXH_ASSERT(customSeed != NULL);
+#else
+    if (customSeed == NULL) return XXH_ERROR;
+#endif
+
+    /* Fill secretBuffer with a copy of customSeed - repeat as needed */
+    {   size_t pos = 0;
+        while (pos < secretSize) {
+            size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
+            memcpy((char*)secretBuffer + pos, customSeed, toCopy);
+            pos += toCopy;
+    }   }
+
+    {   size_t const nbSeg16 = secretSize / 16;
+        size_t n;
+        XXH128_canonical_t scrambler;
+        XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
+        for (n=0; n
-#endif
 #include "compiler.h"
+#include "cpu.h"
 #include "mem.h"
 #include "debug.h"                 /* assert, DEBUGLOG, RAWLOG, g_debuglevel */
 #include "error_private.h"
@@ -30,7 +28,6 @@
 #include "../zstd.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
 #ifndef XXH_STATIC_LINKING_ONLY
 #  define XXH_STATIC_LINKING_ONLY  /* XXH64_state_t */
@@ -60,81 +57,7 @@ extern "C" {
 #undef MAX
 #define MIN(a,b) ((a)<(b) ? (a) : (b))
 #define MAX(a,b) ((a)>(b) ? (a) : (b))
-
-/**
- * Ignore: this is an internal helper.
- *
- * This is a helper function to help force C99-correctness during compilation.
- * Under strict compilation modes, variadic macro arguments can't be empty.
- * However, variadic function arguments can be. Using a function therefore lets
- * us statically check that at least one (string) argument was passed,
- * independent of the compilation flags.
- */
-static INLINE_KEYWORD UNUSED_ATTR
-void _force_has_format_string(const char *format, ...) {
-  (void)format;
-}
-
-/**
- * Ignore: this is an internal helper.
- *
- * We want to force this function invocation to be syntactically correct, but
- * we don't want to force runtime evaluation of its arguments.
- */
-#define _FORCE_HAS_FORMAT_STRING(...) \
-  if (0) { \
-    _force_has_format_string(__VA_ARGS__); \
-  }
-
-/**
- * Return the specified error if the condition evaluates to true.
- *
- * In debug modes, prints additional information.
- * In order to do that (particularly, printing the conditional that failed),
- * this can't just wrap RETURN_ERROR().
- */
-#define RETURN_ERROR_IF(cond, err, ...) \
-  if (cond) { \
-    RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
-           __FILE__, __LINE__, ZSTD_QUOTE(cond), ZSTD_QUOTE(ERROR(err))); \
-    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
-    RAWLOG(3, ": " __VA_ARGS__); \
-    RAWLOG(3, "\n"); \
-    return ERROR(err); \
-  }
-
-/**
- * Unconditionally return the specified error.
- *
- * In debug modes, prints additional information.
- */
-#define RETURN_ERROR(err, ...) \
-  do { \
-    RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
-           __FILE__, __LINE__, ZSTD_QUOTE(ERROR(err))); \
-    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
-    RAWLOG(3, ": " __VA_ARGS__); \
-    RAWLOG(3, "\n"); \
-    return ERROR(err); \
-  } while(0);
-
-/**
- * If the provided expression evaluates to an error code, returns that error code.
- *
- * In debug modes, prints additional information.
- */
-#define FORWARD_IF_ERROR(err, ...) \
-  do { \
-    size_t const err_code = (err); \
-    if (ERR_isError(err_code)) { \
-      RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
-             __FILE__, __LINE__, ZSTD_QUOTE(err), ERR_getErrorName(err_code)); \
-      _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
-      RAWLOG(3, ": " __VA_ARGS__); \
-      RAWLOG(3, "\n"); \
-      return err_code; \
-    } \
-  } while(0);
+#define BOUNDED(min,val,max) (MAX(min,MIN(val,max)))
 
 
 /*-*************************************
@@ -143,7 +66,6 @@ void _force_has_format_string(const char *format, ...) {
 #define ZSTD_OPT_NUM    (1<<12)
 
 #define ZSTD_REP_NUM      3                 /* number of repcodes */
-#define ZSTD_REP_MOVE     (ZSTD_REP_NUM-1)
 static UNUSED_ATTR const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 
 #define KB *(1 <<10)
@@ -170,9 +92,9 @@ typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
 #define ZSTD_FRAMECHECKSUMSIZE 4
 
 #define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
-#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */)   /* for a non-null block */
+#define MIN_LITERALS_FOR_4_STREAMS 6
 
-#define HufLog 12
 typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingType_e;
 
 #define LONGNBSEQ 0x7F00
@@ -180,6 +102,7 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
 #define MINMATCH 3
 
 #define Litbits  8
+#define LitHufLog 11
 #define MaxLit ((1<= 8 || (ovtype == ZSTD_no_overlap && diff <= -WILDCOPY_VECLEN));
-
     if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
         /* Handle short offset copies. */
         do {
@@ -303,12 +237,6 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
          * one COPY16() in the first call. Then, do two calls per loop since
          * at that point it is more likely to have a high trip count.
          */
-#ifdef __aarch64__
-        do {
-            COPY16(op, ip);
-        }
-        while (op < oend);
-#else
         ZSTD_copy16(op, ip);
         if (16 >= length) return;
         op += 16;
@@ -318,7 +246,6 @@ void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e
             COPY16(op, ip);
         }
         while (op < oend);
-#endif
     }
 }
 
@@ -352,9 +279,9 @@ typedef enum {
 *  Private declarations
 *********************************************/
 typedef struct seqDef_s {
-    U32 offset;         /* offset == rawOffset + ZSTD_REP_NUM, or equivalently, offCode + 1 */
+    U32 offBase;   /* offBase == Offset + ZSTD_REP_NUM, or repcode 1,2,3 */
     U16 litLength;
-    U16 matchLength;
+    U16 mlBase;    /* mlBase == matchLength - MINMATCH */
 } seqDef;
 
 /* Controls whether seqStore has a single "long" litLength or matchLength. See seqStore_t. */
@@ -367,11 +294,11 @@ typedef enum {
 typedef struct {
     seqDef* sequencesStart;
     seqDef* sequences;      /* ptr to end of sequences */
-    BYTE* litStart;
-    BYTE* lit;              /* ptr to end of literals */
-    BYTE* llCode;
-    BYTE* mlCode;
-    BYTE* ofCode;
+    BYTE*  litStart;
+    BYTE*  lit;             /* ptr to end of literals */
+    BYTE*  llCode;
+    BYTE*  mlCode;
+    BYTE*  ofCode;
     size_t maxNbSeq;
     size_t maxNbLit;
 
@@ -379,8 +306,8 @@ typedef struct {
      * in the seqStore that has a value larger than U16 (if it exists). To do so, we increment
      * the existing value of the litLength or matchLength by 0x10000.
      */
-    ZSTD_longLengthType_e   longLengthType;
-    U32                     longLengthPos;  /* Index of the sequence to apply long length modification to */
+    ZSTD_longLengthType_e longLengthType;
+    U32                   longLengthPos;  /* Index of the sequence to apply long length modification to */
 } seqStore_t;
 
 typedef struct {
@@ -396,13 +323,13 @@ MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore
 {
     ZSTD_sequenceLength seqLen;
     seqLen.litLength = seq->litLength;
-    seqLen.matchLength = seq->matchLength + MINMATCH;
+    seqLen.matchLength = seq->mlBase + MINMATCH;
     if (seqStore->longLengthPos == (U32)(seq - seqStore->sequencesStart)) {
         if (seqStore->longLengthType == ZSTD_llt_literalLength) {
-            seqLen.litLength += 0xFFFF;
+            seqLen.litLength += 0x10000;
         }
         if (seqStore->longLengthType == ZSTD_llt_matchLength) {
-            seqLen.matchLength += 0xFFFF;
+            seqLen.matchLength += 0x10000;
         }
     }
     return seqLen;
@@ -415,12 +342,13 @@ MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore
  *          `decompressedBound != ZSTD_CONTENTSIZE_ERROR`
  */
 typedef struct {
+    size_t nbBlocks;
     size_t compressedSize;
     unsigned long long decompressedBound;
 } ZSTD_frameSizeInfo;   /* decompress & legacy */
 
 const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBuilder */
-void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
+int ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
 
 /* custom memory allocation functions */
 void* ZSTD_customMalloc(size_t size, ZSTD_customMem customMem);
@@ -428,35 +356,6 @@ void* ZSTD_customCalloc(size_t size, ZSTD_customMem customMem);
 void ZSTD_customFree(void* ptr, ZSTD_customMem customMem);
 
 
-MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
-{
-    assert(val != 0);
-    {
-#   if defined(_MSC_VER)   /* Visual */
-#       if STATIC_BMI2 == 1
-            return _lzcnt_u32(val)^31;
-#       else
-            unsigned long r=0;
-            return _BitScanReverse(&r, val) ? (unsigned)r : 0;
-#       endif
-#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
-        return __builtin_clz (val) ^ 31;
-#   elif defined(__ICCARM__)    /* IAR Intrinsic */
-        return 31 - __CLZ(val);
-#   else   /* Software version */
-        static const U32 DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
-        U32 v = val;
-        v |= v >> 1;
-        v |= v >> 2;
-        v |= v >> 4;
-        v |= v >> 8;
-        v |= v >> 16;
-        return DeBruijnClz[(v * 0x07C4ACDDU) >> 27];
-#   endif
-    }
-}
-
-
 /* ZSTD_invalidateRepCodes() :
  * ensures next compression will not use repcodes from previous block.
  * Note : only works with regular variant;
@@ -482,6 +381,14 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
 size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                        const void* src, size_t srcSize);
 
+/**
+ * @returns true iff the CPU supports dynamic BMI2 dispatch.
+ */
+MEM_STATIC int ZSTD_cpuSupportsBmi2(void)
+{
+    ZSTD_cpuid_t cpuid = ZSTD_cpuid();
+    return ZSTD_cpuid_bmi1(cpuid) && ZSTD_cpuid_bmi2(cpuid);
+}
 
 #if defined (__cplusplus)
 }
diff --git a/priv/zstd/lib/common/zstd_trace.h b/priv/zstd/lib/common/zstd_trace.h
index 2da5640..da20534 100644
--- a/priv/zstd/lib/common/zstd_trace.h
+++ b/priv/zstd/lib/common/zstd_trace.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -17,10 +17,19 @@ extern "C" {
 
 #include 
 
-/* weak symbol support */
-#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && defined(__GNUC__) && \
+/* weak symbol support
+ * For now, enable conservatively:
+ * - Only GNUC
+ * - Only ELF
+ * - Only x86-64, i386 and aarch64
+ * Also, explicitly disable on platforms known not to work so they aren't
+ * forgotten in the future.
+ */
+#if !defined(ZSTD_HAVE_WEAK_SYMBOLS) && \
+    defined(__GNUC__) && defined(__ELF__) && \
+    (defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86) || defined(__aarch64__)) && \
     !defined(__APPLE__) && !defined(_WIN32) && !defined(__MINGW32__) && \
-    !defined(__CYGWIN__)
+    !defined(__CYGWIN__) && !defined(_AIX)
 #  define ZSTD_HAVE_WEAK_SYMBOLS 1
 #else
 #  define ZSTD_HAVE_WEAK_SYMBOLS 0
diff --git a/priv/zstd/lib/compress/clevels.h b/priv/zstd/lib/compress/clevels.h
new file mode 100644
index 0000000..c18da46
--- /dev/null
+++ b/priv/zstd/lib/compress/clevels.h
@@ -0,0 +1,134 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_CLEVELS_H
+#define ZSTD_CLEVELS_H
+
+#define ZSTD_STATIC_LINKING_ONLY  /* ZSTD_compressionParameters  */
+#include "../zstd.h"
+
+/*-=====  Pre-defined compression levels  =====-*/
+
+#define ZSTD_MAX_CLEVEL     22
+
+#ifdef __GNUC__
+__attribute__((__unused__))
+#endif
+
+static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
+{   /* "default" - for any srcSize > 256 KB */
+    /* W,  C,  H,  S,  L, TL, strat */
+    { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
+    { 19, 13, 14,  1,  7,  0, ZSTD_fast    },  /* level  1 */
+    { 20, 15, 16,  1,  6,  0, ZSTD_fast    },  /* level  2 */
+    { 21, 16, 17,  1,  5,  0, ZSTD_dfast   },  /* level  3 */
+    { 21, 18, 18,  1,  5,  0, ZSTD_dfast   },  /* level  4 */
+    { 21, 18, 19,  3,  5,  2, ZSTD_greedy  },  /* level  5 */
+    { 21, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6 */
+    { 21, 19, 20,  4,  5,  8, ZSTD_lazy    },  /* level  7 */
+    { 21, 19, 20,  4,  5, 16, ZSTD_lazy2   },  /* level  8 */
+    { 22, 20, 21,  4,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 10 */
+    { 22, 21, 22,  6,  5, 16, ZSTD_lazy2   },  /* level 11 */
+    { 22, 22, 23,  6,  5, 32, ZSTD_lazy2   },  /* level 12 */
+    { 22, 22, 22,  4,  5, 32, ZSTD_btlazy2 },  /* level 13 */
+    { 22, 22, 23,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
+    { 22, 23, 23,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
+    { 22, 22, 22,  5,  5, 48, ZSTD_btopt   },  /* level 16 */
+    { 23, 23, 22,  5,  4, 64, ZSTD_btopt   },  /* level 17 */
+    { 23, 23, 22,  6,  3, 64, ZSTD_btultra },  /* level 18 */
+    { 23, 24, 22,  7,  3,256, ZSTD_btultra2},  /* level 19 */
+    { 25, 25, 23,  7,  3,256, ZSTD_btultra2},  /* level 20 */
+    { 26, 26, 24,  7,  3,512, ZSTD_btultra2},  /* level 21 */
+    { 27, 27, 25,  9,  3,999, ZSTD_btultra2},  /* level 22 */
+},
+{   /* for srcSize <= 256 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 18, 13, 14,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 18, 14, 14,  1,  5,  0, ZSTD_dfast   },  /* level  2 */
+    { 18, 16, 16,  1,  4,  0, ZSTD_dfast   },  /* level  3 */
+    { 18, 16, 17,  3,  5,  2, ZSTD_greedy  },  /* level  4.*/
+    { 18, 17, 18,  5,  5,  2, ZSTD_greedy  },  /* level  5.*/
+    { 18, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6.*/
+    { 18, 18, 19,  4,  4,  4, ZSTD_lazy    },  /* level  7 */
+    { 18, 18, 19,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 18, 18, 19,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 18, 18, 19,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 18, 18, 19,  5,  4, 12, ZSTD_btlazy2 },  /* level 11.*/
+    { 18, 19, 19,  7,  4, 12, ZSTD_btlazy2 },  /* level 12.*/
+    { 18, 18, 19,  4,  4, 16, ZSTD_btopt   },  /* level 13 */
+    { 18, 18, 19,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
+    { 18, 18, 19,  6,  3,128, ZSTD_btopt   },  /* level 15.*/
+    { 18, 19, 19,  6,  3,128, ZSTD_btultra },  /* level 16.*/
+    { 18, 19, 19,  8,  3,256, ZSTD_btultra },  /* level 17.*/
+    { 18, 19, 19,  6,  3,128, ZSTD_btultra2},  /* level 18.*/
+    { 18, 19, 19,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 18, 19, 19, 10,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 18, 19, 19, 12,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 18, 19, 19, 13,  3,999, ZSTD_btultra2},  /* level 22.*/
+},
+{   /* for srcSize <= 128 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 17, 12, 13,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 15,  1,  5,  0, ZSTD_fast    },  /* level  2 */
+    { 17, 15, 16,  2,  5,  0, ZSTD_dfast   },  /* level  3 */
+    { 17, 17, 17,  2,  4,  0, ZSTD_dfast   },  /* level  4 */
+    { 17, 16, 17,  3,  4,  2, ZSTD_greedy  },  /* level  5 */
+    { 17, 16, 17,  3,  4,  4, ZSTD_lazy    },  /* level  6 */
+    { 17, 16, 17,  3,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 17, 16, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 17, 16, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 17, 16, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 17, 17, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 11 */
+    { 17, 18, 17,  7,  4, 12, ZSTD_btlazy2 },  /* level 12 */
+    { 17, 18, 17,  3,  4, 12, ZSTD_btopt   },  /* level 13.*/
+    { 17, 18, 17,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
+    { 17, 18, 17,  6,  3,256, ZSTD_btopt   },  /* level 15.*/
+    { 17, 18, 17,  6,  3,128, ZSTD_btultra },  /* level 16.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btultra },  /* level 17.*/
+    { 17, 18, 17, 10,  3,512, ZSTD_btultra },  /* level 18.*/
+    { 17, 18, 17,  5,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 17, 18, 17,  7,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 17, 18, 17,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 17, 18, 17, 11,  3,999, ZSTD_btultra2},  /* level 22.*/
+},
+{   /* for srcSize <= 16 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 14, 14, 15,  1,  5,  0, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 15,  1,  4,  0, ZSTD_fast    },  /* level  2 */
+    { 14, 14, 15,  2,  4,  0, ZSTD_dfast   },  /* level  3 */
+    { 14, 14, 14,  4,  4,  2, ZSTD_greedy  },  /* level  4 */
+    { 14, 14, 14,  3,  4,  4, ZSTD_lazy    },  /* level  5.*/
+    { 14, 14, 14,  4,  4,  8, ZSTD_lazy2   },  /* level  6 */
+    { 14, 14, 14,  6,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 14, 14, 14,  8,  4,  8, ZSTD_lazy2   },  /* level  8.*/
+    { 14, 15, 14,  5,  4,  8, ZSTD_btlazy2 },  /* level  9.*/
+    { 14, 15, 14,  9,  4,  8, ZSTD_btlazy2 },  /* level 10.*/
+    { 14, 15, 14,  3,  4, 12, ZSTD_btopt   },  /* level 11.*/
+    { 14, 15, 14,  4,  3, 24, ZSTD_btopt   },  /* level 12.*/
+    { 14, 15, 14,  5,  3, 32, ZSTD_btultra },  /* level 13.*/
+    { 14, 15, 15,  6,  3, 64, ZSTD_btultra },  /* level 14.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btultra },  /* level 15.*/
+    { 14, 15, 15,  5,  3, 48, ZSTD_btultra2},  /* level 16.*/
+    { 14, 15, 15,  6,  3,128, ZSTD_btultra2},  /* level 17.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btultra2},  /* level 18.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 14, 15, 15,  8,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 14, 15, 15,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 14, 15, 15, 10,  3,999, ZSTD_btultra2},  /* level 22.*/
+},
+};
+
+
+
+#endif  /* ZSTD_CLEVELS_H */
diff --git a/priv/zstd/lib/compress/fse_compress.c b/priv/zstd/lib/compress/fse_compress.c
index b4297ec..5d37708 100644
--- a/priv/zstd/lib/compress/fse_compress.c
+++ b/priv/zstd/lib/compress/fse_compress.c
@@ -1,6 +1,6 @@
 /* ******************************************************************
  * FSE : Finite State Entropy encoder
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -26,6 +26,7 @@
 #define ZSTD_DEPS_NEED_MALLOC
 #define ZSTD_DEPS_NEED_MATH64
 #include "../common/zstd_deps.h"  /* ZSTD_malloc, ZSTD_free, ZSTD_memcpy, ZSTD_memset */
+#include "../common/bits.h" /* ZSTD_highbit32 */
 
 
 /* **************************************************************
@@ -75,13 +76,14 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
     void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
     FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
     U32 const step = FSE_TABLESTEP(tableSize);
+    U32 const maxSV1 = maxSymbolValue+1;
 
-    U32* cumul = (U32*)workSpace;
-    FSE_FUNCTION_TYPE* tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSymbolValue + 2));
+    U16* cumul = (U16*)workSpace;   /* size = maxSV1 */
+    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)(cumul + (maxSV1+1));  /* size = tableSize */
 
     U32 highThreshold = tableSize-1;
 
-    if ((size_t)workSpace & 3) return ERROR(GENERIC); /* Must be 4 byte aligned */
+    assert(((size_t)workSpace & 1) == 0);  /* Must be 2 bytes-aligned */
     if (FSE_BUILD_CTABLE_WORKSPACE_SIZE(maxSymbolValue, tableLog) > wkspSize) return ERROR(tableLog_tooLarge);
     /* CTable header */
     tableU16[-2] = (U16) tableLog;
@@ -89,7 +91,7 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
     assert(tableLog < 16);   /* required for threshold strategy to work */
 
     /* For explanations on how to distribute symbol values over the table :
-     * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+     * https://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
 
      #ifdef __clang_analyzer__
      ZSTD_memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
@@ -98,20 +100,61 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
     /* symbol start positions */
     {   U32 u;
         cumul[0] = 0;
-        for (u=1; u <= maxSymbolValue+1; u++) {
+        for (u=1; u <= maxSV1; u++) {
             if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
                 cumul[u] = cumul[u-1] + 1;
                 tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
             } else {
-                cumul[u] = cumul[u-1] + normalizedCounter[u-1];
+                assert(normalizedCounter[u-1] >= 0);
+                cumul[u] = cumul[u-1] + (U16)normalizedCounter[u-1];
+                assert(cumul[u] >= cumul[u-1]);  /* no overflow */
         }   }
-        cumul[maxSymbolValue+1] = tableSize+1;
+        cumul[maxSV1] = (U16)(tableSize+1);
     }
 
     /* Spread symbols */
-    {   U32 position = 0;
+    if (highThreshold == tableSize - 1) {
+        /* Case for no low prob count symbols. Lay down 8 bytes at a time
+         * to reduce branch misses since we are operating on a small block
+         */
+        BYTE* const spread = tableSymbol + tableSize; /* size = tableSize + 8 (may write beyond tableSize) */
+        {   U64 const add = 0x0101010101010101ull;
+            size_t pos = 0;
+            U64 sv = 0;
+            U32 s;
+            for (s=0; s=0);
+                pos += (size_t)n;
+            }
+        }
+        /* Spread symbols across the table. Lack of lowprob symbols means that
+         * we don't need variable sized inner loop, so we can unroll the loop and
+         * reduce branch misses.
+         */
+        {   size_t position = 0;
+            size_t s;
+            size_t const unroll = 2; /* Experimentally determined optimal unroll */
+            assert(tableSize % unroll == 0); /* FSE_MIN_TABLELOG is 5 */
+            for (s = 0; s < (size_t)tableSize; s += unroll) {
+                size_t u;
+                for (u = 0; u < unroll; ++u) {
+                    size_t const uPosition = (position + (u * step)) & tableMask;
+                    tableSymbol[uPosition] = spread[s + u];
+                }
+                position = (position + (unroll * step)) & tableMask;
+            }
+            assert(position == 0);   /* Must have initialized all positions */
+        }
+    } else {
+        U32 position = 0;
         U32 symbol;
-        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+        for (symbol=0; symbol highThreshold)
                     position = (position + step) & tableMask;   /* Low proba area */
         }   }
-
         assert(position==0);  /* Must have initialized all positions */
     }
 
@@ -144,16 +186,17 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
             case -1:
             case  1:
                 symbolTT[s].deltaNbBits = (tableLog << 16) - (1< 1);
+                {   U32 const maxBitsOut = tableLog - ZSTD_highbit32 ((U32)normalizedCounter[s]-1);
+                    U32 const minStatePlus = (U32)normalizedCounter[s] << maxBitsOut;
                     symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
-                    symbolTT[s].deltaFindState = total - normalizedCounter[s];
-                    total +=  normalizedCounter[s];
+                    symbolTT[s].deltaFindState = (int)(total - (unsigned)normalizedCounter[s]);
+                    total +=  (unsigned)normalizedCounter[s];
     }   }   }   }
 
 #if 0  /* debug : symbol costs */
@@ -164,32 +207,26 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct,
                 symbol, normalizedCounter[symbol],
                 FSE_getMaxNbBits(symbolTT, symbol),
                 (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
-        }
-    }
+    }   }
 #endif
 
     return 0;
 }
 
-#ifndef ZSTD_NO_UNUSED_FUNCTIONS
-size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
-{
-    FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE];   /* memset() is not necessary, even if static analyzer complain about it */
-    return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
-}
-#endif
-
 
 
 #ifndef FSE_COMMONDEFS_ONLY
 
-
 /*-**************************************************************
 *  FSE NCount encoding
 ****************************************************************/
 size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
 {
-    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
+    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog
+                                   + 4 /* bitCount initialized at 4 */
+                                   + 2 /* first two symbols may use one additional bit each */) / 8)
+                                    + 1 /* round up to whole nb bytes */
+                                    + 2 /* additional two bytes for bitstream flush */;
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
@@ -306,21 +343,11 @@ size_t FSE_writeNCount (void* buffer, size_t bufferSize,
 *  FSE Compression Code
 ****************************************************************/
 
-FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
-{
-    size_t size;
-    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
-    size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
-    return (FSE_CTable*)ZSTD_malloc(size);
-}
-
-void FSE_freeCTable (FSE_CTable* ct) { ZSTD_free(ct); }
-
 /* provides the minimum logSize to safely represent a distribution */
 static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 {
-    U32 minBitsSrc = BIT_highbit32((U32)(srcSize)) + 1;
-    U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
+    U32 minBitsSrc = ZSTD_highbit32((U32)(srcSize)) + 1;
+    U32 minBitsSymbols = ZSTD_highbit32(maxSymbolValue) + 2;
     U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
     assert(srcSize > 1); /* Not supported, RLE should be used instead */
     return minBits;
@@ -328,7 +355,7 @@ static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
 {
-    U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
+    U32 maxBitsSrc = ZSTD_highbit32((U32)(srcSize - 1)) - minus;
     U32 tableLog = maxTableLog;
     U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
     assert(srcSize > 1); /* Not supported, RLE should be used instead */
@@ -496,40 +523,6 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
     return tableLog;
 }
 
-
-/* fake FSE_CTable, for raw (uncompressed) input */
-size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
-{
-    const unsigned tableSize = 1 << nbBits;
-    const unsigned tableMask = tableSize - 1;
-    const unsigned maxSymbolValue = tableMask;
-    void* const ptr = ct;
-    U16* const tableU16 = ( (U16*) ptr) + 2;
-    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableSize>>1);   /* assumption : tableLog >= 1 */
-    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
-    unsigned s;
-
-    /* Sanity checks */
-    if (nbBits < 1) return ERROR(GENERIC);             /* min size */
-
-    /* header */
-    tableU16[-2] = (U16) nbBits;
-    tableU16[-1] = (U16) maxSymbolValue;
-
-    /* Build table */
-    for (s=0; s not compressible */
-        if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
-    }
-
-    tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
-    CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue, /* useLowProbCount */ srcSize >= 2048) );
-
-    /* Write table description header */
-    {   CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
-        op += nc_err;
-    }
-
-    /* Compress */
-    CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, scratchBufferSize) );
-    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, src, srcSize, CTable) );
-        if (cSize == 0) return 0;   /* not enough space for compressed data */
-        op += cSize;
-    }
-
-    /* check compressibility */
-    if ( (size_t)(op-ostart) >= srcSize-1 ) return 0;
-
-    return op-ostart;
-}
-
-typedef struct {
-    FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
-    union {
-      U32 hist_wksp[HIST_WKSP_SIZE_U32];
-      BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
-    } workspace;
-} fseWkspMax_t;
-
-size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
-{
-    fseWkspMax_t scratchBuffer;
-    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_COMPRESS_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
-    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
-    return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
-}
-
-size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
-{
-    return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
-}
-#endif
-
 #endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/priv/zstd/lib/compress/hist.c b/priv/zstd/lib/compress/hist.c
index 073c57e..e2fb431 100644
--- a/priv/zstd/lib/compress/hist.c
+++ b/priv/zstd/lib/compress/hist.c
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * hist : Histogram functions
  * part of Finite State Entropy project
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
diff --git a/priv/zstd/lib/compress/hist.h b/priv/zstd/lib/compress/hist.h
index 228ed48..887896b 100644
--- a/priv/zstd/lib/compress/hist.h
+++ b/priv/zstd/lib/compress/hist.h
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * hist : Histogram functions
  * part of Finite State Entropy project
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
diff --git a/priv/zstd/lib/compress/huf_compress.c b/priv/zstd/lib/compress/huf_compress.c
index 485906e..2987187 100644
--- a/priv/zstd/lib/compress/huf_compress.c
+++ b/priv/zstd/lib/compress/huf_compress.c
@@ -1,6 +1,6 @@
 /* ******************************************************************
  * Huffman encoder, part of New Generation Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -29,9 +29,9 @@
 #include "hist.h"
 #define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
 #include "../common/fse.h"        /* header compression */
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/error_private.h"
+#include "../common/bits.h"       /* ZSTD_highbit32 */
 
 
 /* **************************************************************
@@ -42,17 +42,93 @@
 
 
 /* **************************************************************
-*  Utils
+*  Required declarations
 ****************************************************************/
-unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+
+#if DEBUGLEVEL >= 2
+
+static size_t showU32(const U32* arr, size_t size)
 {
-    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    size_t u;
+    for (u=0; u= add) {
+        assert(add < align);
+        assert(((size_t)aligned & mask) == 0);
+        *workspaceSizePtr -= add;
+        return aligned;
+    } else {
+        *workspaceSizePtr = 0;
+        return NULL;
+    }
+}
+
+
 /* HUF_compressWeights() :
  * Same as FSE_compress(), but dedicated to huff0's weights compression.
  * The use case needs much less stack memory.
@@ -67,7 +143,10 @@ typedef struct {
     S16 norm[HUF_TABLELOG_MAX+1];
 } HUF_CompressWeightsWksp;
 
-static size_t HUF_compressWeights(void* dst, size_t dstSize, const void* weightTable, size_t wtSize, void* workspace, size_t workspaceSize)
+static size_t
+HUF_compressWeights(void* dst, size_t dstSize,
+              const void* weightTable, size_t wtSize,
+                    void* workspace, size_t workspaceSize)
 {
     BYTE* const ostart = (BYTE*) dst;
     BYTE* op = ostart;
@@ -75,7 +154,7 @@ static size_t HUF_compressWeights(void* dst, size_t dstSize, const void* weightT
 
     unsigned maxSymbolValue = HUF_TABLELOG_MAX;
     U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
-    HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)workspace;
+    HUF_CompressWeightsWksp* wksp = (HUF_CompressWeightsWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
 
     if (workspaceSize < sizeof(HUF_CompressWeightsWksp)) return ERROR(GENERIC);
 
@@ -106,6 +185,40 @@ static size_t HUF_compressWeights(void* dst, size_t dstSize, const void* weightT
     return (size_t)(op-ostart);
 }
 
+static size_t HUF_getNbBits(HUF_CElt elt)
+{
+    return elt & 0xFF;
+}
+
+static size_t HUF_getNbBitsFast(HUF_CElt elt)
+{
+    return elt;
+}
+
+static size_t HUF_getValue(HUF_CElt elt)
+{
+    return elt & ~(size_t)0xFF;
+}
+
+static size_t HUF_getValueFast(HUF_CElt elt)
+{
+    return elt;
+}
+
+static void HUF_setNbBits(HUF_CElt* elt, size_t nbBits)
+{
+    assert(nbBits <= HUF_TABLELOG_ABSOLUTEMAX);
+    *elt = nbBits;
+}
+
+static void HUF_setValue(HUF_CElt* elt, size_t value)
+{
+    size_t const nbBits = HUF_getNbBits(*elt);
+    if (nbBits > 0) {
+        assert((value >> nbBits) == 0);
+        *elt |= value << (sizeof(HUF_CElt) * 8 - nbBits);
+    }
+}
 
 typedef struct {
     HUF_CompressWeightsWksp wksp;
@@ -117,9 +230,12 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
                             const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog,
                             void* workspace, size_t workspaceSize)
 {
+    HUF_CElt const* const ct = CTable + 1;
     BYTE* op = (BYTE*)dst;
     U32 n;
-    HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)workspace;
+    HUF_WriteCTableWksp* wksp = (HUF_WriteCTableWksp*)HUF_alignUpWorkspace(workspace, &workspaceSize, ZSTD_ALIGNOF(U32));
+
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE >= sizeof(HUF_WriteCTableWksp));
 
     /* check conditions */
     if (workspaceSize < sizeof(HUF_WriteCTableWksp)) return ERROR(GENERIC);
@@ -130,9 +246,10 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
     for (n=1; nbitsToWeight[n] = (BYTE)(huffLog + 1 - n);
     for (n=0; nhuffWeight[n] = wksp->bitsToWeight[CTable[n].nbBits];
+        wksp->huffWeight[n] = wksp->bitsToWeight[HUF_getNbBits(ct[n])];
 
     /* attempt weights compression by FSE */
+    if (maxDstSize < 1) return ERROR(dstSize_tooSmall);
     {   CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, wksp->huffWeight, maxSymbolValue, &wksp->wksp, sizeof(wksp->wksp)) );
         if ((hSize>1) & (hSize < maxSymbolValue/2)) {   /* FSE compressed */
             op[0] = (BYTE)hSize;
@@ -149,16 +266,6 @@ size_t HUF_writeCTable_wksp(void* dst, size_t maxDstSize,
     return ((maxSymbolValue+1)/2) + 1;
 }
 
-/*! HUF_writeCTable() :
-    `CTable` : Huffman tree to save, using huf representation.
-    @return : size of saved CTable */
-size_t HUF_writeCTable (void* dst, size_t maxDstSize,
-                        const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog)
-{
-    HUF_WriteCTableWksp wksp;
-    return HUF_writeCTable_wksp(dst, maxDstSize, CTable, maxSymbolValue, huffLog, &wksp, sizeof(wksp));
-}
-
 
 size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
 {
@@ -166,6 +273,7 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
     U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
     U32 tableLog = 0;
     U32 nbSymbols = 0;
+    HUF_CElt* const ct = CTable + 1;
 
     /* get symbol weights */
     CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
@@ -175,6 +283,8 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
     if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
 
+    CTable[0] = tableLog;
+
     /* Prepare base value per rank */
     {   U32 n, nextRankStart = 0;
         for (n=1; n<=tableLog; n++) {
@@ -186,13 +296,13 @@ size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void
     /* fill nbBits */
     {   U32 n; for (n=0; nn=tableLog+1 */
         U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
-        { U32 n; for (n=0; n>= 1;
         }   }
         /* assign value within rank, symbol order */
-        { U32 n; for (n=0; n maxNbBits to be maxNbBits. Then it adjusts
- * the tree to so that it is a valid canonical Huffman tree.
+ * It attempts to convert all nodes with nbBits > @targetNbBits
+ * to employ @targetNbBits instead. Then it adjusts the tree
+ * so that it remains a valid canonical Huffman tree.
  *
  * @pre               The sum of the ranks of each symbol == 2^largestBits,
  *                    where largestBits == huffNode[lastNonNull].nbBits.
  * @post              The sum of the ranks of each symbol == 2^largestBits,
- *                    where largestBits is the return value <= maxNbBits.
+ *                    where largestBits is the return value (expected <= targetNbBits).
  *
- * @param huffNode    The Huffman tree modified in place to enforce maxNbBits.
+ * @param huffNode    The Huffman tree modified in place to enforce targetNbBits.
+ *                    It's presumed sorted, from most frequent to rarest symbol.
  * @param lastNonNull The symbol with the lowest count in the Huffman tree.
- * @param maxNbBits   The maximum allowed number of bits, which the Huffman tree
+ * @param targetNbBits  The allowed number of bits, which the Huffman tree
  *                    may not respect. After this function the Huffman tree will
- *                    respect maxNbBits.
- * @return            The maximum number of bits of the Huffman tree after adjustment,
- *                    necessarily no more than maxNbBits.
+ *                    respect targetNbBits.
+ * @return            The maximum number of bits of the Huffman tree after adjustment.
  */
-static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
+static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 targetNbBits)
 {
     const U32 largestBits = huffNode[lastNonNull].nbBits;
-    /* early exit : no elt > maxNbBits, so the tree is already valid. */
-    if (largestBits <= maxNbBits) return largestBits;
+    /* early exit : no elt > targetNbBits, so the tree is already valid. */
+    if (largestBits <= targetNbBits) return largestBits;
+
+    DEBUGLOG(5, "HUF_setMaxHeight (targetNbBits = %u)", targetNbBits);
 
     /* there are several too large elements (at least >= 2) */
     {   int totalCost = 0;
-        const U32 baseCost = 1 << (largestBits - maxNbBits);
+        const U32 baseCost = 1 << (largestBits - targetNbBits);
         int n = (int)lastNonNull;
 
-        /* Adjust any ranks > maxNbBits to maxNbBits.
+        /* Adjust any ranks > targetNbBits to targetNbBits.
          * Compute totalCost, which is how far the sum of the ranks is
          * we are over 2^largestBits after adjust the offending ranks.
          */
-        while (huffNode[n].nbBits > maxNbBits) {
+        while (huffNode[n].nbBits > targetNbBits) {
             totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
-            huffNode[n].nbBits = (BYTE)maxNbBits;
+            huffNode[n].nbBits = (BYTE)targetNbBits;
             n--;
         }
-        /* n stops at huffNode[n].nbBits <= maxNbBits */
-        assert(huffNode[n].nbBits <= maxNbBits);
-        /* n end at index of smallest symbol using < maxNbBits */
-        while (huffNode[n].nbBits == maxNbBits) --n;
+        /* n stops at huffNode[n].nbBits <= targetNbBits */
+        assert(huffNode[n].nbBits <= targetNbBits);
+        /* n end at index of smallest symbol using < targetNbBits */
+        while (huffNode[n].nbBits == targetNbBits) --n;
 
-        /* renorm totalCost from 2^largestBits to 2^maxNbBits
+        /* renorm totalCost from 2^largestBits to 2^targetNbBits
          * note : totalCost is necessarily a multiple of baseCost */
-        assert((totalCost & (baseCost - 1)) == 0);
-        totalCost >>= (largestBits - maxNbBits);
+        assert(((U32)totalCost & (baseCost - 1)) == 0);
+        totalCost >>= (largestBits - targetNbBits);
         assert(totalCost > 0);
 
         /* repay normalized cost */
@@ -281,19 +387,19 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
 
             /* Get pos of last (smallest = lowest cum. count) symbol per rank */
             ZSTD_memset(rankLast, 0xF0, sizeof(rankLast));
-            {   U32 currentNbBits = maxNbBits;
+            {   U32 currentNbBits = targetNbBits;
                 int pos;
                 for (pos=n ; pos >= 0; pos--) {
                     if (huffNode[pos].nbBits >= currentNbBits) continue;
-                    currentNbBits = huffNode[pos].nbBits;   /* < maxNbBits */
-                    rankLast[maxNbBits-currentNbBits] = (U32)pos;
+                    currentNbBits = huffNode[pos].nbBits;   /* < targetNbBits */
+                    rankLast[targetNbBits-currentNbBits] = (U32)pos;
             }   }
 
             while (totalCost > 0) {
                 /* Try to reduce the next power of 2 above totalCost because we
                  * gain back half the rank.
                  */
-                U32 nBitsToDecrease = BIT_highbit32((U32)totalCost) + 1;
+                U32 nBitsToDecrease = ZSTD_highbit32((U32)totalCost) + 1;
                 for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
                     U32 const highPos = rankLast[nBitsToDecrease];
                     U32 const lowPos = rankLast[nBitsToDecrease-1];
@@ -333,7 +439,7 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
                     rankLast[nBitsToDecrease] = noSymbol;
                 else {
                     rankLast[nBitsToDecrease]--;
-                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease)
+                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != targetNbBits-nBitsToDecrease)
                         rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
                 }
             }   /* while (totalCost > 0) */
@@ -345,11 +451,11 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
              * TODO.
              */
             while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
-                /* special case : no rank 1 symbol (using maxNbBits-1);
-                 * let's create one from largest rank 0 (using maxNbBits).
+                /* special case : no rank 1 symbol (using targetNbBits-1);
+                 * let's create one from largest rank 0 (using targetNbBits).
                  */
                 if (rankLast[1] == noSymbol) {
-                    while (huffNode[n].nbBits == maxNbBits) n--;
+                    while (huffNode[n].nbBits == targetNbBits) n--;
                     huffNode[n+1].nbBits--;
                     assert(n >= 0);
                     rankLast[1] = (U32)(n+1);
@@ -363,26 +469,122 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
         }   /* repay normalized cost */
     }   /* there are several too large elements (at least >= 2) */
 
-    return maxNbBits;
+    return targetNbBits;
 }
 
 typedef struct {
-    U32 base;
-    U32 curr;
+    U16 base;
+    U16 curr;
 } rankPos;
 
-typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
+typedef nodeElt huffNodeTable[2 * (HUF_SYMBOLVALUE_MAX + 1)];
 
-#define RANK_POSITION_TABLE_SIZE 32
+/* Number of buckets available for HUF_sort() */
+#define RANK_POSITION_TABLE_SIZE 192
 
 typedef struct {
   huffNodeTable huffNodeTbl;
   rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
 } HUF_buildCTable_wksp_tables;
 
+/* RANK_POSITION_DISTINCT_COUNT_CUTOFF == Cutoff point in HUF_sort() buckets for which we use log2 bucketing.
+ * Strategy is to use as many buckets as possible for representing distinct
+ * counts while using the remainder to represent all "large" counts.
+ *
+ * To satisfy this requirement for 192 buckets, we can do the following:
+ * Let buckets 0-166 represent distinct counts of [0, 166]
+ * Let buckets 166 to 192 represent all remaining counts up to RANK_POSITION_MAX_COUNT_LOG using log2 bucketing.
+ */
+#define RANK_POSITION_MAX_COUNT_LOG 32
+#define RANK_POSITION_LOG_BUCKETS_BEGIN ((RANK_POSITION_TABLE_SIZE - 1) - RANK_POSITION_MAX_COUNT_LOG - 1 /* == 158 */)
+#define RANK_POSITION_DISTINCT_COUNT_CUTOFF (RANK_POSITION_LOG_BUCKETS_BEGIN + ZSTD_highbit32(RANK_POSITION_LOG_BUCKETS_BEGIN) /* == 166 */)
+
+/* Return the appropriate bucket index for a given count. See definition of
+ * RANK_POSITION_DISTINCT_COUNT_CUTOFF for explanation of bucketing strategy.
+ */
+static U32 HUF_getIndex(U32 const count) {
+    return (count < RANK_POSITION_DISTINCT_COUNT_CUTOFF)
+        ? count
+        : ZSTD_highbit32(count) + RANK_POSITION_LOG_BUCKETS_BEGIN;
+}
+
+/* Helper swap function for HUF_quickSortPartition() */
+static void HUF_swapNodes(nodeElt* a, nodeElt* b) {
+	nodeElt tmp = *a;
+	*a = *b;
+	*b = tmp;
+}
+
+/* Returns 0 if the huffNode array is not sorted by descending count */
+MEM_STATIC int HUF_isSorted(nodeElt huffNode[], U32 const maxSymbolValue1) {
+    U32 i;
+    for (i = 1; i < maxSymbolValue1; ++i) {
+        if (huffNode[i].count > huffNode[i-1].count) {
+            return 0;
+        }
+    }
+    return 1;
+}
+
+/* Insertion sort by descending order */
+HINT_INLINE void HUF_insertionSort(nodeElt huffNode[], int const low, int const high) {
+    int i;
+    int const size = high-low+1;
+    huffNode += low;
+    for (i = 1; i < size; ++i) {
+        nodeElt const key = huffNode[i];
+        int j = i - 1;
+        while (j >= 0 && huffNode[j].count < key.count) {
+            huffNode[j + 1] = huffNode[j];
+            j--;
+        }
+        huffNode[j + 1] = key;
+    }
+}
+
+/* Pivot helper function for quicksort. */
+static int HUF_quickSortPartition(nodeElt arr[], int const low, int const high) {
+    /* Simply select rightmost element as pivot. "Better" selectors like
+     * median-of-three don't experimentally appear to have any benefit.
+     */
+    U32 const pivot = arr[high].count;
+    int i = low - 1;
+    int j = low;
+    for ( ; j < high; j++) {
+        if (arr[j].count > pivot) {
+            i++;
+            HUF_swapNodes(&arr[i], &arr[j]);
+        }
+    }
+    HUF_swapNodes(&arr[i + 1], &arr[high]);
+    return i + 1;
+}
+
+/* Classic quicksort by descending with partially iterative calls
+ * to reduce worst case callstack size.
+ */
+static void HUF_simpleQuickSort(nodeElt arr[], int low, int high) {
+    int const kInsertionSortThreshold = 8;
+    if (high - low < kInsertionSortThreshold) {
+        HUF_insertionSort(arr, low, high);
+        return;
+    }
+    while (low < high) {
+        int const idx = HUF_quickSortPartition(arr, low, high);
+        if (idx - low < high - idx) {
+            HUF_simpleQuickSort(arr, low, idx - 1);
+            low = idx + 1;
+        } else {
+            HUF_simpleQuickSort(arr, idx + 1, high);
+            high = idx - 1;
+        }
+    }
+}
+
 /**
  * HUF_sort():
  * Sorts the symbols [0, maxSymbolValue] by count[symbol] in decreasing order.
+ * This is a typical bucket sorting strategy that uses either quicksort or insertion sort to sort each bucket.
  *
  * @param[out] huffNode       Sorted symbols by decreasing count. Only members `.count` and `.byte` are filled.
  *                            Must have (maxSymbolValue + 1) entries.
@@ -390,42 +592,51 @@ typedef struct {
  * @param[in]  maxSymbolValue Maximum symbol value.
  * @param      rankPosition   This is a scratch workspace. Must have RANK_POSITION_TABLE_SIZE entries.
  */
-static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValue, rankPos* rankPosition)
-{
-    int n;
-    int const maxSymbolValue1 = (int)maxSymbolValue + 1;
+static void HUF_sort(nodeElt huffNode[], const unsigned count[], U32 const maxSymbolValue, rankPos rankPosition[]) {
+    U32 n;
+    U32 const maxSymbolValue1 = maxSymbolValue+1;
 
     /* Compute base and set curr to base.
-     * For symbol s let lowerRank = BIT_highbit32(count[n]+1) and rank = lowerRank + 1.
-     * Then 2^lowerRank <= count[n]+1 <= 2^rank.
+     * For symbol s let lowerRank = HUF_getIndex(count[n]) and rank = lowerRank + 1.
+     * See HUF_getIndex to see bucketing strategy.
      * We attribute each symbol to lowerRank's base value, because we want to know where
      * each rank begins in the output, so for rank R we want to count ranks R+1 and above.
      */
     ZSTD_memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
     for (n = 0; n < maxSymbolValue1; ++n) {
-        U32 lowerRank = BIT_highbit32(count[n] + 1);
+        U32 lowerRank = HUF_getIndex(count[n]);
+        assert(lowerRank < RANK_POSITION_TABLE_SIZE - 1);
         rankPosition[lowerRank].base++;
     }
+
     assert(rankPosition[RANK_POSITION_TABLE_SIZE - 1].base == 0);
+    /* Set up the rankPosition table */
     for (n = RANK_POSITION_TABLE_SIZE - 1; n > 0; --n) {
         rankPosition[n-1].base += rankPosition[n].base;
         rankPosition[n-1].curr = rankPosition[n-1].base;
     }
-    /* Sort */
+
+    /* Insert each symbol into their appropriate bucket, setting up rankPosition table. */
     for (n = 0; n < maxSymbolValue1; ++n) {
         U32 const c = count[n];
-        U32 const r = BIT_highbit32(c+1) + 1;
-        U32 pos = rankPosition[r].curr++;
-        /* Insert into the correct position in the rank.
-         * We have at most 256 symbols, so this insertion should be fine.
-         */
-        while ((pos > rankPosition[r].base) && (c > huffNode[pos-1].count)) {
-            huffNode[pos] = huffNode[pos-1];
-            pos--;
-        }
+        U32 const r = HUF_getIndex(c) + 1;
+        U32 const pos = rankPosition[r].curr++;
+        assert(pos < maxSymbolValue1);
         huffNode[pos].count = c;
         huffNode[pos].byte  = (BYTE)n;
     }
+
+    /* Sort each bucket. */
+    for (n = RANK_POSITION_DISTINCT_COUNT_CUTOFF; n < RANK_POSITION_TABLE_SIZE - 1; ++n) {
+        int const bucketSize = rankPosition[n].curr - rankPosition[n].base;
+        U32 const bucketStartIdx = rankPosition[n].base;
+        if (bucketSize > 1) {
+            assert(bucketStartIdx < maxSymbolValue1);
+            HUF_simpleQuickSort(huffNode + bucketStartIdx, 0, bucketSize-1);
+        }
+    }
+
+    assert(HUF_isSorted(huffNode, maxSymbolValue1));
 }
 
 
@@ -449,6 +660,7 @@ static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
     int lowS, lowN;
     int nodeNb = STARTNODE;
     int n, nodeRoot;
+    DEBUGLOG(5, "HUF_buildTree (alphabet size = %u)", maxSymbolValue + 1);
     /* init for parents */
     nonNullRank = (int)maxSymbolValue;
     while(huffNode[nonNullRank].count == 0) nonNullRank--;
@@ -475,6 +687,8 @@ static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
     for (n=0; n<=nonNullRank; n++)
         huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
 
+    DEBUGLOG(6, "Initial distribution of bits completed (%zu sorted symbols)", showHNodeBits(huffNode, maxSymbolValue+1));
+
     return nonNullRank;
 }
 
@@ -490,6 +704,7 @@ static int HUF_buildTree(nodeElt* huffNode, U32 maxSymbolValue)
  */
 static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, int nonNullRank, U32 maxSymbolValue, U32 maxNbBits)
 {
+    HUF_CElt* const ct = CTable + 1;
     /* fill result into ctable (val, nbBits) */
     int n;
     U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
@@ -505,127 +720,373 @@ static void HUF_buildCTableFromTree(HUF_CElt* CTable, nodeElt const* huffNode, i
             min >>= 1;
     }   }
     for (n=0; nhuffNodeTbl;
     nodeElt* const huffNode = huffNode0+1;
     int nonNullRank;
 
+    HUF_STATIC_ASSERT(HUF_CTABLE_WORKSPACE_SIZE == sizeof(HUF_buildCTable_wksp_tables));
+
+    DEBUGLOG(5, "HUF_buildCTable_wksp (alphabet size = %u)", maxSymbolValue+1);
+
     /* safety checks */
-    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
     if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
-      return ERROR(workSpace_tooSmall);
+        return ERROR(workSpace_tooSmall);
     if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
     if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
-      return ERROR(maxSymbolValue_tooLarge);
+        return ERROR(maxSymbolValue_tooLarge);
     ZSTD_memset(huffNode0, 0, sizeof(huffNodeTable));
 
     /* sort, decreasing order */
     HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
+    DEBUGLOG(6, "sorted symbols completed (%zu symbols)", showHNodeSymbols(huffNode, maxSymbolValue+1));
 
     /* build tree */
     nonNullRank = HUF_buildTree(huffNode, maxSymbolValue);
 
-    /* enforce maxTableLog */
+    /* determine and enforce maxTableLog */
     maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
     if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
 
-    HUF_buildCTableFromTree(tree, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
+    HUF_buildCTableFromTree(CTable, huffNode, nonNullRank, maxSymbolValue, maxNbBits);
 
     return maxNbBits;
 }
 
 size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
 {
+    HUF_CElt const* ct = CTable + 1;
     size_t nbBits = 0;
     int s;
     for (s = 0; s <= (int)maxSymbolValue; ++s) {
-        nbBits += CTable[s].nbBits * count[s];
+        nbBits += HUF_getNbBits(ct[s]) * count[s];
     }
     return nbBits >> 3;
 }
 
 int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+  HUF_CElt const* ct = CTable + 1;
   int bad = 0;
   int s;
   for (s = 0; s <= (int)maxSymbolValue; ++s) {
-    bad |= (count[s] != 0) & (CTable[s].nbBits == 0);
+    bad |= (count[s] != 0) & (HUF_getNbBits(ct[s]) == 0);
   }
   return !bad;
 }
 
 size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
 
+/** HUF_CStream_t:
+ * Huffman uses its own BIT_CStream_t implementation.
+ * There are three major differences from BIT_CStream_t:
+ *   1. HUF_addBits() takes a HUF_CElt (size_t) which is
+ *      the pair (nbBits, value) in the format:
+ *      format:
+ *        - Bits [0, 4)            = nbBits
+ *        - Bits [4, 64 - nbBits)  = 0
+ *        - Bits [64 - nbBits, 64) = value
+ *   2. The bitContainer is built from the upper bits and
+ *      right shifted. E.g. to add a new value of N bits
+ *      you right shift the bitContainer by N, then or in
+ *      the new value into the N upper bits.
+ *   3. The bitstream has two bit containers. You can add
+ *      bits to the second container and merge them into
+ *      the first container.
+ */
+
+#define HUF_BITS_IN_CONTAINER (sizeof(size_t) * 8)
+
+typedef struct {
+    size_t bitContainer[2];
+    size_t bitPos[2];
+
+    BYTE* startPtr;
+    BYTE* ptr;
+    BYTE* endPtr;
+} HUF_CStream_t;
+
+/**! HUF_initCStream():
+ * Initializes the bitstream.
+ * @returns 0 or an error code.
+ */
+static size_t HUF_initCStream(HUF_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
+{
+    ZSTD_memset(bitC, 0, sizeof(*bitC));
+    bitC->startPtr = (BYTE*)startPtr;
+    bitC->ptr = bitC->startPtr;
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer[0]);
+    if (dstCapacity <= sizeof(bitC->bitContainer[0])) return ERROR(dstSize_tooSmall);
+    return 0;
+}
+
+/*! HUF_addBits():
+ * Adds the symbol stored in HUF_CElt elt to the bitstream.
+ *
+ * @param elt   The element we're adding. This is a (nbBits, value) pair.
+ *              See the HUF_CStream_t docs for the format.
+ * @param idx   Insert into the bitstream at this idx.
+ * @param kFast This is a template parameter. If the bitstream is guaranteed
+ *              to have at least 4 unused bits after this call it may be 1,
+ *              otherwise it must be 0. HUF_addBits() is faster when fast is set.
+ */
+FORCE_INLINE_TEMPLATE void HUF_addBits(HUF_CStream_t* bitC, HUF_CElt elt, int idx, int kFast)
+{
+    assert(idx <= 1);
+    assert(HUF_getNbBits(elt) <= HUF_TABLELOG_ABSOLUTEMAX);
+    /* This is efficient on x86-64 with BMI2 because shrx
+     * only reads the low 6 bits of the register. The compiler
+     * knows this and elides the mask. When fast is set,
+     * every operation can use the same value loaded from elt.
+     */
+    bitC->bitContainer[idx] >>= HUF_getNbBits(elt);
+    bitC->bitContainer[idx] |= kFast ? HUF_getValueFast(elt) : HUF_getValue(elt);
+    /* We only read the low 8 bits of bitC->bitPos[idx] so it
+     * doesn't matter that the high bits have noise from the value.
+     */
+    bitC->bitPos[idx] += HUF_getNbBitsFast(elt);
+    assert((bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+    /* The last 4-bits of elt are dirty if fast is set,
+     * so we must not be overwriting bits that have already been
+     * inserted into the bit container.
+     */
+#if DEBUGLEVEL >= 1
+    {
+        size_t const nbBits = HUF_getNbBits(elt);
+        size_t const dirtyBits = nbBits == 0 ? 0 : ZSTD_highbit32((U32)nbBits) + 1;
+        (void)dirtyBits;
+        /* Middle bits are 0. */
+        assert(((elt >> dirtyBits) << (dirtyBits + nbBits)) == 0);
+        /* We didn't overwrite any bits in the bit container. */
+        assert(!kFast || (bitC->bitPos[idx] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+        (void)dirtyBits;
+    }
+#endif
+}
+
+FORCE_INLINE_TEMPLATE void HUF_zeroIndex1(HUF_CStream_t* bitC)
+{
+    bitC->bitContainer[1] = 0;
+    bitC->bitPos[1] = 0;
+}
+
+/*! HUF_mergeIndex1() :
+ * Merges the bit container @ index 1 into the bit container @ index 0
+ * and zeros the bit container @ index 1.
+ */
+FORCE_INLINE_TEMPLATE void HUF_mergeIndex1(HUF_CStream_t* bitC)
+{
+    assert((bitC->bitPos[1] & 0xFF) < HUF_BITS_IN_CONTAINER);
+    bitC->bitContainer[0] >>= (bitC->bitPos[1] & 0xFF);
+    bitC->bitContainer[0] |= bitC->bitContainer[1];
+    bitC->bitPos[0] += bitC->bitPos[1];
+    assert((bitC->bitPos[0] & 0xFF) <= HUF_BITS_IN_CONTAINER);
+}
+
+/*! HUF_flushBits() :
+* Flushes the bits in the bit container @ index 0.
+*
+* @post bitPos will be < 8.
+* @param kFast If kFast is set then we must know a-priori that
+*              the bit container will not overflow.
+*/
+FORCE_INLINE_TEMPLATE void HUF_flushBits(HUF_CStream_t* bitC, int kFast)
+{
+    /* The upper bits of bitPos are noisy, so we must mask by 0xFF. */
+    size_t const nbBits = bitC->bitPos[0] & 0xFF;
+    size_t const nbBytes = nbBits >> 3;
+    /* The top nbBits bits of bitContainer are the ones we need. */
+    size_t const bitContainer = bitC->bitContainer[0] >> (HUF_BITS_IN_CONTAINER - nbBits);
+    /* Mask bitPos to account for the bytes we consumed. */
+    bitC->bitPos[0] &= 7;
+    assert(nbBits > 0);
+    assert(nbBits <= sizeof(bitC->bitContainer[0]) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
+    MEM_writeLEST(bitC->ptr, bitContainer);
+    bitC->ptr += nbBytes;
+    assert(!kFast || bitC->ptr <= bitC->endPtr);
+    if (!kFast && bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
+    /* bitContainer doesn't need to be modified because the leftover
+     * bits are already the top bitPos bits. And we don't care about
+     * noise in the lower values.
+     */
+}
+
+/*! HUF_endMark()
+ * @returns The Huffman stream end mark: A 1-bit value = 1.
+ */
+static HUF_CElt HUF_endMark(void)
+{
+    HUF_CElt endMark;
+    HUF_setNbBits(&endMark, 1);
+    HUF_setValue(&endMark, 1);
+    return endMark;
+}
+
+/*! HUF_closeCStream() :
+ *  @return Size of CStream, in bytes,
+ *          or 0 if it could not fit into dstBuffer */
+static size_t HUF_closeCStream(HUF_CStream_t* bitC)
+{
+    HUF_addBits(bitC, HUF_endMark(), /* idx */ 0, /* kFast */ 0);
+    HUF_flushBits(bitC, /* kFast */ 0);
+    {
+        size_t const nbBits = bitC->bitPos[0] & 0xFF;
+        if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
+        return (size_t)(bitC->ptr - bitC->startPtr) + (nbBits > 0);
+    }
+}
+
 FORCE_INLINE_TEMPLATE void
-HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
+HUF_encodeSymbol(HUF_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable, int idx, int fast)
 {
-    BIT_addBitsFast(bitCPtr, CTable[symbol].val, CTable[symbol].nbBits);
+    HUF_addBits(bitCPtr, CTable[symbol], idx, fast);
 }
 
-#define HUF_FLUSHBITS(s)  BIT_flushBits(s)
+FORCE_INLINE_TEMPLATE void
+HUF_compress1X_usingCTable_internal_body_loop(HUF_CStream_t* bitC,
+                                   const BYTE* ip, size_t srcSize,
+                                   const HUF_CElt* ct,
+                                   int kUnroll, int kFastFlush, int kLastFast)
+{
+    /* Join to kUnroll */
+    int n = (int)srcSize;
+    int rem = n % kUnroll;
+    if (rem > 0) {
+        for (; rem > 0; --rem) {
+            HUF_encodeSymbol(bitC, ip[--n], ct, 0, /* fast */ 0);
+        }
+        HUF_flushBits(bitC, kFastFlush);
+    }
+    assert(n % kUnroll == 0);
+
+    /* Join to 2 * kUnroll */
+    if (n % (2 * kUnroll)) {
+        int u;
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - u], ct, 0, 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, 0, kLastFast);
+        HUF_flushBits(bitC, kFastFlush);
+        n -= kUnroll;
+    }
+    assert(n % (2 * kUnroll) == 0);
 
-#define HUF_FLUSHBITS_1(stream) \
-    if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*2+7) HUF_FLUSHBITS(stream)
+    for (; n>0; n-= 2 * kUnroll) {
+        /* Encode kUnroll symbols into the bitstream @ index 0. */
+        int u;
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - u], ct, /* idx */ 0, /* fast */ 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll], ct, /* idx */ 0, /* fast */ kLastFast);
+        HUF_flushBits(bitC, kFastFlush);
+        /* Encode kUnroll symbols into the bitstream @ index 1.
+         * This allows us to start filling the bit container
+         * without any data dependencies.
+         */
+        HUF_zeroIndex1(bitC);
+        for (u = 1; u < kUnroll; ++u) {
+            HUF_encodeSymbol(bitC, ip[n - kUnroll - u], ct, /* idx */ 1, /* fast */ 1);
+        }
+        HUF_encodeSymbol(bitC, ip[n - kUnroll - kUnroll], ct, /* idx */ 1, /* fast */ kLastFast);
+        /* Merge bitstream @ index 1 into the bitstream @ index 0 */
+        HUF_mergeIndex1(bitC);
+        HUF_flushBits(bitC, kFastFlush);
+    }
+    assert(n == 0);
+
+}
+
+/**
+ * Returns a tight upper bound on the output space needed by Huffman
+ * with 8 bytes buffer to handle over-writes. If the output is at least
+ * this large we don't need to do bounds checks during Huffman encoding.
+ */
+static size_t HUF_tightCompressBound(size_t srcSize, size_t tableLog)
+{
+    return ((srcSize * tableLog) >> 3) + 8;
+}
 
-#define HUF_FLUSHBITS_2(stream) \
-    if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*4+7) HUF_FLUSHBITS(stream)
 
 FORCE_INLINE_TEMPLATE size_t
 HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
                                    const void* src, size_t srcSize,
                                    const HUF_CElt* CTable)
 {
+    U32 const tableLog = (U32)CTable[0];
+    HUF_CElt const* ct = CTable + 1;
     const BYTE* ip = (const BYTE*) src;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
-    size_t n;
-    BIT_CStream_t bitC;
+    HUF_CStream_t bitC;
 
     /* init */
     if (dstSize < 8) return 0;   /* not enough space to compress */
-    { size_t const initErr = BIT_initCStream(&bitC, op, (size_t)(oend-op));
+    { size_t const initErr = HUF_initCStream(&bitC, op, (size_t)(oend-op));
       if (HUF_isError(initErr)) return 0; }
 
-    n = srcSize & ~3;  /* join to mod 4 */
-    switch (srcSize & 3)
-    {
-        case 3 : HUF_encodeSymbol(&bitC, ip[n+ 2], CTable);
-                 HUF_FLUSHBITS_2(&bitC);
-		 /* fall-through */
-        case 2 : HUF_encodeSymbol(&bitC, ip[n+ 1], CTable);
-                 HUF_FLUSHBITS_1(&bitC);
-		 /* fall-through */
-        case 1 : HUF_encodeSymbol(&bitC, ip[n+ 0], CTable);
-                 HUF_FLUSHBITS(&bitC);
-		 /* fall-through */
-        case 0 : /* fall-through */
-        default: break;
-    }
-
-    for (; n>0; n-=4) {  /* note : n&3==0 at this stage */
-        HUF_encodeSymbol(&bitC, ip[n- 1], CTable);
-        HUF_FLUSHBITS_1(&bitC);
-        HUF_encodeSymbol(&bitC, ip[n- 2], CTable);
-        HUF_FLUSHBITS_2(&bitC);
-        HUF_encodeSymbol(&bitC, ip[n- 3], CTable);
-        HUF_FLUSHBITS_1(&bitC);
-        HUF_encodeSymbol(&bitC, ip[n- 4], CTable);
-        HUF_FLUSHBITS(&bitC);
+    if (dstSize < HUF_tightCompressBound(srcSize, (size_t)tableLog) || tableLog > 11)
+        HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ MEM_32bits() ? 2 : 4, /* kFast */ 0, /* kLastFast */ 0);
+    else {
+        if (MEM_32bits()) {
+            switch (tableLog) {
+            case 11:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 10: ZSTD_FALLTHROUGH;
+            case 9: ZSTD_FALLTHROUGH;
+            case 8:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 2, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            case 7: ZSTD_FALLTHROUGH;
+            default:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 3, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            }
+        } else {
+            switch (tableLog) {
+            case 11:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 10:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 5, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            case 9:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 6, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 8:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 7, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 7:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 8, /* kFastFlush */ 1, /* kLastFast */ 0);
+                break;
+            case 6: ZSTD_FALLTHROUGH;
+            default:
+                HUF_compress1X_usingCTable_internal_body_loop(&bitC, ip, srcSize, ct, /* kUnroll */ 9, /* kFastFlush */ 1, /* kLastFast */ 1);
+                break;
+            }
+        }
     }
+    assert(bitC.ptr <= bitC.endPtr);
 
-    return BIT_closeCStream(&bitC);
+    return HUF_closeCStream(&bitC);
 }
 
 #if DYNAMIC_BMI2
 
-static TARGET_ATTRIBUTE("bmi2") size_t
+static BMI2_TARGET_ATTRIBUTE size_t
 HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
                                    const void* src, size_t srcSize,
                                    const HUF_CElt* CTable)
@@ -644,9 +1105,9 @@ HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
 static size_t
 HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
                               const void* src, size_t srcSize,
-                              const HUF_CElt* CTable, const int bmi2)
+                              const HUF_CElt* CTable, const int flags)
 {
-    if (bmi2) {
+    if (flags & HUF_flags_bmi2) {
         return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
     }
     return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
@@ -657,24 +1118,23 @@ HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
 static size_t
 HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
                               const void* src, size_t srcSize,
-                              const HUF_CElt* CTable, const int bmi2)
+                              const HUF_CElt* CTable, const int flags)
 {
-    (void)bmi2;
+    (void)flags;
     return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
 }
 
 #endif
 
-size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
 {
-    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
 }
 
-
 static size_t
 HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
                               const void* src, size_t srcSize,
-                              const HUF_CElt* CTable, int bmi2)
+                              const HUF_CElt* CTable, int flags)
 {
     size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
     const BYTE* ip = (const BYTE*) src;
@@ -688,27 +1148,24 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
     op += 6;   /* jumpTable */
 
     assert(op <= oend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
-        if (cSize==0) return 0;
-        assert(cSize <= 65535);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
         MEM_writeLE16(ostart, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
     assert(op <= oend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
-        if (cSize==0) return 0;
-        assert(cSize <= 65535);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
         MEM_writeLE16(ostart+2, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
     assert(op <= oend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
-        if (cSize==0) return 0;
-        assert(cSize <= 65535);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
         MEM_writeLE16(ostart+4, (U16)cSize);
         op += cSize;
     }
@@ -716,17 +1173,17 @@ HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
     ip += segmentSize;
     assert(op <= oend);
     assert(ip <= iend);
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, bmi2) );
-        if (cSize==0) return 0;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, flags) );
+        if (cSize == 0 || cSize > 65535) return 0;
         op += cSize;
     }
 
     return (size_t)(op-ostart);
 }
 
-size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable, int flags)
 {
-    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, flags);
 }
 
 typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
@@ -734,11 +1191,11 @@ typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
 static size_t HUF_compressCTable_internal(
                 BYTE* const ostart, BYTE* op, BYTE* const oend,
                 const void* src, size_t srcSize,
-                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int bmi2)
+                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int flags)
 {
     size_t const cSize = (nbStreams==HUF_singleStream) ?
-                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, bmi2) :
-                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, bmi2);
+                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags) :
+                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, flags);
     if (HUF_isError(cSize)) { return cSize; }
     if (cSize==0) { return 0; }   /* uncompressible */
     op += cSize;
@@ -750,35 +1207,111 @@ static size_t HUF_compressCTable_internal(
 
 typedef struct {
     unsigned count[HUF_SYMBOLVALUE_MAX + 1];
-    HUF_CElt CTable[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_CElt CTable[HUF_CTABLE_SIZE_ST(HUF_SYMBOLVALUE_MAX)];
     union {
         HUF_buildCTable_wksp_tables buildCTable_wksp;
         HUF_WriteCTableWksp writeCTable_wksp;
+        U32 hist_wksp[HIST_WKSP_SIZE_U32];
     } wksps;
 } HUF_compress_tables_t;
 
+#define SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE 4096
+#define SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO 10  /* Must be >= 2 */
+
+unsigned HUF_cardinality(const unsigned* count, unsigned maxSymbolValue)
+{
+    unsigned cardinality = 0;
+    unsigned i;
+
+    for (i = 0; i < maxSymbolValue + 1; i++) {
+        if (count[i] != 0) cardinality += 1;
+    }
+
+    return cardinality;
+}
+
+unsigned HUF_minTableLog(unsigned symbolCardinality)
+{
+    U32 minBitsSymbols = ZSTD_highbit32(symbolCardinality) + 1;
+    return minBitsSymbols;
+}
+
+unsigned HUF_optimalTableLog(
+            unsigned maxTableLog,
+            size_t srcSize,
+            unsigned maxSymbolValue,
+            void* workSpace, size_t wkspSize,
+            HUF_CElt* table,
+      const unsigned* count,
+            int flags)
+{
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    assert(wkspSize >= sizeof(HUF_buildCTable_wksp_tables));
+
+    if (!(flags & HUF_flags_optimalDepth)) {
+        /* cheap evaluation, based on FSE */
+        return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    }
+
+    {   BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
+        size_t dstSize = wkspSize - sizeof(HUF_WriteCTableWksp);
+        size_t maxBits, hSize, newSize;
+        const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
+        const unsigned minTableLog = HUF_minTableLog(symbolCardinality);
+        size_t optSize = ((size_t) ~0) - 1;
+        unsigned optLog = maxTableLog, optLogGuess;
+
+        DEBUGLOG(6, "HUF_optimalTableLog: probing huf depth (srcSize=%zu)", srcSize);
+
+        /* Search until size increases */
+        for (optLogGuess = minTableLog; optLogGuess <= maxTableLog; optLogGuess++) {
+            DEBUGLOG(7, "checking for huffLog=%u", optLogGuess);
+            maxBits = HUF_buildCTable_wksp(table, count, maxSymbolValue, optLogGuess, workSpace, wkspSize);
+            if (ERR_isError(maxBits)) continue;
+
+            if (maxBits < optLogGuess && optLogGuess > minTableLog) break;
+
+            hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits, workSpace, wkspSize);
+
+            if (ERR_isError(hSize)) continue;
+
+            newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
+
+            if (newSize > optSize + 1) {
+                break;
+            }
+
+            if (newSize < optSize) {
+                optSize = newSize;
+                optLog = optLogGuess;
+            }
+        }
+        assert(optLog <= HUF_TABLELOG_MAX);
+        return optLog;
+    }
+}
+
 /* HUF_compress_internal() :
  * `workSpace_align4` must be aligned on 4-bytes boundaries,
- * and occupies the same space as a table of HUF_WORKSPACE_SIZE_U32 unsigned */
+ * and occupies the same space as a table of HUF_WORKSPACE_SIZE_U64 unsigned */
 static size_t
 HUF_compress_internal (void* dst, size_t dstSize,
                  const void* src, size_t srcSize,
                        unsigned maxSymbolValue, unsigned huffLog,
                        HUF_nbStreams_e nbStreams,
-                       void* workSpace_align4, size_t wkspSize,
-                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
-                 const int bmi2)
+                       void* workSpace, size_t wkspSize,
+                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int flags)
 {
-    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace_align4;
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)HUF_alignUpWorkspace(workSpace, &wkspSize, ZSTD_ALIGNOF(size_t));
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
 
-    HUF_STATIC_ASSERT(sizeof(*table) <= HUF_WORKSPACE_SIZE);
-    assert(((size_t)workSpace_align4 & 3) == 0);   /* must be aligned on 4-bytes boundaries */
+    DEBUGLOG(5, "HUF_compress_internal (srcSize=%zu)", srcSize);
+    HUF_STATIC_ASSERT(sizeof(*table) + HUF_WORKSPACE_MAX_ALIGNMENT <= HUF_WORKSPACE_SIZE);
 
     /* checks & inits */
-    if (wkspSize < HUF_WORKSPACE_SIZE) return ERROR(workSpace_tooSmall);
+    if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
     if (!srcSize) return 0;  /* Uncompressed */
     if (!dstSize) return 0;  /* cannot fit anything within dst budget */
     if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
@@ -788,17 +1321,34 @@ HUF_compress_internal (void* dst, size_t dstSize,
     if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
 
     /* Heuristic : If old table is valid, use it for small inputs */
-    if (preferRepeat && repeat && *repeat == HUF_repeat_valid) {
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat == HUF_repeat_valid) {
         return HUF_compressCTable_internal(ostart, op, oend,
                                            src, srcSize,
-                                           nbStreams, oldHufTable, bmi2);
+                                           nbStreams, oldHufTable, flags);
+    }
+
+    /* If uncompressible data is suspected, do a smaller sampling first */
+    DEBUG_STATIC_ASSERT(SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO >= 2);
+    if ((flags & HUF_flags_suspectUncompressible) && srcSize >= (SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE * SUSPECT_INCOMPRESSIBLE_SAMPLE_RATIO)) {
+        size_t largestTotal = 0;
+        DEBUGLOG(5, "input suspected incompressible : sampling to check");
+        {   unsigned maxSymbolValueBegin = maxSymbolValue;
+            CHECK_V_F(largestBegin, HIST_count_simple (table->count, &maxSymbolValueBegin, (const BYTE*)src, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
+            largestTotal += largestBegin;
+        }
+        {   unsigned maxSymbolValueEnd = maxSymbolValue;
+            CHECK_V_F(largestEnd, HIST_count_simple (table->count, &maxSymbolValueEnd, (const BYTE*)src + srcSize - SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE, SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) );
+            largestTotal += largestEnd;
+        }
+        if (largestTotal <= ((2 * SUSPECT_INCOMPRESSIBLE_SAMPLE_SIZE) >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
     }
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace_align4, wkspSize) );
+    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->wksps.hist_wksp, sizeof(table->wksps.hist_wksp)) );
         if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
         if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
     }
+    DEBUGLOG(6, "histogram detail completed (%zu symbols)", showU32(table->count, maxSymbolValue+1));
 
     /* Check validity of previous table */
     if ( repeat
@@ -807,22 +1357,26 @@ HUF_compress_internal (void* dst, size_t dstSize,
         *repeat = HUF_repeat_none;
     }
     /* Heuristic : use existing table for small inputs */
-    if (preferRepeat && repeat && *repeat != HUF_repeat_none) {
+    if ((flags & HUF_flags_preferRepeat) && repeat && *repeat != HUF_repeat_none) {
         return HUF_compressCTable_internal(ostart, op, oend,
                                            src, srcSize,
-                                           nbStreams, oldHufTable, bmi2);
+                                           nbStreams, oldHufTable, flags);
     }
 
     /* Build Huffman Tree */
-    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, &table->wksps, sizeof(table->wksps), table->CTable, table->count, flags);
     {   size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
                                             maxSymbolValue, huffLog,
                                             &table->wksps.buildCTable_wksp, sizeof(table->wksps.buildCTable_wksp));
         CHECK_F(maxBits);
         huffLog = (U32)maxBits;
-        /* Zero unused symbols in CTable, so we can check it for validity */
-        ZSTD_memset(table->CTable + (maxSymbolValue + 1), 0,
-               sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
+        DEBUGLOG(6, "bit distribution completed (%zu symbols)", showCTableBits(table->CTable + 1, maxSymbolValue+1));
+    }
+    /* Zero unused symbols in CTable, so we can check it for validity */
+    {
+        size_t const ctableSize = HUF_CTABLE_SIZE_ST(maxSymbolValue);
+        size_t const unusedSize = sizeof(table->CTable) - ctableSize * sizeof(HUF_CElt);
+        ZSTD_memset(table->CTable + ctableSize, 0, unusedSize);
     }
 
     /* Write table description header */
@@ -835,7 +1389,7 @@ HUF_compress_internal (void* dst, size_t dstSize,
             if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
                 return HUF_compressCTable_internal(ostart, op, oend,
                                                    src, srcSize,
-                                                   nbStreams, oldHufTable, bmi2);
+                                                   nbStreams, oldHufTable, flags);
         }   }
 
         /* Use the new huffman table */
@@ -847,91 +1401,35 @@ HUF_compress_internal (void* dst, size_t dstSize,
     }
     return HUF_compressCTable_internal(ostart, op, oend,
                                        src, srcSize,
-                                       nbStreams, table->CTable, bmi2);
-}
-
-
-size_t HUF_compress1X_wksp (void* dst, size_t dstSize,
-                      const void* src, size_t srcSize,
-                      unsigned maxSymbolValue, unsigned huffLog,
-                      void* workSpace, size_t wkspSize)
-{
-    return HUF_compress_internal(dst, dstSize, src, srcSize,
-                                 maxSymbolValue, huffLog, HUF_singleStream,
-                                 workSpace, wkspSize,
-                                 NULL, NULL, 0, 0 /*bmi2*/);
+                                       nbStreams, table->CTable, flags);
 }
 
 size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize,
-                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
 {
+    DEBUGLOG(5, "HUF_compress1X_repeat (srcSize = %zu)", srcSize);
     return HUF_compress_internal(dst, dstSize, src, srcSize,
                                  maxSymbolValue, huffLog, HUF_singleStream,
                                  workSpace, wkspSize, hufTable,
-                                 repeat, preferRepeat, bmi2);
-}
-
-/* HUF_compress4X_repeat():
- * compress input using 4 streams.
- * provide workspace to generate compression tables */
-size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
-                      const void* src, size_t srcSize,
-                      unsigned maxSymbolValue, unsigned huffLog,
-                      void* workSpace, size_t wkspSize)
-{
-    return HUF_compress_internal(dst, dstSize, src, srcSize,
-                                 maxSymbolValue, huffLog, HUF_fourStreams,
-                                 workSpace, wkspSize,
-                                 NULL, NULL, 0, 0 /*bmi2*/);
+                                 repeat, flags);
 }
 
 /* HUF_compress4X_repeat():
  * compress input using 4 streams.
+ * consider skipping quickly
  * re-use an existing huffman compression table */
 size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize,
-                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int flags)
 {
+    DEBUGLOG(5, "HUF_compress4X_repeat (srcSize = %zu)", srcSize);
     return HUF_compress_internal(dst, dstSize, src, srcSize,
                                  maxSymbolValue, huffLog, HUF_fourStreams,
                                  workSpace, wkspSize,
-                                 hufTable, repeat, preferRepeat, bmi2);
-}
-
-#ifndef ZSTD_NO_UNUSED_FUNCTIONS
-/** HUF_buildCTable() :
- * @return : maxNbBits
- *  Note : count is used before tree is written, so they can safely overlap
- */
-size_t HUF_buildCTable (HUF_CElt* tree, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits)
-{
-    HUF_buildCTable_wksp_tables workspace;
-    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, &workspace, sizeof(workspace));
-}
-
-size_t HUF_compress1X (void* dst, size_t dstSize,
-                 const void* src, size_t srcSize,
-                 unsigned maxSymbolValue, unsigned huffLog)
-{
-    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
-    return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
+                                 hufTable, repeat, flags);
 }
-
-size_t HUF_compress2 (void* dst, size_t dstSize,
-                const void* src, size_t srcSize,
-                unsigned maxSymbolValue, unsigned huffLog)
-{
-    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
-    return HUF_compress4X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
-}
-
-size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSize)
-{
-    return HUF_compress2(dst, maxDstSize, src, srcSize, 255, HUF_TABLELOG_DEFAULT);
-}
-#endif
diff --git a/priv/zstd/lib/compress/zstd_compress.c b/priv/zstd/lib/compress/zstd_compress.c
index b7ee298..b55f684 100644
--- a/priv/zstd/lib/compress/zstd_compress.c
+++ b/priv/zstd/lib/compress/zstd_compress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -12,12 +12,10 @@
 *  Dependencies
 ***************************************/
 #include "../common/zstd_deps.h"  /* INT_MAX, ZSTD_memset, ZSTD_memcpy */
-#include "../common/cpu.h"
 #include "../common/mem.h"
 #include "hist.h"           /* HIST_countFast_wksp */
 #define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "zstd_compress_internal.h"
 #include "zstd_compress_sequences.h"
@@ -28,6 +26,7 @@
 #include "zstd_opt.h"
 #include "zstd_ldm.h"
 #include "zstd_compress_superblock.h"
+#include  "../common/bits.h"      /* ZSTD_highbit32 */
 
 /* ***************************************************************
 *  Tuning parameters
@@ -42,19 +41,34 @@
 #  define ZSTD_COMPRESS_HEAPMODE 0
 #endif
 
+/*!
+ * ZSTD_HASHLOG3_MAX :
+ * Maximum size of the hash table dedicated to find 3-bytes matches,
+ * in log format, aka 17 => 1 << 17 == 128Ki positions.
+ * This structure is only used in zstd_opt.
+ * Since allocation is centralized for all strategies, it has to be known here.
+ * The actual (selected) size of the hash table is then stored in ZSTD_matchState_t.hashLog3,
+ * so that zstd_opt.c doesn't need to know about this constant.
+ */
+#ifndef ZSTD_HASHLOG3_MAX
+#  define ZSTD_HASHLOG3_MAX 17
+#endif
 
 /*-*************************************
 *  Helper functions
 ***************************************/
 /* ZSTD_compressBound()
- * Note that the result from this function is only compatible with the "normal"
- * full-block strategy.
- * When there are a lot of small blocks due to frequent flush in streaming mode
- * the overhead of headers can make the compressed data to be larger than the
- * return value of ZSTD_compressBound().
+ * Note that the result from this function is only valid for
+ * the one-pass compression functions.
+ * When employing the streaming mode,
+ * if flushes are frequently altering the size of blocks,
+ * the overhead from block headers can make the compressed data larger
+ * than the return value of ZSTD_compressBound().
  */
 size_t ZSTD_compressBound(size_t srcSize) {
-    return ZSTD_COMPRESSBOUND(srcSize);
+    size_t const r = ZSTD_COMPRESSBOUND(srcSize);
+    if (r==0) return ERROR(srcSize_wrong);
+    return r;
 }
 
 
@@ -72,10 +86,10 @@ struct ZSTD_CDict_s {
     ZSTD_customMem customMem;
     U32 dictID;
     int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
-    ZSTD_useRowMatchFinderMode_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
-                                                     * row-based matchfinder. Unless the cdict is reloaded, we will use
-                                                     * the same greedy/lazy matchfinder at compression time.
-                                                     */
+    ZSTD_paramSwitch_e useRowMatchFinder; /* Indicates whether the CDict was created with params that would use
+                                           * row-based matchfinder. Unless the cdict is reloaded, we will use
+                                           * the same greedy/lazy matchfinder at compression time.
+                                           */
 };  /* typedef'd to ZSTD_CDict within "zstd.h" */
 
 ZSTD_CCtx* ZSTD_createCCtx(void)
@@ -88,7 +102,7 @@ static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
     assert(cctx != NULL);
     ZSTD_memset(cctx, 0, sizeof(*cctx));
     cctx->customMem = memManager;
-    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    cctx->bmi2 = ZSTD_cpuSupportsBmi2();
     {   size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
         assert(!ZSTD_isError(err));
         (void)err;
@@ -166,12 +180,9 @@ size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
     if (cctx==NULL) return 0;   /* support free on NULL */
     RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
                     "not compatible with static CCtx");
-    {
-        int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
+    {   int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
         ZSTD_freeCCtxContent(cctx);
-        if (!cctxInWorkspace) {
-            ZSTD_customFree(cctx, cctx->customMem);
-        }
+        if (!cctxInWorkspace) ZSTD_customFree(cctx, cctx->customMem);
     }
     return 0;
 }
@@ -214,55 +225,84 @@ static int ZSTD_rowMatchFinderSupported(const ZSTD_strategy strategy) {
 /* Returns true if the strategy and useRowMatchFinder mode indicate that we will use the row based matchfinder
  * for this compression.
  */
-static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_useRowMatchFinderMode_e mode) {
-    assert(mode != ZSTD_urm_auto);
-    return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_urm_enableRowMatchFinder);
+static int ZSTD_rowMatchFinderUsed(const ZSTD_strategy strategy, const ZSTD_paramSwitch_e mode) {
+    assert(mode != ZSTD_ps_auto);
+    return ZSTD_rowMatchFinderSupported(strategy) && (mode == ZSTD_ps_enable);
 }
 
-/* Returns row matchfinder usage enum given an initial mode and cParams */
-static ZSTD_useRowMatchFinderMode_e ZSTD_resolveRowMatchFinderMode(ZSTD_useRowMatchFinderMode_e mode,
-                                                                   const ZSTD_compressionParameters* const cParams) {
-#if !defined(ZSTD_NO_INTRINSICS) && (defined(__SSE2__) || defined(__ARM_NEON))
+/* Returns row matchfinder usage given an initial mode and cParams */
+static ZSTD_paramSwitch_e ZSTD_resolveRowMatchFinderMode(ZSTD_paramSwitch_e mode,
+                                                         const ZSTD_compressionParameters* const cParams) {
+#if defined(ZSTD_ARCH_X86_SSE2) || defined(ZSTD_ARCH_ARM_NEON)
     int const kHasSIMD128 = 1;
 #else
     int const kHasSIMD128 = 0;
 #endif
-    if (mode != ZSTD_urm_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
-    mode = ZSTD_urm_disableRowMatchFinder;
+    if (mode != ZSTD_ps_auto) return mode; /* if requested enabled, but no SIMD, we still will use row matchfinder */
+    mode = ZSTD_ps_disable;
     if (!ZSTD_rowMatchFinderSupported(cParams->strategy)) return mode;
     if (kHasSIMD128) {
-        if (cParams->windowLog > 14) mode = ZSTD_urm_enableRowMatchFinder;
+        if (cParams->windowLog > 14) mode = ZSTD_ps_enable;
     } else {
-        if (cParams->windowLog > 17) mode = ZSTD_urm_enableRowMatchFinder;
+        if (cParams->windowLog > 17) mode = ZSTD_ps_enable;
     }
     return mode;
 }
 
+/* Returns block splitter usage (generally speaking, when using slower/stronger compression modes) */
+static ZSTD_paramSwitch_e ZSTD_resolveBlockSplitterMode(ZSTD_paramSwitch_e mode,
+                                                        const ZSTD_compressionParameters* const cParams) {
+    if (mode != ZSTD_ps_auto) return mode;
+    return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17) ? ZSTD_ps_enable : ZSTD_ps_disable;
+}
+
 /* Returns 1 if the arguments indicate that we should allocate a chainTable, 0 otherwise */
 static int ZSTD_allocateChainTable(const ZSTD_strategy strategy,
-                                   const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+                                   const ZSTD_paramSwitch_e useRowMatchFinder,
                                    const U32 forDDSDict) {
-    assert(useRowMatchFinder != ZSTD_urm_auto);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
     /* We always should allocate a chaintable if we are allocating a matchstate for a DDS dictionary matchstate.
      * We do not allocate a chaintable if we are using ZSTD_fast, or are using the row-based matchfinder.
      */
     return forDDSDict || ((strategy != ZSTD_fast) && !ZSTD_rowMatchFinderUsed(strategy, useRowMatchFinder));
 }
 
-/* Returns 1 if compression parameters are such that we should
+/* Returns ZSTD_ps_enable if compression parameters are such that we should
  * enable long distance matching (wlog >= 27, strategy >= btopt).
- * Returns 0 otherwise.
+ * Returns ZSTD_ps_disable otherwise.
  */
-static U32 ZSTD_CParams_shouldEnableLdm(const ZSTD_compressionParameters* const cParams) {
-    return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27;
+static ZSTD_paramSwitch_e ZSTD_resolveEnableLdm(ZSTD_paramSwitch_e mode,
+                                 const ZSTD_compressionParameters* const cParams) {
+    if (mode != ZSTD_ps_auto) return mode;
+    return (cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 27) ? ZSTD_ps_enable : ZSTD_ps_disable;
 }
 
-/* Returns 1 if compression parameters are such that we should
- * enable blockSplitter (wlog >= 17, strategy >= btopt).
- * Returns 0 otherwise.
- */
-static U32 ZSTD_CParams_useBlockSplitter(const ZSTD_compressionParameters* const cParams) {
-    return cParams->strategy >= ZSTD_btopt && cParams->windowLog >= 17;
+static int ZSTD_resolveExternalSequenceValidation(int mode) {
+    return mode;
+}
+
+/* Resolves maxBlockSize to the default if no value is present. */
+static size_t ZSTD_resolveMaxBlockSize(size_t maxBlockSize) {
+    if (maxBlockSize == 0) {
+        return ZSTD_BLOCKSIZE_MAX;
+    } else {
+        return maxBlockSize;
+    }
+}
+
+static ZSTD_paramSwitch_e ZSTD_resolveExternalRepcodeSearch(ZSTD_paramSwitch_e value, int cLevel) {
+    if (value != ZSTD_ps_auto) return value;
+    if (cLevel < 10) {
+        return ZSTD_ps_disable;
+    } else {
+        return ZSTD_ps_enable;
+    }
+}
+
+/* Returns 1 if compression parameters are such that CDict hashtable and chaintable indices are tagged.
+ * If so, the tags need to be removed in ZSTD_resetCCtx_byCopyingCDict. */
+static int ZSTD_CDictIndicesAreTagged(const ZSTD_compressionParameters* const cParams) {
+    return cParams->strategy == ZSTD_fast || cParams->strategy == ZSTD_dfast;
 }
 
 static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
@@ -274,21 +314,18 @@ static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
     cctxParams.cParams = cParams;
 
     /* Adjust advanced params according to cParams */
-    if (ZSTD_CParams_shouldEnableLdm(&cParams)) {
-        DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including LDM into cctx params");
-        cctxParams.ldmParams.enableLdm = 1;
-        /* LDM is enabled by default for optimal parser and window size >= 128MB */
+    cctxParams.ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams.ldmParams.enableLdm, &cParams);
+    if (cctxParams.ldmParams.enableLdm == ZSTD_ps_enable) {
         ZSTD_ldm_adjustParameters(&cctxParams.ldmParams, &cParams);
         assert(cctxParams.ldmParams.hashLog >= cctxParams.ldmParams.bucketSizeLog);
         assert(cctxParams.ldmParams.hashRateLog < 32);
     }
-
-    if (ZSTD_CParams_useBlockSplitter(&cParams)) {
-        DEBUGLOG(4, "ZSTD_makeCCtxParamsFromCParams(): Including block splitting into cctx params");
-        cctxParams.splitBlocks = 1;
-    }
-
+    cctxParams.useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams.useBlockSplitter, &cParams);
     cctxParams.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams.useRowMatchFinder, &cParams);
+    cctxParams.validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams.validateSequences);
+    cctxParams.maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams.maxBlockSize);
+    cctxParams.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams.searchForExternalRepcodes,
+                                                                             cctxParams.compressionLevel);
     assert(!ZSTD_checkCParams(cParams));
     return cctxParams;
 }
@@ -334,10 +371,13 @@ size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel)
 #define ZSTD_NO_CLEVEL 0
 
 /**
- * Initializes the cctxParams from params and compressionLevel.
+ * Initializes `cctxParams` from `params` and `compressionLevel`.
  * @param compressionLevel If params are derived from a compression level then that compression level, otherwise ZSTD_NO_CLEVEL.
  */
-static void ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams, ZSTD_parameters const* params, int compressionLevel)
+static void
+ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams,
+                        const ZSTD_parameters* params,
+                              int compressionLevel)
 {
     assert(!ZSTD_checkCParams(params->cParams));
     ZSTD_memset(cctxParams, 0, sizeof(*cctxParams));
@@ -348,7 +388,13 @@ static void ZSTD_CCtxParams_init_internal(ZSTD_CCtx_params* cctxParams, ZSTD_par
      */
     cctxParams->compressionLevel = compressionLevel;
     cctxParams->useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(cctxParams->useRowMatchFinder, ¶ms->cParams);
-    DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d", cctxParams->useRowMatchFinder);
+    cctxParams->useBlockSplitter = ZSTD_resolveBlockSplitterMode(cctxParams->useBlockSplitter, ¶ms->cParams);
+    cctxParams->ldmParams.enableLdm = ZSTD_resolveEnableLdm(cctxParams->ldmParams.enableLdm, ¶ms->cParams);
+    cctxParams->validateSequences = ZSTD_resolveExternalSequenceValidation(cctxParams->validateSequences);
+    cctxParams->maxBlockSize = ZSTD_resolveMaxBlockSize(cctxParams->maxBlockSize);
+    cctxParams->searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(cctxParams->searchForExternalRepcodes, compressionLevel);
+    DEBUGLOG(4, "ZSTD_CCtxParams_init_internal: useRowMatchFinder=%d, useBlockSplitter=%d ldm=%d",
+                cctxParams->useRowMatchFinder, cctxParams->useBlockSplitter, cctxParams->ldmParams.enableLdm);
 }
 
 size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
@@ -361,7 +407,7 @@ size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_paramete
 
 /**
  * Sets cctxParams' cParams and fParams from params, but otherwise leaves them alone.
- * @param param Validated zstd parameters.
+ * @param params Validated zstd parameters.
  */
 static void ZSTD_CCtxParams_setZstdParams(
         ZSTD_CCtx_params* cctxParams, const ZSTD_parameters* params)
@@ -470,8 +516,8 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         return bounds;
 
     case ZSTD_c_enableLongDistanceMatching:
-        bounds.lowerBound = 0;
-        bounds.upperBound = 1;
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
         return bounds;
 
     case ZSTD_c_ldmHashLog:
@@ -518,9 +564,9 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         return bounds;
 
     case ZSTD_c_literalCompressionMode:
-        ZSTD_STATIC_ASSERT(ZSTD_lcm_auto < ZSTD_lcm_huffman && ZSTD_lcm_huffman < ZSTD_lcm_uncompressed);
-        bounds.lowerBound = ZSTD_lcm_auto;
-        bounds.upperBound = ZSTD_lcm_uncompressed;
+        ZSTD_STATIC_ASSERT(ZSTD_ps_auto < ZSTD_ps_enable && ZSTD_ps_enable < ZSTD_ps_disable);
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
         return bounds;
 
     case ZSTD_c_targetCBlockSize:
@@ -549,14 +595,14 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = 1;
         return bounds;
 
-    case ZSTD_c_splitBlocks:
-        bounds.lowerBound = 0;
-        bounds.upperBound = 1;
+    case ZSTD_c_useBlockSplitter:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
         return bounds;
 
     case ZSTD_c_useRowMatchFinder:
-        bounds.lowerBound = (int)ZSTD_urm_auto;
-        bounds.upperBound = (int)ZSTD_urm_enableRowMatchFinder;
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
         return bounds;
 
     case ZSTD_c_deterministicRefPrefix:
@@ -564,6 +610,26 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = 1;
         return bounds;
 
+    case ZSTD_c_prefetchCDictTables:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_maxBlockSize:
+        bounds.lowerBound = ZSTD_BLOCKSIZE_MAX_MIN;
+        bounds.upperBound = ZSTD_BLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_searchForExternalRepcodes:
+        bounds.lowerBound = (int)ZSTD_ps_auto;
+        bounds.upperBound = (int)ZSTD_ps_disable;
+        return bounds;
+
     default:
         bounds.error = ERROR(parameter_unsupported);
         return bounds;
@@ -625,9 +691,13 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_stableOutBuffer:
     case ZSTD_c_blockDelimiters:
     case ZSTD_c_validateSequences:
-    case ZSTD_c_splitBlocks:
+    case ZSTD_c_useBlockSplitter:
     case ZSTD_c_useRowMatchFinder:
     case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_searchForExternalRepcodes:
     default:
         return 0;
     }
@@ -640,7 +710,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
         if (ZSTD_isUpdateAuthorized(param)) {
             cctx->cParamsChanged = 1;
         } else {
-            RETURN_ERROR(stage_wrong, "can only set params in ctx init stage");
+            RETURN_ERROR(stage_wrong, "can only set params in cctx init stage");
     }   }
 
     switch(param)
@@ -680,9 +750,13 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
     case ZSTD_c_stableOutBuffer:
     case ZSTD_c_blockDelimiters:
     case ZSTD_c_validateSequences:
-    case ZSTD_c_splitBlocks:
+    case ZSTD_c_useBlockSplitter:
     case ZSTD_c_useRowMatchFinder:
     case ZSTD_c_deterministicRefPrefix:
+    case ZSTD_c_prefetchCDictTables:
+    case ZSTD_c_enableSeqProducerFallback:
+    case ZSTD_c_maxBlockSize:
+    case ZSTD_c_searchForExternalRepcodes:
         break;
 
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
@@ -738,12 +812,12 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
     case ZSTD_c_minMatch :
         if (value!=0)   /* 0 => use default */
             BOUNDCHECK(ZSTD_c_minMatch, value);
-        CCtxParams->cParams.minMatch = value;
+        CCtxParams->cParams.minMatch = (U32)value;
         return CCtxParams->cParams.minMatch;
 
     case ZSTD_c_targetLength :
         BOUNDCHECK(ZSTD_c_targetLength, value);
-        CCtxParams->cParams.targetLength = value;
+        CCtxParams->cParams.targetLength = (U32)value;
         return CCtxParams->cParams.targetLength;
 
     case ZSTD_c_strategy :
@@ -756,12 +830,12 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         /* Content size written in frame header _when known_ (default:1) */
         DEBUGLOG(4, "set content size flag = %u", (value!=0));
         CCtxParams->fParams.contentSizeFlag = value != 0;
-        return CCtxParams->fParams.contentSizeFlag;
+        return (size_t)CCtxParams->fParams.contentSizeFlag;
 
     case ZSTD_c_checksumFlag :
         /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
         CCtxParams->fParams.checksumFlag = value != 0;
-        return CCtxParams->fParams.checksumFlag;
+        return (size_t)CCtxParams->fParams.checksumFlag;
 
     case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
         DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
@@ -770,18 +844,18 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
 
     case ZSTD_c_forceMaxWindow :
         CCtxParams->forceWindow = (value != 0);
-        return CCtxParams->forceWindow;
+        return (size_t)CCtxParams->forceWindow;
 
     case ZSTD_c_forceAttachDict : {
         const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
-        BOUNDCHECK(ZSTD_c_forceAttachDict, pref);
+        BOUNDCHECK(ZSTD_c_forceAttachDict, (int)pref);
         CCtxParams->attachDictPref = pref;
         return CCtxParams->attachDictPref;
     }
 
     case ZSTD_c_literalCompressionMode : {
-        const ZSTD_literalCompressionMode_e lcm = (ZSTD_literalCompressionMode_e)value;
-        BOUNDCHECK(ZSTD_c_literalCompressionMode, lcm);
+        const ZSTD_paramSwitch_e lcm = (ZSTD_paramSwitch_e)value;
+        BOUNDCHECK(ZSTD_c_literalCompressionMode, (int)lcm);
         CCtxParams->literalCompressionMode = lcm;
         return CCtxParams->literalCompressionMode;
     }
@@ -832,47 +906,48 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
 
     case ZSTD_c_enableDedicatedDictSearch :
         CCtxParams->enableDedicatedDictSearch = (value!=0);
-        return CCtxParams->enableDedicatedDictSearch;
+        return (size_t)CCtxParams->enableDedicatedDictSearch;
 
     case ZSTD_c_enableLongDistanceMatching :
-        CCtxParams->ldmParams.enableLdm = (value!=0);
+        BOUNDCHECK(ZSTD_c_enableLongDistanceMatching, value);
+        CCtxParams->ldmParams.enableLdm = (ZSTD_paramSwitch_e)value;
         return CCtxParams->ldmParams.enableLdm;
 
     case ZSTD_c_ldmHashLog :
         if (value!=0)   /* 0 ==> auto */
             BOUNDCHECK(ZSTD_c_ldmHashLog, value);
-        CCtxParams->ldmParams.hashLog = value;
+        CCtxParams->ldmParams.hashLog = (U32)value;
         return CCtxParams->ldmParams.hashLog;
 
     case ZSTD_c_ldmMinMatch :
         if (value!=0)   /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
-        CCtxParams->ldmParams.minMatchLength = value;
+        CCtxParams->ldmParams.minMatchLength = (U32)value;
         return CCtxParams->ldmParams.minMatchLength;
 
     case ZSTD_c_ldmBucketSizeLog :
         if (value!=0)   /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, value);
-        CCtxParams->ldmParams.bucketSizeLog = value;
+        CCtxParams->ldmParams.bucketSizeLog = (U32)value;
         return CCtxParams->ldmParams.bucketSizeLog;
 
     case ZSTD_c_ldmHashRateLog :
-        RETURN_ERROR_IF(value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN,
-                        parameter_outOfBound, "Param out of bounds!");
-        CCtxParams->ldmParams.hashRateLog = value;
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmHashRateLog, value);
+        CCtxParams->ldmParams.hashRateLog = (U32)value;
         return CCtxParams->ldmParams.hashRateLog;
 
     case ZSTD_c_targetCBlockSize :
         if (value!=0)   /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
-        CCtxParams->targetCBlockSize = value;
+        CCtxParams->targetCBlockSize = (U32)value;
         return CCtxParams->targetCBlockSize;
 
     case ZSTD_c_srcSizeHint :
         if (value!=0)    /* 0 ==> default */
             BOUNDCHECK(ZSTD_c_srcSizeHint, value);
         CCtxParams->srcSizeHint = value;
-        return CCtxParams->srcSizeHint;
+        return (size_t)CCtxParams->srcSizeHint;
 
     case ZSTD_c_stableInBuffer:
         BOUNDCHECK(ZSTD_c_stableInBuffer, value);
@@ -894,14 +969,14 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         CCtxParams->validateSequences = value;
         return CCtxParams->validateSequences;
 
-    case ZSTD_c_splitBlocks:
-        BOUNDCHECK(ZSTD_c_splitBlocks, value);
-        CCtxParams->splitBlocks = value;
-        return CCtxParams->splitBlocks;
+    case ZSTD_c_useBlockSplitter:
+        BOUNDCHECK(ZSTD_c_useBlockSplitter, value);
+        CCtxParams->useBlockSplitter = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->useBlockSplitter;
 
     case ZSTD_c_useRowMatchFinder:
         BOUNDCHECK(ZSTD_c_useRowMatchFinder, value);
-        CCtxParams->useRowMatchFinder = (ZSTD_useRowMatchFinderMode_e)value;
+        CCtxParams->useRowMatchFinder = (ZSTD_paramSwitch_e)value;
         return CCtxParams->useRowMatchFinder;
 
     case ZSTD_c_deterministicRefPrefix:
@@ -909,6 +984,27 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         CCtxParams->deterministicRefPrefix = !!value;
         return CCtxParams->deterministicRefPrefix;
 
+    case ZSTD_c_prefetchCDictTables:
+        BOUNDCHECK(ZSTD_c_prefetchCDictTables, value);
+        CCtxParams->prefetchCDictTables = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->prefetchCDictTables;
+
+    case ZSTD_c_enableSeqProducerFallback:
+        BOUNDCHECK(ZSTD_c_enableSeqProducerFallback, value);
+        CCtxParams->enableMatchFinderFallback = value;
+        return CCtxParams->enableMatchFinderFallback;
+
+    case ZSTD_c_maxBlockSize:
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_maxBlockSize, value);
+        CCtxParams->maxBlockSize = value;
+        return CCtxParams->maxBlockSize;
+
+    case ZSTD_c_searchForExternalRepcodes:
+        BOUNDCHECK(ZSTD_c_searchForExternalRepcodes, value);
+        CCtxParams->searchForExternalRepcodes = (ZSTD_paramSwitch_e)value;
+        return CCtxParams->searchForExternalRepcodes;
+
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
 }
@@ -1032,8 +1128,8 @@ size_t ZSTD_CCtxParams_getParameter(
     case ZSTD_c_validateSequences :
         *value = (int)CCtxParams->validateSequences;
         break;
-    case ZSTD_c_splitBlocks :
-        *value = (int)CCtxParams->splitBlocks;
+    case ZSTD_c_useBlockSplitter :
+        *value = (int)CCtxParams->useBlockSplitter;
         break;
     case ZSTD_c_useRowMatchFinder :
         *value = (int)CCtxParams->useRowMatchFinder;
@@ -1041,6 +1137,18 @@ size_t ZSTD_CCtxParams_getParameter(
     case ZSTD_c_deterministicRefPrefix:
         *value = (int)CCtxParams->deterministicRefPrefix;
         break;
+    case ZSTD_c_prefetchCDictTables:
+        *value = (int)CCtxParams->prefetchCDictTables;
+        break;
+    case ZSTD_c_enableSeqProducerFallback:
+        *value = CCtxParams->enableMatchFinderFallback;
+        break;
+    case ZSTD_c_maxBlockSize:
+        *value = (int)CCtxParams->maxBlockSize;
+        break;
+    case ZSTD_c_searchForExternalRepcodes:
+        *value = (int)CCtxParams->searchForExternalRepcodes;
+        break;
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
     return 0;
@@ -1067,9 +1175,24 @@ size_t ZSTD_CCtx_setParametersUsingCCtxParams(
     return 0;
 }
 
-ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
+size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams)
 {
-    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %u bytes", (U32)pledgedSrcSize);
+    DEBUGLOG(4, "ZSTD_CCtx_setCParams");
+    assert(cctx != NULL);
+    if (cctx->streamStage != zcss_init) {
+        /* All parameters in @cparams are allowed to be updated during MT compression.
+         * This must be signaled, so that MT compression picks up the changes */
+        cctx->cParamsChanged = 1;
+    }
+    /* only update if parameters are valid */
+    FORWARD_IF_ERROR(ZSTD_checkCParams(cparams), "");
+    cctx->requestedParams.cParams = cparams;
+    return 0;
+}
+
+size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %llu bytes", pledgedSrcSize);
     RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
                     "Can't set pledgedSrcSize when not in init stage.");
     cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
@@ -1147,14 +1270,14 @@ size_t ZSTD_CCtx_loadDictionary_advanced(
     return 0;
 }
 
-ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(
+size_t ZSTD_CCtx_loadDictionary_byReference(
       ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
 {
     return ZSTD_CCtx_loadDictionary_advanced(
             cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
 }
 
-ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
 {
     return ZSTD_CCtx_loadDictionary_advanced(
             cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
@@ -1212,6 +1335,7 @@ size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
         RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
                         "Can't reset parameters only when not in init stage.");
         ZSTD_clearAllDicts(cctx);
+        ZSTD_memset(&cctx->externalMatchCtx, 0, sizeof(cctx->externalMatchCtx));
         return ZSTD_CCtxParams_reset(&cctx->requestedParams);
     }
     return 0;
@@ -1308,7 +1432,8 @@ static ZSTD_compressionParameters
 ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
                             unsigned long long srcSize,
                             size_t dictSize,
-                            ZSTD_cParamMode_e mode)
+                            ZSTD_cParamMode_e mode,
+                            ZSTD_paramSwitch_e useRowMatchFinder)
 {
     const U64 minSrcSize = 513; /* (1<<9) + 1 */
     const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
@@ -1324,7 +1449,7 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
         break;
     case ZSTD_cpm_createCDict:
         /* Assume a small source size when creating a dictionary
-         * with an unkown source size.
+         * with an unknown source size.
          */
         if (dictSize && srcSize == ZSTD_CONTENTSIZE_UNKNOWN)
             srcSize = minSrcSize;
@@ -1342,8 +1467,8 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
     }
 
     /* resize windowLog if input is small enough, to use less memory */
-    if ( (srcSize < maxWindowResize)
-      && (dictSize < maxWindowResize) )  {
+    if ( (srcSize <= maxWindowResize)
+      && (dictSize <= maxWindowResize) )  {
         U32 const tSize = (U32)(srcSize + dictSize);
         static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
         U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
@@ -1361,6 +1486,42 @@ ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
     if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
         cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* minimum wlog required for valid frame header */
 
+    /* We can't use more than 32 bits of hash in total, so that means that we require:
+     * (hashLog + 8) <= 32 && (chainLog + 8) <= 32
+     */
+    if (mode == ZSTD_cpm_createCDict && ZSTD_CDictIndicesAreTagged(&cPar)) {
+        U32 const maxShortCacheHashLog = 32 - ZSTD_SHORT_CACHE_TAG_BITS;
+        if (cPar.hashLog > maxShortCacheHashLog) {
+            cPar.hashLog = maxShortCacheHashLog;
+        }
+        if (cPar.chainLog > maxShortCacheHashLog) {
+            cPar.chainLog = maxShortCacheHashLog;
+        }
+    }
+
+
+    /* At this point, we aren't 100% sure if we are using the row match finder.
+     * Unless it is explicitly disabled, conservatively assume that it is enabled.
+     * In this case it will only be disabled for small sources, so shrinking the
+     * hash log a little bit shouldn't result in any ratio loss.
+     */
+    if (useRowMatchFinder == ZSTD_ps_auto)
+        useRowMatchFinder = ZSTD_ps_enable;
+
+    /* We can't hash more than 32-bits in total. So that means that we require:
+     * (hashLog - rowLog + 8) <= 32
+     */
+    if (ZSTD_rowMatchFinderUsed(cPar.strategy, useRowMatchFinder)) {
+        /* Switch to 32-entry rows if searchLog is 5 (or more) */
+        U32 const rowLog = BOUNDED(4, cPar.searchLog, 6);
+        U32 const maxRowHashLog = 32 - ZSTD_ROW_HASH_TAG_BITS;
+        U32 const maxHashLog = maxRowHashLog + rowLog;
+        assert(cPar.hashLog >= rowLog);
+        if (cPar.hashLog > maxHashLog) {
+            cPar.hashLog = maxHashLog;
+        }
+    }
+
     return cPar;
 }
 
@@ -1371,7 +1532,7 @@ ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
 {
     cPar = ZSTD_clampCParams(cPar);   /* resulting cPar is necessarily valid (all parameters within range) */
     if (srcSize == 0) srcSize = ZSTD_CONTENTSIZE_UNKNOWN;
-    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown);
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize, ZSTD_cpm_unknown, ZSTD_ps_auto);
 }
 
 static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
@@ -1398,16 +1559,16 @@ ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
       srcSizeHint = CCtxParams->srcSizeHint;
     }
     cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize, mode);
-    if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    if (CCtxParams->ldmParams.enableLdm == ZSTD_ps_enable) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
     ZSTD_overrideCParams(&cParams, &CCtxParams->cParams);
     assert(!ZSTD_checkCParams(cParams));
     /* srcSizeHint == 0 means 0 */
-    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode);
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize, mode, CCtxParams->useRowMatchFinder);
 }
 
 static size_t
 ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
-                       const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+                       const ZSTD_paramSwitch_e useRowMatchFinder,
                        const U32 enableDedicatedDictSearch,
                        const U32 forCCtx)
 {
@@ -1440,26 +1601,34 @@ ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
 
     /* tables are guaranteed to be sized in multiples of 64 bytes (or 16 uint32_t) */
     ZSTD_STATIC_ASSERT(ZSTD_HASHLOG_MIN >= 4 && ZSTD_WINDOWLOG_MIN >= 4 && ZSTD_CHAINLOG_MIN >= 4);
-    assert(useRowMatchFinder != ZSTD_urm_auto);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
 
     DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
                 (U32)chainSize, (U32)hSize, (U32)h3Size);
     return tableSpace + optSpace + slackSpace + lazyAdditionalSpace;
 }
 
+/* Helper function for calculating memory requirements.
+ * Gives a tighter bound than ZSTD_sequenceBound() by taking minMatch into account. */
+static size_t ZSTD_maxNbSeq(size_t blockSize, unsigned minMatch, int useSequenceProducer) {
+    U32 const divider = (minMatch==3 || useSequenceProducer) ? 3 : 4;
+    return blockSize / divider;
+}
+
 static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
         const ZSTD_compressionParameters* cParams,
         const ldmParams_t* ldmParams,
         const int isStatic,
-        const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+        const ZSTD_paramSwitch_e useRowMatchFinder,
         const size_t buffInSize,
         const size_t buffOutSize,
-        const U64 pledgedSrcSize)
+        const U64 pledgedSrcSize,
+        int useSequenceProducer,
+        size_t maxBlockSize)
 {
-    size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << cParams->windowLog), pledgedSrcSize));
-    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
-    U32    const divider = (cParams->minMatch==3) ? 3 : 4;
-    size_t const maxNbSeq = blockSize / divider;
+    size_t const windowSize = (size_t) BOUNDED(1ULL, 1ULL << cParams->windowLog, pledgedSrcSize);
+    size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(maxBlockSize), windowSize);
+    size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, cParams->minMatch, useSequenceProducer);
     size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
                             + ZSTD_cwksp_aligned_alloc_size(maxNbSeq * sizeof(seqDef))
                             + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
@@ -1469,7 +1638,7 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
 
     size_t const ldmSpace = ZSTD_ldm_getTableSize(*ldmParams);
     size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(*ldmParams, blockSize);
-    size_t const ldmSeqSpace = ldmParams->enableLdm ?
+    size_t const ldmSeqSpace = ldmParams->enableLdm == ZSTD_ps_enable ?
         ZSTD_cwksp_aligned_alloc_size(maxNbLdmSeq * sizeof(rawSeq)) : 0;
 
 
@@ -1478,6 +1647,11 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
 
     size_t const cctxSpace = isStatic ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
 
+    size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+    size_t const externalSeqSpace = useSequenceProducer
+        ? ZSTD_cwksp_aligned_alloc_size(maxNbExternalSeq * sizeof(ZSTD_Sequence))
+        : 0;
+
     size_t const neededSpace =
         cctxSpace +
         entropySpace +
@@ -1486,7 +1660,8 @@ static size_t ZSTD_estimateCCtxSize_usingCCtxParams_internal(
         ldmSeqSpace +
         matchStateSize +
         tokenSpace +
-        bufferSpace;
+        bufferSpace +
+        externalSeqSpace;
 
     DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
     return neededSpace;
@@ -1496,15 +1671,15 @@ size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
 {
     ZSTD_compressionParameters const cParams =
                 ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
-    ZSTD_useRowMatchFinderMode_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder,
-                                                                                         &cParams);
+    ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder,
+                                                                               &cParams);
 
     RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
     /* estimateCCtxSize is for one-shot compression. So no buffers should
      * be needed. However, we still allocate two 0-sized buffers, which can
      * take space under ASAN. */
     return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
-        &cParams, ¶ms->ldmParams, 1, useRowMatchFinder, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN);
+        &cParams, ¶ms->ldmParams, 1, useRowMatchFinder, 0, 0, ZSTD_CONTENTSIZE_UNKNOWN, params->useSequenceProducer, params->maxBlockSize);
 }
 
 size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
@@ -1514,9 +1689,9 @@ size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
         /* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
         size_t noRowCCtxSize;
         size_t rowCCtxSize;
-        initialParams.useRowMatchFinder = ZSTD_urm_disableRowMatchFinder;
+        initialParams.useRowMatchFinder = ZSTD_ps_disable;
         noRowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
-        initialParams.useRowMatchFinder = ZSTD_urm_enableRowMatchFinder;
+        initialParams.useRowMatchFinder = ZSTD_ps_enable;
         rowCCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(&initialParams);
         return MAX(noRowCCtxSize, rowCCtxSize);
     } else {
@@ -1554,18 +1729,18 @@ size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
     RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
     {   ZSTD_compressionParameters const cParams =
                 ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
-        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
+        size_t const blockSize = MIN(ZSTD_resolveMaxBlockSize(params->maxBlockSize), (size_t)1 << cParams.windowLog);
         size_t const inBuffSize = (params->inBufferMode == ZSTD_bm_buffered)
                 ? ((size_t)1 << cParams.windowLog) + blockSize
                 : 0;
         size_t const outBuffSize = (params->outBufferMode == ZSTD_bm_buffered)
                 ? ZSTD_compressBound(blockSize) + 1
                 : 0;
-        ZSTD_useRowMatchFinderMode_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder, ¶ms->cParams);
+        ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params->useRowMatchFinder, ¶ms->cParams);
 
         return ZSTD_estimateCCtxSize_usingCCtxParams_internal(
             &cParams, ¶ms->ldmParams, 1, useRowMatchFinder, inBuffSize, outBuffSize,
-            ZSTD_CONTENTSIZE_UNKNOWN);
+            ZSTD_CONTENTSIZE_UNKNOWN, params->useSequenceProducer, params->maxBlockSize);
     }
 }
 
@@ -1576,9 +1751,9 @@ size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
         /* Pick bigger of not using and using row-based matchfinder for greedy and lazy strategies */
         size_t noRowCCtxSize;
         size_t rowCCtxSize;
-        initialParams.useRowMatchFinder = ZSTD_urm_disableRowMatchFinder;
+        initialParams.useRowMatchFinder = ZSTD_ps_disable;
         noRowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
-        initialParams.useRowMatchFinder = ZSTD_urm_enableRowMatchFinder;
+        initialParams.useRowMatchFinder = ZSTD_ps_enable;
         rowCCtxSize = ZSTD_estimateCStreamSize_usingCCtxParams(&initialParams);
         return MAX(noRowCCtxSize, rowCCtxSize);
     } else {
@@ -1713,7 +1888,7 @@ static size_t
 ZSTD_reset_matchState(ZSTD_matchState_t* ms,
                       ZSTD_cwksp* ws,
                 const ZSTD_compressionParameters* cParams,
-                const ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+                const ZSTD_paramSwitch_e useRowMatchFinder,
                 const ZSTD_compResetPolicy_e crp,
                 const ZSTD_indexResetPolicy_e forceResetIndex,
                 const ZSTD_resetTarget_e forWho)
@@ -1728,7 +1903,7 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
     size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
 
     DEBUGLOG(4, "reset indices : %u", forceResetIndex == ZSTDirp_reset);
-    assert(useRowMatchFinder != ZSTD_urm_auto);
+    assert(useRowMatchFinder != ZSTD_ps_auto);
     if (forceResetIndex == ZSTDirp_reset) {
         ZSTD_window_init(&ms->window);
         ZSTD_cwksp_mark_tables_dirty(ws);
@@ -1774,8 +1949,8 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
             if (ms->tagTable) ZSTD_memset(ms->tagTable, 0, tagTableSize);
         }
         {   /* Switch to 32-entry rows if searchLog is 5 (or more) */
-            U32 const rowLog = cParams->searchLog < 5 ? 4 : 5;
-            assert(cParams->hashLog > rowLog);
+            U32 const rowLog = BOUNDED(4, cParams->searchLog, 6);
+            assert(cParams->hashLog >= rowLog);
             ms->rowHashLog = cParams->hashLog - rowLog;
         }
     }
@@ -1824,8 +1999,8 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                                       ZSTD_buffered_policy_e const zbuff)
 {
     ZSTD_cwksp* const ws = &zc->workspace;
-    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u, useRowMatchFinder=%d",
-                (U32)pledgedSrcSize, params->cParams.windowLog, (int)params->useRowMatchFinder);
+    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u, useRowMatchFinder=%d useBlockSplitter=%d",
+                (U32)pledgedSrcSize, params->cParams.windowLog, (int)params->useRowMatchFinder, (int)params->useBlockSplitter);
     assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
 
     zc->isFirstBlock = 1;
@@ -1836,8 +2011,11 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
     zc->appliedParams = *params;
     params = &zc->appliedParams;
 
-    assert(params->useRowMatchFinder != ZSTD_urm_auto);
-    if (params->ldmParams.enableLdm) {
+    assert(params->useRowMatchFinder != ZSTD_ps_auto);
+    assert(params->useBlockSplitter != ZSTD_ps_auto);
+    assert(params->ldmParams.enableLdm != ZSTD_ps_auto);
+    assert(params->maxBlockSize != 0);
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
         /* Adjust long distance matching parameters */
         ZSTD_ldm_adjustParameters(&zc->appliedParams.ldmParams, ¶ms->cParams);
         assert(params->ldmParams.hashLog >= params->ldmParams.bucketSizeLog);
@@ -1845,9 +2023,8 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
     }
 
     {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params->cParams.windowLog), pledgedSrcSize));
-        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
-        U32    const divider = (params->cParams.minMatch==3) ? 3 : 4;
-        size_t const maxNbSeq = blockSize / divider;
+        size_t const blockSize = MIN(params->maxBlockSize, windowSize);
+        size_t const maxNbSeq = ZSTD_maxNbSeq(blockSize, params->cParams.minMatch, params->useSequenceProducer);
         size_t const buffOutSize = (zbuff == ZSTDb_buffered && params->outBufferMode == ZSTD_bm_buffered)
                 ? ZSTD_compressBound(blockSize) + 1
                 : 0;
@@ -1864,7 +2041,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         size_t const neededSpace =
             ZSTD_estimateCCtxSize_usingCCtxParams_internal(
                 ¶ms->cParams, ¶ms->ldmParams, zc->staticSize != 0, params->useRowMatchFinder,
-                buffInSize, buffOutSize, pledgedSrcSize);
+                buffInSize, buffOutSize, pledgedSrcSize, params->useSequenceProducer, params->maxBlockSize);
         int resizeWorkspace;
 
         FORWARD_IF_ERROR(neededSpace, "cctx size estimate failed!");
@@ -1900,13 +2077,14 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                 zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
                 RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock");
                 zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, ENTROPY_WORKSPACE_SIZE);
-                RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate entropyWorkspace");
+                RETURN_ERROR_IF(zc->entropyWorkspace == NULL, memory_allocation, "couldn't allocate entropyWorkspace");
         }   }
 
         ZSTD_cwksp_clear(ws);
 
         /* init params */
         zc->blockState.matchState.cParams = params->cParams;
+        zc->blockState.matchState.prefetchCDictTables = params->prefetchCDictTables == ZSTD_ps_enable;
         zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
         zc->consumedSrcSize = 0;
         zc->producedCSize = 0;
@@ -1937,7 +2115,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize);
 
         /* ldm bucketOffsets table */
-        if (params->ldmParams.enableLdm) {
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
             /* TODO: avoid memset? */
             size_t const numBuckets =
                   ((size_t)1) << (params->ldmParams.hashLog -
@@ -1964,7 +2142,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             ZSTD_resetTarget_CCtx), "");
 
         /* ldm hash table */
-        if (params->ldmParams.enableLdm) {
+        if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
             /* TODO: avoid memset? */
             size_t const ldmHSize = ((size_t)1) << params->ldmParams.hashLog;
             zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
@@ -1976,8 +2154,16 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             zc->ldmState.loadedDictEnd = 0;
         }
 
-        assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace, resizeWorkspace));
+        /* reserve space for block-level external sequences */
+        if (params->useSequenceProducer) {
+            size_t const maxNbExternalSeq = ZSTD_sequenceBound(blockSize);
+            zc->externalMatchCtx.seqBufferCapacity = maxNbExternalSeq;
+            zc->externalMatchCtx.seqBuffer =
+                (ZSTD_Sequence*)ZSTD_cwksp_reserve_aligned(ws, maxNbExternalSeq * sizeof(ZSTD_Sequence));
+        }
+
         DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
+        assert(ZSTD_cwksp_estimated_space_within_bounds(ws, neededSpace, resizeWorkspace));
 
         zc->initialized = 1;
 
@@ -2049,7 +2235,8 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
         }
 
         params.cParams = ZSTD_adjustCParams_internal(adjusted_cdict_cParams, pledgedSrcSize,
-                                                     cdict->dictContentSize, ZSTD_cpm_attachDict);
+                                                     cdict->dictContentSize, ZSTD_cpm_attachDict,
+                                                     params.useRowMatchFinder);
         params.cParams.windowLog = windowLog;
         params.useRowMatchFinder = cdict->useRowMatchFinder;    /* cdict overrides */
         FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, ¶ms, pledgedSrcSize,
@@ -2088,6 +2275,22 @@ ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
     return 0;
 }
 
+static void ZSTD_copyCDictTableIntoCCtx(U32* dst, U32 const* src, size_t tableSize,
+                                        ZSTD_compressionParameters const* cParams) {
+    if (ZSTD_CDictIndicesAreTagged(cParams)){
+        /* Remove tags from the CDict table if they are present.
+         * See docs on "short cache" in zstd_compress_internal.h for context. */
+        size_t i;
+        for (i = 0; i < tableSize; i++) {
+            U32 const taggedIndex = src[i];
+            U32 const index = taggedIndex >> ZSTD_SHORT_CACHE_TAG_BITS;
+            dst[i] = index;
+        }
+    } else {
+        ZSTD_memcpy(dst, src, tableSize * sizeof(U32));
+    }
+}
+
 static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
                             const ZSTD_CDict* cdict,
                             ZSTD_CCtx_params params,
@@ -2115,7 +2318,7 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
     }
 
     ZSTD_cwksp_mark_tables_dirty(&cctx->workspace);
-    assert(params.useRowMatchFinder != ZSTD_urm_auto);
+    assert(params.useRowMatchFinder != ZSTD_ps_auto);
 
     /* copy tables */
     {   size_t const chainSize = ZSTD_allocateChainTable(cdict_cParams->strategy, cdict->useRowMatchFinder, 0 /* DDS guaranteed disabled */)
@@ -2123,14 +2326,15 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
                                                             : 0;
         size_t const hSize =  (size_t)1 << cdict_cParams->hashLog;
 
-        ZSTD_memcpy(cctx->blockState.matchState.hashTable,
-               cdict->matchState.hashTable,
-               hSize * sizeof(U32));
+        ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.hashTable,
+                                cdict->matchState.hashTable,
+                                hSize, cdict_cParams);
+
         /* Do not copy cdict's chainTable if cctx has parameters such that it would not use chainTable */
         if (ZSTD_allocateChainTable(cctx->appliedParams.cParams.strategy, cctx->appliedParams.useRowMatchFinder, 0 /* forDDSDict */)) {
-            ZSTD_memcpy(cctx->blockState.matchState.chainTable,
-               cdict->matchState.chainTable,
-               chainSize * sizeof(U32));
+            ZSTD_copyCDictTableIntoCCtx(cctx->blockState.matchState.chainTable,
+                                    cdict->matchState.chainTable,
+                                    chainSize, cdict_cParams);
         }
         /* copy tag table */
         if (ZSTD_rowMatchFinderUsed(cdict_cParams->strategy, cdict->useRowMatchFinder)) {
@@ -2209,9 +2413,14 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
     {   ZSTD_CCtx_params params = dstCCtx->requestedParams;
         /* Copy only compression parameters related to tables. */
         params.cParams = srcCCtx->appliedParams.cParams;
-        assert(srcCCtx->appliedParams.useRowMatchFinder != ZSTD_urm_auto);
+        assert(srcCCtx->appliedParams.useRowMatchFinder != ZSTD_ps_auto);
+        assert(srcCCtx->appliedParams.useBlockSplitter != ZSTD_ps_auto);
+        assert(srcCCtx->appliedParams.ldmParams.enableLdm != ZSTD_ps_auto);
         params.useRowMatchFinder = srcCCtx->appliedParams.useRowMatchFinder;
+        params.useBlockSplitter = srcCCtx->appliedParams.useBlockSplitter;
+        params.ldmParams = srcCCtx->appliedParams.ldmParams;
         params.fParams = fParams;
+        params.maxBlockSize = srcCCtx->appliedParams.maxBlockSize;
         ZSTD_resetCCtx_internal(dstCCtx, ¶ms, pledgedSrcSize,
                                 /* loadedDictSize */ 0,
                                 ZSTDcrp_leaveDirty, zbuff);
@@ -2296,6 +2505,8 @@ ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerVa
     int const nbRows = (int)size / ZSTD_ROWSIZE;
     int cellNb = 0;
     int rowNb;
+    /* Protect special index values < ZSTD_WINDOW_START_INDEX. */
+    U32 const reducerThreshold = reducerValue + ZSTD_WINDOW_START_INDEX;
     assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
     assert(size < (1U<<31));   /* can be casted to int */
 
@@ -2315,12 +2526,17 @@ ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerVa
     for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
         int column;
         for (column=0; columnsequencesStart;
     BYTE* const llCodeTable = seqStorePtr->llCode;
@@ -2372,18 +2588,24 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
     BYTE* const mlCodeTable = seqStorePtr->mlCode;
     U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     U32 u;
+    int longOffsets = 0;
     assert(nbSeq <= seqStorePtr->maxNbSeq);
     for (u=0; u= STREAM_ACCUMULATOR_MIN));
+        if (MEM_32bits() && ofCode >= STREAM_ACCUMULATOR_MIN)
+            longOffsets = 1;
     }
     if (seqStorePtr->longLengthType==ZSTD_llt_literalLength)
         llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
     if (seqStorePtr->longLengthType==ZSTD_llt_matchLength)
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
+    return longOffsets;
 }
 
 /* ZSTD_useTargetCBlockSize():
@@ -2399,11 +2621,13 @@ static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
 /* ZSTD_blockSplitterEnabled():
  * Returns if block splitting param is being used
  * If used, compression will do best effort to split a block in order to improve compression ratio.
+ * At the time this function is called, the parameter must be finalized.
  * Returns 1 if true, 0 otherwise. */
 static int ZSTD_blockSplitterEnabled(ZSTD_CCtx_params* cctxParams)
 {
-    DEBUGLOG(5, "ZSTD_blockSplitterEnabled(splitBlocks=%d)", cctxParams->splitBlocks);
-    return (cctxParams->splitBlocks != 0);
+    DEBUGLOG(5, "ZSTD_blockSplitterEnabled (useBlockSplitter=%d)", cctxParams->useBlockSplitter);
+    assert(cctxParams->useBlockSplitter != ZSTD_ps_auto);
+    return (cctxParams->useBlockSplitter == ZSTD_ps_enable);
 }
 
 /* Type returned by ZSTD_buildSequencesStatistics containing finalized symbol encoding types
@@ -2415,6 +2639,7 @@ typedef struct {
     U32 MLtype;
     size_t size;
     size_t lastCountSize; /* Accounts for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
+    int longOffsets;
 } ZSTD_symbolEncodingTypeStats_t;
 
 /* ZSTD_buildSequencesStatistics():
@@ -2425,11 +2650,13 @@ typedef struct {
  * entropyWkspSize must be of size at least ENTROPY_WORKSPACE_SIZE - (MaxSeq + 1)*sizeof(U32)
  */
 static ZSTD_symbolEncodingTypeStats_t
-ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
-                        const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
-                              BYTE* dst, const BYTE* const dstEnd,
-                              ZSTD_strategy strategy, unsigned* countWorkspace,
-                              void* entropyWorkspace, size_t entropyWkspSize) {
+ZSTD_buildSequencesStatistics(
+                const seqStore_t* seqStorePtr, size_t nbSeq,
+                const ZSTD_fseCTables_t* prevEntropy, ZSTD_fseCTables_t* nextEntropy,
+                      BYTE* dst, const BYTE* const dstEnd,
+                      ZSTD_strategy strategy, unsigned* countWorkspace,
+                      void* entropyWorkspace, size_t entropyWkspSize)
+{
     BYTE* const ostart = dst;
     const BYTE* const oend = dstEnd;
     BYTE* op = ostart;
@@ -2443,7 +2670,7 @@ ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
 
     stats.lastCountSize = 0;
     /* convert length/distances into codes */
-    ZSTD_seqToCodes(seqStorePtr);
+    stats.longOffsets = ZSTD_seqToCodes(seqStorePtr);
     assert(op <= oend);
     assert(nbSeq != 0); /* ZSTD_selectEncodingType() divides by nbSeq */
     /* build CTable for Literal Lengths */
@@ -2546,23 +2773,24 @@ ZSTD_buildSequencesStatistics(seqStore_t* seqStorePtr, size_t nbSeq,
  * compresses both literals and sequences
  * Returns compressed size of block, or a zstd error.
  */
+#define SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO 20
 MEM_STATIC size_t
-ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
-                          const ZSTD_entropyCTables_t* prevEntropy,
-                                ZSTD_entropyCTables_t* nextEntropy,
-                          const ZSTD_CCtx_params* cctxParams,
-                                void* dst, size_t dstCapacity,
-                                void* entropyWorkspace, size_t entropyWkspSize,
-                          const int bmi2)
+ZSTD_entropyCompressSeqStore_internal(
+                        const seqStore_t* seqStorePtr,
+                        const ZSTD_entropyCTables_t* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                        const ZSTD_CCtx_params* cctxParams,
+                              void* dst, size_t dstCapacity,
+                              void* entropyWorkspace, size_t entropyWkspSize,
+                        const int bmi2)
 {
-    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
     ZSTD_strategy const strategy = cctxParams->cParams.strategy;
     unsigned* count = (unsigned*)entropyWorkspace;
     FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
     FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
     FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
     const seqDef* const sequences = seqStorePtr->sequencesStart;
-    const size_t nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    const size_t nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     const BYTE* const ofCodeTable = seqStorePtr->ofCode;
     const BYTE* const llCodeTable = seqStorePtr->llCode;
     const BYTE* const mlCodeTable = seqStorePtr->mlCode;
@@ -2570,25 +2798,31 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
     BYTE* const oend = ostart + dstCapacity;
     BYTE* op = ostart;
     size_t lastCountSize;
+    int longOffsets = 0;
 
     entropyWorkspace = count + (MaxSeq + 1);
     entropyWkspSize -= (MaxSeq + 1) * sizeof(*count);
 
-    DEBUGLOG(4, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu)", nbSeq);
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore_internal (nbSeq=%zu, dstCapacity=%zu)", nbSeq, dstCapacity);
     ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<= HUF_WORKSPACE_SIZE);
 
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
+        size_t const numSequences = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+        size_t const numLiterals = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
+        /* Base suspicion of uncompressibility on ratio of literals to sequences */
+        unsigned const suspectUncompressible = (numSequences == 0) || (numLiterals / numSequences >= SUSPECT_UNCOMPRESSIBLE_LITERAL_RATIO);
         size_t const litSize = (size_t)(seqStorePtr->lit - literals);
+
         size_t const cSize = ZSTD_compressLiterals(
-                                    &prevEntropy->huf, &nextEntropy->huf,
-                                    cctxParams->cParams.strategy,
-                                    ZSTD_disableLiteralsCompression(cctxParams),
                                     op, dstCapacity,
                                     literals, litSize,
                                     entropyWorkspace, entropyWkspSize,
-                                    bmi2);
+                                    &prevEntropy->huf, &nextEntropy->huf,
+                                    cctxParams->cParams.strategy,
+                                    ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                    suspectUncompressible, bmi2);
         FORWARD_IF_ERROR(cSize, "ZSTD_compressLiterals failed");
         assert(cSize <= dstCapacity);
         op += cSize;
@@ -2614,11 +2848,10 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
         ZSTD_memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
         return (size_t)(op - ostart);
     }
-    {
-        ZSTD_symbolEncodingTypeStats_t stats;
-        BYTE* seqHead = op++;
+    {   BYTE* const seqHead = op++;
         /* build stats for sequences */
-        stats = ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
+        const ZSTD_symbolEncodingTypeStats_t stats =
+                ZSTD_buildSequencesStatistics(seqStorePtr, nbSeq,
                                              &prevEntropy->fse, &nextEntropy->fse,
                                               op, oend,
                                               strategy, count,
@@ -2627,6 +2860,7 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
         *seqHead = (BYTE)((stats.LLtype<<6) + (stats.Offtype<<4) + (stats.MLtype<<2));
         lastCountSize = stats.lastCountSize;
         op += stats.size;
+        longOffsets = stats.longOffsets;
     }
 
     {   size_t const bitstreamSize = ZSTD_encodeSequences(
@@ -2661,14 +2895,15 @@ ZSTD_entropyCompressSeqStore_internal(seqStore_t* seqStorePtr,
 }
 
 MEM_STATIC size_t
-ZSTD_entropyCompressSeqStore(seqStore_t* seqStorePtr,
-                       const ZSTD_entropyCTables_t* prevEntropy,
-                             ZSTD_entropyCTables_t* nextEntropy,
-                       const ZSTD_CCtx_params* cctxParams,
-                             void* dst, size_t dstCapacity,
-                             size_t srcSize,
-                             void* entropyWorkspace, size_t entropyWkspSize,
-                             int bmi2)
+ZSTD_entropyCompressSeqStore(
+                    const seqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          void* dst, size_t dstCapacity,
+                          size_t srcSize,
+                          void* entropyWorkspace, size_t entropyWkspSize,
+                          int bmi2)
 {
     size_t const cSize = ZSTD_entropyCompressSeqStore_internal(
                             seqStorePtr, prevEntropy, nextEntropy, cctxParams,
@@ -2678,22 +2913,28 @@ ZSTD_entropyCompressSeqStore(seqStore_t* seqStorePtr,
     /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
      * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
      */
-    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
+    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity)) {
+        DEBUGLOG(4, "not enough dstCapacity (%zu) for ZSTD_entropyCompressSeqStore_internal()=> do not compress block", dstCapacity);
         return 0;  /* block not compressed */
+    }
     FORWARD_IF_ERROR(cSize, "ZSTD_entropyCompressSeqStore_internal failed");
 
     /* Check compressibility */
     {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
         if (cSize >= maxCSize) return 0;  /* block not compressed */
     }
-    DEBUGLOG(4, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
+    DEBUGLOG(5, "ZSTD_entropyCompressSeqStore() cSize: %zu", cSize);
+    /* libzstd decoder before  > v1.5.4 is not compatible with compressed blocks of size ZSTD_BLOCKSIZE_MAX exactly.
+     * This restriction is indirectly already fulfilled by respecting ZSTD_minGain() condition above.
+     */
+    assert(cSize < ZSTD_BLOCKSIZE_MAX);
     return cSize;
 }
 
 /* ZSTD_selectBlockCompressor() :
  * Not static, but internal use only (used by long distance matcher)
  * assumption : strat is a valid strategy */
-ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRowMatchFinderMode_e useRowMatchFinder, ZSTD_dictMode_e dictMode)
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e useRowMatchFinder, ZSTD_dictMode_e dictMode)
 {
     static const ZSTD_blockCompressor blockCompressor[4][ZSTD_STRATEGY_MAX+1] = {
         { ZSTD_compressBlock_fast  /* default for 0 */,
@@ -2758,7 +2999,7 @@ ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRow
             ZSTD_compressBlock_lazy2_dedicatedDictSearch_row }
         };
         DEBUGLOG(4, "Selecting a row-based matchfinder");
-        assert(useRowMatchFinder != ZSTD_urm_auto);
+        assert(useRowMatchFinder != ZSTD_ps_auto);
         selectedCompressor = rowBasedBlockCompressors[(int)dictMode][(int)strat - (int)ZSTD_greedy];
     } else {
         selectedCompressor = blockCompressor[(int)dictMode][(int)strat];
@@ -2781,6 +3022,72 @@ void ZSTD_resetSeqStore(seqStore_t* ssPtr)
     ssPtr->longLengthType = ZSTD_llt_none;
 }
 
+/* ZSTD_postProcessSequenceProducerResult() :
+ * Validates and post-processes sequences obtained through the external matchfinder API:
+ *   - Checks whether nbExternalSeqs represents an error condition.
+ *   - Appends a block delimiter to outSeqs if one is not already present.
+ *     See zstd.h for context regarding block delimiters.
+ * Returns the number of sequences after post-processing, or an error code. */
+static size_t ZSTD_postProcessSequenceProducerResult(
+    ZSTD_Sequence* outSeqs, size_t nbExternalSeqs, size_t outSeqsCapacity, size_t srcSize
+) {
+    RETURN_ERROR_IF(
+        nbExternalSeqs > outSeqsCapacity,
+        sequenceProducer_failed,
+        "External sequence producer returned error code %lu",
+        (unsigned long)nbExternalSeqs
+    );
+
+    RETURN_ERROR_IF(
+        nbExternalSeqs == 0 && srcSize > 0,
+        sequenceProducer_failed,
+        "Got zero sequences from external sequence producer for a non-empty src buffer!"
+    );
+
+    if (srcSize == 0) {
+        ZSTD_memset(&outSeqs[0], 0, sizeof(ZSTD_Sequence));
+        return 1;
+    }
+
+    {
+        ZSTD_Sequence const lastSeq = outSeqs[nbExternalSeqs - 1];
+
+        /* We can return early if lastSeq is already a block delimiter. */
+        if (lastSeq.offset == 0 && lastSeq.matchLength == 0) {
+            return nbExternalSeqs;
+        }
+
+        /* This error condition is only possible if the external matchfinder
+         * produced an invalid parse, by definition of ZSTD_sequenceBound(). */
+        RETURN_ERROR_IF(
+            nbExternalSeqs == outSeqsCapacity,
+            sequenceProducer_failed,
+            "nbExternalSeqs == outSeqsCapacity but lastSeq is not a block delimiter!"
+        );
+
+        /* lastSeq is not a block delimiter, so we need to append one. */
+        ZSTD_memset(&outSeqs[nbExternalSeqs], 0, sizeof(ZSTD_Sequence));
+        return nbExternalSeqs + 1;
+    }
+}
+
+/* ZSTD_fastSequenceLengthSum() :
+ * Returns sum(litLen) + sum(matchLen) + lastLits for *seqBuf*.
+ * Similar to another function in zstd_compress.c (determine_blockSize),
+ * except it doesn't check for a block delimiter to end summation.
+ * Removing the early exit allows the compiler to auto-vectorize (https://godbolt.org/z/cY1cajz9P).
+ * This function can be deleted and replaced by determine_blockSize after we resolve issue #3456. */
+static size_t ZSTD_fastSequenceLengthSum(ZSTD_Sequence const* seqBuf, size_t seqBufSize) {
+    size_t matchLenSum, litLenSum, i;
+    matchLenSum = 0;
+    litLenSum = 0;
+    for (i = 0; i < seqBufSize; i++) {
+        litLenSum += seqBuf[i].litLength;
+        matchLenSum += seqBuf[i].matchLength;
+    }
+    return litLenSum + matchLenSum;
+}
+
 typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_buildSeqStore_e;
 
 static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
@@ -2790,7 +3097,9 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
     assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
     /* Assert that we have correctly flushed the ctx params into the ms's copy */
     ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
-    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+    /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+     * additional 1. We need to revisit and change this logic to be more consistent */
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
         if (zc->appliedParams.cParams.strategy >= ZSTD_btopt) {
             ZSTD_ldm_skipRawSeqStoreBytes(&zc->externSeqStore, srcSize);
         } else {
@@ -2825,7 +3134,16 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
                 zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
         }
         if (zc->externSeqStore.pos < zc->externSeqStore.size) {
-            assert(!zc->appliedParams.ldmParams.enableLdm);
+            assert(zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_disable);
+
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                zc->appliedParams.useSequenceProducer,
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
             /* Updates ldmSeqStore.pos */
             lastLLSize =
                 ZSTD_ldm_blockCompress(&zc->externSeqStore,
@@ -2834,9 +3152,17 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
                                        zc->appliedParams.useRowMatchFinder,
                                        src, srcSize);
             assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
-        } else if (zc->appliedParams.ldmParams.enableLdm) {
+        } else if (zc->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
             rawSeqStore_t ldmSeqStore = kNullRawSeqStore;
 
+            /* External matchfinder + LDM is technically possible, just not implemented yet.
+             * We need to revisit soon and implement it. */
+            RETURN_ERROR_IF(
+                zc->appliedParams.useSequenceProducer,
+                parameter_combination_unsupported,
+                "Long-distance matching with external sequence producer enabled is not currently supported."
+            );
+
             ldmSeqStore.seq = zc->ldmSequences;
             ldmSeqStore.capacity = zc->maxNbLdmSequences;
             /* Updates ldmSeqStore.size */
@@ -2851,7 +3177,68 @@ static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
                                        zc->appliedParams.useRowMatchFinder,
                                        src, srcSize);
             assert(ldmSeqStore.pos == ldmSeqStore.size);
-        } else {   /* not long range mode */
+        } else if (zc->appliedParams.useSequenceProducer) {
+            assert(
+                zc->externalMatchCtx.seqBufferCapacity >= ZSTD_sequenceBound(srcSize)
+            );
+            assert(zc->externalMatchCtx.mFinder != NULL);
+
+            {   U32 const windowSize = (U32)1 << zc->appliedParams.cParams.windowLog;
+
+                size_t const nbExternalSeqs = (zc->externalMatchCtx.mFinder)(
+                    zc->externalMatchCtx.mState,
+                    zc->externalMatchCtx.seqBuffer,
+                    zc->externalMatchCtx.seqBufferCapacity,
+                    src, srcSize,
+                    NULL, 0,  /* dict and dictSize, currently not supported */
+                    zc->appliedParams.compressionLevel,
+                    windowSize
+                );
+
+                size_t const nbPostProcessedSeqs = ZSTD_postProcessSequenceProducerResult(
+                    zc->externalMatchCtx.seqBuffer,
+                    nbExternalSeqs,
+                    zc->externalMatchCtx.seqBufferCapacity,
+                    srcSize
+                );
+
+                /* Return early if there is no error, since we don't need to worry about last literals */
+                if (!ZSTD_isError(nbPostProcessedSeqs)) {
+                    ZSTD_sequencePosition seqPos = {0,0,0};
+                    size_t const seqLenSum = ZSTD_fastSequenceLengthSum(zc->externalMatchCtx.seqBuffer, nbPostProcessedSeqs);
+                    RETURN_ERROR_IF(seqLenSum > srcSize, externalSequences_invalid, "External sequences imply too large a block!");
+                    FORWARD_IF_ERROR(
+                        ZSTD_copySequencesToSeqStoreExplicitBlockDelim(
+                            zc, &seqPos,
+                            zc->externalMatchCtx.seqBuffer, nbPostProcessedSeqs,
+                            src, srcSize,
+                            zc->appliedParams.searchForExternalRepcodes
+                        ),
+                        "Failed to copy external sequences to seqStore!"
+                    );
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(5, "Copied %lu sequences from external sequence producer to internal seqStore.", (unsigned long)nbExternalSeqs);
+                    return ZSTDbss_compress;
+                }
+
+                /* Propagate the error if fallback is disabled */
+                if (!zc->appliedParams.enableMatchFinderFallback) {
+                    return nbPostProcessedSeqs;
+                }
+
+                /* Fallback to software matchfinder */
+                {   ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy,
+                                                                                            zc->appliedParams.useRowMatchFinder,
+                                                                                            dictMode);
+                    ms->ldmSeqStore = NULL;
+                    DEBUGLOG(
+                        5,
+                        "External sequence producer returned error code %lu. Falling back to internal parser.",
+                        (unsigned long)nbExternalSeqs
+                    );
+                    lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+            }   }
+        } else {   /* not long range mode and no external matchfinder */
             ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy,
                                                                                     zc->appliedParams.useRowMatchFinder,
                                                                                     dictMode);
@@ -2882,9 +3269,9 @@ static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
     assert(zc->seqCollector.maxSequences >= seqStoreSeqSize + 1);
     ZSTD_memcpy(updatedRepcodes.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
     for (i = 0; i < seqStoreSeqSize; ++i) {
-        U32 rawOffset = seqStoreSeqs[i].offset - ZSTD_REP_NUM;
+        U32 rawOffset = seqStoreSeqs[i].offBase - ZSTD_REP_NUM;
         outSeqs[i].litLength = seqStoreSeqs[i].litLength;
-        outSeqs[i].matchLength = seqStoreSeqs[i].matchLength + MINMATCH;
+        outSeqs[i].matchLength = seqStoreSeqs[i].mlBase + MINMATCH;
         outSeqs[i].rep = 0;
 
         if (i == seqStore->longLengthPos) {
@@ -2895,9 +3282,9 @@ static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
             }
         }
 
-        if (seqStoreSeqs[i].offset <= ZSTD_REP_NUM) {
+        if (seqStoreSeqs[i].offBase <= ZSTD_REP_NUM) {
             /* Derive the correct offset corresponding to a repcode */
-            outSeqs[i].rep = seqStoreSeqs[i].offset;
+            outSeqs[i].rep = seqStoreSeqs[i].offBase;
             if (outSeqs[i].litLength != 0) {
                 rawOffset = updatedRepcodes.rep[outSeqs[i].rep - 1];
             } else {
@@ -2911,9 +3298,9 @@ static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
         outSeqs[i].offset = rawOffset;
         /* seqStoreSeqs[i].offset == offCode+1, and ZSTD_updateRep() expects offCode
            so we provide seqStoreSeqs[i].offset - 1 */
-        updatedRepcodes = ZSTD_updateRep(updatedRepcodes.rep,
-                                         seqStoreSeqs[i].offset - 1,
-                                         seqStoreSeqs[i].litLength == 0);
+        ZSTD_updateRep(updatedRepcodes.rep,
+                       seqStoreSeqs[i].offBase,
+                       seqStoreSeqs[i].litLength == 0);
         literalsRead += outSeqs[i].litLength;
     }
     /* Insert last literals (if any exist) in the block as a sequence with ml == off == 0.
@@ -2928,6 +3315,10 @@ static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
     zc->seqCollector.seqIndex += seqStoreSeqSize;
 }
 
+size_t ZSTD_sequenceBound(size_t srcSize) {
+    return (srcSize / ZSTD_MINMATCH_MIN) + 1;
+}
+
 size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
                               size_t outSeqsSize, const void* src, size_t srcSize)
 {
@@ -2973,19 +3364,17 @@ static int ZSTD_isRLE(const BYTE* src, size_t length) {
     const size_t unrollMask = unrollSize - 1;
     const size_t prefixLength = length & unrollMask;
     size_t i;
-    size_t u;
     if (length == 1) return 1;
     /* Check if prefix is RLE first before using unrolled loop */
     if (prefixLength && ZSTD_count(ip+1, ip, ip+prefixLength) != prefixLength-1) {
         return 0;
     }
     for (i = prefixLength; i != length; i += unrollSize) {
+        size_t u;
         for (u = 0; u < unrollSize; u += sizeof(size_t)) {
             if (MEM_readST(ip + i + u) != valueST) {
                 return 0;
-            }
-        }
-    }
+    }   }   }
     return 1;
 }
 
@@ -3001,7 +3390,8 @@ static int ZSTD_maybeRLE(seqStore_t const* seqStore)
     return nbSeqs < 4 && nbLits < 10;
 }
 
-static void ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* const bs)
+static void
+ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* const bs)
 {
     ZSTD_compressedBlockState_t* const tmp = bs->prevCBlock;
     bs->prevCBlock = bs->nextCBlock;
@@ -3009,7 +3399,9 @@ static void ZSTD_blockState_confirmRepcodesAndEntropyTables(ZSTD_blockState_t* c
 }
 
 /* Writes the block header */
-static void writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock) {
+static void
+writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastBlock)
+{
     U32 const cBlockHeader = cSize == 1 ?
                         lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
                         lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
@@ -3022,13 +3414,16 @@ static void writeBlockHeader(void* op, size_t cSize, size_t blockSize, U32 lastB
  *  Stores literals block type (raw, rle, compressed, repeat) and
  *  huffman description table to hufMetadata.
  *  Requires ENTROPY_WORKSPACE_SIZE workspace
- *  @return : size of huffman description table or error code */
-static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
-                                            const ZSTD_hufCTables_t* prevHuf,
-                                                  ZSTD_hufCTables_t* nextHuf,
-                                                  ZSTD_hufCTablesMetadata_t* hufMetadata,
-                                                  const int disableLiteralsCompression,
-                                                  void* workspace, size_t wkspSize)
+ * @return : size of huffman description table, or an error code
+ */
+static size_t
+ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSize,
+                               const ZSTD_hufCTables_t* prevHuf,
+                                     ZSTD_hufCTables_t* nextHuf,
+                                     ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               const int literalsCompressionIsDisabled,
+                                     void* workspace, size_t wkspSize,
+                                     int hufFlags)
 {
     BYTE* const wkspStart = (BYTE*)workspace;
     BYTE* const wkspEnd = wkspStart + wkspSize;
@@ -3036,16 +3431,16 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
     unsigned* const countWksp = (unsigned*)workspace;
     const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
     BYTE* const nodeWksp = countWkspStart + countWkspSize;
-    const size_t nodeWkspSize = wkspEnd-nodeWksp;
+    const size_t nodeWkspSize = (size_t)(wkspEnd - nodeWksp);
     unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
-    unsigned huffLog = HUF_TABLELOG_DEFAULT;
+    unsigned huffLog = LitHufLog;
     HUF_repeat repeat = prevHuf->repeatMode;
     DEBUGLOG(5, "ZSTD_buildBlockEntropyStats_literals (srcSize=%zu)", srcSize);
 
     /* Prepare nextEntropy assuming reusing the existing table */
     ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
 
-    if (disableLiteralsCompression) {
+    if (literalsCompressionIsDisabled) {
         DEBUGLOG(5, "set_basic - disabled");
         hufMetadata->hType = set_basic;
         return 0;
@@ -3053,73 +3448,77 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
 
     /* small ? don't even attempt compression (speed opt) */
 #ifndef COMPRESS_LITERALS_SIZE_MIN
-#define COMPRESS_LITERALS_SIZE_MIN 63
+# define COMPRESS_LITERALS_SIZE_MIN 63  /* heuristic */
 #endif
     {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
         if (srcSize <= minLitSize) {
             DEBUGLOG(5, "set_basic - too small");
             hufMetadata->hType = set_basic;
             return 0;
-        }
-    }
+    }   }
 
     /* Scan input and build symbol stats */
-    {   size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)src, srcSize, workspace, wkspSize);
+    {   size_t const largest =
+            HIST_count_wksp (countWksp, &maxSymbolValue,
+                            (const BYTE*)src, srcSize,
+                            workspace, wkspSize);
         FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
         if (largest == srcSize) {
+            /* only one literal symbol */
             DEBUGLOG(5, "set_rle");
             hufMetadata->hType = set_rle;
             return 0;
         }
         if (largest <= (srcSize >> 7)+4) {
+            /* heuristic: likely not compressible */
             DEBUGLOG(5, "set_basic - no gain");
             hufMetadata->hType = set_basic;
             return 0;
-        }
-    }
+    }   }
 
     /* Validate the previous Huffman table */
-    if (repeat == HUF_repeat_check && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
+    if (repeat == HUF_repeat_check
+      && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
         repeat = HUF_repeat_none;
     }
 
     /* Build Huffman Tree */
     ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
-    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue, nodeWksp, nodeWkspSize, nextHuf->CTable, countWksp, hufFlags);
+    assert(huffLog <= LitHufLog);
     {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
                                                     maxSymbolValue, huffLog,
                                                     nodeWksp, nodeWkspSize);
         FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
         huffLog = (U32)maxBits;
-        {   /* Build and write the CTable */
-            size_t const newCSize = HUF_estimateCompressedSize(
-                    (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
-            size_t const hSize = HUF_writeCTable_wksp(
-                    hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
-                    (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
-                    nodeWksp, nodeWkspSize);
-            /* Check against repeating the previous CTable */
-            if (repeat != HUF_repeat_none) {
-                size_t const oldCSize = HUF_estimateCompressedSize(
-                        (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
-                if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
-                    DEBUGLOG(5, "set_repeat - smaller");
-                    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-                    hufMetadata->hType = set_repeat;
-                    return 0;
-                }
-            }
-            if (newCSize + hSize >= srcSize) {
-                DEBUGLOG(5, "set_basic - no gains");
+    }
+    {   /* Build and write the CTable */
+        size_t const newCSize = HUF_estimateCompressedSize(
+                (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
+        size_t const hSize = HUF_writeCTable_wksp(
+                hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
+                (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog,
+                nodeWksp, nodeWkspSize);
+        /* Check against repeating the previous CTable */
+        if (repeat != HUF_repeat_none) {
+            size_t const oldCSize = HUF_estimateCompressedSize(
+                    (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
+            if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
+                DEBUGLOG(5, "set_repeat - smaller");
                 ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-                hufMetadata->hType = set_basic;
+                hufMetadata->hType = set_repeat;
                 return 0;
-            }
-            DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
-            hufMetadata->hType = set_compressed;
-            nextHuf->repeatMode = HUF_repeat_check;
-            return hSize;
+        }   }
+        if (newCSize + hSize >= srcSize) {
+            DEBUGLOG(5, "set_basic - no gains");
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            hufMetadata->hType = set_basic;
+            return 0;
         }
+        DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
+        hufMetadata->hType = set_compressed;
+        nextHuf->repeatMode = HUF_repeat_check;
+        return hSize;
     }
 }
 
@@ -3129,8 +3528,9 @@ static size_t ZSTD_buildBlockEntropyStats_literals(void* const src, size_t srcSi
  * and updates nextEntropy to the appropriate repeatMode.
  */
 static ZSTD_symbolEncodingTypeStats_t
-ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy) {
-    ZSTD_symbolEncodingTypeStats_t stats = {set_basic, set_basic, set_basic, 0, 0};
+ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy)
+{
+    ZSTD_symbolEncodingTypeStats_t stats = {set_basic, set_basic, set_basic, 0, 0, 0};
     nextEntropy->litlength_repeatMode = FSE_repeat_none;
     nextEntropy->offcode_repeatMode = FSE_repeat_none;
     nextEntropy->matchlength_repeatMode = FSE_repeat_none;
@@ -3141,16 +3541,18 @@ ZSTD_buildDummySequencesStatistics(ZSTD_fseCTables_t* nextEntropy) {
  *  Builds entropy for the sequences.
  *  Stores symbol compression modes and fse table to fseMetadata.
  *  Requires ENTROPY_WORKSPACE_SIZE wksp.
- *  @return : size of fse tables or error code */
-static size_t ZSTD_buildBlockEntropyStats_sequences(seqStore_t* seqStorePtr,
-                                              const ZSTD_fseCTables_t* prevEntropy,
-                                                    ZSTD_fseCTables_t* nextEntropy,
-                                              const ZSTD_CCtx_params* cctxParams,
-                                                    ZSTD_fseCTablesMetadata_t* fseMetadata,
-                                                    void* workspace, size_t wkspSize)
+ * @return : size of fse tables or error code */
+static size_t
+ZSTD_buildBlockEntropyStats_sequences(
+                const seqStore_t* seqStorePtr,
+                const ZSTD_fseCTables_t* prevEntropy,
+                      ZSTD_fseCTables_t* nextEntropy,
+                const ZSTD_CCtx_params* cctxParams,
+                      ZSTD_fseCTablesMetadata_t* fseMetadata,
+                      void* workspace, size_t wkspSize)
 {
     ZSTD_strategy const strategy = cctxParams->cParams.strategy;
-    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     BYTE* const ostart = fseMetadata->fseTablesBuffer;
     BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
     BYTE* op = ostart;
@@ -3177,23 +3579,28 @@ static size_t ZSTD_buildBlockEntropyStats_sequences(seqStore_t* seqStorePtr,
 /** ZSTD_buildBlockEntropyStats() :
  *  Builds entropy for the block.
  *  Requires workspace size ENTROPY_WORKSPACE_SIZE
- *
- *  @return : 0 on success or error code
+ * @return : 0 on success, or an error code
+ *  Note : also employed in superblock
  */
-size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
-                             const ZSTD_entropyCTables_t* prevEntropy,
-                                   ZSTD_entropyCTables_t* nextEntropy,
-                             const ZSTD_CCtx_params* cctxParams,
-                                   ZSTD_entropyCTablesMetadata_t* entropyMetadata,
-                                   void* workspace, size_t wkspSize)
-{
-    size_t const litSize = seqStorePtr->lit - seqStorePtr->litStart;
+size_t ZSTD_buildBlockEntropyStats(
+            const seqStore_t* seqStorePtr,
+            const ZSTD_entropyCTables_t* prevEntropy,
+                  ZSTD_entropyCTables_t* nextEntropy,
+            const ZSTD_CCtx_params* cctxParams,
+                  ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                  void* workspace, size_t wkspSize)
+{
+    size_t const litSize = (size_t)(seqStorePtr->lit - seqStorePtr->litStart);
+    int const huf_useOptDepth = (cctxParams->cParams.strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD);
+    int const hufFlags = huf_useOptDepth ? HUF_flags_optimalDepth : 0;
+
     entropyMetadata->hufMetadata.hufDesSize =
         ZSTD_buildBlockEntropyStats_literals(seqStorePtr->litStart, litSize,
                                             &prevEntropy->huf, &nextEntropy->huf,
                                             &entropyMetadata->hufMetadata,
-                                            ZSTD_disableLiteralsCompression(cctxParams),
-                                            workspace, wkspSize);
+                                            ZSTD_literalsCompressionIsDisabled(cctxParams),
+                                            workspace, wkspSize, hufFlags);
+
     FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildBlockEntropyStats_literals failed");
     entropyMetadata->fseMetadata.fseTablesSize =
         ZSTD_buildBlockEntropyStats_sequences(seqStorePtr,
@@ -3206,11 +3613,12 @@ size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
 }
 
 /* Returns the size estimate for the literals section (header + content) of a block */
-static size_t ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
-                                                const ZSTD_hufCTables_t* huf,
-                                                const ZSTD_hufCTablesMetadata_t* hufMetadata,
-                                                void* workspace, size_t wkspSize,
-                                                int writeEntropy)
+static size_t
+ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSize,
+                               const ZSTD_hufCTables_t* huf,
+                               const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                               void* workspace, size_t wkspSize,
+                               int writeEntropy)
 {
     unsigned* const countWksp = (unsigned*)workspace;
     unsigned maxSymbolValue = HUF_SYMBOLVALUE_MAX;
@@ -3232,12 +3640,13 @@ static size_t ZSTD_estimateBlockSize_literal(const BYTE* literals, size_t litSiz
 }
 
 /* Returns the size estimate for the FSE-compressed symbols (of, ml, ll) of a block */
-static size_t ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
-                        const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
-                        const FSE_CTable* fseCTable,
-                        const U32* additionalBits,
-                        short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
-                        void* workspace, size_t wkspSize)
+static size_t
+ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
+                    const BYTE* codeTable, size_t nbSeq, unsigned maxCode,
+                    const FSE_CTable* fseCTable,
+                    const U8* additionalBits,
+                    short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                    void* workspace, size_t wkspSize)
 {
     unsigned* const countWksp = (unsigned*)workspace;
     const BYTE* ctp = codeTable;
@@ -3269,98 +3678,107 @@ static size_t ZSTD_estimateBlockSize_symbolType(symbolEncodingType_e type,
 }
 
 /* Returns the size estimate for the sequences section (header + content) of a block */
-static size_t ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
-                                                  const BYTE* llCodeTable,
-                                                  const BYTE* mlCodeTable,
-                                                  size_t nbSeq,
-                                                  const ZSTD_fseCTables_t* fseTables,
-                                                  const ZSTD_fseCTablesMetadata_t* fseMetadata,
-                                                  void* workspace, size_t wkspSize,
-                                                  int writeEntropy)
+static size_t
+ZSTD_estimateBlockSize_sequences(const BYTE* ofCodeTable,
+                                 const BYTE* llCodeTable,
+                                 const BYTE* mlCodeTable,
+                                 size_t nbSeq,
+                                 const ZSTD_fseCTables_t* fseTables,
+                                 const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                 void* workspace, size_t wkspSize,
+                                 int writeEntropy)
 {
     size_t sequencesSectionHeaderSize = 1 /* seqHead */ + 1 /* min seqSize size */ + (nbSeq >= 128) + (nbSeq >= LONGNBSEQ);
     size_t cSeqSizeEstimate = 0;
     cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, nbSeq, MaxOff,
-                                         fseTables->offcodeCTable, NULL,
-                                         OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
-                                         workspace, wkspSize);
+                                    fseTables->offcodeCTable, NULL,
+                                    OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                    workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->llType, llCodeTable, nbSeq, MaxLL,
-                                         fseTables->litlengthCTable, LL_bits,
-                                         LL_defaultNorm, LL_defaultNormLog, MaxLL,
-                                         workspace, wkspSize);
+                                    fseTables->litlengthCTable, LL_bits,
+                                    LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                    workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, nbSeq, MaxML,
-                                         fseTables->matchlengthCTable, ML_bits,
-                                         ML_defaultNorm, ML_defaultNormLog, MaxML,
-                                         workspace, wkspSize);
+                                    fseTables->matchlengthCTable, ML_bits,
+                                    ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                    workspace, wkspSize);
     if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
     return cSeqSizeEstimate + sequencesSectionHeaderSize;
 }
 
 /* Returns the size estimate for a given stream of literals, of, ll, ml */
-static size_t ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
-                                     const BYTE* ofCodeTable,
-                                     const BYTE* llCodeTable,
-                                     const BYTE* mlCodeTable,
-                                     size_t nbSeq,
-                                     const ZSTD_entropyCTables_t* entropy,
-                                     const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
-                                     void* workspace, size_t wkspSize,
-                                     int writeLitEntropy, int writeSeqEntropy) {
+static size_t
+ZSTD_estimateBlockSize(const BYTE* literals, size_t litSize,
+                       const BYTE* ofCodeTable,
+                       const BYTE* llCodeTable,
+                       const BYTE* mlCodeTable,
+                       size_t nbSeq,
+                       const ZSTD_entropyCTables_t* entropy,
+                       const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                       void* workspace, size_t wkspSize,
+                       int writeLitEntropy, int writeSeqEntropy)
+{
     size_t const literalsSize = ZSTD_estimateBlockSize_literal(literals, litSize,
-                                                         &entropy->huf, &entropyMetadata->hufMetadata,
-                                                         workspace, wkspSize, writeLitEntropy);
+                                    &entropy->huf, &entropyMetadata->hufMetadata,
+                                    workspace, wkspSize, writeLitEntropy);
     size_t const seqSize = ZSTD_estimateBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
-                                                         nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
-                                                         workspace, wkspSize, writeSeqEntropy);
+                                    nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
+                                    workspace, wkspSize, writeSeqEntropy);
     return seqSize + literalsSize + ZSTD_blockHeaderSize;
 }
 
 /* Builds entropy statistics and uses them for blocksize estimation.
  *
- * Returns the estimated compressed size of the seqStore, or a zstd error.
+ * @return: estimated compressed size of the seqStore, or a zstd error.
  */
-static size_t ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, const ZSTD_CCtx* zc) {
-    ZSTD_entropyCTablesMetadata_t entropyMetadata;
+static size_t
+ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(seqStore_t* seqStore, ZSTD_CCtx* zc)
+{
+    ZSTD_entropyCTablesMetadata_t* const entropyMetadata = &zc->blockSplitCtx.entropyMetadata;
+    DEBUGLOG(6, "ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize()");
     FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(seqStore,
                     &zc->blockState.prevCBlock->entropy,
                     &zc->blockState.nextCBlock->entropy,
                     &zc->appliedParams,
-                    &entropyMetadata,
-                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
-    return ZSTD_estimateBlockSize(seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
+                    entropyMetadata,
+                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE), "");
+    return ZSTD_estimateBlockSize(
+                    seqStore->litStart, (size_t)(seqStore->lit - seqStore->litStart),
                     seqStore->ofCode, seqStore->llCode, seqStore->mlCode,
                     (size_t)(seqStore->sequences - seqStore->sequencesStart),
-                    &zc->blockState.nextCBlock->entropy, &entropyMetadata, zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
-                    (int)(entropyMetadata.hufMetadata.hType == set_compressed), 1);
+                    &zc->blockState.nextCBlock->entropy,
+                    entropyMetadata,
+                    zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE,
+                    (int)(entropyMetadata->hufMetadata.hType == set_compressed), 1);
 }
 
 /* Returns literals bytes represented in a seqStore */
-static size_t ZSTD_countSeqStoreLiteralsBytes(const seqStore_t* const seqStore) {
+static size_t ZSTD_countSeqStoreLiteralsBytes(const seqStore_t* const seqStore)
+{
     size_t literalsBytes = 0;
-    size_t const nbSeqs = seqStore->sequences - seqStore->sequencesStart;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
     size_t i;
     for (i = 0; i < nbSeqs; ++i) {
-        seqDef seq = seqStore->sequencesStart[i];
+        seqDef const seq = seqStore->sequencesStart[i];
         literalsBytes += seq.litLength;
         if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_literalLength) {
             literalsBytes += 0x10000;
-        }
-    }
+    }   }
     return literalsBytes;
 }
 
 /* Returns match bytes represented in a seqStore */
-static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore) {
+static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore)
+{
     size_t matchBytes = 0;
-    size_t const nbSeqs = seqStore->sequences - seqStore->sequencesStart;
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
     size_t i;
     for (i = 0; i < nbSeqs; ++i) {
         seqDef seq = seqStore->sequencesStart[i];
-        matchBytes += seq.matchLength + MINMATCH;
+        matchBytes += seq.mlBase + MINMATCH;
         if (i == seqStore->longLengthPos && seqStore->longLengthType == ZSTD_llt_matchLength) {
             matchBytes += 0x10000;
-        }
-    }
+    }   }
     return matchBytes;
 }
 
@@ -3369,15 +3787,12 @@ static size_t ZSTD_countSeqStoreMatchBytes(const seqStore_t* const seqStore) {
  */
 static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
                                const seqStore_t* originalSeqStore,
-                                     size_t startIdx, size_t endIdx) {
-    BYTE* const litEnd = originalSeqStore->lit;
-    size_t literalsBytes;
-    size_t literalsBytesPreceding = 0;
-
+                                     size_t startIdx, size_t endIdx)
+{
     *resultSeqStore = *originalSeqStore;
     if (startIdx > 0) {
         resultSeqStore->sequences = originalSeqStore->sequencesStart + startIdx;
-        literalsBytesPreceding = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+        resultSeqStore->litStart += ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
     }
 
     /* Move longLengthPos into the correct position if necessary */
@@ -3390,13 +3805,12 @@ static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
     }
     resultSeqStore->sequencesStart = originalSeqStore->sequencesStart + startIdx;
     resultSeqStore->sequences = originalSeqStore->sequencesStart + endIdx;
-    literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
-    resultSeqStore->litStart += literalsBytesPreceding;
     if (endIdx == (size_t)(originalSeqStore->sequences - originalSeqStore->sequencesStart)) {
         /* This accounts for possible last literals if the derived chunk reaches the end of the block */
-        resultSeqStore->lit = litEnd;
+        assert(resultSeqStore->lit == originalSeqStore->lit);
     } else {
-        resultSeqStore->lit = resultSeqStore->litStart+literalsBytes;
+        size_t const literalsBytes = ZSTD_countSeqStoreLiteralsBytes(resultSeqStore);
+        resultSeqStore->lit = resultSeqStore->litStart + literalsBytes;
     }
     resultSeqStore->llCode += startIdx;
     resultSeqStore->mlCode += startIdx;
@@ -3404,52 +3818,67 @@ static void ZSTD_deriveSeqStoreChunk(seqStore_t* resultSeqStore,
 }
 
 /**
- * Returns the raw offset represented by the combination of offCode, ll0, and repcode history.
- * offCode must be an offCode representing a repcode, therefore in the range of [0, 2].
+ * Returns the raw offset represented by the combination of offBase, ll0, and repcode history.
+ * offBase must represent a repcode in the numeric representation of ZSTD_storeSeq().
  */
-static U32 ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offCode, const U32 ll0) {
-    U32 const adjustedOffCode = offCode + ll0;
-    assert(offCode < ZSTD_REP_NUM);
-    if (adjustedOffCode == ZSTD_REP_NUM) {
-        /* litlength == 0 and offCode == 2 implies selection of first repcode - 1 */
-        assert(rep[0] > 0);
+static U32
+ZSTD_resolveRepcodeToRawOffset(const U32 rep[ZSTD_REP_NUM], const U32 offBase, const U32 ll0)
+{
+    U32 const adjustedRepCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;  /* [ 0 - 3 ] */
+    assert(OFFBASE_IS_REPCODE(offBase));
+    if (adjustedRepCode == ZSTD_REP_NUM) {
+        assert(ll0);
+        /* litlength == 0 and offCode == 2 implies selection of first repcode - 1
+         * This is only valid if it results in a valid offset value, aka > 0.
+         * Note : it may happen that `rep[0]==1` in exceptional circumstances.
+         * In which case this function will return 0, which is an invalid offset.
+         * It's not an issue though, since this value will be
+         * compared and discarded within ZSTD_seqStore_resolveOffCodes().
+         */
         return rep[0] - 1;
     }
-    return rep[adjustedOffCode];
+    return rep[adjustedRepCode];
 }
 
 /**
  * ZSTD_seqStore_resolveOffCodes() reconciles any possible divergences in offset history that may arise
- * due to emission of RLE/raw blocks that disturb the offset history, and replaces any repcodes within
- * the seqStore that may be invalid.
+ * due to emission of RLE/raw blocks that disturb the offset history,
+ * and replaces any repcodes within the seqStore that may be invalid.
+ *
+ * dRepcodes are updated as would be on the decompression side.
+ * cRepcodes are updated exactly in accordance with the seqStore.
  *
- * dRepcodes are updated as would be on the decompression side. cRepcodes are updated exactly in
- * accordance with the seqStore.
+ * Note : this function assumes seq->offBase respects the following numbering scheme :
+ *        0 : invalid
+ *        1-3 : repcode 1-3
+ *        4+ : real_offset+3
  */
-static void ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_t* const cRepcodes,
-                                          seqStore_t* const seqStore, U32 const nbSeq) {
+static void
+ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_t* const cRepcodes,
+                        const seqStore_t* const seqStore, U32 const nbSeq)
+{
     U32 idx = 0;
     for (; idx < nbSeq; ++idx) {
         seqDef* const seq = seqStore->sequencesStart + idx;
         U32 const ll0 = (seq->litLength == 0);
-        U32 offCode = seq->offset - 1;
-        assert(seq->offset > 0);
-        if (offCode <= ZSTD_REP_MOVE) {
-            U32 const dRawOffset = ZSTD_resolveRepcodeToRawOffset(dRepcodes->rep, offCode, ll0);
-            U32 const cRawOffset = ZSTD_resolveRepcodeToRawOffset(cRepcodes->rep, offCode, ll0);
+        U32 const offBase = seq->offBase;
+        assert(offBase > 0);
+        if (OFFBASE_IS_REPCODE(offBase)) {
+            U32 const dRawOffset = ZSTD_resolveRepcodeToRawOffset(dRepcodes->rep, offBase, ll0);
+            U32 const cRawOffset = ZSTD_resolveRepcodeToRawOffset(cRepcodes->rep, offBase, ll0);
             /* Adjust simulated decompression repcode history if we come across a mismatch. Replace
              * the repcode with the offset it actually references, determined by the compression
              * repcode history.
              */
             if (dRawOffset != cRawOffset) {
-                seq->offset = cRawOffset + ZSTD_REP_NUM;
+                seq->offBase = OFFSET_TO_OFFBASE(cRawOffset);
             }
         }
         /* Compression repcode history is always updated with values directly from the unmodified seqStore.
          * Decompression repcode history may use modified seq->offset value taken from compression repcode history.
          */
-        *dRepcodes = ZSTD_updateRep(dRepcodes->rep, seq->offset - 1, ll0);
-        *cRepcodes = ZSTD_updateRep(cRepcodes->rep, offCode, ll0);
+        ZSTD_updateRep(dRepcodes->rep, seq->offBase, ll0);
+        ZSTD_updateRep(cRepcodes->rep, offBase, ll0);
     }
 }
 
@@ -3458,11 +3887,14 @@ static void ZSTD_seqStore_resolveOffCodes(repcodes_t* const dRepcodes, repcodes_
  *
  * Returns the total size of that block (including header) or a ZSTD error code.
  */
-static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const seqStore,
-                                                repcodes_t* const dRep, repcodes_t* const cRep,
-                                                void* dst, size_t dstCapacity,
-                                                const void* src, size_t srcSize,
-                                                U32 lastBlock, U32 isPartition) {
+static size_t
+ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc,
+                            const seqStore_t* const seqStore,
+                                  repcodes_t* const dRep, repcodes_t* const cRep,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                                  U32 lastBlock, U32 isPartition)
+{
     const U32 rleMaxLength = 25;
     BYTE* op = (BYTE*)dst;
     const BYTE* ip = (const BYTE*)src;
@@ -3471,9 +3903,11 @@ static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const
 
     /* In case of an RLE or raw block, the simulated decompression repcode history must be reset */
     repcodes_t const dRepOriginal = *dRep;
+    DEBUGLOG(5, "ZSTD_compressSeqStore_singleBlock");
     if (isPartition)
         ZSTD_seqStore_resolveOffCodes(dRep, cRep, seqStore, (U32)(seqStore->sequences - seqStore->sequencesStart));
 
+    RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "Block header doesn't fit");
     cSeqsSize = ZSTD_entropyCompressSeqStore(seqStore,
                 &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
                 &zc->appliedParams,
@@ -3499,9 +3933,6 @@ static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const
         return 0;
     }
 
-    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
-        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
-
     if (cSeqsSize == 0) {
         cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
         FORWARD_IF_ERROR(cSize, "Nocompress block failed");
@@ -3518,6 +3949,10 @@ static size_t ZSTD_compressSeqStore_singleBlock(ZSTD_CCtx* zc, seqStore_t* const
         cSize = ZSTD_blockHeaderSize + cSeqsSize;
         DEBUGLOG(4, "Writing out compressed block, size: %zu", cSize);
     }
+
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
     return cSize;
 }
 
@@ -3528,45 +3963,52 @@ typedef struct {
 } seqStoreSplits;
 
 #define MIN_SEQUENCES_BLOCK_SPLITTING 300
-#define MAX_NB_SPLITS 196
 
 /* Helper function to perform the recursive search for block splits.
  * Estimates the cost of seqStore prior to split, and estimates the cost of splitting the sequences in half.
- * If advantageous to split, then we recurse down the two sub-blocks. If not, or if an error occurred in estimation, then
- * we do not recurse.
+ * If advantageous to split, then we recurse down the two sub-blocks.
+ * If not, or if an error occurred in estimation, then we do not recurse.
  *
- * Note: The recursion depth is capped by a heuristic minimum number of sequences, defined by MIN_SEQUENCES_BLOCK_SPLITTING.
+ * Note: The recursion depth is capped by a heuristic minimum number of sequences,
+ * defined by MIN_SEQUENCES_BLOCK_SPLITTING.
  * In theory, this means the absolute largest recursion depth is 10 == log2(maxNbSeqInBlock/MIN_SEQUENCES_BLOCK_SPLITTING).
  * In practice, recursion depth usually doesn't go beyond 4.
  *
- * Furthermore, the number of splits is capped by MAX_NB_SPLITS. At MAX_NB_SPLITS == 196 with the current existing blockSize
+ * Furthermore, the number of splits is capped by ZSTD_MAX_NB_BLOCK_SPLITS.
+ * At ZSTD_MAX_NB_BLOCK_SPLITS == 196 with the current existing blockSize
  * maximum of 128 KB, this value is actually impossible to reach.
  */
-static void ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
-                                         const ZSTD_CCtx* zc, const seqStore_t* origSeqStore) {
-    seqStore_t fullSeqStoreChunk;
-    seqStore_t firstHalfSeqStore;
-    seqStore_t secondHalfSeqStore;
+static void
+ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx, size_t endIdx,
+                             ZSTD_CCtx* zc, const seqStore_t* origSeqStore)
+{
+    seqStore_t* const fullSeqStoreChunk = &zc->blockSplitCtx.fullSeqStoreChunk;
+    seqStore_t* const firstHalfSeqStore = &zc->blockSplitCtx.firstHalfSeqStore;
+    seqStore_t* const secondHalfSeqStore = &zc->blockSplitCtx.secondHalfSeqStore;
     size_t estimatedOriginalSize;
     size_t estimatedFirstHalfSize;
     size_t estimatedSecondHalfSize;
     size_t midIdx = (startIdx + endIdx)/2;
 
-    if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= MAX_NB_SPLITS) {
+    DEBUGLOG(5, "ZSTD_deriveBlockSplitsHelper: startIdx=%zu endIdx=%zu", startIdx, endIdx);
+    assert(endIdx >= startIdx);
+    if (endIdx - startIdx < MIN_SEQUENCES_BLOCK_SPLITTING || splits->idx >= ZSTD_MAX_NB_BLOCK_SPLITS) {
+        DEBUGLOG(6, "ZSTD_deriveBlockSplitsHelper: Too few sequences (%zu)", endIdx - startIdx);
         return;
     }
-    ZSTD_deriveSeqStoreChunk(&fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
-    ZSTD_deriveSeqStoreChunk(&firstHalfSeqStore, origSeqStore, startIdx, midIdx);
-    ZSTD_deriveSeqStoreChunk(&secondHalfSeqStore, origSeqStore, midIdx, endIdx);
-    estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&fullSeqStoreChunk, zc);
-    estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&firstHalfSeqStore, zc);
-    estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&secondHalfSeqStore, zc);
+    ZSTD_deriveSeqStoreChunk(fullSeqStoreChunk, origSeqStore, startIdx, endIdx);
+    ZSTD_deriveSeqStoreChunk(firstHalfSeqStore, origSeqStore, startIdx, midIdx);
+    ZSTD_deriveSeqStoreChunk(secondHalfSeqStore, origSeqStore, midIdx, endIdx);
+    estimatedOriginalSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(fullSeqStoreChunk, zc);
+    estimatedFirstHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(firstHalfSeqStore, zc);
+    estimatedSecondHalfSize = ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(secondHalfSeqStore, zc);
     DEBUGLOG(5, "Estimated original block size: %zu -- First half split: %zu -- Second half split: %zu",
              estimatedOriginalSize, estimatedFirstHalfSize, estimatedSecondHalfSize);
     if (ZSTD_isError(estimatedOriginalSize) || ZSTD_isError(estimatedFirstHalfSize) || ZSTD_isError(estimatedSecondHalfSize)) {
         return;
     }
     if (estimatedFirstHalfSize + estimatedSecondHalfSize < estimatedOriginalSize) {
+        DEBUGLOG(5, "split decided at seqNb:%zu", midIdx);
         ZSTD_deriveBlockSplitsHelper(splits, startIdx, midIdx, zc, origSeqStore);
         splits->splitLocations[splits->idx] = (U32)midIdx;
         splits->idx++;
@@ -3574,14 +4016,18 @@ static void ZSTD_deriveBlockSplitsHelper(seqStoreSplits* splits, size_t startIdx
     }
 }
 
-/* Base recursive function. Populates a table with intra-block partition indices that can improve compression ratio.
+/* Base recursive function.
+ * Populates a table with intra-block partition indices that can improve compression ratio.
  *
- * Returns the number of splits made (which equals the size of the partition table - 1).
+ * @return: number of splits made (which equals the size of the partition table - 1).
  */
-static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq) {
-    seqStoreSplits splits = {partitions, 0};
+static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq)
+{
+    seqStoreSplits splits;
+    splits.splitLocations = partitions;
+    splits.idx = 0;
     if (nbSeq <= 4) {
-        DEBUGLOG(4, "ZSTD_deriveBlockSplits: Too few sequences to split");
+        DEBUGLOG(5, "ZSTD_deriveBlockSplits: Too few sequences to split (%u <= 4)", nbSeq);
         /* Refuse to try and split anything with less than 4 sequences */
         return 0;
     }
@@ -3596,17 +4042,21 @@ static size_t ZSTD_deriveBlockSplits(ZSTD_CCtx* zc, U32 partitions[], U32 nbSeq)
  *
  * Returns combined size of all blocks (which includes headers), or a ZSTD error code.
  */
-static size_t ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapacity,
-                                                     const void* src, size_t blockSize, U32 lastBlock, U32 nbSeq) {
+static size_t
+ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc,
+                                    void* dst, size_t dstCapacity,
+                              const void* src, size_t blockSize,
+                                    U32 lastBlock, U32 nbSeq)
+{
     size_t cSize = 0;
     const BYTE* ip = (const BYTE*)src;
     BYTE* op = (BYTE*)dst;
-    U32 partitions[MAX_NB_SPLITS];
     size_t i = 0;
     size_t srcBytesTotal = 0;
-    size_t numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
-    seqStore_t nextSeqStore;
-    seqStore_t currSeqStore;
+    U32* const partitions = zc->blockSplitCtx.partitions; /* size == ZSTD_MAX_NB_BLOCK_SPLITS */
+    seqStore_t* const nextSeqStore = &zc->blockSplitCtx.nextSeqStore;
+    seqStore_t* const currSeqStore = &zc->blockSplitCtx.currSeqStore;
+    size_t const numSplits = ZSTD_deriveBlockSplits(zc, partitions, nbSeq);
 
     /* If a block is split and some partitions are emitted as RLE/uncompressed, then repcode history
      * may become invalid. In order to reconcile potentially invalid repcodes, we keep track of two
@@ -3626,77 +4076,81 @@ static size_t ZSTD_compressBlock_splitBlock_internal(ZSTD_CCtx* zc, void* dst, s
     repcodes_t cRep;
     ZSTD_memcpy(dRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
     ZSTD_memcpy(cRep.rep, zc->blockState.prevCBlock->rep, sizeof(repcodes_t));
+    ZSTD_memset(nextSeqStore, 0, sizeof(seqStore_t));
 
-    DEBUGLOG(4, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+    DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
                 (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
                 (unsigned)zc->blockState.matchState.nextToUpdate);
 
     if (numSplits == 0) {
-        size_t cSizeSingleBlock = ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore,
-                                                                   &dRep, &cRep,
-                                                                    op, dstCapacity,
-                                                                    ip, blockSize,
-                                                                    lastBlock, 0 /* isPartition */);
+        size_t cSizeSingleBlock =
+            ZSTD_compressSeqStore_singleBlock(zc, &zc->seqStore,
+                                            &dRep, &cRep,
+                                            op, dstCapacity,
+                                            ip, blockSize,
+                                            lastBlock, 0 /* isPartition */);
         FORWARD_IF_ERROR(cSizeSingleBlock, "Compressing single block from splitBlock_internal() failed!");
         DEBUGLOG(5, "ZSTD_compressBlock_splitBlock_internal: No splits");
-        assert(cSizeSingleBlock <= ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize);
+        assert(zc->blockSize <= ZSTD_BLOCKSIZE_MAX);
+        assert(cSizeSingleBlock <= zc->blockSize + ZSTD_blockHeaderSize);
         return cSizeSingleBlock;
     }
 
-    ZSTD_deriveSeqStoreChunk(&currSeqStore, &zc->seqStore, 0, partitions[0]);
+    ZSTD_deriveSeqStoreChunk(currSeqStore, &zc->seqStore, 0, partitions[0]);
     for (i = 0; i <= numSplits; ++i) {
-        size_t srcBytes;
         size_t cSizeChunk;
         U32 const lastPartition = (i == numSplits);
         U32 lastBlockEntireSrc = 0;
 
-        srcBytes = ZSTD_countSeqStoreLiteralsBytes(&currSeqStore) + ZSTD_countSeqStoreMatchBytes(&currSeqStore);
+        size_t srcBytes = ZSTD_countSeqStoreLiteralsBytes(currSeqStore) + ZSTD_countSeqStoreMatchBytes(currSeqStore);
         srcBytesTotal += srcBytes;
         if (lastPartition) {
             /* This is the final partition, need to account for possible last literals */
             srcBytes += blockSize - srcBytesTotal;
             lastBlockEntireSrc = lastBlock;
         } else {
-            ZSTD_deriveSeqStoreChunk(&nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
+            ZSTD_deriveSeqStoreChunk(nextSeqStore, &zc->seqStore, partitions[i], partitions[i+1]);
         }
 
-        cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, &currSeqStore,
+        cSizeChunk = ZSTD_compressSeqStore_singleBlock(zc, currSeqStore,
                                                       &dRep, &cRep,
                                                        op, dstCapacity,
                                                        ip, srcBytes,
                                                        lastBlockEntireSrc, 1 /* isPartition */);
-        DEBUGLOG(5, "Estimated size: %zu actual size: %zu", ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(&currSeqStore, zc), cSizeChunk);
+        DEBUGLOG(5, "Estimated size: %zu vs %zu : actual size",
+                    ZSTD_buildEntropyStatisticsAndEstimateSubBlockSize(currSeqStore, zc), cSizeChunk);
         FORWARD_IF_ERROR(cSizeChunk, "Compressing chunk failed!");
 
         ip += srcBytes;
         op += cSizeChunk;
         dstCapacity -= cSizeChunk;
         cSize += cSizeChunk;
-        currSeqStore = nextSeqStore;
-        assert(cSizeChunk <= ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize);
+        *currSeqStore = *nextSeqStore;
+        assert(cSizeChunk <= zc->blockSize + ZSTD_blockHeaderSize);
     }
-    /* cRep and dRep may have diverged during the compression. If so, we use the dRep repcodes
-     * for the next block.
+    /* cRep and dRep may have diverged during the compression.
+     * If so, we use the dRep repcodes for the next block.
      */
     ZSTD_memcpy(zc->blockState.prevCBlock->rep, dRep.rep, sizeof(repcodes_t));
     return cSize;
 }
 
-static size_t ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
-                                        void* dst, size_t dstCapacity,
-                                        const void* src, size_t srcSize, U32 lastBlock) {
-    const BYTE* ip = (const BYTE*)src;
-    BYTE* op = (BYTE*)dst;
+static size_t
+ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
+                              void* dst, size_t dstCapacity,
+                              const void* src, size_t srcSize, U32 lastBlock)
+{
     U32 nbSeq;
     size_t cSize;
     DEBUGLOG(4, "ZSTD_compressBlock_splitBlock");
+    assert(zc->appliedParams.useBlockSplitter == ZSTD_ps_enable);
 
     {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
         FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
         if (bss == ZSTDbss_noCompress) {
             if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
                 zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
-            cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, srcSize, lastBlock);
+            cSize = ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
             FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
             DEBUGLOG(4, "ZSTD_compressBlock_splitBlock: Nocompress block");
             return cSize;
@@ -3704,19 +4158,19 @@ static size_t ZSTD_compressBlock_splitBlock(ZSTD_CCtx* zc,
         nbSeq = (U32)(zc->seqStore.sequences - zc->seqStore.sequencesStart);
     }
 
-    assert(zc->appliedParams.splitBlocks == 1);
     cSize = ZSTD_compressBlock_splitBlock_internal(zc, dst, dstCapacity, src, srcSize, lastBlock, nbSeq);
     FORWARD_IF_ERROR(cSize, "Splitting blocks failed!");
     return cSize;
 }
 
-static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
-                                        void* dst, size_t dstCapacity,
-                                        const void* src, size_t srcSize, U32 frame)
+static size_t
+ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                            void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize, U32 frame)
 {
-    /* This the upper bound for the length of an rle block.
-     * This isn't the actual upper bound. Finding the real threshold
-     * needs further investigation.
+    /* This is an estimated upper bound for the length of an rle block.
+     * This isn't the actual upper bound.
+     * Finding the real threshold needs further investigation.
      */
     const U32 rleMaxLength = 25;
     size_t cSize;
@@ -3746,12 +4200,6 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
             zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
             zc->bmi2);
 
-    if (zc->seqCollector.collectSequences) {
-        ZSTD_copyBlockSequences(zc);
-        return 0;
-    }
-
-
     if (frame &&
         /* We don't want to emit our first block as a RLE even if it qualifies because
          * doing so will cause the decoder (cli only) to throw a "should consume all input error."
@@ -3814,10 +4262,11 @@ static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
          *   * cSize >= blockBound(srcSize): We have expanded the block too much so
          *     emit an uncompressed block.
          */
-        {
-            size_t const cSize = ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
+        {   size_t const cSize =
+                ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
             if (cSize != ERROR(dstSize_tooSmall)) {
-                size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
+                size_t const maxCSize =
+                    srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
                 FORWARD_IF_ERROR(cSize, "ZSTD_compressSuperBlock failed");
                 if (cSize != 0 && cSize < maxCSize + ZSTD_blockHeaderSize) {
                     ZSTD_blockState_confirmRepcodesAndEntropyTables(&zc->blockState);
@@ -3825,7 +4274,7 @@ static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
                 }
             }
         }
-    }
+    } /* if (bss == ZSTDbss_compress)*/
 
     DEBUGLOG(6, "Resorting to ZSTD_noCompressBlock()");
     /* Superblock compression failed, attempt to emit a single no compress block.
@@ -3883,7 +4332,7 @@ static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
 *   All blocks will be terminated, all input will be consumed.
 *   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
 *   Frame is supposed already started (header already produced)
-*   @return : compressed size, or an error code
+*  @return : compressed size, or an error code
 */
 static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
                                      void* dst, size_t dstCapacity,
@@ -3907,7 +4356,9 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
         ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
         U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
 
-        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE,
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE + 1,
                         dstSize_tooSmall,
                         "not enough space to store compressed block");
         if (remaining < blockSize) blockSize = remaining;
@@ -3915,6 +4366,7 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
         ZSTD_overflowCorrectIfNeeded(
             ms, &cctx->workspace, &cctx->appliedParams, ip, ip + blockSize);
         ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+        ZSTD_window_enforceMaxDist(&ms->window, ip, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
 
         /* Ensure hash/chain table insertion resumes no sooner than lowlimit */
         if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
@@ -3945,7 +4397,7 @@ static size_t ZSTD_compress_frameChunk(ZSTD_CCtx* cctx,
                     MEM_writeLE24(op, cBlockHeader);
                     cSize += ZSTD_blockHeaderSize;
                 }
-            }
+            }  /* if (ZSTD_useTargetCBlockSize(&cctx->appliedParams))*/
 
 
             ip += blockSize;
@@ -3991,7 +4443,9 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
     if (!singleSegment) op[pos++] = windowLogByte;
     switch(dictIDSizeCode)
     {
-        default:  assert(0); /* impossible */
+        default:
+            assert(0); /* impossible */
+            ZSTD_FALLTHROUGH;
         case 0 : break;
         case 1 : op[pos] = (BYTE)(dictID); pos++; break;
         case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
@@ -3999,7 +4453,9 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
     }
     switch(fcsCode)
     {
-        default:  assert(0); /* impossible */
+        default:
+            assert(0); /* impossible */
+            ZSTD_FALLTHROUGH;
         case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
         case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
         case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
@@ -4047,7 +4503,7 @@ size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSe
 {
     RETURN_ERROR_IF(cctx->stage != ZSTDcs_init, stage_wrong,
                     "wrong cctx stage");
-    RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm,
+    RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable,
                     parameter_unsupported,
                     "incompatible with ldm");
     cctx->externSeqStore.seq = seq;
@@ -4088,7 +4544,7 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
         ms->forceNonContiguous = 0;
         ms->nextToUpdate = ms->window.dictLimit;
     }
-    if (cctx->appliedParams.ldmParams.enableLdm) {
+    if (cctx->appliedParams.ldmParams.enableLdm == ZSTD_ps_enable) {
         ZSTD_window_update(&cctx->ldmState.window, src, srcSize, /* forceNonContiguous */ 0);
     }
 
@@ -4133,7 +4589,7 @@ size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
 {
     ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
     assert(!ZSTD_checkCParams(cParams));
-    return MIN (ZSTD_BLOCKSIZE_MAX, (U32)1 << cParams.windowLog);
+    return MIN(cctx->appliedParams.maxBlockSize, (size_t)1 << cParams.windowLog);
 }
 
 size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
@@ -4153,31 +4609,47 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
                                          ZSTD_cwksp* ws,
                                          ZSTD_CCtx_params const* params,
                                          const void* src, size_t srcSize,
-                                         ZSTD_dictTableLoadMethod_e dtlm)
+                                         ZSTD_dictTableLoadMethod_e dtlm,
+                                         ZSTD_tableFillPurpose_e tfp)
 {
     const BYTE* ip = (const BYTE*) src;
     const BYTE* const iend = ip + srcSize;
-    int const loadLdmDict = params->ldmParams.enableLdm && ls != NULL;
+    int const loadLdmDict = params->ldmParams.enableLdm == ZSTD_ps_enable && ls != NULL;
 
-    /* Assert that we the ms params match the params we're being given */
+    /* Assert that the ms params match the params we're being given */
     ZSTD_assertEqualCParams(params->cParams, ms->cParams);
 
-    if (srcSize > ZSTD_CHUNKSIZE_MAX) {
+    {   /* Ensure large dictionaries can't cause index overflow */
+
         /* Allow the dictionary to set indices up to exactly ZSTD_CURRENT_MAX.
          * Dictionaries right at the edge will immediately trigger overflow
          * correction, but I don't want to insert extra constraints here.
          */
-        U32 const maxDictSize = ZSTD_CURRENT_MAX - 1;
-        /* We must have cleared our windows when our source is this large. */
-        assert(ZSTD_window_isEmpty(ms->window));
-        if (loadLdmDict)
-            assert(ZSTD_window_isEmpty(ls->window));
+        U32 maxDictSize = ZSTD_CURRENT_MAX - ZSTD_WINDOW_START_INDEX;
+
+        int const CDictTaggedIndices = ZSTD_CDictIndicesAreTagged(¶ms->cParams);
+        if (CDictTaggedIndices && tfp == ZSTD_tfp_forCDict) {
+            /* Some dictionary matchfinders in zstd use "short cache",
+             * which treats the lower ZSTD_SHORT_CACHE_TAG_BITS of each
+             * CDict hashtable entry as a tag rather than as part of an index.
+             * When short cache is used, we need to truncate the dictionary
+             * so that its indices don't overlap with the tag. */
+            U32 const shortCacheMaxDictSize = (1u << (32 - ZSTD_SHORT_CACHE_TAG_BITS)) - ZSTD_WINDOW_START_INDEX;
+            maxDictSize = MIN(maxDictSize, shortCacheMaxDictSize);
+            assert(!loadLdmDict);
+        }
+
         /* If the dictionary is too large, only load the suffix of the dictionary. */
         if (srcSize > maxDictSize) {
             ip = iend - maxDictSize;
             src = ip;
             srcSize = maxDictSize;
-        }
+    }   }
+
+    if (srcSize > ZSTD_CHUNKSIZE_MAX) {
+        /* We must have cleared our windows when our source is this large. */
+        assert(ZSTD_window_isEmpty(ms->window));
+        if (loadLdmDict) assert(ZSTD_window_isEmpty(ls->window));
     }
 
     DEBUGLOG(4, "ZSTD_loadDictionaryContent(): useRowMatchFinder=%d", (int)params->useRowMatchFinder);
@@ -4200,10 +4672,10 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
     switch(params->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(ms, iend, dtlm);
+        ZSTD_fillHashTable(ms, iend, dtlm, tfp);
         break;
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, iend, dtlm);
+        ZSTD_fillDoubleHashTable(ms, iend, dtlm, tfp);
         break;
 
     case ZSTD_greedy:
@@ -4214,8 +4686,8 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
             assert(ms->chainTable != NULL);
             ZSTD_dedicatedDictSearch_lazy_loadDictionary(ms, iend-HASH_READ_SIZE);
         } else {
-            assert(params->useRowMatchFinder != ZSTD_urm_auto);
-            if (params->useRowMatchFinder == ZSTD_urm_enableRowMatchFinder) {
+            assert(params->useRowMatchFinder != ZSTD_ps_auto);
+            if (params->useRowMatchFinder == ZSTD_ps_enable) {
                 size_t const tagTableSize = ((size_t)1 << params->cParams.hashLog) * sizeof(U16);
                 ZSTD_memset(ms->tagTable, 0, tagTableSize);
                 ZSTD_row_update(ms, iend-HASH_READ_SIZE);
@@ -4369,6 +4841,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
                                       ZSTD_CCtx_params const* params,
                                       const void* dict, size_t dictSize,
                                       ZSTD_dictTableLoadMethod_e dtlm,
+                                      ZSTD_tableFillPurpose_e tfp,
                                       void* workspace)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
@@ -4387,7 +4860,7 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
     {
         size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
         FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
-            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm), "");
+            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm, tfp), "");
     }
     return dictID;
 }
@@ -4403,6 +4876,7 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
                          const void* dict, size_t dictSize,
                                ZSTD_dictContentType_e dictContentType,
                                ZSTD_dictTableLoadMethod_e dtlm,
+                               ZSTD_tableFillPurpose_e tfp,
                                void* workspace)
 {
     DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
@@ -4415,13 +4889,13 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
 
     /* dict restricted modes */
     if (dictContentType == ZSTD_dct_rawContent)
-        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm);
+        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm, tfp);
 
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
         if (dictContentType == ZSTD_dct_auto) {
             DEBUGLOG(4, "raw content dictionary detected");
             return ZSTD_loadDictionaryContent(
-                ms, ls, ws, params, dict, dictSize, dtlm);
+                ms, ls, ws, params, dict, dictSize, dtlm, tfp);
         }
         RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
         assert(0);   /* impossible */
@@ -4429,13 +4903,14 @@ ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
 
     /* dict as full zstd dictionary */
     return ZSTD_loadZstdDictionary(
-        bs, ms, ws, params, dict, dictSize, dtlm, workspace);
+        bs, ms, ws, params, dict, dictSize, dtlm, tfp, workspace);
 }
 
 #define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
 #define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6ULL)
 
 /*! ZSTD_compressBegin_internal() :
+ * Assumption : either @dict OR @cdict (or none) is non-NULL, never both
  * @return : 0, or an error code */
 static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                                     const void* dict, size_t dictSize,
@@ -4471,11 +4946,11 @@ static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                         cctx->blockState.prevCBlock, &cctx->blockState.matchState,
                         &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, cdict->dictContent,
                         cdict->dictContentSize, cdict->dictContentType, dtlm,
-                        cctx->entropyWorkspace)
+                        ZSTD_tfp_forCCtx, cctx->entropyWorkspace)
               : ZSTD_compress_insertDictionary(
                         cctx->blockState.prevCBlock, &cctx->blockState.matchState,
                         &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, dict, dictSize,
-                        dictContentType, dtlm, cctx->entropyWorkspace);
+                        dictContentType, dtlm, ZSTD_tfp_forCCtx, cctx->entropyWorkspace);
         FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
         assert(dictID <= UINT_MAX);
         cctx->dictID = (U32)dictID;
@@ -4516,11 +4991,11 @@ size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
                                             &cctxParams, pledgedSrcSize);
 }
 
-size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+size_t
+ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
 {
     ZSTD_CCtx_params cctxParams;
-    {
-        ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
+    {   ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize, ZSTD_cpm_noAttachDict);
         ZSTD_CCtxParams_init_internal(&cctxParams, ¶ms, (compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : compressionLevel);
     }
     DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
@@ -4715,7 +5190,7 @@ size_t ZSTD_estimateCDictSize_advanced(
          + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
          /* enableDedicatedDictSearch == 1 ensures that CDict estimation will not be too small
           * in case we are using DDS with row-hash. */
-         + ZSTD_sizeof_matchState(&cParams, ZSTD_resolveRowMatchFinderMode(ZSTD_urm_auto, &cParams),
+         + ZSTD_sizeof_matchState(&cParams, ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams),
                                   /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0)
          + (dictLoadMethod == ZSTD_dlm_byRef ? 0
             : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void *))));
@@ -4779,7 +5254,7 @@ static size_t ZSTD_initCDict_internal(
         {   size_t const dictID = ZSTD_compress_insertDictionary(
                     &cdict->cBlockState, &cdict->matchState, NULL, &cdict->workspace,
                     ¶ms, cdict->dictContent, cdict->dictContentSize,
-                    dictContentType, ZSTD_dtlm_full, cdict->entropyWorkspace);
+                    dictContentType, ZSTD_dtlm_full, ZSTD_tfp_forCDict, cdict->entropyWorkspace);
             FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
             assert(dictID <= (size_t)(U32)-1);
             cdict->dictID = (U32)dictID;
@@ -4792,7 +5267,7 @@ static size_t ZSTD_initCDict_internal(
 static ZSTD_CDict* ZSTD_createCDict_advanced_internal(size_t dictSize,
                                       ZSTD_dictLoadMethod_e dictLoadMethod,
                                       ZSTD_compressionParameters cParams,
-                                      ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+                                      ZSTD_paramSwitch_e useRowMatchFinder,
                                       U32 enableDedicatedDictSearch,
                                       ZSTD_customMem customMem)
 {
@@ -4842,7 +5317,7 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
         &cctxParams, customMem);
 }
 
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced2(
+ZSTD_CDict* ZSTD_createCDict_advanced2(
         const void* dict, size_t dictSize,
         ZSTD_dictLoadMethod_e dictLoadMethod,
         ZSTD_dictContentType_e dictContentType,
@@ -4947,7 +5422,7 @@ const ZSTD_CDict* ZSTD_initStaticCDict(
                                  ZSTD_dictContentType_e dictContentType,
                                  ZSTD_compressionParameters cParams)
 {
-    ZSTD_useRowMatchFinderMode_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(ZSTD_urm_auto, &cParams);
+    ZSTD_paramSwitch_e const useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(ZSTD_ps_auto, &cParams);
     /* enableDedicatedDictSearch == 1 ensures matchstate is not too small in case this CDict will be used for DDS + row hash */
     size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, useRowMatchFinder, /* enableDedicatedDictSearch */ 1, /* forCCtx */ 0);
     size_t const neededSize = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
@@ -5267,30 +5742,41 @@ size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
 
 static size_t ZSTD_nextInputSizeHint(const ZSTD_CCtx* cctx)
 {
-    size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
-    if (hintInSize==0) hintInSize = cctx->blockSize;
-    return hintInSize;
+    if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        return cctx->blockSize - cctx->stableIn_notConsumed;
+    }
+    assert(cctx->appliedParams.inBufferMode == ZSTD_bm_buffered);
+    {   size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
+        if (hintInSize==0) hintInSize = cctx->blockSize;
+        return hintInSize;
+    }
 }
 
 /** ZSTD_compressStream_generic():
  *  internal function for all *compressStream*() variants
- *  non-static, because can be called from zstdmt_compress.c
- * @return : hint size for next input */
+ * @return : hint size for next input to complete ongoing block */
 static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                                           ZSTD_outBuffer* output,
                                           ZSTD_inBuffer* input,
                                           ZSTD_EndDirective const flushMode)
 {
-    const char* const istart = (const char*)input->src;
-    const char* const iend = input->size != 0 ? istart + input->size : istart;
-    const char* ip = input->pos != 0 ? istart + input->pos : istart;
-    char* const ostart = (char*)output->dst;
-    char* const oend = output->size != 0 ? ostart + output->size : ostart;
-    char* op = output->pos != 0 ? ostart + output->pos : ostart;
+    const char* const istart = (assert(input != NULL), (const char*)input->src);
+    const char* const iend = (istart != NULL) ? istart + input->size : istart;
+    const char* ip = (istart != NULL) ? istart + input->pos : istart;
+    char* const ostart = (assert(output != NULL), (char*)output->dst);
+    char* const oend = (ostart != NULL) ? ostart + output->size : ostart;
+    char* op = (ostart != NULL) ? ostart + output->pos : ostart;
     U32 someMoreWork = 1;
 
     /* check expectations */
-    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (unsigned)flushMode);
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%i, srcSize = %zu", (int)flushMode, input->size - input->pos);
+    assert(zcs != NULL);
+    if (zcs->appliedParams.inBufferMode == ZSTD_bm_stable) {
+        assert(input->pos >= zcs->stableIn_notConsumed);
+        input->pos -= zcs->stableIn_notConsumed;
+        ip -= zcs->stableIn_notConsumed;
+        zcs->stableIn_notConsumed = 0;
+    }
     if (zcs->appliedParams.inBufferMode == ZSTD_bm_buffered) {
         assert(zcs->inBuff != NULL);
         assert(zcs->inBuffSize > 0);
@@ -5299,8 +5785,10 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
         assert(zcs->outBuff !=  NULL);
         assert(zcs->outBuffSize > 0);
     }
-    assert(output->pos <= output->size);
+    if (input->src == NULL) assert(input->size == 0);
     assert(input->pos <= input->size);
+    if (output->dst == NULL) assert(output->size == 0);
+    assert(output->pos <= output->size);
     assert((U32)flushMode <= (U32)ZSTD_e_end);
 
     while (someMoreWork) {
@@ -5332,8 +5820,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                                         zcs->inBuff + zcs->inBuffPos, toLoad,
                                         ip, iend-ip);
                 zcs->inBuffPos += loaded;
-                if (loaded != 0)
-                    ip += loaded;
+                if (ip) ip += loaded;
                 if ( (flushMode == ZSTD_e_continue)
                   && (zcs->inBuffPos < zcs->inBuffTarget) ) {
                     /* not enough input to fill full block : stop here */
@@ -5344,6 +5831,20 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                     /* empty */
                     someMoreWork = 0; break;
                 }
+            } else {
+                assert(zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+                if ( (flushMode == ZSTD_e_continue)
+                  && ( (size_t)(iend - ip) < zcs->blockSize) ) {
+                    /* can't compress a full block : stop here */
+                    zcs->stableIn_notConsumed = (size_t)(iend - ip);
+                    ip = iend;  /* pretend to have consumed input */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (ip == iend) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
             }
             /* compress current block (note : this stage cannot be stopped in the middle) */
             DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
@@ -5351,9 +5852,8 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 void* cDst;
                 size_t cSize;
                 size_t oSize = oend-op;
-                size_t const iSize = inputBuffered
-                    ? zcs->inBuffPos - zcs->inToCompress
-                    : MIN((size_t)(iend - ip), zcs->blockSize);
+                size_t const iSize = inputBuffered ? zcs->inBuffPos - zcs->inToCompress
+                                                   : MIN((size_t)(iend - ip), zcs->blockSize);
                 if (oSize >= ZSTD_compressBound(iSize) || zcs->appliedParams.outBufferMode == ZSTD_bm_stable)
                     cDst = op;   /* compress into output buffer, to skip flush stage */
                 else
@@ -5376,19 +5876,16 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                     if (!lastBlock)
                         assert(zcs->inBuffTarget <= zcs->inBuffSize);
                     zcs->inToCompress = zcs->inBuffPos;
-                } else {
-                    unsigned const lastBlock = (ip + iSize == iend);
-                    assert(flushMode == ZSTD_e_end /* Already validated */);
+                } else { /* !inputBuffered, hence ZSTD_bm_stable */
+                    unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip + iSize == iend);
                     cSize = lastBlock ?
                             ZSTD_compressEnd(zcs, cDst, oSize, ip, iSize) :
                             ZSTD_compressContinue(zcs, cDst, oSize, ip, iSize);
                     /* Consume the input prior to error checking to mirror buffered mode. */
-                    if (iSize > 0)
-                        ip += iSize;
+                    if (ip) ip += iSize;
                     FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
                     zcs->frameEnded = lastBlock;
-                    if (lastBlock)
-                        assert(ip == iend);
+                    if (lastBlock) assert(ip == iend);
                 }
                 if (cDst == op) {  /* no need to flush */
                     op += cSize;
@@ -5403,7 +5900,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 zcs->outBuffFlushedSize = 0;
                 zcs->streamStage = zcss_flush; /* pass-through to flush stage */
             }
-	    /* fall-through */
+	    ZSTD_FALLTHROUGH;
         case zcss_flush:
             DEBUGLOG(5, "flush stage");
             assert(zcs->appliedParams.outBufferMode == ZSTD_bm_buffered);
@@ -5464,8 +5961,10 @@ size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuf
 /* After a compression call set the expected input/output buffer.
  * This is validated at the start of the next compression call.
  */
-static void ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, ZSTD_outBuffer const* output, ZSTD_inBuffer const* input)
+static void
+ZSTD_setBufferExpectations(ZSTD_CCtx* cctx, const ZSTD_outBuffer* output, const ZSTD_inBuffer* input)
 {
+    DEBUGLOG(5, "ZSTD_setBufferExpectations (for advanced stable in/out modes)");
     if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
         cctx->expectedInBuffer = *input;
     }
@@ -5484,22 +5983,22 @@ static size_t ZSTD_checkBufferStability(ZSTD_CCtx const* cctx,
 {
     if (cctx->appliedParams.inBufferMode == ZSTD_bm_stable) {
         ZSTD_inBuffer const expect = cctx->expectedInBuffer;
-        if (expect.src != input->src || expect.pos != input->pos || expect.size != input->size)
-            RETURN_ERROR(srcBuffer_wrong, "ZSTD_c_stableInBuffer enabled but input differs!");
-        if (endOp != ZSTD_e_end)
-            RETURN_ERROR(srcBuffer_wrong, "ZSTD_c_stableInBuffer can only be used with ZSTD_e_end!");
+        if (expect.src != input->src || expect.pos != input->pos)
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableInBuffer enabled but input differs!");
     }
+    (void)endOp;
     if (cctx->appliedParams.outBufferMode == ZSTD_bm_stable) {
         size_t const outBufferSize = output->size - output->pos;
         if (cctx->expectedOutBufferSize != outBufferSize)
-            RETURN_ERROR(dstBuffer_wrong, "ZSTD_c_stableOutBuffer enabled but output size differs!");
+            RETURN_ERROR(stabilityCondition_notRespected, "ZSTD_c_stableOutBuffer enabled but output size differs!");
     }
     return 0;
 }
 
 static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
                                              ZSTD_EndDirective endOp,
-                                             size_t inSize) {
+                                             size_t inSize)
+{
     ZSTD_CCtx_params params = cctx->requestedParams;
     ZSTD_prefixDict const prefixDict = cctx->prefixDict;
     FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
@@ -5513,9 +6012,9 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
         params.compressionLevel = cctx->cdict->compressionLevel;
     }
     DEBUGLOG(4, "ZSTD_compressStream2 : transparent init stage");
-    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-fix pledgedSrcSize */
-    {
-        size_t const dictSize = prefixDict.dict
+    if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = inSize + 1;  /* auto-determine pledgedSrcSize */
+
+    {   size_t const dictSize = prefixDict.dict
                 ? prefixDict.dictSize
                 : (cctx->cdict ? cctx->cdict->dictContentSize : 0);
         ZSTD_cParamMode_e const mode = ZSTD_getCParamMode(cctx->cdict, ¶ms, cctx->pledgedSrcSizePlusOne - 1);
@@ -5524,20 +6023,21 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
                 dictSize, mode);
     }
 
-    if (ZSTD_CParams_shouldEnableLdm(¶ms.cParams)) {
-        /* Enable LDM by default for optimal parser and window size >= 128MB */
-        DEBUGLOG(4, "LDM enabled by default (window size >= 128MB, strategy >= btopt)");
-        params.ldmParams.enableLdm = 1;
-    }
-
-    if (ZSTD_CParams_useBlockSplitter(¶ms.cParams)) {
-        DEBUGLOG(4, "Block splitter enabled by default (window size >= 128K, strategy >= btopt)");
-        params.splitBlocks = 1;
-    }
-
+    params.useBlockSplitter = ZSTD_resolveBlockSplitterMode(params.useBlockSplitter, ¶ms.cParams);
+    params.ldmParams.enableLdm = ZSTD_resolveEnableLdm(params.ldmParams.enableLdm, ¶ms.cParams);
     params.useRowMatchFinder = ZSTD_resolveRowMatchFinderMode(params.useRowMatchFinder, ¶ms.cParams);
+    params.validateSequences = ZSTD_resolveExternalSequenceValidation(params.validateSequences);
+    params.maxBlockSize = ZSTD_resolveMaxBlockSize(params.maxBlockSize);
+    params.searchForExternalRepcodes = ZSTD_resolveExternalRepcodeSearch(params.searchForExternalRepcodes, params.compressionLevel);
 
 #ifdef ZSTD_MULTITHREAD
+    /* If external matchfinder is enabled, make sure to fail before checking job size (for consistency) */
+    RETURN_ERROR_IF(
+        params.useSequenceProducer == 1 && params.nbWorkers >= 1,
+        parameter_combination_unsupported,
+        "External sequence producer isn't supported with nbWorkers >= 1"
+    );
+
     if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
         params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
     }
@@ -5565,7 +6065,7 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
         cctx->streamStage = zcss_load;
         cctx->appliedParams = params;
     } else
-#endif
+#endif  /* ZSTD_MULTITHREAD */
     {   U64 const pledgedSrcSize = cctx->pledgedSrcSizePlusOne - 1;
         assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
         FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
@@ -5591,6 +6091,8 @@ static size_t ZSTD_CCtx_init_compressStream2(ZSTD_CCtx* cctx,
     return 0;
 }
 
+/* @return provides a minimum amount of data remaining to be flushed from internal buffers
+ */
 size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
                              ZSTD_outBuffer* output,
                              ZSTD_inBuffer* input,
@@ -5605,8 +6107,27 @@ size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
 
     /* transparent initialization stage */
     if (cctx->streamStage == zcss_init) {
-        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, input->size), "CompressStream2 initialization failed");
-        ZSTD_setBufferExpectations(cctx, output, input);    /* Set initial buffer expectations now that we've initialized */
+        size_t const inputSize = input->size - input->pos;  /* no obligation to start from pos==0 */
+        size_t const totalInputSize = inputSize + cctx->stableIn_notConsumed;
+        if ( (cctx->requestedParams.inBufferMode == ZSTD_bm_stable) /* input is presumed stable, across invocations */
+          && (endOp == ZSTD_e_continue)                             /* no flush requested, more input to come */
+          && (totalInputSize < ZSTD_BLOCKSIZE_MAX) ) {              /* not even reached one block yet */
+            if (cctx->stableIn_notConsumed) {  /* not the first time */
+                /* check stable source guarantees */
+                RETURN_ERROR_IF(input->src != cctx->expectedInBuffer.src, stabilityCondition_notRespected, "stableInBuffer condition not respected: wrong src pointer");
+                RETURN_ERROR_IF(input->pos != cctx->expectedInBuffer.size, stabilityCondition_notRespected, "stableInBuffer condition not respected: externally modified pos");
+            }
+            /* pretend input was consumed, to give a sense forward progress */
+            input->pos = input->size;
+            /* save stable inBuffer, for later control, and flush/end */
+            cctx->expectedInBuffer = *input;
+            /* but actually input wasn't consumed, so keep track of position from where compression shall resume */
+            cctx->stableIn_notConsumed += inputSize;
+            /* don't initialize yet, wait for the first block of flush() order, for better parameters adaptation */
+            return ZSTD_FRAMEHEADERSIZE_MIN(cctx->requestedParams.format);  /* at least some header to produce */
+        }
+        FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, endOp, totalInputSize), "compressStream2 initialization failed");
+        ZSTD_setBufferExpectations(cctx, output, input);   /* Set initial buffer expectations now that we've initialized */
     }
     /* end of transparent initialization stage */
 
@@ -5619,6 +6140,13 @@ size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
             ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
             cctx->cParamsChanged = 0;
         }
+        if (cctx->stableIn_notConsumed) {
+            assert(cctx->appliedParams.inBufferMode == ZSTD_bm_stable);
+            /* some early data was skipped - make it available for consumption */
+            assert(input->pos >= cctx->stableIn_notConsumed);
+            input->pos -= cctx->stableIn_notConsumed;
+            cctx->stableIn_notConsumed = 0;
+        }
         for (;;) {
             size_t const ipos = input->pos;
             size_t const opos = output->pos;
@@ -5657,7 +6185,7 @@ size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
         ZSTD_setBufferExpectations(cctx, output, input);
         return flushMin;
     }
-#endif
+#endif /* ZSTD_MULTITHREAD */
     FORWARD_IF_ERROR( ZSTD_compressStream_generic(cctx, output, input, endOp) , "");
     DEBUGLOG(5, "completed ZSTD_compressStream2");
     ZSTD_setBufferExpectations(cctx, output, input);
@@ -5670,13 +6198,20 @@ size_t ZSTD_compressStream2_simpleArgs (
                       const void* src, size_t srcSize, size_t* srcPos,
                             ZSTD_EndDirective endOp)
 {
-    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
-    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
     /* ZSTD_compressStream2() will check validity of dstPos and srcPos */
-    size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
-    *dstPos = output.pos;
-    *srcPos = input.pos;
-    return cErr;
+    {   size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
 }
 
 size_t ZSTD_compress2(ZSTD_CCtx* cctx,
@@ -5699,6 +6234,7 @@ size_t ZSTD_compress2(ZSTD_CCtx* cctx,
         /* Reset to the original values. */
         cctx->requestedParams.inBufferMode = originalInBufferMode;
         cctx->requestedParams.outBufferMode = originalOutBufferMode;
+
         FORWARD_IF_ERROR(result, "ZSTD_compressStream2_simpleArgs failed");
         if (result != 0) {  /* compression not completed, due to lack of output space */
             assert(oPos == dstCapacity);
@@ -5709,64 +6245,60 @@ size_t ZSTD_compress2(ZSTD_CCtx* cctx,
     }
 }
 
-typedef struct {
-    U32 idx;             /* Index in array of ZSTD_Sequence */
-    U32 posInSequence;   /* Position within sequence at idx */
-    size_t posInSrc;        /* Number of bytes given by sequences provided so far */
-} ZSTD_sequencePosition;
-
-/* Returns a ZSTD error code if sequence is not valid */
-static size_t ZSTD_validateSequence(U32 offCode, U32 matchLength,
-                                    size_t posInSrc, U32 windowLog, size_t dictSize, U32 minMatch) {
-    size_t offsetBound;
-    U32 windowSize = 1 << windowLog;
-    /* posInSrc represents the amount of data the the decoder would decode up to this point.
+/* ZSTD_validateSequence() :
+ * @offCode : is presumed to follow format required by ZSTD_storeSeq()
+ * @returns a ZSTD error code if sequence is not valid
+ */
+static size_t
+ZSTD_validateSequence(U32 offCode, U32 matchLength, U32 minMatch,
+                      size_t posInSrc, U32 windowLog, size_t dictSize, int useSequenceProducer)
+{
+    U32 const windowSize = 1u << windowLog;
+    /* posInSrc represents the amount of data the decoder would decode up to this point.
      * As long as the amount of data decoded is less than or equal to window size, offsets may be
      * larger than the total length of output decoded in order to reference the dict, even larger than
      * window size. After output surpasses windowSize, we're limited to windowSize offsets again.
      */
-    offsetBound = posInSrc > windowSize ? (size_t)windowSize : posInSrc + (size_t)dictSize;
-    RETURN_ERROR_IF(offCode > offsetBound + ZSTD_REP_MOVE, corruption_detected, "Offset too large!");
-    RETURN_ERROR_IF(matchLength < minMatch, corruption_detected, "Matchlength too small");
+    size_t const offsetBound = posInSrc > windowSize ? (size_t)windowSize : posInSrc + (size_t)dictSize;
+    size_t const matchLenLowerBound = (minMatch == 3 || useSequenceProducer) ? 3 : 4;
+    RETURN_ERROR_IF(offCode > OFFSET_TO_OFFBASE(offsetBound), externalSequences_invalid, "Offset too large!");
+    /* Validate maxNbSeq is large enough for the given matchLength and minMatch */
+    RETURN_ERROR_IF(matchLength < matchLenLowerBound, externalSequences_invalid, "Matchlength too small for the minMatch");
     return 0;
 }
 
 /* Returns an offset code, given a sequence's raw offset, the ongoing repcode array, and whether litLength == 0 */
-static U32 ZSTD_finalizeOffCode(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0) {
-    U32 offCode = rawOffset + ZSTD_REP_MOVE;
-    U32 repCode = 0;
+static U32 ZSTD_finalizeOffBase(U32 rawOffset, const U32 rep[ZSTD_REP_NUM], U32 ll0)
+{
+    U32 offBase = OFFSET_TO_OFFBASE(rawOffset);
 
     if (!ll0 && rawOffset == rep[0]) {
-        repCode = 1;
+        offBase = REPCODE1_TO_OFFBASE;
     } else if (rawOffset == rep[1]) {
-        repCode = 2 - ll0;
+        offBase = REPCODE_TO_OFFBASE(2 - ll0);
     } else if (rawOffset == rep[2]) {
-        repCode = 3 - ll0;
+        offBase = REPCODE_TO_OFFBASE(3 - ll0);
     } else if (ll0 && rawOffset == rep[0] - 1) {
-        repCode = 3;
+        offBase = REPCODE3_TO_OFFBASE;
     }
-    if (repCode) {
-        /* ZSTD_storeSeq expects a number in the range [0, 2] to represent a repcode */
-        offCode = repCode - 1;
-    }
-    return offCode;
+    return offBase;
 }
 
-/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
- * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
- */
-static size_t ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
-                                                             const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
-                                                             const void* src, size_t blockSize) {
+size_t
+ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
+                                              ZSTD_sequencePosition* seqPos,
+                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                        const void* src, size_t blockSize,
+                                        ZSTD_paramSwitch_e externalRepSearch)
+{
     U32 idx = seqPos->idx;
+    U32 const startIdx = idx;
     BYTE const* ip = (BYTE const*)(src);
     const BYTE* const iend = ip + blockSize;
     repcodes_t updatedRepcodes;
     U32 dictSize;
-    U32 litLength;
-    U32 matchLength;
-    U32 ll0;
-    U32 offCode;
+
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStoreExplicitBlockDelim (blockSize = %zu)", blockSize);
 
     if (cctx->cdict) {
         dictSize = (U32)cctx->cdict->dictContentSize;
@@ -5776,26 +6308,55 @@ static size_t ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx, ZS
         dictSize = 0;
     }
     ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
-    for (; (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0) && idx < inSeqsSize; ++idx) {
-        litLength = inSeqs[idx].litLength;
-        matchLength = inSeqs[idx].matchLength;
-        ll0 = litLength == 0;
-        offCode = ZSTD_finalizeOffCode(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
-        updatedRepcodes = ZSTD_updateRep(updatedRepcodes.rep, offCode, ll0);
-
-        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offCode, matchLength, litLength);
+    for (; idx < inSeqsSize && (inSeqs[idx].matchLength != 0 || inSeqs[idx].offset != 0); ++idx) {
+        U32 const litLength = inSeqs[idx].litLength;
+        U32 const matchLength = inSeqs[idx].matchLength;
+        U32 offBase;
+
+        if (externalRepSearch == ZSTD_ps_disable) {
+            offBase = OFFSET_TO_OFFBASE(inSeqs[idx].offset);
+        } else {
+            U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(inSeqs[idx].offset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
+        }
+
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
         if (cctx->appliedParams.validateSequences) {
             seqPos->posInSrc += litLength + matchLength;
-            FORWARD_IF_ERROR(ZSTD_validateSequence(offCode, matchLength, seqPos->posInSrc,
-                                                cctx->appliedParams.cParams.windowLog, dictSize,
-                                                cctx->appliedParams.cParams.minMatch),
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                cctx->appliedParams.cParams.windowLog, dictSize, cctx->appliedParams.useSequenceProducer),
                                                 "Sequence validation failed");
         }
-        RETURN_ERROR_IF(idx - seqPos->idx > cctx->seqStore.maxNbSeq, memory_allocation,
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
                         "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
-        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offCode, matchLength - MINMATCH);
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
         ip += matchLength + litLength;
     }
+
+    /* If we skipped repcode search while parsing, we need to update repcodes now */
+    assert(externalRepSearch != ZSTD_ps_auto);
+    assert(idx >= startIdx);
+    if (externalRepSearch == ZSTD_ps_disable && idx != startIdx) {
+        U32* const rep = updatedRepcodes.rep;
+        U32 lastSeqIdx = idx - 1; /* index of last non-block-delimiter sequence */
+
+        if (lastSeqIdx >= startIdx + 2) {
+            rep[2] = inSeqs[lastSeqIdx - 2].offset;
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else if (lastSeqIdx == startIdx + 1) {
+            rep[2] = rep[0];
+            rep[1] = inSeqs[lastSeqIdx - 1].offset;
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        } else {
+            assert(lastSeqIdx == startIdx);
+            rep[2] = rep[1];
+            rep[1] = rep[0];
+            rep[0] = inSeqs[lastSeqIdx].offset;
+        }
+    }
+
     ZSTD_memcpy(cctx->blockState.nextCBlock->rep, updatedRepcodes.rep, sizeof(repcodes_t));
 
     if (inSeqs[idx].litLength) {
@@ -5804,25 +6365,16 @@ static size_t ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx, ZS
         ip += inSeqs[idx].litLength;
         seqPos->posInSrc += inSeqs[idx].litLength;
     }
-    RETURN_ERROR_IF(ip != iend, corruption_detected, "Blocksize doesn't agree with block delimiter!");
+    RETURN_ERROR_IF(ip != iend, externalSequences_invalid, "Blocksize doesn't agree with block delimiter!");
     seqPos->idx = idx+1;
     return 0;
 }
 
-/* Returns the number of bytes to move the current read position back by. Only non-zero
- * if we ended up splitting a sequence. Otherwise, it may return a ZSTD error if something
- * went wrong.
- *
- * This function will attempt to scan through blockSize bytes represented by the sequences
- * in inSeqs, storing any (partial) sequences.
- *
- * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
- * avoid splitting a match, or to avoid splitting a match such that it would produce a match
- * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
- */
-static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
-                                                       const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
-                                                       const void* src, size_t blockSize) {
+size_t
+ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                   const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                   const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch)
+{
     U32 idx = seqPos->idx;
     U32 startPosInSequence = seqPos->posInSequence;
     U32 endPosInSequence = seqPos->posInSequence + (U32)blockSize;
@@ -5832,10 +6384,9 @@ static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_seq
     repcodes_t updatedRepcodes;
     U32 bytesAdjustment = 0;
     U32 finalMatchSplit = 0;
-    U32 litLength;
-    U32 matchLength;
-    U32 rawOffset;
-    U32 offCode;
+
+    /* TODO(embg) support fast parsing mode in noBlockDelim mode */
+    (void)externalRepSearch;
 
     if (cctx->cdict) {
         dictSize = cctx->cdict->dictContentSize;
@@ -5844,14 +6395,15 @@ static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_seq
     } else {
         dictSize = 0;
     }
-    DEBUGLOG(5, "ZSTD_copySequencesToSeqStore: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
+    DEBUGLOG(5, "ZSTD_copySequencesToSeqStoreNoBlockDelim: idx: %u PIS: %u blockSize: %zu", idx, startPosInSequence, blockSize);
     DEBUGLOG(5, "Start seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
     ZSTD_memcpy(updatedRepcodes.rep, cctx->blockState.prevCBlock->rep, sizeof(repcodes_t));
     while (endPosInSequence && idx < inSeqsSize && !finalMatchSplit) {
         const ZSTD_Sequence currSeq = inSeqs[idx];
-        litLength = currSeq.litLength;
-        matchLength = currSeq.matchLength;
-        rawOffset = currSeq.offset;
+        U32 litLength = currSeq.litLength;
+        U32 matchLength = currSeq.matchLength;
+        U32 const rawOffset = currSeq.offset;
+        U32 offBase;
 
         /* Modify the sequence depending on where endPosInSequence lies */
         if (endPosInSequence >= currSeq.litLength + currSeq.matchLength) {
@@ -5865,7 +6417,6 @@ static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_seq
             /* Move to the next sequence */
             endPosInSequence -= currSeq.litLength + currSeq.matchLength;
             startPosInSequence = 0;
-            idx++;
         } else {
             /* This is the final (partial) sequence we're adding from inSeqs, and endPosInSequence
                does not reach the end of the match. So, we have to split the sequence */
@@ -5904,23 +6455,24 @@ static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_seq
             }
         }
         /* Check if this offset can be represented with a repcode */
-        {   U32 ll0 = (litLength == 0);
-            offCode = ZSTD_finalizeOffCode(rawOffset, updatedRepcodes.rep, ll0);
-            updatedRepcodes = ZSTD_updateRep(updatedRepcodes.rep, offCode, ll0);
+        {   U32 const ll0 = (litLength == 0);
+            offBase = ZSTD_finalizeOffBase(rawOffset, updatedRepcodes.rep, ll0);
+            ZSTD_updateRep(updatedRepcodes.rep, offBase, ll0);
         }
 
         if (cctx->appliedParams.validateSequences) {
             seqPos->posInSrc += litLength + matchLength;
-            FORWARD_IF_ERROR(ZSTD_validateSequence(offCode, matchLength, seqPos->posInSrc,
-                                                   cctx->appliedParams.cParams.windowLog, dictSize,
-                                                   cctx->appliedParams.cParams.minMatch),
+            FORWARD_IF_ERROR(ZSTD_validateSequence(offBase, matchLength, cctx->appliedParams.cParams.minMatch, seqPos->posInSrc,
+                                                   cctx->appliedParams.cParams.windowLog, dictSize, cctx->appliedParams.useSequenceProducer),
                                                    "Sequence validation failed");
         }
-        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offCode, matchLength, litLength);
-        RETURN_ERROR_IF(idx - seqPos->idx > cctx->seqStore.maxNbSeq, memory_allocation,
+        DEBUGLOG(6, "Storing sequence: (of: %u, ml: %u, ll: %u)", offBase, matchLength, litLength);
+        RETURN_ERROR_IF(idx - seqPos->idx >= cctx->seqStore.maxNbSeq, externalSequences_invalid,
                         "Not enough memory allocated. Try adjusting ZSTD_c_minMatch.");
-        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offCode, matchLength - MINMATCH);
+        ZSTD_storeSeq(&cctx->seqStore, litLength, ip, iend, offBase, matchLength);
         ip += matchLength + litLength;
+        if (!finalMatchSplit)
+            idx++; /* Next Sequence */
     }
     DEBUGLOG(5, "Ending seq: idx: %u (of: %u ml: %u ll: %u)", idx, inSeqs[idx].offset, inSeqs[idx].matchLength, inSeqs[idx].litLength);
     assert(idx == inSeqsSize || endPosInSequence <= inSeqs[idx].litLength + inSeqs[idx].matchLength);
@@ -5943,8 +6495,9 @@ static size_t ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_seq
 
 typedef size_t (*ZSTD_sequenceCopier) (ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
-                                       const void* src, size_t blockSize);
-static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode) {
+                                       const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode)
+{
     ZSTD_sequenceCopier sequenceCopier = NULL;
     assert(ZSTD_cParam_withinBounds(ZSTD_c_blockDelimiters, mode));
     if (mode == ZSTD_sf_explicitBlockDelimiters) {
@@ -5956,24 +6509,75 @@ static ZSTD_sequenceCopier ZSTD_selectSequenceCopier(ZSTD_sequenceFormat_e mode)
     return sequenceCopier;
 }
 
+/* Discover the size of next block by searching for the delimiter.
+ * Note that a block delimiter **must** exist in this mode,
+ * otherwise it's an input error.
+ * The block size retrieved will be later compared to ensure it remains within bounds */
+static size_t
+blockSize_explicitDelimiter(const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_sequencePosition seqPos)
+{
+    int end = 0;
+    size_t blockSize = 0;
+    size_t spos = seqPos.idx;
+    DEBUGLOG(6, "blockSize_explicitDelimiter : seq %zu / %zu", spos, inSeqsSize);
+    assert(spos <= inSeqsSize);
+    while (spos < inSeqsSize) {
+        end = (inSeqs[spos].offset == 0);
+        blockSize += inSeqs[spos].litLength + inSeqs[spos].matchLength;
+        if (end) {
+            if (inSeqs[spos].matchLength != 0)
+                RETURN_ERROR(externalSequences_invalid, "delimiter format error : both matchlength and offset must be == 0");
+            break;
+        }
+        spos++;
+    }
+    if (!end)
+        RETURN_ERROR(externalSequences_invalid, "Reached end of sequences without finding a block delimiter");
+    return blockSize;
+}
+
+/* More a "target" block size */
+static size_t blockSize_noDelimiter(size_t blockSize, size_t remaining)
+{
+    int const lastBlock = (remaining <= blockSize);
+    return lastBlock ? remaining : blockSize;
+}
+
+static size_t determine_blockSize(ZSTD_sequenceFormat_e mode,
+                           size_t blockSize, size_t remaining,
+                     const ZSTD_Sequence* inSeqs, size_t inSeqsSize, ZSTD_sequencePosition seqPos)
+{
+    DEBUGLOG(6, "determine_blockSize : remainingSize = %zu", remaining);
+    if (mode == ZSTD_sf_noBlockDelimiters)
+        return blockSize_noDelimiter(blockSize, remaining);
+    {   size_t const explicitBlockSize = blockSize_explicitDelimiter(inSeqs, inSeqsSize, seqPos);
+        FORWARD_IF_ERROR(explicitBlockSize, "Error while determining block size with explicit delimiters");
+        if (explicitBlockSize > blockSize)
+            RETURN_ERROR(externalSequences_invalid, "sequences incorrectly define a too large block");
+        if (explicitBlockSize > remaining)
+            RETURN_ERROR(externalSequences_invalid, "sequences define a frame longer than source");
+        return explicitBlockSize;
+    }
+}
+
 /* Compress, block-by-block, all of the sequences given.
  *
- * Returns the cumulative size of all compressed blocks (including their headers), otherwise a ZSTD error.
+ * Returns the cumulative size of all compressed blocks (including their headers),
+ * otherwise a ZSTD error.
  */
-static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
-                                              void* dst, size_t dstCapacity,
-                                              const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
-                                              const void* src, size_t srcSize) {
+static size_t
+ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                          const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                          const void* src, size_t srcSize)
+{
     size_t cSize = 0;
-    U32 lastBlock;
-    size_t blockSize;
-    size_t compressedSeqsSize;
     size_t remaining = srcSize;
     ZSTD_sequencePosition seqPos = {0, 0, 0};
 
     BYTE const* ip = (BYTE const*)src;
     BYTE* op = (BYTE*)dst;
-    ZSTD_sequenceCopier sequenceCopier = ZSTD_selectSequenceCopier(cctx->appliedParams.blockDelimiters);
+    ZSTD_sequenceCopier const sequenceCopier = ZSTD_selectSequenceCopier(cctx->appliedParams.blockDelimiters);
 
     DEBUGLOG(4, "ZSTD_compressSequences_internal srcSize: %zu, inSeqsSize: %zu", srcSize, inSeqsSize);
     /* Special case: empty frame */
@@ -5987,22 +6591,29 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
     }
 
     while (remaining) {
+        size_t compressedSeqsSize;
         size_t cBlockSize;
         size_t additionalByteAdjustment;
-        lastBlock = remaining <= cctx->blockSize;
-        blockSize = lastBlock ? (U32)remaining : (U32)cctx->blockSize;
+        size_t blockSize = determine_blockSize(cctx->appliedParams.blockDelimiters,
+                                        cctx->blockSize, remaining,
+                                        inSeqs, inSeqsSize, seqPos);
+        U32 const lastBlock = (blockSize == remaining);
+        FORWARD_IF_ERROR(blockSize, "Error while trying to determine block size");
+        assert(blockSize <= remaining);
         ZSTD_resetSeqStore(&cctx->seqStore);
-        DEBUGLOG(4, "Working on new block. Blocksize: %zu", blockSize);
+        DEBUGLOG(5, "Working on new block. Blocksize: %zu (total:%zu)", blockSize, (ip - (const BYTE*)src) + blockSize);
 
-        additionalByteAdjustment = sequenceCopier(cctx, &seqPos, inSeqs, inSeqsSize, ip, blockSize);
+        additionalByteAdjustment = sequenceCopier(cctx, &seqPos, inSeqs, inSeqsSize, ip, blockSize, cctx->appliedParams.searchForExternalRepcodes);
         FORWARD_IF_ERROR(additionalByteAdjustment, "Bad sequence copy");
         blockSize -= additionalByteAdjustment;
 
         /* If blocks are too small, emit as a nocompress block */
-        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+        /* TODO: See 3090. We reduced MIN_CBLOCK_SIZE from 3 to 2 so to compensate we are adding
+         * additional 1. We need to revisit and change this logic to be more consistent */
+        if (blockSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1+1) {
             cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
             FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
-            DEBUGLOG(4, "Block too small, writing out nocompress block: cSize: %zu", cBlockSize);
+            DEBUGLOG(5, "Block too small, writing out nocompress block: cSize: %zu", cBlockSize);
             cSize += cBlockSize;
             ip += blockSize;
             op += cBlockSize;
@@ -6011,6 +6622,7 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
             continue;
         }
 
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall, "not enough dstCapacity to write a new compressed block");
         compressedSeqsSize = ZSTD_entropyCompressSeqStore(&cctx->seqStore,
                                 &cctx->blockState.prevCBlock->entropy, &cctx->blockState.nextCBlock->entropy,
                                 &cctx->appliedParams,
@@ -6019,11 +6631,11 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
                                 cctx->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
                                 cctx->bmi2);
         FORWARD_IF_ERROR(compressedSeqsSize, "Compressing sequences of block failed");
-        DEBUGLOG(4, "Compressed sequences size: %zu", compressedSeqsSize);
+        DEBUGLOG(5, "Compressed sequences size: %zu", compressedSeqsSize);
 
         if (!cctx->isFirstBlock &&
             ZSTD_maybeRLE(&cctx->seqStore) &&
-            ZSTD_isRLE((BYTE const*)src, srcSize)) {
+            ZSTD_isRLE(ip, blockSize)) {
             /* We don't want to emit our first block as a RLE even if it qualifies because
             * doing so will cause the decoder (cli only) to throw a "should consume all input error."
             * This is only an issue for zstd <= v1.4.3
@@ -6034,12 +6646,12 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
         if (compressedSeqsSize == 0) {
             /* ZSTD_noCompressBlock writes the block header as well */
             cBlockSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
-            FORWARD_IF_ERROR(cBlockSize, "Nocompress block failed");
-            DEBUGLOG(4, "Writing out nocompress block, size: %zu", cBlockSize);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_noCompressBlock failed");
+            DEBUGLOG(5, "Writing out nocompress block, size: %zu", cBlockSize);
         } else if (compressedSeqsSize == 1) {
             cBlockSize = ZSTD_rleCompressBlock(op, dstCapacity, *ip, blockSize, lastBlock);
-            FORWARD_IF_ERROR(cBlockSize, "RLE compress block failed");
-            DEBUGLOG(4, "Writing out RLE block, size: %zu", cBlockSize);
+            FORWARD_IF_ERROR(cBlockSize, "ZSTD_rleCompressBlock failed");
+            DEBUGLOG(5, "Writing out RLE block, size: %zu", cBlockSize);
         } else {
             U32 cBlockHeader;
             /* Error checking and repcodes update */
@@ -6051,11 +6663,10 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
             cBlockHeader = lastBlock + (((U32)bt_compressed)<<1) + (U32)(compressedSeqsSize << 3);
             MEM_writeLE24(op, cBlockHeader);
             cBlockSize = ZSTD_blockHeaderSize + compressedSeqsSize;
-            DEBUGLOG(4, "Writing out compressed block, size: %zu", cBlockSize);
+            DEBUGLOG(5, "Writing out compressed block, size: %zu", cBlockSize);
         }
 
         cSize += cBlockSize;
-        DEBUGLOG(4, "cSize running total: %zu", cSize);
 
         if (lastBlock) {
             break;
@@ -6066,21 +6677,25 @@ static size_t ZSTD_compressSequences_internal(ZSTD_CCtx* cctx,
             dstCapacity -= cBlockSize;
             cctx->isFirstBlock = 0;
         }
+        DEBUGLOG(5, "cSize running total: %zu (remaining dstCapacity=%zu)", cSize, dstCapacity);
     }
 
+    DEBUGLOG(4, "cSize final total: %zu", cSize);
     return cSize;
 }
 
-size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstCapacity,
+size_t ZSTD_compressSequences(ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
                               const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
-                              const void* src, size_t srcSize) {
+                              const void* src, size_t srcSize)
+{
     BYTE* op = (BYTE*)dst;
     size_t cSize = 0;
     size_t compressedBlocksSize = 0;
     size_t frameHeaderSize = 0;
 
     /* Transparent initialization stage, same as compressStream2() */
-    DEBUGLOG(3, "ZSTD_compressSequences()");
+    DEBUGLOG(4, "ZSTD_compressSequences (dstCapacity=%zu)", dstCapacity);
     assert(cctx != NULL);
     FORWARD_IF_ERROR(ZSTD_CCtx_init_compressStream2(cctx, ZSTD_e_end, srcSize), "CCtx initialization failed");
     /* Begin writing output, starting with frame header */
@@ -6108,26 +6723,34 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstCapaci
         cSize += 4;
     }
 
-    DEBUGLOG(3, "Final compressed size: %zu", cSize);
+    DEBUGLOG(4, "Final compressed size: %zu", cSize);
     return cSize;
 }
 
 /*======   Finalize   ======*/
 
+static ZSTD_inBuffer inBuffer_forEndFlush(const ZSTD_CStream* zcs)
+{
+    const ZSTD_inBuffer nullInput = { NULL, 0, 0 };
+    const int stableInput = (zcs->appliedParams.inBufferMode == ZSTD_bm_stable);
+    return stableInput ? zcs->expectedInBuffer : nullInput;
+}
+
 /*! ZSTD_flushStream() :
  * @return : amount of data remaining to flush */
 size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
-    ZSTD_inBuffer input = { NULL, 0, 0 };
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
+    input.size = input.pos; /* do not ingest more input during flush */
     return ZSTD_compressStream2(zcs, output, &input, ZSTD_e_flush);
 }
 
 
 size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
-    ZSTD_inBuffer input = { NULL, 0, 0 };
+    ZSTD_inBuffer input = inBuffer_forEndFlush(zcs);
     size_t const remainingToFlush = ZSTD_compressStream2(zcs, output, &input, ZSTD_e_end);
-    FORWARD_IF_ERROR( remainingToFlush , "ZSTD_compressStream2 failed");
+    FORWARD_IF_ERROR(remainingToFlush , "ZSTD_compressStream2(,,ZSTD_e_end) failed");
     if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush;   /* minimal estimation */
     /* single thread mode : attempt to calculate remaining to flush more precisely */
     {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
@@ -6140,119 +6763,12 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 
 
 /*-=====  Pre-defined compression levels  =====-*/
+#include "clevels.h"
 
-#define ZSTD_MAX_CLEVEL     22
 int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
 int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
 int ZSTD_defaultCLevel(void) { return ZSTD_CLEVEL_DEFAULT; }
 
-static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
-{   /* "default" - for any srcSize > 256 KB */
-    /* W,  C,  H,  S,  L, TL, strat */
-    { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
-    { 19, 13, 14,  1,  7,  0, ZSTD_fast    },  /* level  1 */
-    { 20, 15, 16,  1,  6,  0, ZSTD_fast    },  /* level  2 */
-    { 21, 16, 17,  1,  5,  0, ZSTD_dfast   },  /* level  3 */
-    { 21, 18, 18,  1,  5,  0, ZSTD_dfast   },  /* level  4 */
-    { 21, 18, 19,  2,  5,  2, ZSTD_greedy  },  /* level  5 */
-    { 21, 19, 19,  3,  5,  4, ZSTD_greedy  },  /* level  6 */
-    { 21, 19, 19,  3,  5,  8, ZSTD_lazy    },  /* level  7 */
-    { 21, 19, 19,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
-    { 21, 19, 20,  4,  5, 16, ZSTD_lazy2   },  /* level  9 */
-    { 22, 20, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
-    { 22, 21, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
-    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
-    { 22, 21, 22,  5,  5, 32, ZSTD_btlazy2 },  /* level 13 */
-    { 22, 22, 23,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
-    { 22, 23, 23,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
-    { 22, 22, 22,  5,  5, 48, ZSTD_btopt   },  /* level 16 */
-    { 23, 23, 22,  5,  4, 64, ZSTD_btopt   },  /* level 17 */
-    { 23, 23, 22,  6,  3, 64, ZSTD_btultra },  /* level 18 */
-    { 23, 24, 22,  7,  3,256, ZSTD_btultra2},  /* level 19 */
-    { 25, 25, 23,  7,  3,256, ZSTD_btultra2},  /* level 20 */
-    { 26, 26, 24,  7,  3,512, ZSTD_btultra2},  /* level 21 */
-    { 27, 27, 25,  9,  3,999, ZSTD_btultra2},  /* level 22 */
-},
-{   /* for srcSize <= 256 KB */
-    /* W,  C,  H,  S,  L,  T, strat */
-    { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
-    { 18, 13, 14,  1,  6,  0, ZSTD_fast    },  /* level  1 */
-    { 18, 14, 14,  1,  5,  0, ZSTD_dfast   },  /* level  2 */
-    { 18, 16, 16,  1,  4,  0, ZSTD_dfast   },  /* level  3 */
-    { 18, 16, 17,  2,  5,  2, ZSTD_greedy  },  /* level  4.*/
-    { 18, 18, 18,  3,  5,  2, ZSTD_greedy  },  /* level  5.*/
-    { 18, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6.*/
-    { 18, 18, 19,  4,  4,  4, ZSTD_lazy    },  /* level  7 */
-    { 18, 18, 19,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
-    { 18, 18, 19,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
-    { 18, 18, 19,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
-    { 18, 18, 19,  5,  4, 12, ZSTD_btlazy2 },  /* level 11.*/
-    { 18, 19, 19,  7,  4, 12, ZSTD_btlazy2 },  /* level 12.*/
-    { 18, 18, 19,  4,  4, 16, ZSTD_btopt   },  /* level 13 */
-    { 18, 18, 19,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
-    { 18, 18, 19,  6,  3,128, ZSTD_btopt   },  /* level 15.*/
-    { 18, 19, 19,  6,  3,128, ZSTD_btultra },  /* level 16.*/
-    { 18, 19, 19,  8,  3,256, ZSTD_btultra },  /* level 17.*/
-    { 18, 19, 19,  6,  3,128, ZSTD_btultra2},  /* level 18.*/
-    { 18, 19, 19,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
-    { 18, 19, 19, 10,  3,512, ZSTD_btultra2},  /* level 20.*/
-    { 18, 19, 19, 12,  3,512, ZSTD_btultra2},  /* level 21.*/
-    { 18, 19, 19, 13,  3,999, ZSTD_btultra2},  /* level 22.*/
-},
-{   /* for srcSize <= 128 KB */
-    /* W,  C,  H,  S,  L,  T, strat */
-    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
-    { 17, 12, 13,  1,  6,  0, ZSTD_fast    },  /* level  1 */
-    { 17, 13, 15,  1,  5,  0, ZSTD_fast    },  /* level  2 */
-    { 17, 15, 16,  2,  5,  0, ZSTD_dfast   },  /* level  3 */
-    { 17, 17, 17,  2,  4,  0, ZSTD_dfast   },  /* level  4 */
-    { 17, 16, 17,  3,  4,  2, ZSTD_greedy  },  /* level  5 */
-    { 17, 17, 17,  3,  4,  4, ZSTD_lazy    },  /* level  6 */
-    { 17, 17, 17,  3,  4,  8, ZSTD_lazy2   },  /* level  7 */
-    { 17, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
-    { 17, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
-    { 17, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
-    { 17, 17, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 11 */
-    { 17, 18, 17,  7,  4, 12, ZSTD_btlazy2 },  /* level 12 */
-    { 17, 18, 17,  3,  4, 12, ZSTD_btopt   },  /* level 13.*/
-    { 17, 18, 17,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
-    { 17, 18, 17,  6,  3,256, ZSTD_btopt   },  /* level 15.*/
-    { 17, 18, 17,  6,  3,128, ZSTD_btultra },  /* level 16.*/
-    { 17, 18, 17,  8,  3,256, ZSTD_btultra },  /* level 17.*/
-    { 17, 18, 17, 10,  3,512, ZSTD_btultra },  /* level 18.*/
-    { 17, 18, 17,  5,  3,256, ZSTD_btultra2},  /* level 19.*/
-    { 17, 18, 17,  7,  3,512, ZSTD_btultra2},  /* level 20.*/
-    { 17, 18, 17,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
-    { 17, 18, 17, 11,  3,999, ZSTD_btultra2},  /* level 22.*/
-},
-{   /* for srcSize <= 16 KB */
-    /* W,  C,  H,  S,  L,  T, strat */
-    { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
-    { 14, 14, 15,  1,  5,  0, ZSTD_fast    },  /* level  1 */
-    { 14, 14, 15,  1,  4,  0, ZSTD_fast    },  /* level  2 */
-    { 14, 14, 15,  2,  4,  0, ZSTD_dfast   },  /* level  3 */
-    { 14, 14, 14,  4,  4,  2, ZSTD_greedy  },  /* level  4 */
-    { 14, 14, 14,  3,  4,  4, ZSTD_lazy    },  /* level  5.*/
-    { 14, 14, 14,  4,  4,  8, ZSTD_lazy2   },  /* level  6 */
-    { 14, 14, 14,  6,  4,  8, ZSTD_lazy2   },  /* level  7 */
-    { 14, 14, 14,  8,  4,  8, ZSTD_lazy2   },  /* level  8.*/
-    { 14, 15, 14,  5,  4,  8, ZSTD_btlazy2 },  /* level  9.*/
-    { 14, 15, 14,  9,  4,  8, ZSTD_btlazy2 },  /* level 10.*/
-    { 14, 15, 14,  3,  4, 12, ZSTD_btopt   },  /* level 11.*/
-    { 14, 15, 14,  4,  3, 24, ZSTD_btopt   },  /* level 12.*/
-    { 14, 15, 14,  5,  3, 32, ZSTD_btultra },  /* level 13.*/
-    { 14, 15, 15,  6,  3, 64, ZSTD_btultra },  /* level 14.*/
-    { 14, 15, 15,  7,  3,256, ZSTD_btultra },  /* level 15.*/
-    { 14, 15, 15,  5,  3, 48, ZSTD_btultra2},  /* level 16.*/
-    { 14, 15, 15,  6,  3,128, ZSTD_btultra2},  /* level 17.*/
-    { 14, 15, 15,  7,  3,256, ZSTD_btultra2},  /* level 18.*/
-    { 14, 15, 15,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
-    { 14, 15, 15,  8,  3,512, ZSTD_btultra2},  /* level 20.*/
-    { 14, 15, 15,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
-    { 14, 15, 15, 10,  3,999, ZSTD_btultra2},  /* level 22.*/
-},
-};
-
 static ZSTD_compressionParameters ZSTD_dedicatedDictSearch_getCParams(int const compressionLevel, size_t const dictSize)
 {
     ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, 0, dictSize, ZSTD_cpm_createCDict);
@@ -6356,7 +6872,7 @@ static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel,
             cp.targetLength = (unsigned)(-clampedCompressionLevel);
         }
         /* refine parameters based on srcSize & dictSize */
-        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode);
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize, mode, ZSTD_ps_auto);
     }
 }
 
@@ -6391,3 +6907,21 @@ ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeH
     if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
     return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize, ZSTD_cpm_unknown);
 }
+
+void ZSTD_registerSequenceProducer(
+    ZSTD_CCtx* zc, void* mState,
+    ZSTD_sequenceProducer_F* mFinder
+) {
+    if (mFinder != NULL) {
+        ZSTD_externalMatchCtx emctx;
+        emctx.mState = mState;
+        emctx.mFinder = mFinder;
+        emctx.seqBuffer = NULL;
+        emctx.seqBufferCapacity = 0;
+        zc->externalMatchCtx = emctx;
+        zc->requestedParams.useSequenceProducer = 1;
+    } else {
+        ZSTD_memset(&zc->externalMatchCtx, 0, sizeof(zc->externalMatchCtx));
+        zc->requestedParams.useSequenceProducer = 0;
+    }
+}
diff --git a/priv/zstd/lib/compress/zstd_compress_internal.h b/priv/zstd/lib/compress/zstd_compress_internal.h
index 3b04fd0..cbb85e5 100644
--- a/priv/zstd/lib/compress/zstd_compress_internal.h
+++ b/priv/zstd/lib/compress/zstd_compress_internal.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -23,6 +23,7 @@
 #ifdef ZSTD_MULTITHREAD
 #  include "zstdmt_compress.h"
 #endif
+#include "../common/bits.h" /* ZSTD_highbit32, ZSTD_NbCommonBytes */
 
 #if defined (__cplusplus)
 extern "C" {
@@ -63,7 +64,7 @@ typedef struct {
 } ZSTD_localDict;
 
 typedef struct {
-    HUF_CElt CTable[HUF_CTABLE_SIZE_U32(255)];
+    HUF_CElt CTable[HUF_CTABLE_SIZE_ST(255)];
     HUF_repeat repeatMode;
 } ZSTD_hufCTables_t;
 
@@ -117,19 +118,20 @@ typedef struct {
 /** ZSTD_buildBlockEntropyStats() :
  *  Builds entropy for the block.
  *  @return : 0 on success or error code */
-size_t ZSTD_buildBlockEntropyStats(seqStore_t* seqStorePtr,
-                             const ZSTD_entropyCTables_t* prevEntropy,
-                                   ZSTD_entropyCTables_t* nextEntropy,
-                             const ZSTD_CCtx_params* cctxParams,
-                                   ZSTD_entropyCTablesMetadata_t* entropyMetadata,
-                                   void* workspace, size_t wkspSize);
+size_t ZSTD_buildBlockEntropyStats(
+                    const seqStore_t* seqStorePtr,
+                    const ZSTD_entropyCTables_t* prevEntropy,
+                          ZSTD_entropyCTables_t* nextEntropy,
+                    const ZSTD_CCtx_params* cctxParams,
+                          ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                          void* workspace, size_t wkspSize);
 
 /*********************************
 *  Compression internals structs *
 *********************************/
 
 typedef struct {
-    U32 off;            /* Offset code (offset + ZSTD_REP_MOVE) for the match */
+    U32 off;            /* Offset sumtype code for the match, using ZSTD_storeSeq() format */
     U32 len;            /* Raw length of match */
 } ZSTD_match_t;
 
@@ -148,6 +150,12 @@ typedef struct {
   size_t capacity;      /* The capacity starting from `seq` pointer */
 } rawSeqStore_t;
 
+typedef struct {
+    U32 idx;            /* Index in array of ZSTD_Sequence */
+    U32 posInSequence;  /* Position within sequence at idx */
+    size_t posInSrc;    /* Number of bytes given by sequences provided so far */
+} ZSTD_sequencePosition;
+
 UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
 
 typedef struct {
@@ -179,7 +187,7 @@ typedef struct {
     U32  offCodeSumBasePrice;    /* to compare to log2(offreq)  */
     ZSTD_OptPrice_e priceType;   /* prices can be determined dynamically, or follow a pre-defined cost structure */
     const ZSTD_entropyCTables_t* symbolCosts;  /* pre-calculated dictionary statistics */
-    ZSTD_literalCompressionMode_e literalCompressionMode;
+    ZSTD_paramSwitch_e literalCompressionMode;
 } optState_t;
 
 typedef struct {
@@ -199,6 +207,8 @@ typedef struct {
                                 */
 } ZSTD_window_t;
 
+#define ZSTD_WINDOW_START_INDEX 2
+
 typedef struct ZSTD_matchState_t ZSTD_matchState_t;
 
 #define ZSTD_ROW_HASH_CACHE_SIZE 8       /* Size of prefetching hash cache for row-based matchfinder */
@@ -232,6 +242,11 @@ struct ZSTD_matchState_t {
     const ZSTD_matchState_t* dictMatchState;
     ZSTD_compressionParameters cParams;
     const rawSeqStore_t* ldmSeqStore;
+
+    /* Controls prefetching in some dictMatchState matchfinders.
+     * This behavior is controlled from the cctx ms.
+     * This parameter has no effect in the cdict ms. */
+    int prefetchCDictTables;
 };
 
 typedef struct {
@@ -264,7 +279,7 @@ typedef struct {
 } ldmState_t;
 
 typedef struct {
-    U32 enableLdm;          /* 1 if enable long distance matching */
+    ZSTD_paramSwitch_e enableLdm; /* ZSTD_ps_enable to enable LDM. ZSTD_ps_auto by default */
     U32 hashLog;            /* Log size of hashTable */
     U32 bucketSizeLog;      /* Log bucket size for collision resolution, at most 8 */
     U32 minMatchLength;     /* Minimum match length */
@@ -295,7 +310,7 @@ struct ZSTD_CCtx_params_s {
                                 * There is no guarantee that hint is close to actual source size */
 
     ZSTD_dictAttachPref_e attachDictPref;
-    ZSTD_literalCompressionMode_e literalCompressionMode;
+    ZSTD_paramSwitch_e literalCompressionMode;
 
     /* Multithreading: used to pass parameters to mtctx */
     int nbWorkers;
@@ -318,16 +333,34 @@ struct ZSTD_CCtx_params_s {
     int validateSequences;
 
     /* Block splitting */
-    int splitBlocks;
+    ZSTD_paramSwitch_e useBlockSplitter;
 
     /* Param for deciding whether to use row-based matchfinder */
-    ZSTD_useRowMatchFinderMode_e useRowMatchFinder;
+    ZSTD_paramSwitch_e useRowMatchFinder;
 
     /* Always load a dictionary in ext-dict mode (not prefix mode)? */
     int deterministicRefPrefix;
 
     /* Internal use, for createCCtxParams() and freeCCtxParams() only */
     ZSTD_customMem customMem;
+
+    /* Controls prefetching in some dictMatchState matchfinders */
+    ZSTD_paramSwitch_e prefetchCDictTables;
+
+    /* Controls whether zstd will fall back to an internal matchfinder
+     * if the external matchfinder returns an error code. */
+    int enableMatchFinderFallback;
+
+    /* Indicates whether an external matchfinder has been referenced.
+     * Users can't set this externally.
+     * It is set internally in ZSTD_registerSequenceProducer(). */
+    int useSequenceProducer;
+
+    /* Adjust the max block size*/
+    size_t maxBlockSize;
+
+    /* Controls repcode search in external sequence parsing */
+    ZSTD_paramSwitch_e searchForExternalRepcodes;
 };  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
 
 #define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
@@ -343,6 +376,30 @@ typedef enum {
     ZSTDb_buffered
 } ZSTD_buffered_policy_e;
 
+/**
+ * Struct that contains all elements of block splitter that should be allocated
+ * in a wksp.
+ */
+#define ZSTD_MAX_NB_BLOCK_SPLITS 196
+typedef struct {
+    seqStore_t fullSeqStoreChunk;
+    seqStore_t firstHalfSeqStore;
+    seqStore_t secondHalfSeqStore;
+    seqStore_t currSeqStore;
+    seqStore_t nextSeqStore;
+
+    U32 partitions[ZSTD_MAX_NB_BLOCK_SPLITS];
+    ZSTD_entropyCTablesMetadata_t entropyMetadata;
+} ZSTD_blockSplitCtx;
+
+/* Context for block-level external matchfinder API */
+typedef struct {
+  void* mState;
+  ZSTD_sequenceProducer_F* mFinder;
+  ZSTD_Sequence* seqBuffer;
+  size_t seqBufferCapacity;
+} ZSTD_externalMatchCtx;
+
 struct ZSTD_CCtx_s {
     ZSTD_compressionStage_e stage;
     int cParamsChanged;                  /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
@@ -374,7 +431,7 @@ struct ZSTD_CCtx_s {
     ZSTD_blockState_t blockState;
     U32* entropyWorkspace;  /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
 
-    /* Wether we are streaming or not */
+    /* Whether we are streaming or not */
     ZSTD_buffered_policy_e bufferedPolicy;
 
     /* streaming */
@@ -392,6 +449,7 @@ struct ZSTD_CCtx_s {
 
     /* Stable in/out buffer verification */
     ZSTD_inBuffer expectedInBuffer;
+    size_t stableIn_notConsumed; /* nb bytes within stable input buffer that are said to be consumed but are not */
     size_t expectedOutBufferSize;
 
     /* Dictionary */
@@ -408,9 +466,16 @@ struct ZSTD_CCtx_s {
 #if ZSTD_TRACE
     ZSTD_TraceCtx traceCtx;
 #endif
+
+    /* Workspace for block splitter */
+    ZSTD_blockSplitCtx blockSplitCtx;
+
+    /* Workspace for external matchfinder */
+    ZSTD_externalMatchCtx externalMatchCtx;
 };
 
 typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
+typedef enum { ZSTD_tfp_forCCtx, ZSTD_tfp_forCDict } ZSTD_tableFillPurpose_e;
 
 typedef enum {
     ZSTD_noDict = 0,
@@ -432,7 +497,7 @@ typedef enum {
                                  * In this mode we take both the source size and the dictionary size
                                  * into account when selecting and adjusting the parameters.
                                  */
-    ZSTD_cpm_unknown = 3,       /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
+    ZSTD_cpm_unknown = 3        /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
                                  * We don't know what these parameters are for. We default to the legacy
                                  * behavior of taking both the source size and the dict size into account
                                  * when selecting and adjusting parameters.
@@ -442,7 +507,7 @@ typedef enum {
 typedef size_t (*ZSTD_blockCompressor) (
         ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_useRowMatchFinderMode_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
 
 
 MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
@@ -476,31 +541,6 @@ MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
     return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
 }
 
-typedef struct repcodes_s {
-    U32 rep[3];
-} repcodes_t;
-
-MEM_STATIC repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
-{
-    repcodes_t newReps;
-    if (offset >= ZSTD_REP_NUM) {  /* full offset */
-        newReps.rep[2] = rep[1];
-        newReps.rep[1] = rep[0];
-        newReps.rep[0] = offset - ZSTD_REP_MOVE;
-    } else {   /* repcode */
-        U32 const repCode = offset + ll0;
-        if (repCode > 0) {  /* note : if repCode==0, no change */
-            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
-            newReps.rep[2] = (repCode >= 2) ? rep[1] : rep[2];
-            newReps.rep[1] = rep[0];
-            newReps.rep[0] = currentOffset;
-        } else {   /* repCode == 0 */
-            ZSTD_memcpy(&newReps, rep, sizeof(newReps));
-        }
-    }
-    return newReps;
-}
-
 /* ZSTD_cParam_withinBounds:
  * @return 1 if value is within cParam bounds,
  * 0 otherwise */
@@ -516,9 +556,11 @@ MEM_STATIC int ZSTD_cParam_withinBounds(ZSTD_cParameter cParam, int value)
 /* ZSTD_noCompressBlock() :
  * Writes uncompressed block to dst buffer from given src.
  * Returns the size of the block */
-MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
+MEM_STATIC size_t
+ZSTD_noCompressBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
 {
     U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
+    DEBUGLOG(5, "ZSTD_noCompressBlock (srcSize=%zu, dstCapacity=%zu)", srcSize, dstCapacity);
     RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
                     dstSize_tooSmall, "dst buf too small for uncompressed block");
     MEM_writeLE24(dst, cBlockHeader24);
@@ -526,7 +568,8 @@ MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const voi
     return ZSTD_blockHeaderSize + srcSize;
 }
 
-MEM_STATIC size_t ZSTD_rleCompressBlock (void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
+MEM_STATIC size_t
+ZSTD_rleCompressBlock(void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
 {
     BYTE* const op = (BYTE*)dst;
     U32 const cBlockHeader = lastBlock + (((U32)bt_rle)<<1) + (U32)(srcSize << 3);
@@ -545,21 +588,21 @@ MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
 {
     U32 const minlog = (strat>=ZSTD_btultra) ? (U32)(strat) - 1 : 6;
     ZSTD_STATIC_ASSERT(ZSTD_btultra == 8);
-    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, strat));
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, (int)strat));
     return (srcSize >> minlog) + 2;
 }
 
-MEM_STATIC int ZSTD_disableLiteralsCompression(const ZSTD_CCtx_params* cctxParams)
+MEM_STATIC int ZSTD_literalsCompressionIsDisabled(const ZSTD_CCtx_params* cctxParams)
 {
     switch (cctxParams->literalCompressionMode) {
-    case ZSTD_lcm_huffman:
+    case ZSTD_ps_enable:
         return 0;
-    case ZSTD_lcm_uncompressed:
+    case ZSTD_ps_disable:
         return 1;
     default:
         assert(0 /* impossible: pre-validated */);
-        /* fall-through */
-    case ZSTD_lcm_auto:
+        ZSTD_FALLTHROUGH;
+    case ZSTD_ps_auto:
         return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
     }
 }
@@ -569,7 +612,9 @@ MEM_STATIC int ZSTD_disableLiteralsCompression(const ZSTD_CCtx_params* cctxParam
  *  Only called when the sequence ends past ilimit_w, so it only needs to be optimized for single
  *  large copies.
  */
-static void ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE const* ilimit_w) {
+static void
+ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE const* ilimit_w)
+{
     assert(iend > ilimit_w);
     if (ip <= ilimit_w) {
         ZSTD_wildcopy(op, ip, ilimit_w - ip, ZSTD_no_overlap);
@@ -579,14 +624,28 @@ static void ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const ie
     while (ip < iend) *op++ = *ip++;
 }
 
+
+#define REPCODE1_TO_OFFBASE REPCODE_TO_OFFBASE(1)
+#define REPCODE2_TO_OFFBASE REPCODE_TO_OFFBASE(2)
+#define REPCODE3_TO_OFFBASE REPCODE_TO_OFFBASE(3)
+#define REPCODE_TO_OFFBASE(r) (assert((r)>=1), assert((r)<=ZSTD_REP_NUM), (r)) /* accepts IDs 1,2,3 */
+#define OFFSET_TO_OFFBASE(o)  (assert((o)>0), o + ZSTD_REP_NUM)
+#define OFFBASE_IS_OFFSET(o)  ((o) > ZSTD_REP_NUM)
+#define OFFBASE_IS_REPCODE(o) ( 1 <= (o) && (o) <= ZSTD_REP_NUM)
+#define OFFBASE_TO_OFFSET(o)  (assert(OFFBASE_IS_OFFSET(o)), (o) - ZSTD_REP_NUM)
+#define OFFBASE_TO_REPCODE(o) (assert(OFFBASE_IS_REPCODE(o)), (o))  /* returns ID 1,2,3 */
+
 /*! ZSTD_storeSeq() :
- *  Store a sequence (litlen, litPtr, offCode and mlBase) into seqStore_t.
- *  `offCode` : distance to match + ZSTD_REP_MOVE (values <= ZSTD_REP_MOVE are repCodes).
- *  `mlBase` : matchLength - MINMATCH
- *  Allowed to overread literals up to litLimit.
+ *  Store a sequence (litlen, litPtr, offBase and matchLength) into seqStore_t.
+ *  @offBase : Users should employ macros REPCODE_TO_OFFBASE() and OFFSET_TO_OFFBASE().
+ *  @matchLength : must be >= MINMATCH
+ *  Allowed to over-read literals up to litLimit.
 */
-HINT_INLINE UNUSED_ATTR
-void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals, const BYTE* litLimit, U32 offCode, size_t mlBase)
+HINT_INLINE UNUSED_ATTR void
+ZSTD_storeSeq(seqStore_t* seqStorePtr,
+              size_t litLength, const BYTE* literals, const BYTE* litLimit,
+              U32 offBase,
+              size_t matchLength)
 {
     BYTE const* const litLimit_w = litLimit - WILDCOPY_OVERLENGTH;
     BYTE const* const litEnd = literals + litLength;
@@ -594,8 +653,8 @@ void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* litera
     static const BYTE* g_start = NULL;
     if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
     {   U32 const pos = (U32)((const BYTE*)literals - g_start);
-        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offCode%7u",
-               pos, (U32)litLength, (U32)mlBase+MINMATCH, (U32)offCode);
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offBase%7u",
+               pos, (U32)litLength, (U32)matchLength, (U32)offBase);
     }
 #endif
     assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
@@ -605,9 +664,9 @@ void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* litera
     assert(literals + litLength <= litLimit);
     if (litEnd <= litLimit_w) {
         /* Common case we can use wildcopy.
-	 * First copy 16 bytes, because literals are likely short.
-	 */
-        assert(WILDCOPY_OVERLENGTH >= 16);
+         * First copy 16 bytes, because literals are likely short.
+         */
+        ZSTD_STATIC_ASSERT(WILDCOPY_OVERLENGTH >= 16);
         ZSTD_copy16(seqStorePtr->lit, literals);
         if (litLength > 16) {
             ZSTD_wildcopy(seqStorePtr->lit+16, literals+16, (ptrdiff_t)litLength-16, ZSTD_no_overlap);
@@ -626,96 +685,63 @@ void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* litera
     seqStorePtr->sequences[0].litLength = (U16)litLength;
 
     /* match offset */
-    seqStorePtr->sequences[0].offset = offCode + 1;
+    seqStorePtr->sequences[0].offBase = offBase;
 
     /* match Length */
-    if (mlBase>0xFFFF) {
-        assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
-        seqStorePtr->longLengthType = ZSTD_llt_matchLength;
-        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    assert(matchLength >= MINMATCH);
+    {   size_t const mlBase = matchLength - MINMATCH;
+        if (mlBase>0xFFFF) {
+            assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
+            seqStorePtr->longLengthType = ZSTD_llt_matchLength;
+            seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+        }
+        seqStorePtr->sequences[0].mlBase = (U16)mlBase;
     }
-    seqStorePtr->sequences[0].matchLength = (U16)mlBase;
 
     seqStorePtr->sequences++;
 }
 
-
-/*-*************************************
-*  Match length counter
-***************************************/
-static unsigned ZSTD_NbCommonBytes (size_t val)
+/* ZSTD_updateRep() :
+ * updates in-place @rep (array of repeat offsets)
+ * @offBase : sum-type, using numeric representation of ZSTD_storeSeq()
+ */
+MEM_STATIC void
+ZSTD_updateRep(U32 rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
 {
-    if (MEM_isLittleEndian()) {
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-#           if STATIC_BMI2
-                return _tzcnt_u64(val) >> 3;
-#           else
-                unsigned long r = 0;
-                return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
-#           endif
-#       elif defined(__GNUC__) && (__GNUC__ >= 4)
-            return (__builtin_ctzll((U64)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
-                                                     0, 3, 1, 3, 1, 4, 2, 7,
-                                                     0, 2, 3, 6, 1, 5, 3, 5,
-                                                     1, 3, 4, 4, 2, 5, 6, 7,
-                                                     7, 0, 1, 2, 3, 3, 4, 6,
-                                                     2, 6, 5, 5, 3, 4, 5, 6,
-                                                     7, 1, 2, 4, 6, 4, 4, 5,
-                                                     7, 2, 6, 5, 7, 6, 7, 7 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r=0;
-            return _BitScanForward( &r, (U32)val ) ? (unsigned)(r >> 3) : 0;
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctz((U32)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
-                                                     3, 2, 2, 1, 3, 2, 0, 1,
-                                                     3, 3, 1, 2, 2, 2, 2, 0,
-                                                     3, 1, 2, 0, 1, 0, 1, 1 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
+    if (OFFBASE_IS_OFFSET(offBase)) {  /* full offset */
+        rep[2] = rep[1];
+        rep[1] = rep[0];
+        rep[0] = OFFBASE_TO_OFFSET(offBase);
+    } else {   /* repcode */
+        U32 const repCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;
+        if (repCode > 0) {  /* note : if repCode==0, no change */
+            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            rep[2] = (repCode >= 2) ? rep[1] : rep[2];
+            rep[1] = rep[0];
+            rep[0] = currentOffset;
+        } else {   /* repCode == 0 */
+            /* nothing to do */
         }
-    } else {  /* Big Endian CPU */
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-#           if STATIC_BMI2
-			    return _lzcnt_u64(val) >> 3;
-#           else
-			    unsigned long r = 0;
-			    return _BitScanReverse64(&r, (U64)val) ? (unsigned)(r >> 3) : 0;
-#           endif
-#       elif defined(__GNUC__) && (__GNUC__ >= 4)
-            return (__builtin_clzll(val) >> 3);
-#       else
-            unsigned r;
-            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
-            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
-            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r = 0;
-            return _BitScanReverse( &r, (unsigned long)val ) ? (unsigned)(r >> 3) : 0;
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clz((U32)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-    }   }
+    }
+}
+
+typedef struct repcodes_s {
+    U32 rep[3];
+} repcodes_t;
+
+MEM_STATIC repcodes_t
+ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
+{
+    repcodes_t newReps;
+    ZSTD_memcpy(&newReps, rep, sizeof(newReps));
+    ZSTD_updateRep(newReps.rep, offBase, ll0);
+    return newReps;
 }
 
 
+/*-*************************************
+*  Match length counter
+***************************************/
 MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
 {
     const BYTE* const pStart = pIn;
@@ -761,32 +787,36 @@ ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
  *  Hashes
  ***************************************/
 static const U32 prime3bytes = 506832829U;
-static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
+static U32    ZSTD_hash3(U32 u, U32 h) { assert(h <= 32); return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
 MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); } /* only in zstd_opt.h */
 
 static const U32 prime4bytes = 2654435761U;
-static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
-static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
+static U32    ZSTD_hash4(U32 u, U32 h) { assert(h <= 32); return (u * prime4bytes) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_readLE32(ptr), h); }
 
 static const U64 prime5bytes = 889523592379ULL;
-static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
+static size_t ZSTD_hash5(U64 u, U32 h) { assert(h <= 64); return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
 static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
 
 static const U64 prime6bytes = 227718039650203ULL;
-static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6(U64 u, U32 h) { assert(h <= 64); return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
 static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
 
 static const U64 prime7bytes = 58295818150454627ULL;
-static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
+static size_t ZSTD_hash7(U64 u, U32 h) { assert(h <= 64); return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
 static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
 
 static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
-static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8(U64 u, U32 h) { assert(h <= 64); return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
 static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
 
 MEM_STATIC FORCE_INLINE_ATTR
 size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
 {
+    /* Although some of these hashes do support hBits up to 64, some do not.
+     * To be on the safe side, always avoid hBits > 32. */
+    assert(hBits <= 32);
+
     switch(mls)
     {
     default:
@@ -884,9 +914,9 @@ MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
 
 MEM_STATIC U32 ZSTD_window_isEmpty(ZSTD_window_t const window)
 {
-    return window.dictLimit == 1 &&
-           window.lowLimit == 1 &&
-           (window.nextSrc - window.base) == 1;
+    return window.dictLimit == ZSTD_WINDOW_START_INDEX &&
+           window.lowLimit == ZSTD_WINDOW_START_INDEX &&
+           (window.nextSrc - window.base) == ZSTD_WINDOW_START_INDEX;
 }
 
 /**
@@ -937,7 +967,9 @@ MEM_STATIC U32 ZSTD_window_canOverflowCorrect(ZSTD_window_t const window,
 {
     U32 const cycleSize = 1u << cycleLog;
     U32 const curr = (U32)((BYTE const*)src - window.base);
-    U32 const minIndexToOverflowCorrect = cycleSize + MAX(maxDist, cycleSize);
+    U32 const minIndexToOverflowCorrect = cycleSize
+                                        + MAX(maxDist, cycleSize)
+                                        + ZSTD_WINDOW_START_INDEX;
 
     /* Adjust the min index to backoff the overflow correction frequency,
      * so we don't waste too much CPU in overflow correction. If this
@@ -1012,10 +1044,14 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
     U32 const cycleSize = 1u << cycleLog;
     U32 const cycleMask = cycleSize - 1;
     U32 const curr = (U32)((BYTE const*)src - window->base);
-    U32 const currentCycle0 = curr & cycleMask;
-    /* Exclude zero so that newCurrent - maxDist >= 1. */
-    U32 const currentCycle1 = currentCycle0 == 0 ? cycleSize : currentCycle0;
-    U32 const newCurrent = currentCycle1 + MAX(maxDist, cycleSize);
+    U32 const currentCycle = curr & cycleMask;
+    /* Ensure newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX. */
+    U32 const currentCycleCorrection = currentCycle < ZSTD_WINDOW_START_INDEX
+                                     ? MAX(cycleSize, ZSTD_WINDOW_START_INDEX)
+                                     : 0;
+    U32 const newCurrent = currentCycle
+                         + currentCycleCorrection
+                         + MAX(maxDist, cycleSize);
     U32 const correction = curr - newCurrent;
     /* maxDist must be a power of two so that:
      *   (newCurrent & cycleMask) == (curr & cycleMask)
@@ -1031,14 +1067,20 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
 
     window->base += correction;
     window->dictBase += correction;
-    if (window->lowLimit <= correction) window->lowLimit = 1;
-    else window->lowLimit -= correction;
-    if (window->dictLimit <= correction) window->dictLimit = 1;
-    else window->dictLimit -= correction;
+    if (window->lowLimit < correction + ZSTD_WINDOW_START_INDEX) {
+        window->lowLimit = ZSTD_WINDOW_START_INDEX;
+    } else {
+        window->lowLimit -= correction;
+    }
+    if (window->dictLimit < correction + ZSTD_WINDOW_START_INDEX) {
+        window->dictLimit = ZSTD_WINDOW_START_INDEX;
+    } else {
+        window->dictLimit -= correction;
+    }
 
     /* Ensure we can still reference the full window. */
     assert(newCurrent >= maxDist);
-    assert(newCurrent - maxDist >= 1);
+    assert(newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX);
     /* Ensure that lowLimit and dictLimit didn't underflow. */
     assert(window->lowLimit <= newCurrent);
     assert(window->dictLimit <= newCurrent);
@@ -1133,10 +1175,15 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
                     (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
         assert(blockEndIdx >= loadedDictEnd);
 
-        if (blockEndIdx > loadedDictEnd + maxDist) {
+        if (blockEndIdx > loadedDictEnd + maxDist || loadedDictEnd != window->dictLimit) {
             /* On reaching window size, dictionaries are invalidated.
              * For simplification, if window size is reached anywhere within next block,
              * the dictionary is invalidated for the full block.
+             *
+             * We also have to invalidate the dictionary if ZSTD_window_update() has detected
+             * non-contiguous segments, which means that loadedDictEnd != window->dictLimit.
+             * loadedDictEnd may be 0, if forceWindow is true, but in that case we never use
+             * dictMatchState, so setting it to NULL is not a problem.
              */
             DEBUGLOG(6, "invalidating dictionary for current block (distance > windowSize)");
             *loadedDictEndPtr = 0;
@@ -1149,11 +1196,12 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
 
 MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
     ZSTD_memset(window, 0, sizeof(*window));
-    window->base = (BYTE const*)"";
-    window->dictBase = (BYTE const*)"";
-    window->dictLimit = 1;    /* start from 1, so that 1st position is valid */
-    window->lowLimit = 1;     /* it ensures first and later CCtx usages compress the same */
-    window->nextSrc = window->base + 1;   /* see issue #1241 */
+    window->base = (BYTE const*)" ";
+    window->dictBase = (BYTE const*)" ";
+    ZSTD_STATIC_ASSERT(ZSTD_DUBT_UNSORTED_MARK < ZSTD_WINDOW_START_INDEX); /* Start above ZSTD_DUBT_UNSORTED_MARK */
+    window->dictLimit = ZSTD_WINDOW_START_INDEX;    /* start from >0, so that 1st position is valid */
+    window->lowLimit = ZSTD_WINDOW_START_INDEX;     /* it ensures first and later CCtx usages compress the same */
+    window->nextSrc = window->base + ZSTD_WINDOW_START_INDEX;   /* see issue #1241 */
     window->nbOverflowCorrections = 0;
 }
 
@@ -1206,15 +1254,15 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
  */
 MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
 {
-    U32    const maxDistance = 1U << windowLog;
-    U32    const lowestValid = ms->window.lowLimit;
-    U32    const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
-    U32    const isDictionary = (ms->loadedDictEnd != 0);
+    U32 const maxDistance = 1U << windowLog;
+    U32 const lowestValid = ms->window.lowLimit;
+    U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
+    U32 const isDictionary = (ms->loadedDictEnd != 0);
     /* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
      * is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
      * valid for the entire block. So this check is sufficient to find the lowest valid match index.
      */
-    U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
+    U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
     return matchLowest;
 }
 
@@ -1267,6 +1315,42 @@ MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
 
 #endif
 
+/* Short Cache */
+
+/* Normally, zstd matchfinders follow this flow:
+ *     1. Compute hash at ip
+ *     2. Load index from hashTable[hash]
+ *     3. Check if *ip == *(base + index)
+ * In dictionary compression, loading *(base + index) is often an L2 or even L3 miss.
+ *
+ * Short cache is an optimization which allows us to avoid step 3 most of the time
+ * when the data doesn't actually match. With short cache, the flow becomes:
+ *     1. Compute (hash, currentTag) at ip. currentTag is an 8-bit independent hash at ip.
+ *     2. Load (index, matchTag) from hashTable[hash]. See ZSTD_writeTaggedIndex to understand how this works.
+ *     3. Only if currentTag == matchTag, check *ip == *(base + index). Otherwise, continue.
+ *
+ * Currently, short cache is only implemented in CDict hashtables. Thus, its use is limited to
+ * dictMatchState matchfinders.
+ */
+#define ZSTD_SHORT_CACHE_TAG_BITS 8
+#define ZSTD_SHORT_CACHE_TAG_MASK ((1u << ZSTD_SHORT_CACHE_TAG_BITS) - 1)
+
+/* Helper function for ZSTD_fillHashTable and ZSTD_fillDoubleHashTable.
+ * Unpacks hashAndTag into (hash, tag), then packs (index, tag) into hashTable[hash]. */
+MEM_STATIC void ZSTD_writeTaggedIndex(U32* const hashTable, size_t hashAndTag, U32 index) {
+    size_t const hash = hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
+    U32 const tag = (U32)(hashAndTag & ZSTD_SHORT_CACHE_TAG_MASK);
+    assert(index >> (32 - ZSTD_SHORT_CACHE_TAG_BITS) == 0);
+    hashTable[hash] = (index << ZSTD_SHORT_CACHE_TAG_BITS) | tag;
+}
+
+/* Helper function for short cache matchfinders.
+ * Unpacks tag1 and tag2 from lower bits of packedTag1 and packedTag2, then checks if the tags match. */
+MEM_STATIC int ZSTD_comparePackedTags(size_t packedTag1, size_t packedTag2) {
+    U32 const tag1 = packedTag1 & ZSTD_SHORT_CACHE_TAG_MASK;
+    U32 const tag2 = packedTag2 & ZSTD_SHORT_CACHE_TAG_MASK;
+    return tag1 == tag2;
+}
 
 #if defined (__cplusplus)
 }
@@ -1364,4 +1448,31 @@ U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat);
  */
 void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize);
 
+/* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
+ * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
+ * Note that the block delimiter must include the last literals of the block.
+ */
+size_t
+ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
+                                              ZSTD_sequencePosition* seqPos,
+                                        const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                        const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+
+/* Returns the number of bytes to move the current read position back by.
+ * Only non-zero if we ended up splitting a sequence.
+ * Otherwise, it may return a ZSTD error if something went wrong.
+ *
+ * This function will attempt to scan through blockSize bytes
+ * represented by the sequences in @inSeqs,
+ * storing any (partial) sequences.
+ *
+ * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
+ * avoid splitting a match, or to avoid splitting a match such that it would produce a match
+ * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
+ */
+size_t
+ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
+                                   const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
+                                   const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
+
 #endif /* ZSTD_COMPRESS_H */
diff --git a/priv/zstd/lib/compress/zstd_compress_literals.c b/priv/zstd/lib/compress/zstd_compress_literals.c
index 008337b..bfd4f11 100644
--- a/priv/zstd/lib/compress/zstd_compress_literals.c
+++ b/priv/zstd/lib/compress/zstd_compress_literals.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -13,11 +13,36 @@
  ***************************************/
 #include "zstd_compress_literals.h"
 
+
+/* **************************************************************
+*  Debug Traces
+****************************************************************/
+#if DEBUGLEVEL >= 2
+
+static size_t showHexa(const void* src, size_t srcSize)
+{
+    const BYTE* const ip = (const BYTE*)src;
+    size_t u;
+    for (u=0; u31) + (srcSize>4095);
 
+    DEBUGLOG(5, "ZSTD_noCompressLiterals: srcSize=%zu, dstCapacity=%zu", srcSize, dstCapacity);
+
     RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
 
     switch(flSize)
@@ -36,16 +61,30 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
     }
 
     ZSTD_memcpy(ostart + flSize, src, srcSize);
-    DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
+    DEBUGLOG(5, "Raw (uncompressed) literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
     return srcSize + flSize;
 }
 
+static int allBytesIdentical(const void* src, size_t srcSize)
+{
+    assert(srcSize >= 1);
+    assert(src != NULL);
+    {   const BYTE b = ((const BYTE*)src)[0];
+        size_t p;
+        for (p=1; p31) + (srcSize>4095);
 
-    (void)dstCapacity;  /* dstCapacity already guaranteed to be >=4, hence large enough */
+    assert(dstCapacity >= 4); (void)dstCapacity;
+    assert(allBytesIdentical(src, srcSize));
 
     switch(flSize)
     {
@@ -63,27 +102,51 @@ size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void*
     }
 
     ostart[flSize] = *(const BYTE*)src;
-    DEBUGLOG(5, "RLE literals: %u -> %u", (U32)srcSize, (U32)flSize + 1);
+    DEBUGLOG(5, "RLE : Repeated Literal (%02X: %u times) -> %u bytes encoded", ((const BYTE*)src)[0], (U32)srcSize, (U32)flSize + 1);
     return flSize+1;
 }
 
-size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
-                              ZSTD_hufCTables_t* nextHuf,
-                              ZSTD_strategy strategy, int disableLiteralCompression,
-                              void* dst, size_t dstCapacity,
-                        const void* src, size_t srcSize,
-                              void* entropyWorkspace, size_t entropyWorkspaceSize,
-                        const int bmi2)
+/* ZSTD_minLiteralsToCompress() :
+ * returns minimal amount of literals
+ * for literal compression to even be attempted.
+ * Minimum is made tighter as compression strategy increases.
+ */
+static size_t
+ZSTD_minLiteralsToCompress(ZSTD_strategy strategy, HUF_repeat huf_repeat)
+{
+    assert((int)strategy >= 0);
+    assert((int)strategy <= 9);
+    /* btultra2 : min 8 bytes;
+     * then 2x larger for each successive compression strategy
+     * max threshold 64 bytes */
+    {   int const shift = MIN(9-(int)strategy, 3);
+        size_t const mintc = (huf_repeat == HUF_repeat_valid) ? 6 : (size_t)8 << shift;
+        DEBUGLOG(7, "minLiteralsToCompress = %zu", mintc);
+        return mintc;
+    }
+}
+
+size_t ZSTD_compressLiterals (
+                  void* dst, size_t dstCapacity,
+            const void* src, size_t srcSize,
+                  void* entropyWorkspace, size_t entropyWorkspaceSize,
+            const ZSTD_hufCTables_t* prevHuf,
+                  ZSTD_hufCTables_t* nextHuf,
+                  ZSTD_strategy strategy,
+                  int disableLiteralCompression,
+                  int suspectUncompressible,
+                  int bmi2)
 {
-    size_t const minGain = ZSTD_minGain(srcSize, strategy);
     size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
     BYTE*  const ostart = (BYTE*)dst;
     U32 singleStream = srcSize < 256;
     symbolEncodingType_e hType = set_compressed;
     size_t cLitSize;
 
-    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i srcSize=%u)",
-                disableLiteralCompression, (U32)srcSize);
+    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i, srcSize=%u, dstCapacity=%zu)",
+                disableLiteralCompression, (U32)srcSize, dstCapacity);
+
+    DEBUGLOG(6, "Completed literals listing (%zu bytes)", showHexa(src, srcSize));
 
     /* Prepare nextEntropy assuming reusing the existing table */
     ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
@@ -91,40 +154,51 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
     if (disableLiteralCompression)
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
 
-    /* small ? don't even attempt compression (speed opt) */
-#   define COMPRESS_LITERALS_SIZE_MIN 63
-    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
-        if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    }
+    /* if too small, don't even attempt compression (speed opt) */
+    if (srcSize < ZSTD_minLiteralsToCompress(strategy, prevHuf->repeatMode))
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
 
     RETURN_ERROR_IF(dstCapacity < lhSize+1, dstSize_tooSmall, "not enough space for compression");
     {   HUF_repeat repeat = prevHuf->repeatMode;
-        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        int const flags = 0
+            | (bmi2 ? HUF_flags_bmi2 : 0)
+            | (strategy < ZSTD_lazy && srcSize <= 1024 ? HUF_flags_preferRepeat : 0)
+            | (strategy >= HUF_OPTIMAL_DEPTH_THRESHOLD ? HUF_flags_optimalDepth : 0)
+            | (suspectUncompressible ? HUF_flags_suspectUncompressible : 0);
+
+        typedef size_t (*huf_compress_f)(void*, size_t, const void*, size_t, unsigned, unsigned, void*, size_t, HUF_CElt*, HUF_repeat*, int);
+        huf_compress_f huf_compress;
         if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
-        cLitSize = singleStream ?
-            HUF_compress1X_repeat(
-                ostart+lhSize, dstCapacity-lhSize, src, srcSize,
-                HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
-                (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2) :
-            HUF_compress4X_repeat(
-                ostart+lhSize, dstCapacity-lhSize, src, srcSize,
-                HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
-                (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
+        huf_compress = singleStream ? HUF_compress1X_repeat : HUF_compress4X_repeat;
+        cLitSize = huf_compress(ostart+lhSize, dstCapacity-lhSize,
+                                src, srcSize,
+                                HUF_SYMBOLVALUE_MAX, LitHufLog,
+                                entropyWorkspace, entropyWorkspaceSize,
+                                (HUF_CElt*)nextHuf->CTable,
+                                &repeat, flags);
+        DEBUGLOG(5, "%zu literals compressed into %zu bytes (before header)", srcSize, cLitSize);
         if (repeat != HUF_repeat_none) {
             /* reused the existing table */
-            DEBUGLOG(5, "Reusing previous huffman table");
+            DEBUGLOG(5, "reusing statistics from previous huffman block");
             hType = set_repeat;
         }
     }
 
-    if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
-        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    }
+    {   size_t const minGain = ZSTD_minGain(srcSize, strategy);
+        if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }   }
     if (cLitSize==1) {
-        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
-        return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
-    }
+        /* A return value of 1 signals that the alphabet consists of a single symbol.
+         * However, in some rare circumstances, it could be the compressed size (a single byte).
+         * For that outcome to have a chance to happen, it's necessary that `srcSize < 8`.
+         * (it's also necessary to not generate statistics).
+         * Therefore, in such a case, actively check that all bytes are identical. */
+        if ((srcSize >= 8) || allBytesIdentical(src, srcSize)) {
+            ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+            return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }   }
 
     if (hType == set_compressed) {
         /* using a newly constructed table */
@@ -135,16 +209,19 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
     switch(lhSize)
     {
     case 3: /* 2 - 2 - 10 - 10 */
-        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
+        if (!singleStream) assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
+        {   U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
             MEM_writeLE24(ostart, lhc);
             break;
         }
     case 4: /* 2 - 2 - 14 - 14 */
+        assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
         {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
             MEM_writeLE32(ostart, lhc);
             break;
         }
     case 5: /* 2 - 2 - 18 - 18 */
+        assert(srcSize >= MIN_LITERALS_FOR_4_STREAMS);
         {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
             MEM_writeLE32(ostart, lhc);
             ostart[4] = (BYTE)(cLitSize >> 10);
diff --git a/priv/zstd/lib/compress/zstd_compress_literals.h b/priv/zstd/lib/compress/zstd_compress_literals.h
index 9904c0c..b060c8a 100644
--- a/priv/zstd/lib/compress/zstd_compress_literals.h
+++ b/priv/zstd/lib/compress/zstd_compress_literals.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -16,14 +16,24 @@
 
 size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
+/* ZSTD_compressRleLiteralsBlock() :
+ * Conditions :
+ * - All bytes in @src are identical
+ * - dstCapacity >= 4 */
 size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
-size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
-                              ZSTD_hufCTables_t* nextHuf,
-                              ZSTD_strategy strategy, int disableLiteralCompression,
-                              void* dst, size_t dstCapacity,
+/* ZSTD_compressLiterals():
+ * @entropyWorkspace: must be aligned on 4-bytes boundaries
+ * @entropyWorkspaceSize : must be >= HUF_WORKSPACE_SIZE
+ * @suspectUncompressible: sampling checks, to potentially skip huffman coding
+ */
+size_t ZSTD_compressLiterals (void* dst, size_t dstCapacity,
                         const void* src, size_t srcSize,
                               void* entropyWorkspace, size_t entropyWorkspaceSize,
-                        const int bmi2);
+                        const ZSTD_hufCTables_t* prevHuf,
+                              ZSTD_hufCTables_t* nextHuf,
+                              ZSTD_strategy strategy, int disableLiteralCompression,
+                              int suspectUncompressible,
+                              int bmi2);
 
 #endif /* ZSTD_COMPRESS_LITERALS_H */
diff --git a/priv/zstd/lib/compress/zstd_compress_sequences.c b/priv/zstd/lib/compress/zstd_compress_sequences.c
index 611eabd..8872d4d 100644
--- a/priv/zstd/lib/compress/zstd_compress_sequences.c
+++ b/priv/zstd/lib/compress/zstd_compress_sequences.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -58,7 +58,7 @@ static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
 {
     /* Heuristic: This should cover most blocks <= 16K and
      * start to fade out after 16K to about 32K depending on
-     * comprssibility.
+     * compressibility.
      */
     return nbSeq >= 2048;
 }
@@ -166,7 +166,7 @@ ZSTD_selectEncodingType(
     if (mostFrequent == nbSeq) {
         *repeatMode = FSE_repeat_none;
         if (isDefaultAllowed && nbSeq <= 2) {
-            /* Prefer set_basic over set_rle when there are 2 or less symbols,
+            /* Prefer set_basic over set_rle when there are 2 or fewer symbols,
              * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
              * If basic encoding isn't possible, always choose RLE.
              */
@@ -275,10 +275,11 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
         assert(nbSeq_1 > 1);
         assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
         (void)entropyWorkspaceSize;
-        FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "");
-        {   size_t const NCountSize = FSE_writeNCount(op, oend - op, wksp->norm, max, tableLog);   /* overflow protected */
+        FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "FSE_normalizeCount failed");
+        assert(oend >= op);
+        {   size_t const NCountSize = FSE_writeNCount(op, (size_t)(oend - op), wksp->norm, max, tableLog);   /* overflow protected */
             FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
-            FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "");
+            FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "FSE_buildCTable_wksp failed");
             return NCountSize;
         }
     }
@@ -312,19 +313,19 @@ ZSTD_encodeSequences_body(
     FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
     BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
     if (MEM_32bits()) BIT_flushBits(&blockStream);
-    BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].mlBase, ML_bits[mlCodeTable[nbSeq-1]]);
     if (MEM_32bits()) BIT_flushBits(&blockStream);
     if (longOffsets) {
         U32 const ofBits = ofCodeTable[nbSeq-1];
         unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
         if (extraBits) {
-            BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, extraBits);
             BIT_flushBits(&blockStream);
         }
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase >> extraBits,
                     ofBits - extraBits);
     } else {
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, ofCodeTable[nbSeq-1]);
     }
     BIT_flushBits(&blockStream);
 
@@ -338,8 +339,8 @@ ZSTD_encodeSequences_body(
             U32  const mlBits = ML_bits[mlCode];
             DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
                         (unsigned)sequences[n].litLength,
-                        (unsigned)sequences[n].matchLength + MINMATCH,
-                        (unsigned)sequences[n].offset);
+                        (unsigned)sequences[n].mlBase + MINMATCH,
+                        (unsigned)sequences[n].offBase);
                                                                             /* 32b*/  /* 64b*/
                                                                             /* (7)*/  /* (7)*/
             FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
@@ -350,18 +351,18 @@ ZSTD_encodeSequences_body(
                 BIT_flushBits(&blockStream);                                /* (7)*/
             BIT_addBits(&blockStream, sequences[n].litLength, llBits);
             if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
-            BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
+            BIT_addBits(&blockStream, sequences[n].mlBase, mlBits);
             if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
             if (longOffsets) {
                 unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
                 if (extraBits) {
-                    BIT_addBits(&blockStream, sequences[n].offset, extraBits);
+                    BIT_addBits(&blockStream, sequences[n].offBase, extraBits);
                     BIT_flushBits(&blockStream);                            /* (7)*/
                 }
-                BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
+                BIT_addBits(&blockStream, sequences[n].offBase >> extraBits,
                             ofBits - extraBits);                            /* 31 */
             } else {
-                BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
+                BIT_addBits(&blockStream, sequences[n].offBase, ofBits);     /* 31 */
             }
             BIT_flushBits(&blockStream);                                    /* (7)*/
             DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
@@ -398,7 +399,7 @@ ZSTD_encodeSequences_default(
 
 #if DYNAMIC_BMI2
 
-static TARGET_ATTRIBUTE("bmi2") size_t
+static BMI2_TARGET_ATTRIBUTE size_t
 ZSTD_encodeSequences_bmi2(
             void* dst, size_t dstCapacity,
             FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
diff --git a/priv/zstd/lib/compress/zstd_compress_sequences.h b/priv/zstd/lib/compress/zstd_compress_sequences.h
index 7991364..4a3a05d 100644
--- a/priv/zstd/lib/compress/zstd_compress_sequences.h
+++ b/priv/zstd/lib/compress/zstd_compress_sequences.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/compress/zstd_compress_superblock.c b/priv/zstd/lib/compress/zstd_compress_superblock.c
index e4e4506..638c4ac 100644
--- a/priv/zstd/lib/compress/zstd_compress_superblock.c
+++ b/priv/zstd/lib/compress/zstd_compress_superblock.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -36,13 +36,14 @@
  *      If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
  *      and the following sub-blocks' literals sections will be Treeless_Literals_Block.
  *  @return : compressed size of literals section of a sub-block
- *            Or 0 if it unable to compress.
+ *            Or 0 if unable to compress.
  *            Or error code */
-static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
-                                    const ZSTD_hufCTablesMetadata_t* hufMetadata,
-                                    const BYTE* literals, size_t litSize,
-                                    void* dst, size_t dstSize,
-                                    const int bmi2, int writeEntropy, int* entropyWritten)
+static size_t
+ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
+                              const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                              const BYTE* literals, size_t litSize,
+                              void* dst, size_t dstSize,
+                              const int bmi2, int writeEntropy, int* entropyWritten)
 {
     size_t const header = writeEntropy ? 200 : 0;
     size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
@@ -53,8 +54,6 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
     size_t cLitSize = 0;
 
-    (void)bmi2; /* TODO bmi2... */
-
     DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
 
     *entropyWritten = 0;
@@ -76,9 +75,9 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
         DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
     }
 
-    /* TODO bmi2 */
-    {   const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable)
-                                          : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable);
+    {   int const flags = bmi2 ? HUF_flags_bmi2 : 0;
+        const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable, flags)
+                                          : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable, flags);
         op += cSize;
         cLitSize += cSize;
         if (cSize == 0 || ERR_isError(cSize)) {
@@ -126,12 +125,17 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     return op-ostart;
 }
 
-static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef* sequences, size_t nbSeq, size_t litSize, int lastSequence) {
+static size_t
+ZSTD_seqDecompressedSize(seqStore_t const* seqStore,
+                   const seqDef* sequences, size_t nbSeq,
+                         size_t litSize, int lastSequence)
+{
     const seqDef* const sstart = sequences;
     const seqDef* const send = sequences + nbSeq;
     const seqDef* sp = sstart;
     size_t matchLengthSum = 0;
     size_t litLengthSum = 0;
+    (void)(litLengthSum); /* suppress unused variable warning on some environments */
     while (send-sp > 0) {
         ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
         litLengthSum += seqLen.litLength;
@@ -155,13 +159,14 @@ static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef*
  *  @return : compressed size of sequences section of a sub-block
  *            Or 0 if it is unable to compress
  *            Or error code. */
-static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
-                                              const ZSTD_fseCTablesMetadata_t* fseMetadata,
-                                              const seqDef* sequences, size_t nbSeq,
-                                              const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
-                                              const ZSTD_CCtx_params* cctxParams,
-                                              void* dst, size_t dstCapacity,
-                                              const int bmi2, int writeEntropy, int* entropyWritten)
+static size_t
+ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
+                                const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                const seqDef* sequences, size_t nbSeq,
+                                const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
+                                const ZSTD_CCtx_params* cctxParams,
+                                void* dst, size_t dstCapacity,
+                                const int bmi2, int writeEntropy, int* entropyWritten)
 {
     const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
     BYTE* const ostart = (BYTE*)dst;
@@ -324,7 +329,7 @@ static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t lit
 static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
                         const BYTE* codeTable, unsigned maxCode,
                         size_t nbSeq, const FSE_CTable* fseCTable,
-                        const U32* additionalBits,
+                        const U8* additionalBits,
                         short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
                         void* workspace, size_t wkspSize)
 {
@@ -474,7 +479,7 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
         /* I think there is an optimization opportunity here.
          * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
          * since it recalculates estimate from scratch.
-         * For example, it would recount literal distribution and symbol codes everytime.
+         * For example, it would recount literal distribution and symbol codes every time.
          */
         cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
                                                        &nextCBlock->entropy, entropyMetadata,
@@ -538,7 +543,7 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
             repcodes_t rep;
             ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
             for (seq = sstart; seq < sp; ++seq) {
-                rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
+                ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
             }
             ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
         }
diff --git a/priv/zstd/lib/compress/zstd_compress_superblock.h b/priv/zstd/lib/compress/zstd_compress_superblock.h
index 176f9b1..8e494f0 100644
--- a/priv/zstd/lib/compress/zstd_compress_superblock.h
+++ b/priv/zstd/lib/compress/zstd_compress_superblock.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/compress/zstd_cwksp.h b/priv/zstd/lib/compress/zstd_cwksp.h
index 2656d26..9767669 100644
--- a/priv/zstd/lib/compress/zstd_cwksp.h
+++ b/priv/zstd/lib/compress/zstd_cwksp.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -219,7 +219,7 @@ MEM_STATIC size_t ZSTD_cwksp_aligned_alloc_size(size_t size) {
 MEM_STATIC size_t ZSTD_cwksp_slack_space_required(void) {
     /* For alignment, the wksp will always allocate an additional n_1=[1, 64] bytes
      * to align the beginning of tables section, as well as another n_2=[0, 63] bytes
-     * to align the beginning of the aligned secion.
+     * to align the beginning of the aligned section.
      *
      * n_1 + n_2 == 64 bytes if the cwksp is freshly allocated, due to tables and
      * aligneds being sized in multiples of 64 bytes.
@@ -243,12 +243,14 @@ MEM_STATIC size_t ZSTD_cwksp_bytes_to_align_ptr(void* ptr, const size_t alignByt
 
 /**
  * Internal function. Do not use directly.
- * Reserves the given number of bytes within the aligned/buffer segment of the wksp, which
- * counts from the end of the wksp. (as opposed to the object/table segment)
+ * Reserves the given number of bytes within the aligned/buffer segment of the wksp,
+ * which counts from the end of the wksp (as opposed to the object/table segment).
  *
  * Returns a pointer to the beginning of that space.
  */
-MEM_STATIC void* ZSTD_cwksp_reserve_internal_buffer_space(ZSTD_cwksp* ws, size_t const bytes) {
+MEM_STATIC void*
+ZSTD_cwksp_reserve_internal_buffer_space(ZSTD_cwksp* ws, size_t const bytes)
+{
     void* const alloc = (BYTE*)ws->allocStart - bytes;
     void* const bottom = ws->tableEnd;
     DEBUGLOG(5, "cwksp: reserving %p %zd bytes, %zd bytes remaining",
@@ -260,6 +262,8 @@ MEM_STATIC void* ZSTD_cwksp_reserve_internal_buffer_space(ZSTD_cwksp* ws, size_t
         ws->allocFailed = 1;
         return NULL;
     }
+    /* the area is reserved from the end of wksp.
+     * If it overlaps with tableValidEnd, it voids guarantees on values' range */
     if (alloc < ws->tableValidEnd) {
         ws->tableValidEnd = alloc;
     }
@@ -269,10 +273,12 @@ MEM_STATIC void* ZSTD_cwksp_reserve_internal_buffer_space(ZSTD_cwksp* ws, size_t
 
 /**
  * Moves the cwksp to the next phase, and does any necessary allocations.
+ * cwksp initialization must necessarily go through each phase in order.
  * Returns a 0 on success, or zstd error
  */
-MEM_STATIC size_t ZSTD_cwksp_internal_advance_phase(
-        ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase) {
+MEM_STATIC size_t
+ZSTD_cwksp_internal_advance_phase(ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase)
+{
     assert(phase >= ws->phase);
     if (phase > ws->phase) {
         /* Going from allocating objects to allocating buffers */
@@ -295,15 +301,15 @@ MEM_STATIC size_t ZSTD_cwksp_internal_advance_phase(
             {   /* Align the start of the tables to 64 bytes. Use [0, 63] bytes */
                 void* const alloc = ws->objectEnd;
                 size_t const bytesToAlign = ZSTD_cwksp_bytes_to_align_ptr(alloc, ZSTD_CWKSP_ALIGNMENT_BYTES);
-                void* const end = (BYTE*)alloc + bytesToAlign;
+                void* const objectEnd = (BYTE*)alloc + bytesToAlign;
                 DEBUGLOG(5, "reserving table alignment addtl space: %zu", bytesToAlign);
-                RETURN_ERROR_IF(end > ws->workspaceEnd, memory_allocation,
+                RETURN_ERROR_IF(objectEnd > ws->workspaceEnd, memory_allocation,
                                 "table phase - alignment initial allocation failed!");
-                ws->objectEnd = end;
-                ws->tableEnd = end;
-                ws->tableValidEnd = end;
-            }
-        }
+                ws->objectEnd = objectEnd;
+                ws->tableEnd = objectEnd;  /* table area starts being empty */
+                if (ws->tableValidEnd < ws->tableEnd) {
+                    ws->tableValidEnd = ws->tableEnd;
+        }   }   }
         ws->phase = phase;
         ZSTD_cwksp_assert_internal_consistency(ws);
     }
@@ -313,15 +319,17 @@ MEM_STATIC size_t ZSTD_cwksp_internal_advance_phase(
 /**
  * Returns whether this object/buffer/etc was allocated in this workspace.
  */
-MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr) {
+MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr)
+{
     return (ptr != NULL) && (ws->workspace <= ptr) && (ptr <= ws->workspaceEnd);
 }
 
 /**
  * Internal function. Do not use directly.
  */
-MEM_STATIC void* ZSTD_cwksp_reserve_internal(
-        ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase) {
+MEM_STATIC void*
+ZSTD_cwksp_reserve_internal(ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase)
+{
     void* alloc;
     if (ZSTD_isError(ZSTD_cwksp_internal_advance_phase(ws, phase)) || bytes == 0) {
         return NULL;
@@ -340,7 +348,9 @@ MEM_STATIC void* ZSTD_cwksp_reserve_internal(
     if (alloc) {
         alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
         if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
-            __asan_unpoison_memory_region(alloc, bytes);
+            /* We need to keep the redzone poisoned while unpoisoning the bytes that
+             * are actually allocated. */
+            __asan_unpoison_memory_region(alloc, bytes - 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE);
         }
     }
 #endif
@@ -351,14 +361,16 @@ MEM_STATIC void* ZSTD_cwksp_reserve_internal(
 /**
  * Reserves and returns unaligned memory.
  */
-MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes) {
+MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes)
+{
     return (BYTE*)ZSTD_cwksp_reserve_internal(ws, bytes, ZSTD_cwksp_alloc_buffers);
 }
 
 /**
  * Reserves and returns memory sized on and aligned on ZSTD_CWKSP_ALIGNMENT_BYTES (64 bytes).
  */
-MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes) {
+MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes)
+{
     void* ptr = ZSTD_cwksp_reserve_internal(ws, ZSTD_cwksp_align(bytes, ZSTD_CWKSP_ALIGNMENT_BYTES),
                                             ZSTD_cwksp_alloc_aligned);
     assert(((size_t)ptr & (ZSTD_CWKSP_ALIGNMENT_BYTES-1))== 0);
@@ -370,7 +382,8 @@ MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes) {
  * their values remain constrained, allowing us to re-use them without
  * memset()-ing them.
  */
-MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes) {
+MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes)
+{
     const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned;
     void* alloc;
     void* end;
@@ -408,9 +421,11 @@ MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes) {
 
 /**
  * Aligned on sizeof(void*).
+ * Note : should happen only once, at workspace first initialization
  */
-MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
-    size_t roundedBytes = ZSTD_cwksp_align(bytes, sizeof(void*));
+MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes)
+{
+    size_t const roundedBytes = ZSTD_cwksp_align(bytes, sizeof(void*));
     void* alloc = ws->objectEnd;
     void* end = (BYTE*)alloc + roundedBytes;
 
@@ -419,15 +434,15 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
     end = (BYTE *)end + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
 #endif
 
-    DEBUGLOG(5,
+    DEBUGLOG(4,
         "cwksp: reserving %p object %zd bytes (rounded to %zd), %zd bytes remaining",
         alloc, bytes, roundedBytes, ZSTD_cwksp_available_space(ws) - roundedBytes);
-    assert(((size_t)alloc & (sizeof(void*)-1)) == 0);
-    assert((bytes & (sizeof(void*)-1)) == 0);
+    assert((size_t)alloc % ZSTD_ALIGNOF(void*) == 0);
+    assert(bytes % ZSTD_ALIGNOF(void*) == 0);
     ZSTD_cwksp_assert_internal_consistency(ws);
     /* we must be in the first phase, no advance is possible */
     if (ws->phase != ZSTD_cwksp_alloc_objects || end > ws->workspaceEnd) {
-        DEBUGLOG(4, "cwksp: object alloc failed!");
+        DEBUGLOG(3, "cwksp: object alloc failed!");
         ws->allocFailed = 1;
         return NULL;
     }
@@ -438,7 +453,7 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
 #if ZSTD_ADDRESS_SANITIZER && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
     /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
      * either size. */
-    alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+    alloc = (BYTE*)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
     if (ws->isStatic == ZSTD_cwksp_dynamic_alloc) {
         __asan_unpoison_memory_region(alloc, bytes);
     }
@@ -447,7 +462,8 @@ MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
     return alloc;
 }
 
-MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws) {
+MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws)
+{
     DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_dirty");
 
 #if ZSTD_MEMORY_SANITIZER && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
@@ -485,7 +501,7 @@ MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
     assert(ws->tableValidEnd >= ws->objectEnd);
     assert(ws->tableValidEnd <= ws->allocStart);
     if (ws->tableValidEnd < ws->tableEnd) {
-        ZSTD_memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
+        ZSTD_memset(ws->tableValidEnd, 0, (size_t)((BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd));
     }
     ZSTD_cwksp_mark_tables_clean(ws);
 }
diff --git a/priv/zstd/lib/compress/zstd_double_fast.c b/priv/zstd/lib/compress/zstd_double_fast.c
index d0d3a78..0ad88ff 100644
--- a/priv/zstd/lib/compress/zstd_double_fast.c
+++ b/priv/zstd/lib/compress/zstd_double_fast.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,8 +11,43 @@
 #include "zstd_compress_internal.h"
 #include "zstd_double_fast.h"
 
+static void ZSTD_fillDoubleHashTableForCDict(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
 
-void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0) {
+                ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
+            }
+            if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
+                ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
+            }
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+    }   }
+}
+
+static void ZSTD_fillDoubleHashTableForCCtx(ZSTD_matchState_t* ms,
                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
@@ -43,15 +78,237 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
             /* Only load extra positions for ZSTD_dtlm_full */
             if (dtlm == ZSTD_dtlm_fast)
                 break;
-    }   }
+        }   }
+}
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_doubleFast_noDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* anchor = istart;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    /* presumes that, if there is a dictionary, it must be using Attach mode */
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    size_t mLength;
+    U32 offset;
+    U32 curr;
+
+    /* how many positions to search before increasing step size */
+    const size_t kStepIncr = 1 << kSearchStrength;
+    /* the position at which to increment the step size if no match is found */
+    const BYTE* nextStep;
+    size_t step; /* the current step size */
+
+    size_t hl0; /* the long hash at ip */
+    size_t hl1; /* the long hash at ip1 */
+
+    U32 idxl0; /* the long match index for ip */
+    U32 idxl1; /* the long match index for ip1 */
+
+    const BYTE* matchl0; /* the long match for ip */
+    const BYTE* matchs0; /* the short match for ip */
+    const BYTE* matchl1; /* the long match for ip1 */
+
+    const BYTE* ip = istart; /* the current position */
+    const BYTE* ip1; /* the next position */
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
+
+    /* init */
+    ip += ((ip - prefixLowest) == 0);
+    {
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+
+    /* Outer Loop: one iteration per match found and stored */
+    while (1) {
+        step = 1;
+        nextStep = ip + kStepIncr;
+        ip1 = ip + step;
+
+        if (ip1 > ilimit) {
+            goto _cleanup;
+        }
+
+        hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
+        idxl0 = hashLong[hl0];
+        matchl0 = base + idxl0;
+
+        /* Inner Loop: one iteration per search / position */
+        do {
+            const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
+            const U32 idxs0 = hashSmall[hs0];
+            curr = (U32)(ip-base);
+            matchs0 = base + idxs0;
+
+            hashLong[hl0] = hashSmall[hs0] = curr;   /* update hash tables */
+
+            /* check noDict repcode */
+            if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+                mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+                ip++;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+                goto _match_stored;
+            }
+
+            hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
+
+            if (idxl0 > prefixLowestIndex) {
+                /* check prefix long match */
+                if (MEM_read64(matchl0) == MEM_read64(ip)) {
+                    mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
+                    offset = (U32)(ip-matchl0);
+                    while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
+                    goto _match_found;
+                }
+            }
+
+            idxl1 = hashLong[hl1];
+            matchl1 = base + idxl1;
+
+            if (idxs0 > prefixLowestIndex) {
+                /* check prefix short match */
+                if (MEM_read32(matchs0) == MEM_read32(ip)) {
+                    goto _search_next_long;
+                }
+            }
+
+            if (ip1 >= nextStep) {
+                PREFETCH_L1(ip1 + 64);
+                PREFETCH_L1(ip1 + 128);
+                step++;
+                nextStep += kStepIncr;
+            }
+            ip = ip1;
+            ip1 += step;
+
+            hl0 = hl1;
+            idxl0 = idxl1;
+            matchl0 = matchl1;
+    #if defined(__aarch64__)
+            PREFETCH_L1(ip+256);
+    #endif
+        } while (ip1 <= ilimit);
+
+_cleanup:
+        /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+         * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+        offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+        /* save reps for next block */
+        rep[0] = offset_1 ? offset_1 : offsetSaved1;
+        rep[1] = offset_2 ? offset_2 : offsetSaved2;
+
+        /* Return the last literals size */
+        return (size_t)(iend - anchor);
+
+_search_next_long:
+
+        /* check prefix long +1 match */
+        if (idxl1 > prefixLowestIndex) {
+            if (MEM_read64(matchl1) == MEM_read64(ip1)) {
+                ip = ip1;
+                mLength = ZSTD_count(ip+8, matchl1+8, iend) + 8;
+                offset = (U32)(ip-matchl1);
+                while (((ip>anchor) & (matchl1>prefixLowest)) && (ip[-1] == matchl1[-1])) { ip--; matchl1--; mLength++; } /* catch up */
+                goto _match_found;
+            }
+        }
+
+        /* if no long +1 match, explore the short match we found */
+        mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
+        offset = (U32)(ip - matchs0);
+        while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* catch up */
+
+        /* fall-through */
+
+_match_found: /* requires ip, offset, mLength */
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        if (step < 4) {
+            /* It is unsafe to write this value back to the hashtable when ip1 is
+             * greater than or equal to the new ip we will have after we're done
+             * processing this match. Rather than perform that test directly
+             * (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
+             * more predictable test. The minmatch even if we take a short match is
+             * 4 bytes, so as long as step, the distance between ip and ip1
+             * (initially) is less than 4, we know ip1 < new ip. */
+            hashLong[hl1] = (U32)(ip1 - base);
+        }
+
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+
+_match_stored:
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = curr+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                    & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
+                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+        }
+    }
 }
 
 
 FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_doubleFast_generic(
+size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize,
-        U32 const mls /* template */, ZSTD_dictMode_e const dictMode)
+        U32 const mls /* template */)
 {
     ZSTD_compressionParameters const* cParams = &ms->cParams;
     U32* const hashLong = ms->hashTable;
@@ -69,57 +326,39 @@ size_t ZSTD_compressBlock_doubleFast_generic(
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
-    U32 offsetSaved = 0;
 
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
-    const ZSTD_compressionParameters* const dictCParams =
-                                     dictMode == ZSTD_dictMatchState ?
-                                     &dms->cParams : NULL;
-    const U32* const dictHashLong  = dictMode == ZSTD_dictMatchState ?
-                                     dms->hashTable : NULL;
-    const U32* const dictHashSmall = dictMode == ZSTD_dictMatchState ?
-                                     dms->chainTable : NULL;
-    const U32 dictStartIndex       = dictMode == ZSTD_dictMatchState ?
-                                     dms->window.dictLimit : 0;
-    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
-                                     dms->window.base : NULL;
-    const BYTE* const dictStart    = dictMode == ZSTD_dictMatchState ?
-                                     dictBase + dictStartIndex : NULL;
-    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
-                                     dms->window.nextSrc : NULL;
-    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
-                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
-                                     0;
-    const U32 dictHBitsL           = dictMode == ZSTD_dictMatchState ?
-                                     dictCParams->hashLog : hBitsL;
-    const U32 dictHBitsS           = dictMode == ZSTD_dictMatchState ?
-                                     dictCParams->chainLog : hBitsS;
+    const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
+    const U32* const dictHashLong  = dms->hashTable;
+    const U32* const dictHashSmall = dms->chainTable;
+    const U32 dictStartIndex       = dms->window.dictLimit;
+    const BYTE* const dictBase     = dms->window.base;
+    const BYTE* const dictStart    = dictBase + dictStartIndex;
+    const BYTE* const dictEnd      = dms->window.nextSrc;
+    const U32 dictIndexDelta       = prefixLowestIndex - (U32)(dictEnd - dictBase);
+    const U32 dictHBitsL           = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    const U32 dictHBitsS           = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
     const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
 
-    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
-
-    assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
 
     /* if a dictionary is attached, it must be within window range */
-    if (dictMode == ZSTD_dictMatchState) {
-        assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
+    assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
+
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashLong, hashTableBytes)
+        PREFETCH_AREA(dictHashSmall, chainTableBytes)
     }
 
     /* init */
     ip += (dictAndPrefixLength == 0);
-    if (dictMode == ZSTD_noDict) {
-        U32 const curr = (U32)(ip - base);
-        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
-        U32 const maxRep = curr - windowLow;
-        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
-    }
-    if (dictMode == ZSTD_dictMatchState) {
-        /* dictMatchState repCode checks don't currently handle repCode == 0
-         * disabling. */
-        assert(offset_1 <= dictAndPrefixLength);
-        assert(offset_2 <= dictAndPrefixLength);
-    }
+
+    /* dictMatchState repCode checks don't currently handle repCode == 0
+     * disabling. */
+    assert(offset_1 <= dictAndPrefixLength);
+    assert(offset_2 <= dictAndPrefixLength);
 
     /* Main Search Loop */
     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
@@ -127,37 +366,30 @@ size_t ZSTD_compressBlock_doubleFast_generic(
         U32 offset;
         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
-        size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
-        size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
+        size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
+        size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
+        U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
+        U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
+        int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
         U32 const curr = (U32)(ip-base);
         U32 const matchIndexL = hashLong[h2];
         U32 matchIndexS = hashSmall[h];
         const BYTE* matchLong = base + matchIndexL;
         const BYTE* match = base + matchIndexS;
         const U32 repIndex = curr + 1 - offset_1;
-        const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
-                            && repIndex < prefixLowestIndex) ?
+        const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
                                dictBase + (repIndex - dictIndexDelta) :
                                base + repIndex;
         hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
 
-        /* check dictMatchState repcode */
-        if (dictMode == ZSTD_dictMatchState
-            && ((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+        /* check repcode */
+        if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
             && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
             const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
-            goto _match_stored;
-        }
-
-        /* check noDict repcode */
-        if ( dictMode == ZSTD_noDict
-          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
-            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
             goto _match_stored;
         }
 
@@ -169,9 +401,9 @@ size_t ZSTD_compressBlock_doubleFast_generic(
                 while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
                 goto _match_found;
             }
-        } else if (dictMode == ZSTD_dictMatchState) {
+        } else if (dictTagsMatchL) {
             /* check dictMatchState long match */
-            U32 const dictMatchIndexL = dictHashLong[dictHL];
+            U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
             const BYTE* dictMatchL = dictBase + dictMatchIndexL;
             assert(dictMatchL < dictEnd);
 
@@ -187,9 +419,9 @@ size_t ZSTD_compressBlock_doubleFast_generic(
             if (MEM_read32(match) == MEM_read32(ip)) {
                 goto _search_next_long;
             }
-        } else if (dictMode == ZSTD_dictMatchState) {
+        } else if (dictTagsMatchS) {
             /* check dictMatchState short match */
-            U32 const dictMatchIndexS = dictHashSmall[dictHS];
+            U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
             match = dictBase + dictMatchIndexS;
             matchIndexS = dictMatchIndexS + dictIndexDelta;
 
@@ -204,10 +436,11 @@ size_t ZSTD_compressBlock_doubleFast_generic(
         continue;
 
 _search_next_long:
-
         {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-            size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
+            size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
             U32 const matchIndexL3 = hashLong[hl3];
+            U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
             const BYTE* matchL3 = base + matchIndexL3;
             hashLong[hl3] = curr + 1;
 
@@ -220,9 +453,9 @@ size_t ZSTD_compressBlock_doubleFast_generic(
                     while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
                     goto _match_found;
                 }
-            } else if (dictMode == ZSTD_dictMatchState) {
+            } else if (dictTagsMatchL3) {
                 /* check dict long +1 match */
-                U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
+                U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
                 const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
                 assert(dictMatchL3 < dictEnd);
                 if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
@@ -234,7 +467,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
         }   }   }
 
         /* if no long +1 match, explore the short match we found */
-        if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
+        if (matchIndexS < prefixLowestIndex) {
             mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
             offset = (U32)(curr - matchIndexS);
             while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
@@ -244,13 +477,11 @@ size_t ZSTD_compressBlock_doubleFast_generic(
             while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
         }
 
-        /* fall-through */
-
 _match_found:
         offset_2 = offset_1;
         offset_1 = offset;
 
-        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
 _match_stored:
         /* match found */
@@ -268,53 +499,55 @@ size_t ZSTD_compressBlock_doubleFast_generic(
             }
 
             /* check immediate repcode */
-            if (dictMode == ZSTD_dictMatchState) {
-                while (ip <= ilimit) {
-                    U32 const current2 = (U32)(ip-base);
-                    U32 const repIndex2 = current2 - offset_2;
-                    const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
-                        && repIndex2 < prefixLowestIndex ?
-                            dictBase + repIndex2 - dictIndexDelta :
-                            base + repIndex2;
-                    if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
-                       && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                        const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
-                        size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
-                        U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                        ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
-                        hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
-                        hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
-                        ip += repLength2;
-                        anchor = ip;
-                        continue;
-                    }
-                    break;
-            }   }
-
-            if (dictMode == ZSTD_noDict) {
-                while ( (ip <= ilimit)
-                     && ( (offset_2>0)
-                        & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                    /* store sequence */
-                    size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                    U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
-                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
-                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, rLength-MINMATCH);
-                    ip += rLength;
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
+                        dictBase + repIndex2 - dictIndexDelta :
+                        base + repIndex2;
+                if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
                     anchor = ip;
-                    continue;   /* faster when present ... (?) */
-        }   }   }
+                    continue;
+                }
+                break;
+            }
+        }
     }   /* while (ip < ilimit) */
 
     /* save reps for next block */
-    rep[0] = offset_1 ? offset_1 : offsetSaved;
-    rep[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
 }
 
+#define ZSTD_GEN_DFAST_FN(dictMode, mls)                                                                 \
+    static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls(                                      \
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                          \
+            void const* src, size_t srcSize)                                                             \
+    {                                                                                                    \
+        return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
+    }
+
+ZSTD_GEN_DFAST_FN(noDict, 4)
+ZSTD_GEN_DFAST_FN(noDict, 5)
+ZSTD_GEN_DFAST_FN(noDict, 6)
+ZSTD_GEN_DFAST_FN(noDict, 7)
+
+ZSTD_GEN_DFAST_FN(dictMatchState, 4)
+ZSTD_GEN_DFAST_FN(dictMatchState, 5)
+ZSTD_GEN_DFAST_FN(dictMatchState, 6)
+ZSTD_GEN_DFAST_FN(dictMatchState, 7)
+
 
 size_t ZSTD_compressBlock_doubleFast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -325,13 +558,13 @@ size_t ZSTD_compressBlock_doubleFast(
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
+        return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
+        return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
+        return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
+        return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
     }
 }
 
@@ -345,13 +578,13 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState(
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
+        return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
+        return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
+        return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
+        return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
     }
 }
 
@@ -387,7 +620,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
 
     /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
     if (prefixStartIndex == dictStartIndex)
-        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, mls, ZSTD_noDict);
+        return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
 
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
@@ -409,12 +642,12 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
         hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
 
         if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
-            & (offset_1 < curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
+            & (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
         } else {
             if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
                 const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
@@ -425,7 +658,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
             } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
@@ -450,7 +683,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 }
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
 
             } else {
                 ip += ((ip-anchor) >> kSearchStrength) + 1;
@@ -477,12 +710,12 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 U32 const repIndex2 = current2 - offset_2;
                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
                 if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3)   /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
-                    & (offset_2 < current2 - dictStartIndex))
+                    & (offset_2 <= current2 - dictStartIndex))
                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                     U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
                     ip += repLength2;
@@ -500,6 +733,10 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
     return (size_t)(iend - anchor);
 }
 
+ZSTD_GEN_DFAST_FN(extDict, 4)
+ZSTD_GEN_DFAST_FN(extDict, 5)
+ZSTD_GEN_DFAST_FN(extDict, 6)
+ZSTD_GEN_DFAST_FN(extDict, 7)
 
 size_t ZSTD_compressBlock_doubleFast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -510,12 +747,12 @@ size_t ZSTD_compressBlock_doubleFast_extDict(
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
+        return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
+        return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
+        return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
+        return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
     }
 }
diff --git a/priv/zstd/lib/compress/zstd_double_fast.h b/priv/zstd/lib/compress/zstd_double_fast.h
index e16b7b0..6f0047c 100644
--- a/priv/zstd/lib/compress/zstd_double_fast.h
+++ b/priv/zstd/lib/compress/zstd_double_fast.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -19,7 +19,8 @@ extern "C" {
 #include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
 
 void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
-                              void const* end, ZSTD_dictTableLoadMethod_e dtlm);
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm,
+                              ZSTD_tableFillPurpose_e tfp);
 size_t ZSTD_compressBlock_doubleFast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
diff --git a/priv/zstd/lib/compress/zstd_fast.c b/priv/zstd/lib/compress/zstd_fast.c
index 4edc04d..5f2c6a2 100644
--- a/priv/zstd/lib/compress/zstd_fast.c
+++ b/priv/zstd/lib/compress/zstd_fast.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,8 +11,42 @@
 #include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
 #include "zstd_fast.h"
 
+static void ZSTD_fillHashTableForCDict(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
+    U32  const mls = cParams->minMatch;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
 
-void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+    /* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
+     * Feel free to remove this assert if there's a good reason! */
+    assert(dtlm == ZSTD_dtlm_full);
+
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
+        U32 const curr = (U32)(ip - base);
+        {   size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
+            ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr);   }
+
+        if (dtlm == ZSTD_dtlm_fast) continue;
+        /* Only load extra positions for ZSTD_dtlm_full */
+        {   U32 p;
+            for (p = 1; p < fastHashFillStep; ++p) {
+                size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
+                if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {  /* not yet filled */
+                    ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
+                }   }   }   }
+}
+
+static void ZSTD_fillHashTableForCCtx(ZSTD_matchState_t* ms,
                         const void* const end,
                         ZSTD_dictTableLoadMethod_e dtlm)
 {
@@ -25,6 +59,10 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
     const U32 fastHashFillStep = 3;
 
+    /* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
+     * Feel free to remove this assert if there's a good reason! */
+    assert(dtlm == ZSTD_dtlm_fast);
+
     /* Always insert every fastHashFillStep position into the hash table.
      * Insert the other positions if their hash entry is empty.
      */
@@ -42,146 +80,344 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
     }   }   }   }
 }
 
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp)
+{
+    if (tfp == ZSTD_tfp_forCDict) {
+        ZSTD_fillHashTableForCDict(ms, end, dtlm);
+    } else {
+        ZSTD_fillHashTableForCCtx(ms, end, dtlm);
+    }
+}
+
 
+/**
+ * If you squint hard enough (and ignore repcodes), the search operation at any
+ * given position is broken into 4 stages:
+ *
+ * 1. Hash   (map position to hash value via input read)
+ * 2. Lookup (map hash val to index via hashtable read)
+ * 3. Load   (map index to value at that position via input read)
+ * 4. Compare
+ *
+ * Each of these steps involves a memory read at an address which is computed
+ * from the previous step. This means these steps must be sequenced and their
+ * latencies are cumulative.
+ *
+ * Rather than do 1->2->3->4 sequentially for a single position before moving
+ * onto the next, this implementation interleaves these operations across the
+ * next few positions:
+ *
+ * R = Repcode Read & Compare
+ * H = Hash
+ * T = Table Lookup
+ * M = Match Read & Compare
+ *
+ * Pos | Time -->
+ * ----+-------------------
+ * N   | ... M
+ * N+1 | ...   TM
+ * N+2 |    R H   T M
+ * N+3 |         H    TM
+ * N+4 |           R H   T M
+ * N+5 |                H   ...
+ * N+6 |                  R ...
+ *
+ * This is very much analogous to the pipelining of execution in a CPU. And just
+ * like a CPU, we have to dump the pipeline when we find a match (i.e., take a
+ * branch).
+ *
+ * When this happens, we throw away our current state, and do the following prep
+ * to re-enter the loop:
+ *
+ * Pos | Time -->
+ * ----+-------------------
+ * N   | H T
+ * N+1 |  H
+ *
+ * This is also the work we do at the beginning to enter the loop initially.
+ */
 FORCE_INLINE_TEMPLATE size_t
-ZSTD_compressBlock_fast_generic(
+ZSTD_compressBlock_fast_noDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize,
-        U32 const mls)
+        U32 const mls, U32 const hasStep)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
     U32 const hlog = cParams->hashLog;
     /* support stepSize of 0 */
-    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
+    size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2;
     const BYTE* const base = ms->window.base;
     const BYTE* const istart = (const BYTE*)src;
-    /* We check ip0 (ip + 0) and ip1 (ip + 1) each loop */
-    const BYTE* ip0 = istart;
-    const BYTE* ip1;
-    const BYTE* anchor = istart;
     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
     const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
     const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=rep[0], offset_2=rep[1];
-    U32 offsetSaved = 0;
 
-    /* init */
+    const BYTE* anchor = istart;
+    const BYTE* ip0 = istart;
+    const BYTE* ip1;
+    const BYTE* ip2;
+    const BYTE* ip3;
+    U32 current0;
+
+    U32 rep_offset1 = rep[0];
+    U32 rep_offset2 = rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    size_t hash0; /* hash for ip0 */
+    size_t hash1; /* hash for ip1 */
+    U32 idx; /* match idx for ip0 */
+    U32 mval; /* src value at match idx */
+
+    U32 offcode;
+    const BYTE* match0;
+    size_t mLength;
+
+    /* ip0 and ip1 are always adjacent. The targetLength skipping and
+     * uncompressibility acceleration is applied to every other position,
+     * matching the behavior of #1562. step therefore represents the gap
+     * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */
+    size_t step;
+    const BYTE* nextStep;
+    const size_t kStepIncr = (1 << (kSearchStrength - 1));
+
     DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
     ip0 += (ip0 == prefixStart);
-    ip1 = ip0 + 1;
     {   U32 const curr = (U32)(ip0 - base);
         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
         U32 const maxRep = curr - windowLow;
-        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+        if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
+        if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
     }
 
-    /* Main Search Loop */
-#ifdef __INTEL_COMPILER
-    /* From intel 'The vector pragma indicates that the loop should be
-     * vectorized if it is legal to do so'. Can be used together with
-     * #pragma ivdep (but have opted to exclude that because intel
-     * warns against using it).*/
-    #pragma vector always
-#endif
-    while (ip1 < ilimit) {   /* < instead of <=, because check at ip0+2 */
-        size_t mLength;
-        BYTE const* ip2 = ip0 + 2;
-        size_t const h0 = ZSTD_hashPtr(ip0, hlog, mls);
-        U32 const val0 = MEM_read32(ip0);
-        size_t const h1 = ZSTD_hashPtr(ip1, hlog, mls);
-        U32 const val1 = MEM_read32(ip1);
-        U32 const current0 = (U32)(ip0-base);
-        U32 const current1 = (U32)(ip1-base);
-        U32 const matchIndex0 = hashTable[h0];
-        U32 const matchIndex1 = hashTable[h1];
-        BYTE const* repMatch = ip2 - offset_1;
-        const BYTE* match0 = base + matchIndex0;
-        const BYTE* match1 = base + matchIndex1;
-        U32 offcode;
-
-#if defined(__aarch64__)
-        PREFETCH_L1(ip0+256);
-#endif
-
-        hashTable[h0] = current0;   /* update hash table */
-        hashTable[h1] = current1;   /* update hash table */
-
-        assert(ip0 + 1 == ip1);
-
-        if ((offset_1 > 0) & (MEM_read32(repMatch) == MEM_read32(ip2))) {
-            mLength = (ip2[-1] == repMatch[-1]) ? 1 : 0;
-            ip0 = ip2 - mLength;
-            match0 = repMatch - mLength;
+    /* start each op */
+_start: /* Requires: ip0 */
+
+    step = stepSize;
+    nextStep = ip0 + kStepIncr;
+
+    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
+    ip1 = ip0 + 1;
+    ip2 = ip0 + step;
+    ip3 = ip2 + 1;
+
+    if (ip3 >= ilimit) {
+        goto _cleanup;
+    }
+
+    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+    hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+
+    idx = hashTable[hash0];
+
+    do {
+        /* load repcode match for ip[2]*/
+        const U32 rval = MEM_read32(ip2 - rep_offset1);
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        /* check repcode at ip[2] */
+        if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
+            ip0 = ip2;
+            match0 = ip0 - rep_offset1;
+            mLength = ip0[-1] == match0[-1];
+            ip0 -= mLength;
+            match0 -= mLength;
+            offcode = REPCODE1_TO_OFFBASE;
             mLength += 4;
-            offcode = 0;
+
+            /* First write next hash table entry; we've already calculated it.
+             * This write is known to be safe because the ip1 is before the
+             * repcode (ip2). */
+            hashTable[hash1] = (U32)(ip1 - base);
+
             goto _match;
         }
-        if ((matchIndex0 > prefixStartIndex) && MEM_read32(match0) == val0) {
-            /* found a regular match */
-            goto _offset;
+
+        /* load match for ip[0] */
+        if (idx >= prefixStartIndex) {
+            mval = MEM_read32(base + idx);
+        } else {
+            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
         }
-        if ((matchIndex1 > prefixStartIndex) && MEM_read32(match1) == val1) {
-            /* found a regular match after one literal */
-            ip0 = ip1;
-            match0 = match1;
+
+        /* check match at ip[0] */
+        if (MEM_read32(ip0) == mval) {
+            /* found a match! */
+
+            /* First write next hash table entry; we've already calculated it.
+             * This write is known to be safe because the ip1 == ip0 + 1, so
+             * we know we will resume searching after ip1 */
+            hashTable[hash1] = (U32)(ip1 - base);
+
             goto _offset;
         }
-        {   size_t const step = ((size_t)(ip0-anchor) >> (kSearchStrength - 1)) + stepSize;
-            assert(step >= 2);
-            ip0 += step;
-            ip1 += step;
-            continue;
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip3;
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        /* load match for ip[0] */
+        if (idx >= prefixStartIndex) {
+            mval = MEM_read32(base + idx);
+        } else {
+            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
         }
-_offset: /* Requires: ip0, match0 */
-        /* Compute the offset code */
-        offset_2 = offset_1;
-        offset_1 = (U32)(ip0-match0);
-        offcode = offset_1 + ZSTD_REP_MOVE;
-        mLength = 4;
-        /* Count the backwards match length */
-        while (((ip0>anchor) & (match0>prefixStart))
-             && (ip0[-1] == match0[-1])) { ip0--; match0--; mLength++; } /* catch up */
 
-_match: /* Requires: ip0, match0, offcode */
-        /* Count the forward length */
-        mLength += ZSTD_count(ip0+mLength, match0+mLength, iend);
-        ZSTD_storeSeq(seqStore, (size_t)(ip0-anchor), anchor, iend, offcode, mLength-MINMATCH);
-        /* match found */
-        ip0 += mLength;
-        anchor = ip0;
+        /* check match at ip[0] */
+        if (MEM_read32(ip0) == mval) {
+            /* found a match! */
+
+            /* first write next hash table entry; we've already calculated it */
+            if (step <= 4) {
+                /* We need to avoid writing an index into the hash table >= the
+                 * position at which we will pick up our searching after we've
+                 * taken this match.
+                 *
+                 * The minimum possible match has length 4, so the earliest ip0
+                 * can be after we take this match will be the current ip0 + 4.
+                 * ip1 is ip0 + step - 1. If ip1 is >= ip0 + 4, we can't safely
+                 * write this position.
+                 */
+                hashTable[hash1] = (U32)(ip1 - base);
+            }
 
-        if (ip0 <= ilimit) {
-            /* Fill Table */
-            assert(base+current0+2 > istart);  /* check base overflow */
-            hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
-            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+            goto _offset;
+        }
 
-            if (offset_2 > 0) { /* offset_2==0 means offset_2 is invalidated */
-                while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - offset_2)) ) {
-                    /* store sequence */
-                    size_t const rLength = ZSTD_count(ip0+4, ip0+4-offset_2, iend) + 4;
-                    { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
-                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
-                    ip0 += rLength;
-                    ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, 0 /*offCode*/, rLength-MINMATCH);
-                    anchor = ip0;
-                    continue;   /* faster when present (confirmed on gcc-8) ... (?) */
-        }   }   }
-        ip1 = ip0 + 1;
-    }
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip0 + step;
+        ip3 = ip1 + step;
+
+        /* calculate step */
+        if (ip2 >= nextStep) {
+            step++;
+            PREFETCH_L1(ip1 + 64);
+            PREFETCH_L1(ip1 + 128);
+            nextStep += kStepIncr;
+        }
+    } while (ip3 < ilimit);
+
+_cleanup:
+    /* Note that there are probably still a couple positions we could search.
+     * However, it seems to be a meaningful performance hit to try to search
+     * them. So let's not. */
+
+    /* When the repcodes are outside of the prefix, we set them to zero before the loop.
+     * When the offsets are still zero, we need to restore them after the block to have a correct
+     * repcode history. If only one offset was invalid, it is easy. The tricky case is when both
+     * offsets were invalid. We need to figure out which offset to refill with.
+     *     - If both offsets are zero they are in the same order.
+     *     - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
+     *     - If only one is zero, we need to decide which offset to restore.
+     *         - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
+     *         - It is impossible for rep_offset2 to be non-zero.
+     *
+     * So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
+     * set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
+     */
+    offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;
 
     /* save reps for next block */
-    rep[0] = offset_1 ? offset_1 : offsetSaved;
-    rep[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
+    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
+
+_offset: /* Requires: ip0, idx */
+
+    /* Compute the offset code. */
+    match0 = base + idx;
+    rep_offset2 = rep_offset1;
+    rep_offset1 = (U32)(ip0-match0);
+    offcode = OFFSET_TO_OFFBASE(rep_offset1);
+    mLength = 4;
+
+    /* Count the backwards match length. */
+    while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
+        ip0--;
+        match0--;
+        mLength++;
+    }
+
+_match: /* Requires: ip0, match0, offcode */
+
+    /* Count the forward length. */
+    mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);
+
+    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
+
+    ip0 += mLength;
+    anchor = ip0;
+
+    /* Fill table and check for immediate repcode. */
+    if (ip0 <= ilimit) {
+        /* Fill Table */
+        assert(base+current0+2 > istart);  /* check base overflow */
+        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
+        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+        if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
+            while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
+                { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
+                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                ip0 += rLength;
+                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
+                anchor = ip0;
+                continue;   /* faster when present (confirmed on gcc-8) ... (?) */
+    }   }   }
+
+    goto _start;
 }
 
+#define ZSTD_GEN_FAST_FN(dictMode, mls, step)                                                            \
+    static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step(                                      \
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \
+            void const* src, size_t srcSize)                                                       \
+    {                                                                                              \
+        return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \
+    }
+
+ZSTD_GEN_FAST_FN(noDict, 4, 1)
+ZSTD_GEN_FAST_FN(noDict, 5, 1)
+ZSTD_GEN_FAST_FN(noDict, 6, 1)
+ZSTD_GEN_FAST_FN(noDict, 7, 1)
+
+ZSTD_GEN_FAST_FN(noDict, 4, 0)
+ZSTD_GEN_FAST_FN(noDict, 5, 0)
+ZSTD_GEN_FAST_FN(noDict, 6, 0)
+ZSTD_GEN_FAST_FN(noDict, 7, 0)
 
 size_t ZSTD_compressBlock_fast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -189,24 +425,40 @@ size_t ZSTD_compressBlock_fast(
 {
     U32 const mls = ms->cParams.minMatch;
     assert(ms->dictMatchState == NULL);
-    switch(mls)
-    {
-    default: /* includes case 3 */
-    case 4 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4);
-    case 5 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5);
-    case 6 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6);
-    case 7 :
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7);
+    if (ms->cParams.targetLength > 1) {
+        switch(mls)
+        {
+        default: /* includes case 3 */
+        case 4 :
+            return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);
+        case 5 :
+            return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);
+        case 6 :
+            return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);
+        case 7 :
+            return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);
+        }
+    } else {
+        switch(mls)
+        {
+        default: /* includes case 3 */
+        case 4 :
+            return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);
+        case 5 :
+            return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);
+        case 6 :
+            return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);
+        case 7 :
+            return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);
+        }
+
     }
 }
 
 FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_fast_dictMatchState_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        void const* src, size_t srcSize, U32 const mls)
+        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
@@ -215,14 +467,14 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
     const BYTE* const base = ms->window.base;
     const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
+    const BYTE* ip0 = istart;
+    const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
     const BYTE* anchor = istart;
     const U32   prefixStartIndex = ms->window.dictLimit;
     const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
     U32 offset_1=rep[0], offset_2=rep[1];
-    U32 offsetSaved = 0;
 
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
     const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
@@ -232,125 +484,182 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     const BYTE* const dictStart    = dictBase + dictStartIndex;
     const BYTE* const dictEnd      = dms->window.nextSrc;
     const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
-    const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);
-    const U32 dictHLog             = dictCParams->hashLog;
+    const U32 dictAndPrefixLength  = (U32)(istart - prefixStart + dictEnd - dictStart);
+    const U32 dictHBits            = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
 
     /* if a dictionary is still attached, it necessarily means that
      * it is within window size. So we just check it. */
     const U32 maxDistance = 1U << cParams->windowLog;
-    const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
     assert(endIndex - prefixStartIndex <= maxDistance);
     (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */
 
+    (void)hasStep; /* not currently specialized on whether it's accelerated */
+
     /* ensure there will be no underflow
      * when translating a dict index into a local index */
     assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
 
+    if (ms->prefetchCDictTables) {
+        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
+        PREFETCH_AREA(dictHashTable, hashTableBytes)
+    }
+
     /* init */
     DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
-    ip += (dictAndPrefixLength == 0);
+    ip0 += (dictAndPrefixLength == 0);
     /* dictMatchState repCode checks don't currently handle repCode == 0
      * disabling. */
     assert(offset_1 <= dictAndPrefixLength);
     assert(offset_2 <= dictAndPrefixLength);
 
-    /* Main Search Loop */
-    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+    /* Outer search loop */
+    assert(stepSize >= 1);
+    while (ip1 <= ilimit) {   /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
         size_t mLength;
-        size_t const h = ZSTD_hashPtr(ip, hlog, mls);
-        U32 const curr = (U32)(ip-base);
-        U32 const matchIndex = hashTable[h];
-        const BYTE* match = base + matchIndex;
-        const U32 repIndex = curr + 1 - offset_1;
-        const BYTE* repMatch = (repIndex < prefixStartIndex) ?
-                               dictBase + (repIndex - dictIndexDelta) :
-                               base + repIndex;
-        hashTable[h] = curr;   /* update hash table */
-
-        if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
-          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
-        } else if ( (matchIndex <= prefixStartIndex) ) {
-            size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);
-            U32 const dictMatchIndex = dictHashTable[dictHash];
-            const BYTE* dictMatch = dictBase + dictMatchIndex;
-            if (dictMatchIndex <= dictStartIndex ||
-                MEM_read32(dictMatch) != MEM_read32(ip)) {
-                assert(stepSize >= 1);
-                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
-                continue;
-            } else {
-                /* found a dict match */
-                U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta);
-                mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
-                while (((ip>anchor) & (dictMatch>dictStart))
-                     && (ip[-1] == dictMatch[-1])) {
-                    ip--; dictMatch--; mLength++;
+        size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+
+        size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
+        U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
+        int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);
+
+        U32 matchIndex = hashTable[hash0];
+        U32 curr = (U32)(ip0 - base);
+        size_t step = stepSize;
+        const size_t kStepIncr = 1 << kSearchStrength;
+        const BYTE* nextStep = ip0 + kStepIncr;
+
+        /* Inner search loop */
+        while (1) {
+            const BYTE* match = base + matchIndex;
+            const U32 repIndex = curr + 1 - offset_1;
+            const BYTE* repMatch = (repIndex < prefixStartIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+            size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
+            hashTable[hash0] = curr;   /* update hash table */
+
+            if (((U32) ((prefixStartIndex - 1) - repIndex) >=
+                 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
+                && (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
+                const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+                mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
+                ip0++;
+                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
+                break;
+            }
+
+            if (dictTagsMatch) {
+                /* Found a possible dict match */
+                const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
+                const BYTE* dictMatch = dictBase + dictMatchIndex;
+                if (dictMatchIndex > dictStartIndex &&
+                    MEM_read32(dictMatch) == MEM_read32(ip0)) {
+                    /* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
+                    if (matchIndex <= prefixStartIndex) {
+                        U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
+                        mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
+                        while (((ip0 > anchor) & (dictMatch > dictStart))
+                            && (ip0[-1] == dictMatch[-1])) {
+                            ip0--;
+                            dictMatch--;
+                            mLength++;
+                        } /* catch up */
+                        offset_2 = offset_1;
+                        offset_1 = offset;
+                        ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+                        break;
+                    }
+                }
+            }
+
+            if (matchIndex > prefixStartIndex && MEM_read32(match) == MEM_read32(ip0)) {
+                /* found a regular match */
+                U32 const offset = (U32) (ip0 - match);
+                mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
+                while (((ip0 > anchor) & (match > prefixStart))
+                       && (ip0[-1] == match[-1])) {
+                    ip0--;
+                    match--;
+                    mLength++;
                 } /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
+                break;
             }
-        } else if (MEM_read32(match) != MEM_read32(ip)) {
-            /* it's not a match, and we're not going to check the dictionary */
-            assert(stepSize >= 1);
-            ip += ((ip-anchor) >> kSearchStrength) + stepSize;
-            continue;
-        } else {
-            /* found a regular match */
-            U32 const offset = (U32)(ip-match);
-            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-            while (((ip>anchor) & (match>prefixStart))
-                 && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
-            offset_2 = offset_1;
-            offset_1 = offset;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-        }
+
+            /* Prepare for next iteration */
+            dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
+            dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
+            matchIndex = hashTable[hash1];
+
+            if (ip1 >= nextStep) {
+                step++;
+                nextStep += kStepIncr;
+            }
+            ip0 = ip1;
+            ip1 = ip1 + step;
+            if (ip1 > ilimit) goto _cleanup;
+
+            curr = (U32)(ip0 - base);
+            hash0 = hash1;
+        }   /* end inner search loop */
 
         /* match found */
-        ip += mLength;
-        anchor = ip;
+        assert(mLength);
+        ip0 += mLength;
+        anchor = ip0;
 
-        if (ip <= ilimit) {
+        if (ip0 <= ilimit) {
             /* Fill Table */
             assert(base+curr+2 > istart);  /* check base overflow */
             hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
-            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
 
             /* check immediate repcode */
-            while (ip <= ilimit) {
-                U32 const current2 = (U32)(ip-base);
+            while (ip0 <= ilimit) {
+                U32 const current2 = (U32)(ip0-base);
                 U32 const repIndex2 = current2 - offset_2;
                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
                         dictBase - dictIndexDelta + repIndex2 :
                         base + repIndex2;
                 if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                   && (MEM_read32(repMatch2) == MEM_read32(ip0))) {
                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
-                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
-                    ip += repLength2;
-                    anchor = ip;
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
+                    ip0 += repLength2;
+                    anchor = ip0;
                     continue;
                 }
                 break;
             }
         }
+
+        /* Prepare for next iteration */
+        assert(ip0 == anchor);
+        ip1 = ip0 + stepSize;
     }
 
+_cleanup:
     /* save reps for next block */
-    rep[0] = offset_1 ? offset_1 : offsetSaved;
-    rep[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
 }
 
+
+ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)
+ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)
+ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)
+ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)
+
 size_t ZSTD_compressBlock_fast_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
@@ -361,30 +670,29 @@ size_t ZSTD_compressBlock_fast_dictMatchState(
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 4);
+        return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);
     case 5 :
-        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 5);
+        return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);
     case 6 :
-        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 6);
+        return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);
     case 7 :
-        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 7);
+        return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);
     }
 }
 
 
 static size_t ZSTD_compressBlock_fast_extDict_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-        void const* src, size_t srcSize, U32 const mls)
+        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
     U32 const hlog = cParams->hashLog;
     /* support stepSize of 0 */
-    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
+    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
     const BYTE* const base = ms->window.base;
     const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
@@ -397,100 +705,256 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
     U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
+
+    const BYTE* ip0 = istart;
+    const BYTE* ip1;
+    const BYTE* ip2;
+    const BYTE* ip3;
+    U32 current0;
+
+
+    size_t hash0; /* hash for ip0 */
+    size_t hash1; /* hash for ip1 */
+    U32 idx; /* match idx for ip0 */
+    const BYTE* idxBase; /* base pointer for idx */
+
+    U32 offcode;
+    const BYTE* match0;
+    size_t mLength;
+    const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */
+
+    size_t step;
+    const BYTE* nextStep;
+    const size_t kStepIncr = (1 << (kSearchStrength - 1));
+
+    (void)hasStep; /* not currently specialized on whether it's accelerated */
 
     DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);
 
     /* switch to "regular" variant if extDict is invalidated due to maxDistance */
     if (prefixStartIndex == dictStartIndex)
-        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, mls);
-
-    /* Search Loop */
-    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
-        const size_t h = ZSTD_hashPtr(ip, hlog, mls);
-        const U32    matchIndex = hashTable[h];
-        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
-        const BYTE*  match = matchBase + matchIndex;
-        const U32    curr = (U32)(ip-base);
-        const U32    repIndex = curr + 1 - offset_1;
-        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
-        const BYTE* const repMatch = repBase + repIndex;
-        hashTable[h] = curr;   /* update hash table */
-        DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);
-
-        if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */
-             & (offset_1 < curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */
-           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
-            size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;
-            ip++;
-            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, rLength-MINMATCH);
-            ip += rLength;
-            anchor = ip;
-        } else {
-            if ( (matchIndex < dictStartIndex) ||
-                 (MEM_read32(match) != MEM_read32(ip)) ) {
-                assert(stepSize >= 1);
-                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
-                continue;
+        return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
+
+    {   U32 const curr = (U32)(ip0 - base);
+        U32 const maxRep = curr - dictStartIndex;
+        if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
+    }
+
+    /* start each op */
+_start: /* Requires: ip0 */
+
+    step = stepSize;
+    nextStep = ip0 + kStepIncr;
+
+    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
+    ip1 = ip0 + 1;
+    ip2 = ip0 + step;
+    ip3 = ip2 + 1;
+
+    if (ip3 >= ilimit) {
+        goto _cleanup;
+    }
+
+    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
+    hash1 = ZSTD_hashPtr(ip1, hlog, mls);
+
+    idx = hashTable[hash0];
+    idxBase = idx < prefixStartIndex ? dictBase : base;
+
+    do {
+        {   /* load repcode match for ip[2] */
+            U32 const current2 = (U32)(ip2 - base);
+            U32 const repIndex = current2 - offset_1;
+            const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+            U32 rval;
+            if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
+                 & (offset_1 > 0) ) {
+                rval = MEM_read32(repBase + repIndex);
+            } else {
+                rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
             }
-            {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
-                const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
-                U32 const offset = curr - matchIndex;
-                size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
-                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
-                offset_2 = offset_1; offset_1 = offset;  /* update offset history */
-                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-                ip += mLength;
-                anchor = ip;
+
+            /* write back hash table entry */
+            current0 = (U32)(ip0 - base);
+            hashTable[hash0] = current0;
+
+            /* check repcode at ip[2] */
+            if (MEM_read32(ip2) == rval) {
+                ip0 = ip2;
+                match0 = repBase + repIndex;
+                matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+                assert((match0 != prefixStart) & (match0 != dictStart));
+                mLength = ip0[-1] == match0[-1];
+                ip0 -= mLength;
+                match0 -= mLength;
+                offcode = REPCODE1_TO_OFFBASE;
+                mLength += 4;
+                goto _match;
         }   }
 
-        if (ip <= ilimit) {
-            /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;
-            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
-            /* check immediate repcode */
-            while (ip <= ilimit) {
-                U32 const current2 = (U32)(ip-base);
-                U32 const repIndex2 = current2 - offset_2;
-                const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 < curr - dictStartIndex))  /* intentional overflow */
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
-                    { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, 0 /*offcode*/, repLength2-MINMATCH);
-                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
-                    ip += repLength2;
-                    anchor = ip;
-                    continue;
-                }
-                break;
-    }   }   }
+        {   /* load match for ip[0] */
+            U32 const mval = idx >= dictStartIndex ?
+                    MEM_read32(idxBase + idx) :
+                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */
+
+            /* check match at ip[0] */
+            if (MEM_read32(ip0) == mval) {
+                /* found a match! */
+                goto _offset;
+        }   }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+        idxBase = idx < prefixStartIndex ? dictBase : base;
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip3;
+
+        /* write back hash table entry */
+        current0 = (U32)(ip0 - base);
+        hashTable[hash0] = current0;
+
+        {   /* load match for ip[0] */
+            U32 const mval = idx >= dictStartIndex ?
+                    MEM_read32(idxBase + idx) :
+                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */
+
+            /* check match at ip[0] */
+            if (MEM_read32(ip0) == mval) {
+                /* found a match! */
+                goto _offset;
+        }   }
+
+        /* lookup ip[1] */
+        idx = hashTable[hash1];
+        idxBase = idx < prefixStartIndex ? dictBase : base;
+
+        /* hash ip[2] */
+        hash0 = hash1;
+        hash1 = ZSTD_hashPtr(ip2, hlog, mls);
+
+        /* advance to next positions */
+        ip0 = ip1;
+        ip1 = ip2;
+        ip2 = ip0 + step;
+        ip3 = ip1 + step;
+
+        /* calculate step */
+        if (ip2 >= nextStep) {
+            step++;
+            PREFETCH_L1(ip1 + 64);
+            PREFETCH_L1(ip1 + 128);
+            nextStep += kStepIncr;
+        }
+    } while (ip3 < ilimit);
+
+_cleanup:
+    /* Note that there are probably still a couple positions we could search.
+     * However, it seems to be a meaningful performance hit to try to search
+     * them. So let's not. */
+
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
 
     /* save reps for next block */
-    rep[0] = offset_1;
-    rep[1] = offset_2;
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
+
+_offset: /* Requires: ip0, idx, idxBase */
+
+    /* Compute the offset code. */
+    {   U32 const offset = current0 - idx;
+        const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
+        matchEnd = idx < prefixStartIndex ? dictEnd : iend;
+        match0 = idxBase + idx;
+        offset_2 = offset_1;
+        offset_1 = offset;
+        offcode = OFFSET_TO_OFFBASE(offset);
+        mLength = 4;
+
+        /* Count the backwards match length. */
+        while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
+            ip0--;
+            match0--;
+            mLength++;
+    }   }
+
+_match: /* Requires: ip0, match0, offcode, matchEnd */
+
+    /* Count the forward length. */
+    assert(matchEnd != 0);
+    mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);
+
+    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
+
+    ip0 += mLength;
+    anchor = ip0;
+
+    /* write next hash table entry */
+    if (ip1 < ip0) {
+        hashTable[hash1] = (U32)(ip1 - base);
+    }
+
+    /* Fill table and check for immediate repcode. */
+    if (ip0 <= ilimit) {
+        /* Fill Table */
+        assert(base+current0+2 > istart);  /* check base overflow */
+        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
+        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+        while (ip0 <= ilimit) {
+            U32 const repIndex2 = (U32)(ip0-base) - offset_2;
+            const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+            if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 > 0))  /* intentional underflow */
+                 && (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
+                const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
+                ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
+                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                ip0 += repLength2;
+                anchor = ip0;
+                continue;
+            }
+            break;
+    }   }
+
+    goto _start;
 }
 
+ZSTD_GEN_FAST_FN(extDict, 4, 0)
+ZSTD_GEN_FAST_FN(extDict, 5, 0)
+ZSTD_GEN_FAST_FN(extDict, 6, 0)
+ZSTD_GEN_FAST_FN(extDict, 7, 0)
 
 size_t ZSTD_compressBlock_fast_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
 {
     U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState == NULL);
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
+        return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);
     case 5 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
+        return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);
     case 6 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
+        return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);
     case 7 :
-        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
+        return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);
     }
 }
diff --git a/priv/zstd/lib/compress/zstd_fast.h b/priv/zstd/lib/compress/zstd_fast.h
index 0d4a0c1..9e4236b 100644
--- a/priv/zstd/lib/compress/zstd_fast.h
+++ b/priv/zstd/lib/compress/zstd_fast.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -19,7 +19,8 @@ extern "C" {
 #include "zstd_compress_internal.h"
 
 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
-                        void const* end, ZSTD_dictTableLoadMethod_e dtlm);
+                        void const* end, ZSTD_dictTableLoadMethod_e dtlm,
+                        ZSTD_tableFillPurpose_e tfp);
 size_t ZSTD_compressBlock_fast(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
diff --git a/priv/zstd/lib/compress/zstd_lazy.c b/priv/zstd/lib/compress/zstd_lazy.c
index 3d523e8..a247342 100644
--- a/priv/zstd/lib/compress/zstd_lazy.c
+++ b/priv/zstd/lib/compress/zstd_lazy.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -10,6 +10,7 @@
 
 #include "zstd_compress_internal.h"
 #include "zstd_lazy.h"
+#include "../common/bits.h" /* ZSTD_countTrailingZeros64 */
 
 
 /*-*************************************
@@ -61,7 +62,7 @@ ZSTD_updateDUBT(ZSTD_matchState_t* ms,
  *  assumption : curr >= btlow == (curr - btmask)
  *  doesn't fail */
 static void
-ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
+ZSTD_insertDUBT1(const ZSTD_matchState_t* ms,
                  U32 curr, const BYTE* inputEnd,
                  U32 nbCompares, U32 btLow,
                  const ZSTD_dictMode_e dictMode)
@@ -93,7 +94,7 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
     assert(curr >= btLow);
     assert(ip < iend);   /* condition for ZSTD_count */
 
-    while (nbCompares-- && (matchIndex > windowLow)) {
+    for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         assert(matchIndex < curr);
@@ -151,7 +152,7 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
 
 static size_t
 ZSTD_DUBT_findBetterDictMatch (
-        ZSTD_matchState_t* ms,
+        const ZSTD_matchState_t* ms,
         const BYTE* const ip, const BYTE* const iend,
         size_t* offsetPtr,
         size_t bestLength,
@@ -185,7 +186,7 @@ ZSTD_DUBT_findBetterDictMatch (
     (void)dictMode;
     assert(dictMode == ZSTD_dictMatchState);
 
-    while (nbCompares-- && (dictMatchIndex > dictLowLimit)) {
+    for (; nbCompares && (dictMatchIndex > dictLowLimit); --nbCompares) {
         U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         const BYTE* match = dictBase + dictMatchIndex;
@@ -197,8 +198,8 @@ ZSTD_DUBT_findBetterDictMatch (
             U32 matchIndex = dictMatchIndex + dictIndexDelta;
             if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
                 DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
-                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, ZSTD_REP_MOVE + curr - matchIndex, dictMatchIndex, matchIndex);
-                bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + curr - matchIndex;
+                    curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, OFFSET_TO_OFFBASE(curr - matchIndex), dictMatchIndex, matchIndex);
+                bestLength = matchLength, *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
             }
             if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
                 break;   /* drop, to guarantee consistency (miss a little bit of compression) */
@@ -218,7 +219,7 @@ ZSTD_DUBT_findBetterDictMatch (
     }
 
     if (bestLength >= MINMATCH) {
-        U32 const mIndex = curr - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+        U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offsetPtr); (void)mIndex;
         DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
                     curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
     }
@@ -230,7 +231,7 @@ ZSTD_DUBT_findBetterDictMatch (
 static size_t
 ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iend,
-                        size_t* offsetPtr,
+                        size_t* offBasePtr,
                         U32 const mls,
                         const ZSTD_dictMode_e dictMode)
 {
@@ -309,7 +310,7 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
         matchIndex  = hashTable[h];
         hashTable[h] = curr;   /* Update Hash Table */
 
-        while (nbCompares-- && (matchIndex > windowLow)) {
+        for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
             U32* const nextPtr = bt + 2*(matchIndex & btMask);
             size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
             const BYTE* match;
@@ -327,8 +328,8 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
             if (matchLength > bestLength) {
                 if (matchLength > matchEndIdx - matchIndex)
                     matchEndIdx = matchIndex + (U32)matchLength;
-                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
-                    bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + curr - matchIndex;
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr - matchIndex + 1) - ZSTD_highbit32((U32)*offBasePtr)) )
+                    bestLength = matchLength, *offBasePtr = OFFSET_TO_OFFBASE(curr - matchIndex);
                 if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
                     if (dictMode == ZSTD_dictMatchState) {
                         nbCompares = 0; /* in addition to avoiding checking any
@@ -357,19 +358,20 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
 
         *smallerPtr = *largerPtr = 0;
 
+        assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
         if (dictMode == ZSTD_dictMatchState && nbCompares) {
             bestLength = ZSTD_DUBT_findBetterDictMatch(
                     ms, ip, iend,
-                    offsetPtr, bestLength, nbCompares,
+                    offBasePtr, bestLength, nbCompares,
                     mls, dictMode);
         }
 
         assert(matchEndIdx > curr+8); /* ensure nextToUpdate is increased */
         ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
         if (bestLength >= MINMATCH) {
-            U32 const mIndex = curr - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+            U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offBasePtr); (void)mIndex;
             DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
-                        curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
+                        curr, (U32)bestLength, (U32)*offBasePtr, mIndex);
         }
         return bestLength;
     }
@@ -380,62 +382,14 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
 FORCE_INLINE_TEMPLATE size_t
 ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iLimit,
-                      size_t* offsetPtr,
+                      size_t* offBasePtr,
                 const U32 mls /* template */,
                 const ZSTD_dictMode_e dictMode)
 {
     DEBUGLOG(7, "ZSTD_BtFindBestMatch");
     if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
     ZSTD_updateDUBT(ms, ip, iLimit, mls);
-    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offsetPtr, mls, dictMode);
-}
-
-
-static size_t
-ZSTD_BtFindBestMatch_selectMLS (  ZSTD_matchState_t* ms,
-                            const BYTE* ip, const BYTE* const iLimit,
-                                  size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
-    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
-    case 7 :
-    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
-    }
-}
-
-
-static size_t ZSTD_BtFindBestMatch_dictMatchState_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
-    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
-    case 7 :
-    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
-    }
-}
-
-
-static size_t ZSTD_BtFindBestMatch_extDict_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
-    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
-    case 7 :
-    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
-    }
+    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offBasePtr, mls, dictMode);
 }
 
 /***********************************
@@ -450,7 +404,7 @@ void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const B
     U32* const chainTable = ms->chainTable;
     U32 const chainSize = 1 << ms->cParams.chainLog;
     U32 idx = ms->nextToUpdate;
-    U32 const minChain = chainSize < target ? target - chainSize : idx;
+    U32 const minChain = chainSize < target - idx ? target - chainSize : idx;
     U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG;
     U32 const cacheSize = bucketSize - 1;
     U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize;
@@ -464,7 +418,7 @@ void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_matchState_t* ms, const B
     U32 const hashLog = ms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
     U32* const tmpHashTable = hashTable;
     U32* const tmpChainTable = hashTable + ((size_t)1 << hashLog);
-    U32 const tmpChainSize = ((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog;
+    U32 const tmpChainSize = (U32)((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog;
     U32 const tmpMinChain = tmpChainSize < target ? target - tmpChainSize : idx;
     U32 hashIdx;
 
@@ -608,7 +562,7 @@ size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nb
         /* save best solution */
         if (currentMl > ml) {
             ml = currentMl;
-            *offsetPtr = curr - (matchIndex + ddsIndexDelta) + ZSTD_REP_MOVE;
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
             if (ip+currentMl == iLimit) {
                 /* best possible, avoids read overflow on next attempt */
                 return ml;
@@ -645,7 +599,7 @@ size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nb
             /* save best solution */
             if (currentMl > ml) {
                 ml = currentMl;
-                *offsetPtr = curr - (matchIndex + ddsIndexDelta) + ZSTD_REP_MOVE;
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
         }
@@ -692,7 +646,7 @@ U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
 
 /* inlining is important to hardwire a hot branch (template emulation) */
 FORCE_INLINE_TEMPLATE
-size_t ZSTD_HcFindBestMatch_generic (
+size_t ZSTD_HcFindBestMatch(
                         ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iLimit,
                         size_t* offsetPtr,
@@ -738,7 +692,8 @@ size_t ZSTD_HcFindBestMatch_generic (
         if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
             const BYTE* const match = base + matchIndex;
             assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
-            if (match[ml] == ip[ml])   /* potentially better */
+            /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+            if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
                 currentMl = ZSTD_count(ip, match, iLimit);
         } else {
             const BYTE* const match = dictBase + matchIndex;
@@ -750,7 +705,7 @@ size_t ZSTD_HcFindBestMatch_generic (
         /* save best solution */
         if (currentMl > ml) {
             ml = currentMl;
-            *offsetPtr = curr - matchIndex + ZSTD_REP_MOVE;
+            *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
             if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
         }
 
@@ -758,6 +713,7 @@ size_t ZSTD_HcFindBestMatch_generic (
         matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
     }
 
+    assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
     if (dictMode == ZSTD_dedicatedDictSearch) {
         ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts, dms,
                                                   ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
@@ -784,7 +740,8 @@ size_t ZSTD_HcFindBestMatch_generic (
             /* save best solution */
             if (currentMl > ml) {
                 ml = currentMl;
-                *offsetPtr = curr - (matchIndex + dmsIndexDelta) + ZSTD_REP_MOVE;
+                assert(curr > matchIndex + dmsIndexDelta);
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
 
@@ -797,310 +754,49 @@ size_t ZSTD_HcFindBestMatch_generic (
     return ml;
 }
 
-
-FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
-    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
-    case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
-    }
-}
-
-
-static size_t ZSTD_HcFindBestMatch_dictMatchState_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
-    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
-    case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
-    }
-}
-
-
-static size_t ZSTD_HcFindBestMatch_dedicatedDictSearch_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dedicatedDictSearch);
-    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dedicatedDictSearch);
-    case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dedicatedDictSearch);
-    }
-}
-
-
-FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
-    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
-    case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
-    }
-}
-
 /* *********************************
 * (SIMD) Row-based matchfinder
 ***********************************/
 /* Constants for row-based hash */
-#define ZSTD_ROW_HASH_TAG_OFFSET 1                               /* byte offset of hashes in the match state's tagTable from the beginning of a row */
-#define ZSTD_ROW_HASH_TAG_BITS 8                                 /* nb bits to use for the tag */
+#define ZSTD_ROW_HASH_TAG_OFFSET 16     /* byte offset of hashes in the match state's tagTable from the beginning of a row */
 #define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1)
+#define ZSTD_ROW_HASH_MAX_ENTRIES 64    /* absolute maximum number of entries per row, for all configurations */
 
 #define ZSTD_ROW_HASH_CACHE_MASK (ZSTD_ROW_HASH_CACHE_SIZE - 1)
 
-typedef U32 ZSTD_VecMask;   /* Clarifies when we are interacting with a U32 representing a mask of matches */
-
-#if !defined(ZSTD_NO_INTRINSICS) && defined(__SSE2__) /* SIMD SSE version */
-
-#include 
-typedef __m128i ZSTD_Vec128;
-
-/* Returns a 128-bit container with 128-bits from src */
-static ZSTD_Vec128 ZSTD_Vec128_read(const void* const src) {
-  return _mm_loadu_si128((ZSTD_Vec128 const*)src);
-}
-
-/* Returns a ZSTD_Vec128 with the byte "val" packed 16 times */
-static ZSTD_Vec128 ZSTD_Vec128_set8(BYTE val) {
-  return _mm_set1_epi8((char)val);
-}
-
-/* Do byte-by-byte comparison result of x and y. Then collapse 128-bit resultant mask
- * into a 32-bit mask that is the MSB of each byte.
- * */
-static ZSTD_VecMask ZSTD_Vec128_cmpMask8(ZSTD_Vec128 x, ZSTD_Vec128 y) {
-  return (ZSTD_VecMask)_mm_movemask_epi8(_mm_cmpeq_epi8(x, y));
-}
-
-typedef struct {
-  __m128i fst;
-  __m128i snd;
-} ZSTD_Vec256;
-
-static ZSTD_Vec256 ZSTD_Vec256_read(const void* const ptr) {
-  ZSTD_Vec256 v;
-  v.fst = ZSTD_Vec128_read(ptr);
-  v.snd = ZSTD_Vec128_read((ZSTD_Vec128 const*)ptr + 1);
-  return v;
-}
-
-static ZSTD_Vec256 ZSTD_Vec256_set8(BYTE val) {
-  ZSTD_Vec256 v;
-  v.fst = ZSTD_Vec128_set8(val);
-  v.snd = ZSTD_Vec128_set8(val);
-  return v;
-}
-
-static ZSTD_VecMask ZSTD_Vec256_cmpMask8(ZSTD_Vec256 x, ZSTD_Vec256 y) {
-  ZSTD_VecMask fstMask;
-  ZSTD_VecMask sndMask;
-  fstMask = ZSTD_Vec128_cmpMask8(x.fst, y.fst);
-  sndMask = ZSTD_Vec128_cmpMask8(x.snd, y.snd);
-  return fstMask | (sndMask << 16);
-}
-
-#elif !defined(ZSTD_NO_INTRINSICS) && defined(__ARM_NEON) /* SIMD ARM NEON Version */
-
-#include 
-typedef uint8x16_t ZSTD_Vec128;
-
-static ZSTD_Vec128 ZSTD_Vec128_read(const void* const src) {
-  return vld1q_u8((const BYTE* const)src);
-}
-
-static ZSTD_Vec128 ZSTD_Vec128_set8(BYTE val) {
-  return vdupq_n_u8(val);
-}
-
-/* Mimics '_mm_movemask_epi8()' from SSE */
-static U32 ZSTD_vmovmaskq_u8(ZSTD_Vec128 val) {
-    /* Shift out everything but the MSB bits in each byte */
-    uint16x8_t highBits = vreinterpretq_u16_u8(vshrq_n_u8(val, 7));
-    /* Merge the even lanes together with vsra (right shift and add) */
-    uint32x4_t paired16 = vreinterpretq_u32_u16(vsraq_n_u16(highBits, highBits, 7));
-    uint64x2_t paired32 = vreinterpretq_u64_u32(vsraq_n_u32(paired16, paired16, 14));
-    uint8x16_t paired64 = vreinterpretq_u8_u64(vsraq_n_u64(paired32, paired32, 28));
-    /* Extract the low 8 bits from each lane, merge */
-    return vgetq_lane_u8(paired64, 0) | ((U32)vgetq_lane_u8(paired64, 8) << 8);
-}
-
-static ZSTD_VecMask ZSTD_Vec128_cmpMask8(ZSTD_Vec128 x, ZSTD_Vec128 y) {
-  return (ZSTD_VecMask)ZSTD_vmovmaskq_u8(vceqq_u8(x, y));
-}
-
-typedef struct {
-    uint8x16_t fst;
-    uint8x16_t snd;
-} ZSTD_Vec256;
-
-static ZSTD_Vec256 ZSTD_Vec256_read(const void* const ptr) {
-  ZSTD_Vec256 v;
-  v.fst = ZSTD_Vec128_read(ptr);
-  v.snd = ZSTD_Vec128_read((ZSTD_Vec128 const*)ptr + 1);
-  return v;
-}
-
-static ZSTD_Vec256 ZSTD_Vec256_set8(BYTE val) {
-  ZSTD_Vec256 v;
-  v.fst = ZSTD_Vec128_set8(val);
-  v.snd = ZSTD_Vec128_set8(val);
-  return v;
-}
-
-static ZSTD_VecMask ZSTD_Vec256_cmpMask8(ZSTD_Vec256 x, ZSTD_Vec256 y) {
-  ZSTD_VecMask fstMask;
-  ZSTD_VecMask sndMask;
-  fstMask = ZSTD_Vec128_cmpMask8(x.fst, y.fst);
-  sndMask = ZSTD_Vec128_cmpMask8(x.snd, y.snd);
-  return fstMask | (sndMask << 16);
-}
-
-#else /* Scalar fallback version */
-
-#define VEC128_NB_SIZE_T (16 / sizeof(size_t))
-typedef struct {
-    size_t vec[VEC128_NB_SIZE_T];
-} ZSTD_Vec128;
-
-static ZSTD_Vec128 ZSTD_Vec128_read(const void* const src) {
-    ZSTD_Vec128 ret;
-    ZSTD_memcpy(ret.vec, src, VEC128_NB_SIZE_T*sizeof(size_t));
-    return ret;
-}
-
-static ZSTD_Vec128 ZSTD_Vec128_set8(BYTE val) {
-    ZSTD_Vec128 ret = { {0} };
-    int startBit = sizeof(size_t) * 8 - 8;
-    for (;startBit >= 0; startBit -= 8) {
-        unsigned j = 0;
-        for (;j < VEC128_NB_SIZE_T; ++j) {
-            ret.vec[j] |= ((size_t)val << startBit);
-        }
-    }
-    return ret;
-}
-
-/* Compare x to y, byte by byte, generating a "matches" bitfield */
-static ZSTD_VecMask ZSTD_Vec128_cmpMask8(ZSTD_Vec128 x, ZSTD_Vec128 y) {
-    ZSTD_VecMask res = 0;
-    unsigned i = 0;
-    unsigned l = 0;
-    for (; i < VEC128_NB_SIZE_T; ++i) {
-        const size_t cmp1 = x.vec[i];
-        const size_t cmp2 = y.vec[i];
-        unsigned j = 0;
-        for (; j < sizeof(size_t); ++j, ++l) {
-            if (((cmp1 >> j*8) & 0xFF) == ((cmp2 >> j*8) & 0xFF)) {
-                res |= ((U32)1 << (j+i*sizeof(size_t)));
-            }
-        }
-    }
-    return res;
-}
-
-#define VEC256_NB_SIZE_T 2*VEC128_NB_SIZE_T
-typedef struct {
-    size_t vec[VEC256_NB_SIZE_T];
-} ZSTD_Vec256;
-
-static ZSTD_Vec256 ZSTD_Vec256_read(const void* const src) {
-    ZSTD_Vec256 ret;
-    ZSTD_memcpy(ret.vec, src, VEC256_NB_SIZE_T*sizeof(size_t));
-    return ret;
-}
-
-static ZSTD_Vec256 ZSTD_Vec256_set8(BYTE val) {
-    ZSTD_Vec256 ret = { {0} };
-    int startBit = sizeof(size_t) * 8 - 8;
-    for (;startBit >= 0; startBit -= 8) {
-        unsigned j = 0;
-        for (;j < VEC256_NB_SIZE_T; ++j) {
-            ret.vec[j] |= ((size_t)val << startBit);
-        }
-    }
-    return ret;
-}
-
-/* Compare x to y, byte by byte, generating a "matches" bitfield */
-static ZSTD_VecMask ZSTD_Vec256_cmpMask8(ZSTD_Vec256 x, ZSTD_Vec256 y) {
-    ZSTD_VecMask res = 0;
-    unsigned i = 0;
-    unsigned l = 0;
-    for (; i < VEC256_NB_SIZE_T; ++i) {
-        const size_t cmp1 = x.vec[i];
-        const size_t cmp2 = y.vec[i];
-        unsigned j = 0;
-        for (; j < sizeof(size_t); ++j, ++l) {
-            if (((cmp1 >> j*8) & 0xFF) == ((cmp2 >> j*8) & 0xFF)) {
-                res |= ((U32)1 << (j+i*sizeof(size_t)));
-            }
-        }
-    }
-    return res;
-}
-
-#endif /* !defined(ZSTD_NO_INTRINSICS) && defined(__SSE2__) */
+typedef U64 ZSTD_VecMask;   /* Clarifies when we are interacting with a U64 representing a mask of matches */
 
 /* ZSTD_VecMask_next():
  * Starting from the LSB, returns the idx of the next non-zero bit.
  * Basically counting the nb of trailing zeroes.
  */
-static U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
-#   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    return _BitScanForward(&r, val) ? (U32)r : 0;
-#   elif defined(__GNUC__) && (__GNUC__ >= 3)
-    return (U32)__builtin_ctz(val);
-#   else
-    /* Software ctz version: http://graphics.stanford.edu/~seander/bithacks.html#ZerosOnRightMultLookup */
-    static const U32 multiplyDeBruijnBitPosition[32] =
-    {
-        0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
-		31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9
-    };
-	return multiplyDeBruijnBitPosition[((U32)((v & -(int)v) * 0x077CB531U)) >> 27];
-#   endif
+MEM_STATIC U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
+    return ZSTD_countTrailingZeros64(val);
 }
 
-/* ZSTD_VecMask_rotateRight():
- * Rotates a bitfield to the right by "rotation" bits.
- * If the rotation is greater than totalBits, the returned mask is 0.
+/* ZSTD_rotateRight_*():
+ * Rotates a bitfield to the right by "count" bits.
+ * https://en.wikipedia.org/w/index.php?title=Circular_shift&oldid=991635599#Implementing_circular_shifts
  */
-FORCE_INLINE_TEMPLATE ZSTD_VecMask
-ZSTD_VecMask_rotateRight(ZSTD_VecMask mask, U32 const rotation, U32 const totalBits) {
-  if (rotation == 0)
-    return mask;
-  switch (totalBits) {
-    default:
-      assert(0);
-    case 16:
-      return (mask >> rotation) | (U16)(mask << (16 - rotation));
-    case 32:
-      return (mask >> rotation) | (U32)(mask << (32 - rotation));
-  }
+FORCE_INLINE_TEMPLATE
+U64 ZSTD_rotateRight_U64(U64 const value, U32 count) {
+    assert(count < 64);
+    count &= 0x3F; /* for fickle pattern recognition */
+    return (value >> count) | (U64)(value << ((0U - count) & 0x3F));
+}
+
+FORCE_INLINE_TEMPLATE
+U32 ZSTD_rotateRight_U32(U32 const value, U32 count) {
+    assert(count < 32);
+    count &= 0x1F; /* for fickle pattern recognition */
+    return (value >> count) | (U32)(value << ((0U - count) & 0x1F));
+}
+
+FORCE_INLINE_TEMPLATE
+U16 ZSTD_rotateRight_U16(U16 const value, U32 count) {
+    assert(count < 16);
+    count &= 0x0F; /* for fickle pattern recognition */
+    return (value >> count) | (U16)(value << ((0U - count) & 0x0F));
 }
 
 /* ZSTD_row_nextIndex():
@@ -1126,20 +822,24 @@ MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) {
  */
 FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, U16 const* tagTable, U32 const relRow, U32 const rowLog) {
     PREFETCH_L1(hashTable + relRow);
-    if (rowLog == 5) {
+    if (rowLog >= 5) {
         PREFETCH_L1(hashTable + relRow + 16);
+        /* Note: prefetching more of the hash table does not appear to be beneficial for 128-entry rows */
     }
     PREFETCH_L1(tagTable + relRow);
-    assert(rowLog == 4 || rowLog == 5);
+    if (rowLog == 6) {
+        PREFETCH_L1(tagTable + relRow + 32);
+    }
+    assert(rowLog == 4 || rowLog == 5 || rowLog == 6);
     assert(ZSTD_isAligned(hashTable + relRow, 64));                 /* prefetched hash row always 64-byte aligned */
-    assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on a multiple of 32 or 64 bytes */
+    assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on correct multiple of bytes (32,64,128) */
 }
 
 /* ZSTD_row_fillHashCache():
  * Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries,
  * but not beyond iLimit.
  */
-static void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
+FORCE_INLINE_TEMPLATE void ZSTD_row_fillHashCache(ZSTD_matchState_t* ms, const BYTE* base,
                                    U32 const rowLog, U32 const mls,
                                    U32 idx, const BYTE* const iLimit)
 {
@@ -1179,35 +879,65 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTab
     }
 }
 
-/* ZSTD_row_update_internal():
- * Inserts the byte at ip into the appropriate position in the hash table.
- * Determines the relative row, and the position within the {16, 32} entry row to insert at.
+/* ZSTD_row_update_internalImpl():
+ * Updates the hash table with positions starting from updateStartIdx until updateEndIdx.
  */
-FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
-                                                    U32 const mls, U32 const rowLog,
-                                                    U32 const rowMask, U32 const useCache)
+FORCE_INLINE_TEMPLATE void ZSTD_row_update_internalImpl(ZSTD_matchState_t* ms,
+                                                        U32 updateStartIdx, U32 const updateEndIdx,
+                                                        U32 const mls, U32 const rowLog,
+                                                        U32 const rowMask, U32 const useCache)
 {
     U32* const hashTable = ms->hashTable;
     U16* const tagTable = ms->tagTable;
     U32 const hashLog = ms->rowHashLog;
     const BYTE* const base = ms->window.base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = ms->nextToUpdate;
 
-    DEBUGLOG(6, "ZSTD_row_update_internal(): nextToUpdate=%u, current=%u", idx, target);
-    for (; idx < target; ++idx) {
-        U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, idx, hashLog, rowLog, mls)
-                                  : (U32)ZSTD_hashPtr(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
+    DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx);
+    for (; updateStartIdx < updateEndIdx; ++updateStartIdx) {
+        U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls)
+                                  : (U32)ZSTD_hashPtr(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
         U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
         U32* const row = hashTable + relRow;
         BYTE* tagRow = (BYTE*)(tagTable + relRow);  /* Though tagTable is laid out as a table of U16, each tag is only 1 byte.
                                                        Explicit cast allows us to get exact desired position within each row */
         U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
 
-        assert(hash == ZSTD_hashPtr(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls));
+        assert(hash == ZSTD_hashPtr(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls));
         ((BYTE*)tagRow)[pos + ZSTD_ROW_HASH_TAG_OFFSET] = hash & ZSTD_ROW_HASH_TAG_MASK;
-        row[pos] = idx;
+        row[pos] = updateStartIdx;
     }
+}
+
+/* ZSTD_row_update_internal():
+ * Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate.
+ * Skips sections of long matches as is necessary.
+ */
+FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const BYTE* ip,
+                                                    U32 const mls, U32 const rowLog,
+                                                    U32 const rowMask, U32 const useCache)
+{
+    U32 idx = ms->nextToUpdate;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    const U32 kSkipThreshold = 384;
+    const U32 kMaxMatchStartPositionsToUpdate = 96;
+    const U32 kMaxMatchEndPositionsToUpdate = 32;
+
+    if (useCache) {
+        /* Only skip positions when using hash cache, i.e.
+         * if we are loading a dict, don't skip anything.
+         * If we decide to skip, then we only update a set number
+         * of positions at the beginning and end of the match.
+         */
+        if (UNLIKELY(target - idx > kSkipThreshold)) {
+            U32 const bound = idx + kMaxMatchStartPositionsToUpdate;
+            ZSTD_row_update_internalImpl(ms, idx, bound, mls, rowLog, rowMask, useCache);
+            idx = target - kMaxMatchEndPositionsToUpdate;
+            ZSTD_row_fillHashCache(ms, base, rowLog, mls, idx, ip+1);
+        }
+    }
+    assert(target >= idx);
+    ZSTD_row_update_internalImpl(ms, idx, target, mls, rowLog, rowMask, useCache);
     ms->nextToUpdate = target;
 }
 
@@ -1216,34 +946,178 @@ FORCE_INLINE_TEMPLATE void ZSTD_row_update_internal(ZSTD_matchState_t* ms, const
  * processing.
  */
 void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip) {
-    const U32 rowLog = ms->cParams.searchLog < 5 ? 4 : 5;
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
     const U32 rowMask = (1u << rowLog) - 1;
     const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */);
 
     DEBUGLOG(5, "ZSTD_row_update(), rowLog=%u", rowLog);
-    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* dont use cache */);
+    ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* don't use cache */);
 }
 
-/* Returns a ZSTD_VecMask (U32) that has the nth bit set to 1 if the newly-computed "tag" matches
- * the hash at the nth position in a row of the tagTable.
+/* Returns the mask width of bits group of which will be set to 1. Given not all
+ * architectures have easy movemask instruction, this helps to iterate over
+ * groups of bits easier and faster.
  */
-FORCE_INLINE_TEMPLATE
-ZSTD_VecMask ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 head, const U32 rowEntries) {
-    ZSTD_VecMask matches = 0;
+FORCE_INLINE_TEMPLATE U32
+ZSTD_row_matchMaskGroupWidth(const U32 rowEntries)
+{
+    assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
+    assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
+    (void)rowEntries;
+#if defined(ZSTD_ARCH_ARM_NEON)
+    /* NEON path only works for little endian */
+    if (!MEM_isLittleEndian()) {
+        return 1;
+    }
     if (rowEntries == 16) {
-        ZSTD_Vec128 hashes        = ZSTD_Vec128_read(tagRow + ZSTD_ROW_HASH_TAG_OFFSET);
-        ZSTD_Vec128 expandedTags  = ZSTD_Vec128_set8(tag);
-        matches                   = ZSTD_Vec128_cmpMask8(hashes, expandedTags);
+        return 4;
+    }
+    if (rowEntries == 32) {
+        return 2;
+    }
+    if (rowEntries == 64) {
+        return 1;
+    }
+#endif
+    return 1;
+}
+
+#if defined(ZSTD_ARCH_X86_SSE2)
+FORCE_INLINE_TEMPLATE ZSTD_VecMask
+ZSTD_row_getSSEMask(int nbChunks, const BYTE* const src, const BYTE tag, const U32 head)
+{
+    const __m128i comparisonMask = _mm_set1_epi8((char)tag);
+    int matches[4] = {0};
+    int i;
+    assert(nbChunks == 1 || nbChunks == 2 || nbChunks == 4);
+    for (i=0; i> chunkSize;
+            do {
+                size_t chunk = MEM_readST(&src[i]);
+                chunk ^= splatChar;
+                chunk = (((chunk | x80) - x01) | chunk) & x80;
+                matches <<= chunkSize;
+                matches |= (chunk * extractMagic) >> shiftAmount;
+                i -= chunkSize;
+            } while (i >= 0);
+        } else { /* big endian: reverse bits during extraction */
+            const size_t msb = xFF ^ (xFF >> 1);
+            const size_t extractMagic = (msb / 0x1FF) | msb;
+            do {
+                size_t chunk = MEM_readST(&src[i]);
+                chunk ^= splatChar;
+                chunk = (((chunk | x80) - x01) | chunk) & x80;
+                matches <<= chunkSize;
+                matches |= ((chunk >> 7) * extractMagic) >> shiftAmount;
+                i -= chunkSize;
+            } while (i >= 0);
+        }
+        matches = ~matches;
+        if (rowEntries == 16) {
+            return ZSTD_rotateRight_U16((U16)matches, headGrouped);
+        } else if (rowEntries == 32) {
+            return ZSTD_rotateRight_U32((U32)matches, headGrouped);
+        } else {
+            return ZSTD_rotateRight_U64((U64)matches, headGrouped);
+        }
+    }
+#endif
 }
 
 /* The high-level approach of the SIMD row based match finder is as follows:
@@ -1262,7 +1136,7 @@ ZSTD_VecMask ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, con
  * - Pick the longest match.
  */
 FORCE_INLINE_TEMPLATE
-size_t ZSTD_RowFindBestMatch_generic (
+size_t ZSTD_RowFindBestMatch(
                         ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iLimit,
                         size_t* offsetPtr,
@@ -1288,16 +1162,19 @@ size_t ZSTD_RowFindBestMatch_generic (
     const U32 rowEntries = (1U << rowLog);
     const U32 rowMask = rowEntries - 1;
     const U32 cappedSearchLog = MIN(cParams->searchLog, rowLog); /* nb of searches is capped at nb entries per row */
+    const U32 groupWidth = ZSTD_row_matchMaskGroupWidth(rowEntries);
     U32 nbAttempts = 1U << cappedSearchLog;
     size_t ml=4-1;
 
     /* DMS/DDS variables that may be referenced laster */
     const ZSTD_matchState_t* const dms = ms->dictMatchState;
-    size_t ddsIdx;
-    U32 ddsExtraAttempts; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
-    U32 dmsTag;
-    U32* dmsRow;
-    BYTE* dmsTagRow;
+
+    /* Initialize the following variables to satisfy static analyzer */
+    size_t ddsIdx = 0;
+    U32 ddsExtraAttempts = 0; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
+    U32 dmsTag = 0;
+    U32* dmsRow = NULL;
+    BYTE* dmsTagRow = NULL;
 
     if (dictMode == ZSTD_dedicatedDictSearch) {
         const U32 ddsHashLog = dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
@@ -1328,15 +1205,15 @@ size_t ZSTD_RowFindBestMatch_generic (
         U32 const tag = hash & ZSTD_ROW_HASH_TAG_MASK;
         U32* const row = hashTable + relRow;
         BYTE* tagRow = (BYTE*)(tagTable + relRow);
-        U32 const head = *tagRow & rowMask;
-        U32 matchBuffer[32 /* maximum nb entries per row */];
+        U32 const headGrouped = (*tagRow & rowMask) * groupWidth;
+        U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
         size_t numMatches = 0;
         size_t currMatch = 0;
-        ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, head, rowEntries);
+        ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, headGrouped, rowEntries);
 
         /* Cycle through the matches and prefetch */
         for (; (matches > 0) && (nbAttempts > 0); --nbAttempts, matches &= (matches - 1)) {
-            U32 const matchPos = (head + ZSTD_VecMask_next(matches)) & rowMask;
+            U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
             U32 const matchIndex = row[matchPos];
             assert(numMatches < rowEntries);
             if (matchIndex < lowLimit)
@@ -1367,7 +1244,8 @@ size_t ZSTD_RowFindBestMatch_generic (
             if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
                 const BYTE* const match = base + matchIndex;
                 assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
-                if (match[ml] == ip[ml])   /* potentially better */
+                /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
+                if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
                     currentMl = ZSTD_count(ip, match, iLimit);
             } else {
                 const BYTE* const match = dictBase + matchIndex;
@@ -1379,12 +1257,13 @@ size_t ZSTD_RowFindBestMatch_generic (
             /* Save best solution */
             if (currentMl > ml) {
                 ml = currentMl;
-                *offsetPtr = curr - matchIndex + ZSTD_REP_MOVE;
+                *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
             }
         }
     }
 
+    assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
     if (dictMode == ZSTD_dedicatedDictSearch) {
         ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts + ddsExtraAttempts, dms,
                                                   ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
@@ -1396,14 +1275,14 @@ size_t ZSTD_RowFindBestMatch_generic (
         const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
         const U32 dmsIndexDelta        = dictLimit - dmsSize;
 
-        {   U32 const head = *dmsTagRow & rowMask;
-            U32 matchBuffer[32 /* maximum nb row entries */];
+        {   U32 const headGrouped = (*dmsTagRow & rowMask) * groupWidth;
+            U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
             size_t numMatches = 0;
             size_t currMatch = 0;
-            ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, head, rowEntries);
+            ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, headGrouped, rowEntries);
 
             for (; (matches > 0) && (nbAttempts > 0); --nbAttempts, matches &= (matches - 1)) {
-                U32 const matchPos = (head + ZSTD_VecMask_next(matches)) & rowMask;
+                U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
                 U32 const matchIndex = dmsRow[matchPos];
                 if (matchIndex < dmsLowestIndex)
                     break;
@@ -1426,7 +1305,8 @@ size_t ZSTD_RowFindBestMatch_generic (
 
                 if (currentMl > ml) {
                     ml = currentMl;
-                    *offsetPtr = curr - (matchIndex + dmsIndexDelta) + ZSTD_REP_MOVE;
+                    assert(curr > matchIndex + dmsIndexDelta);
+                    *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
                     if (ip+currentMl == iLimit) break;
                 }
             }
@@ -1435,83 +1315,183 @@ size_t ZSTD_RowFindBestMatch_generic (
     return ml;
 }
 
-/* Inlining is important to hardwire a hot branch (template emulation) */
-FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_selectMLS (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        const ZSTD_dictMode_e dictMode, size_t* offsetPtr, const U32 rowLog)
-{
-    switch(ms->cParams.minMatch)
-    {
-    default : /* includes case 3 */
-    case 4 : return ZSTD_RowFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, dictMode, rowLog);
-    case 5 : return ZSTD_RowFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, dictMode, rowLog);
-    case 7 :
-    case 6 : return ZSTD_RowFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, dictMode, rowLog);
-    }
-}
 
-FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_selectRowLog (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
-    switch(cappedSearchLog)
-    {
-    default :
-    case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_noDict, offsetPtr, 4);
-    case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_noDict, offsetPtr, 5);
-    }
-}
+/**
+ * Generate search functions templated on (dictMode, mls, rowLog).
+ * These functions are outlined for code size & compilation time.
+ * ZSTD_searchMax() dispatches to the correct implementation function.
+ *
+ * TODO: The start of the search function involves loading and calculating a
+ * bunch of constants from the ZSTD_matchState_t. These computations could be
+ * done in an initialization function, and saved somewhere in the match state.
+ * Then we could pass a pointer to the saved state instead of the match state,
+ * and avoid duplicate computations.
+ *
+ * TODO: Move the match re-winding into searchMax. This improves compression
+ * ratio, and unlocks further simplifications with the next TODO.
+ *
+ * TODO: Try moving the repcode search into searchMax. After the re-winding
+ * and repcode search are in searchMax, there is no more logic in the match
+ * finder loop that requires knowledge about the dictMode. So we should be
+ * able to avoid force inlining it, and we can join the extDict loop with
+ * the single segment loop. It should go in searchMax instead of its own
+ * function to avoid having multiple virtual function calls per search.
+ */
 
-FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_dictMatchState_selectRowLog(
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
-    switch(cappedSearchLog)
-    {
-    default :
-    case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dictMatchState, offsetPtr, 4);
-    case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dictMatchState, offsetPtr, 5);
-    }
-}
+#define ZSTD_BT_SEARCH_FN(dictMode, mls) ZSTD_BtFindBestMatch_##dictMode##_##mls
+#define ZSTD_HC_SEARCH_FN(dictMode, mls) ZSTD_HcFindBestMatch_##dictMode##_##mls
+#define ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog) ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog
+
+#define ZSTD_SEARCH_FN_ATTRS FORCE_NOINLINE
+
+#define GEN_ZSTD_BT_SEARCH_FN(dictMode, mls)                                           \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_BT_SEARCH_FN(dictMode, mls)(                      \
+            ZSTD_matchState_t* ms,                                                     \
+            const BYTE* ip, const BYTE* const iLimit,                                  \
+            size_t* offBasePtr)                                                        \
+    {                                                                                  \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                           \
+        return ZSTD_BtFindBestMatch(ms, ip, iLimit, offBasePtr, mls, ZSTD_##dictMode); \
+    }                                                                                  \
+
+#define GEN_ZSTD_HC_SEARCH_FN(dictMode, mls)                                          \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_HC_SEARCH_FN(dictMode, mls)(                     \
+            ZSTD_matchState_t* ms,                                                    \
+            const BYTE* ip, const BYTE* const iLimit,                                 \
+            size_t* offsetPtr)                                                        \
+    {                                                                                 \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                          \
+        return ZSTD_HcFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \
+    }                                                                                 \
+
+#define GEN_ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)                                          \
+    ZSTD_SEARCH_FN_ATTRS size_t ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(                     \
+            ZSTD_matchState_t* ms,                                                             \
+            const BYTE* ip, const BYTE* const iLimit,                                          \
+            size_t* offsetPtr)                                                                 \
+    {                                                                                          \
+        assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                                   \
+        assert(MAX(4, MIN(6, ms->cParams.searchLog)) == rowLog);                               \
+        return ZSTD_RowFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode, rowLog); \
+    }                                                                                          \
+
+#define ZSTD_FOR_EACH_ROWLOG(X, dictMode, mls) \
+    X(dictMode, mls, 4)                        \
+    X(dictMode, mls, 5)                        \
+    X(dictMode, mls, 6)
+
+#define ZSTD_FOR_EACH_MLS_ROWLOG(X, dictMode) \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 4)      \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 5)      \
+    ZSTD_FOR_EACH_ROWLOG(X, dictMode, 6)
+
+#define ZSTD_FOR_EACH_MLS(X, dictMode) \
+    X(dictMode, 4)                     \
+    X(dictMode, 5)                     \
+    X(dictMode, 6)
+
+#define ZSTD_FOR_EACH_DICT_MODE(X, ...) \
+    X(__VA_ARGS__, noDict)              \
+    X(__VA_ARGS__, extDict)             \
+    X(__VA_ARGS__, dictMatchState)      \
+    X(__VA_ARGS__, dedicatedDictSearch)
+
+/* Generate row search fns for each combination of (dictMode, mls, rowLog) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS_ROWLOG, GEN_ZSTD_ROW_SEARCH_FN)
+/* Generate binary Tree search fns for each combination of (dictMode, mls) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_BT_SEARCH_FN)
+/* Generate hash chain search fns for each combination of (dictMode, mls) */
+ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_HC_SEARCH_FN)
 
-FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_dedicatedDictSearch_selectRowLog(
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
-{
-    const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
-    switch(cappedSearchLog)
-    {
-    default :
-    case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dedicatedDictSearch, offsetPtr, 4);
-    case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_dedicatedDictSearch, offsetPtr, 5);
+typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e;
+
+#define GEN_ZSTD_CALL_BT_SEARCH_FN(dictMode, mls)                         \
+    case mls:                                                             \
+        return ZSTD_BT_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
+#define GEN_ZSTD_CALL_HC_SEARCH_FN(dictMode, mls)                         \
+    case mls:                                                             \
+        return ZSTD_HC_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
+#define GEN_ZSTD_CALL_ROW_SEARCH_FN(dictMode, mls, rowLog)                         \
+    case rowLog:                                                                   \
+        return ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(ms, ip, iend, offsetPtr);
+
+#define ZSTD_SWITCH_MLS(X, dictMode)   \
+    switch (mls) {                     \
+        ZSTD_FOR_EACH_MLS(X, dictMode) \
     }
-}
 
-FORCE_INLINE_TEMPLATE size_t ZSTD_RowFindBestMatch_extDict_selectRowLog (
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr)
+#define ZSTD_SWITCH_ROWLOG(dictMode, mls)                                    \
+    case mls:                                                                \
+        switch (rowLog) {                                                    \
+            ZSTD_FOR_EACH_ROWLOG(GEN_ZSTD_CALL_ROW_SEARCH_FN, dictMode, mls) \
+        }                                                                    \
+        ZSTD_UNREACHABLE;                                                    \
+        break;
+
+#define ZSTD_SWITCH_SEARCH_METHOD(dictMode)                       \
+    switch (searchMethod) {                                       \
+        case search_hashChain:                                    \
+            ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_HC_SEARCH_FN, dictMode) \
+            break;                                                \
+        case search_binaryTree:                                   \
+            ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_BT_SEARCH_FN, dictMode) \
+            break;                                                \
+        case search_rowHash:                                      \
+            ZSTD_SWITCH_MLS(ZSTD_SWITCH_ROWLOG, dictMode)         \
+            break;                                                \
+    }                                                             \
+    ZSTD_UNREACHABLE;
+
+/**
+ * Searches for the longest match at @p ip.
+ * Dispatches to the correct implementation function based on the
+ * (searchMethod, dictMode, mls, rowLog). We use switch statements
+ * here instead of using an indirect function call through a function
+ * pointer because after Spectre and Meltdown mitigations, indirect
+ * function calls can be very costly, especially in the kernel.
+ *
+ * NOTE: dictMode and searchMethod should be templated, so those switch
+ * statements should be optimized out. Only the mls & rowLog switches
+ * should be left.
+ *
+ * @param ms The match state.
+ * @param ip The position to search at.
+ * @param iend The end of the input data.
+ * @param[out] offsetPtr Stores the match offset into this pointer.
+ * @param mls The minimum search length, in the range [4, 6].
+ * @param rowLog The row log (if applicable), in the range [4, 6].
+ * @param searchMethod The search method to use (templated).
+ * @param dictMode The dictMode (templated).
+ *
+ * @returns The length of the longest match found, or < mls if no match is found.
+ * If a match is found its offset is stored in @p offsetPtr.
+ */
+FORCE_INLINE_TEMPLATE size_t ZSTD_searchMax(
+    ZSTD_matchState_t* ms,
+    const BYTE* ip,
+    const BYTE* iend,
+    size_t* offsetPtr,
+    U32 const mls,
+    U32 const rowLog,
+    searchMethod_e const searchMethod,
+    ZSTD_dictMode_e const dictMode)
 {
-    const U32 cappedSearchLog = MIN(ms->cParams.searchLog, 5);
-    switch(cappedSearchLog)
-    {
-    default :
-    case 4 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_extDict, offsetPtr, 4);
-    case 5 : return ZSTD_RowFindBestMatch_selectMLS(ms, ip, iLimit, ZSTD_extDict, offsetPtr, 5);
+    if (dictMode == ZSTD_noDict) {
+        ZSTD_SWITCH_SEARCH_METHOD(noDict)
+    } else if (dictMode == ZSTD_extDict) {
+        ZSTD_SWITCH_SEARCH_METHOD(extDict)
+    } else if (dictMode == ZSTD_dictMatchState) {
+        ZSTD_SWITCH_SEARCH_METHOD(dictMatchState)
+    } else if (dictMode == ZSTD_dedicatedDictSearch) {
+        ZSTD_SWITCH_SEARCH_METHOD(dedicatedDictSearch)
     }
+    ZSTD_UNREACHABLE;
+    return 0;
 }
 
-
 /* *******************************
 *  Common parser - lazy strategy
 *********************************/
-typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e;
 
 FORCE_INLINE_TEMPLATE size_t
 ZSTD_compressBlock_lazy_generic(
@@ -1525,47 +1505,15 @@ ZSTD_compressBlock_lazy_generic(
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = searchMethod == search_rowHash ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
+    const BYTE* const ilimit = (searchMethod == search_rowHash) ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
     const BYTE* const base = ms->window.base;
     const U32 prefixLowestIndex = ms->window.dictLimit;
     const BYTE* const prefixLowest = base + prefixLowestIndex;
-    const U32 rowLog = ms->cParams.searchLog < 5 ? 4 : 5;
-
-    typedef size_t (*searchMax_f)(
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
 
-    /**
-     * This table is indexed first by the four ZSTD_dictMode_e values, and then
-     * by the two searchMethod_e values. NULLs are placed for configurations
-     * that should never occur (extDict modes go to the other implementation
-     * below and there is no DDSS for binary tree search yet).
-     */
-    const searchMax_f searchFuncs[4][3] = {
-        {
-            ZSTD_HcFindBestMatch_selectMLS,
-            ZSTD_BtFindBestMatch_selectMLS,
-            ZSTD_RowFindBestMatch_selectRowLog
-        },
-        {
-            NULL,
-            NULL,
-            NULL
-        },
-        {
-            ZSTD_HcFindBestMatch_dictMatchState_selectMLS,
-            ZSTD_BtFindBestMatch_dictMatchState_selectMLS,
-            ZSTD_RowFindBestMatch_dictMatchState_selectRowLog
-        },
-        {
-            ZSTD_HcFindBestMatch_dedicatedDictSearch_selectMLS,
-            NULL,
-            ZSTD_RowFindBestMatch_dedicatedDictSearch_selectRowLog
-        }
-    };
-
-    searchMax_f const searchMax = searchFuncs[dictMode][(int)searchMethod];
-    U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+    U32 offsetSaved1 = 0, offsetSaved2 = 0;
 
     const int isDMS = dictMode == ZSTD_dictMatchState;
     const int isDDS = dictMode == ZSTD_dedicatedDictSearch;
@@ -1580,16 +1528,14 @@ ZSTD_compressBlock_lazy_generic(
                                      0;
     const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
 
-    assert(searchMax != NULL);
-
     DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u) (searchFunc=%u)", (U32)dictMode, (U32)searchMethod);
     ip += (dictAndPrefixLength == 0);
     if (dictMode == ZSTD_noDict) {
         U32 const curr = (U32)(ip - base);
         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog);
         U32 const maxRep = curr - windowLow;
-        if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
-        if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
+        if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
     }
     if (isDxS) {
         /* dictMatchState repCode checks don't currently handle repCode == 0
@@ -1613,8 +1559,9 @@ ZSTD_compressBlock_lazy_generic(
 #endif
     while (ip < ilimit) {
         size_t matchLength=0;
-        size_t offset=0;
+        size_t offBase = REPCODE1_TO_OFFBASE;
         const BYTE* start=ip+1;
+        DEBUGLOG(7, "search baseline (depth 0)");
 
         /* check repCode */
         if (isDxS) {
@@ -1637,10 +1584,10 @@ ZSTD_compressBlock_lazy_generic(
         }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 999999999;
-            size_t const ml2 = searchMax(ms, ip, iend, &offsetFound);
+        {   size_t offbaseFound = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offbaseFound, mls, rowLog, searchMethod, dictMode);
             if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offset=offsetFound;
+                matchLength = ml2, start = ip, offBase = offbaseFound;
         }
 
         if (matchLength < 4) {
@@ -1651,14 +1598,15 @@ ZSTD_compressBlock_lazy_generic(
         /* let's try to find a better solution */
         if (depth>=1)
         while (ip0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+              && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
                 size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
                 int const gain2 = (int)(mlRep * 3);
-                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
                 if ((mlRep >= 4) && (gain2 > gain1))
-                    matchLength = mlRep, offset = 0, start = ip;
+                    matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
             }
             if (isDxS) {
                 const U32 repIndex = (U32)(ip - base) - offset_1;
@@ -1670,30 +1618,31 @@ ZSTD_compressBlock_lazy_generic(
                     const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
                     size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
                     int const gain2 = (int)(mlRep * 3);
-                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
                     if ((mlRep >= 4) && (gain2 > gain1))
-                        matchLength = mlRep, offset = 0, start = ip;
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
                 }
             }
-            {   size_t offset2=999999999;
-                size_t const ml2 = searchMax(ms, ip, iend, &offset2);
-                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+            {   size_t ofbCandidate=999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
-                    matchLength = ml2, offset = offset2, start = ip;
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
                     continue;   /* search a better one */
             }   }
 
             /* let's find an even better one */
             if ((depth==2) && (ip0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                  && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
                     size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
                     int const gain2 = (int)(mlRep * 4);
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
                     if ((mlRep >= 4) && (gain2 > gain1))
-                        matchLength = mlRep, offset = 0, start = ip;
+                        matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
                 }
                 if (isDxS) {
                     const U32 repIndex = (U32)(ip - base) - offset_1;
@@ -1705,46 +1654,45 @@ ZSTD_compressBlock_lazy_generic(
                         const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
                         size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
                         int const gain2 = (int)(mlRep * 4);
-                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
                         if ((mlRep >= 4) && (gain2 > gain1))
-                            matchLength = mlRep, offset = 0, start = ip;
+                            matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
                     }
                 }
-                {   size_t offset2=999999999;
-                    size_t const ml2 = searchMax(ms, ip, iend, &offset2);
-                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                {   size_t ofbCandidate=999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
-                        matchLength = ml2, offset = offset2, start = ip;
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
                         continue;
             }   }   }
             break;  /* nothing found : store previous solution */
         }
 
         /* NOTE:
-         * start[-offset+ZSTD_REP_MOVE-1] is undefined behavior.
-         * (-offset+ZSTD_REP_MOVE-1) is unsigned, and is added to start, which
-         * overflows the pointer, which is undefined behavior.
+         * Pay attention that `start[-value]` can lead to strange undefined behavior
+         * notably if `value` is unsigned, resulting in a large positive `-value`.
          */
         /* catch up */
-        if (offset) {
+        if (OFFBASE_IS_OFFSET(offBase)) {
             if (dictMode == ZSTD_noDict) {
-                while ( ((start > anchor) & (start - (offset-ZSTD_REP_MOVE) > prefixLowest))
-                     && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
+                while ( ((start > anchor) & (start - OFFBASE_TO_OFFSET(offBase) > prefixLowest))
+                     && (start[-1] == (start-OFFBASE_TO_OFFSET(offBase))[-1]) )  /* only search for offset within prefix */
                     { start--; matchLength++; }
             }
             if (isDxS) {
-                U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+                U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
                 const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
                 const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
                 while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
             }
-            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
         }
         /* store sequence */
 _storeSequence:
-        {   size_t const litLength = start - anchor;
-            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offset, matchLength-MINMATCH);
+        {   size_t const litLength = (size_t)(start - anchor);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
             anchor = ip = start + matchLength;
         }
 
@@ -1760,8 +1708,8 @@ ZSTD_compressBlock_lazy_generic(
                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
                     const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
                     matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
-                    offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, matchLength-MINMATCH);
+                    offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
                     ip += matchLength;
                     anchor = ip;
                     continue;
@@ -1775,16 +1723,20 @@ ZSTD_compressBlock_lazy_generic(
                  && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
                 /* store sequence */
                 matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
-                ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, matchLength-MINMATCH);
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap repcodes */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
                 ip += matchLength;
                 anchor = ip;
                 continue;   /* faster when present ... (?) */
     }   }   }
 
-    /* Save reps for next block */
-    rep[0] = offset_1 ? offset_1 : savedOffset;
-    rep[1] = offset_2 ? offset_2 : savedOffset;
+    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
+     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
+    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved1;
+    rep[1] = offset_2 ? offset_2 : offsetSaved2;
 
     /* Return the last literals size */
     return (size_t)(iend - anchor);
@@ -1953,17 +1905,9 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
     const BYTE* const dictEnd  = dictBase + dictLimit;
     const BYTE* const dictStart  = dictBase + ms->window.lowLimit;
     const U32 windowLog = ms->cParams.windowLog;
-    const U32 rowLog = ms->cParams.searchLog < 5 ? 4 : 5;
+    const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
+    const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
 
-    typedef size_t (*searchMax_f)(
-                        ZSTD_matchState_t* ms,
-                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
-    const searchMax_f searchFuncs[3] = {
-        ZSTD_HcFindBestMatch_extDict_selectMLS,
-        ZSTD_BtFindBestMatch_extDict_selectMLS,
-        ZSTD_RowFindBestMatch_extDict_selectRowLog
-    };
-    searchMax_f searchMax = searchFuncs[(int)searchMethod];
     U32 offset_1 = rep[0], offset_2 = rep[1];
 
     DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod);
@@ -1985,7 +1929,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
 #endif
     while (ip < ilimit) {
         size_t matchLength=0;
-        size_t offset=0;
+        size_t offBase = REPCODE1_TO_OFFBASE;
         const BYTE* start=ip+1;
         U32 curr = (U32)(ip-base);
 
@@ -1995,7 +1939,7 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
             if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow */
-               & (offset_1 < curr+1 - windowLow) ) /* note: we are searching at curr+1 */
+               & (offset_1 <= curr+1 - windowLow) ) /* note: we are searching at curr+1 */
             if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -2004,13 +1948,13 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
         }   }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 999999999;
-            size_t const ml2 = searchMax(ms, ip, iend, &offsetFound);
+        {   size_t ofbCandidate = 999999999;
+            size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
             if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offset=offsetFound;
+                matchLength = ml2, start = ip, offBase = ofbCandidate;
         }
 
-         if (matchLength < 4) {
+        if (matchLength < 4) {
             ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
             continue;
         }
@@ -2021,30 +1965,30 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             ip ++;
             curr++;
             /* check repCode */
-            if (offset) {
+            if (offBase) {
                 const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
                 const U32 repIndex = (U32)(curr - offset_1);
                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                 const BYTE* const repMatch = repBase + repIndex;
                 if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments  */
-                   & (offset_1 < curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+                   & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
                 if (MEM_read32(ip) == MEM_read32(repMatch)) {
                     /* repcode detected */
                     const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                     size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
                     int const gain2 = (int)(repLength * 3);
-                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
                     if ((repLength >= 4) && (gain2 > gain1))
-                        matchLength = repLength, offset = 0, start = ip;
+                        matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
             }   }
 
             /* search match, depth 1 */
-            {   size_t offset2=999999999;
-                size_t const ml2 = searchMax(ms, ip, iend, &offset2);
-                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+            {   size_t ofbCandidate = 999999999;
+                size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
-                    matchLength = ml2, offset = offset2, start = ip;
+                    matchLength = ml2, offBase = ofbCandidate, start = ip;
                     continue;   /* search a better one */
             }   }
 
@@ -2053,48 +1997,48 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
                 ip ++;
                 curr++;
                 /* check repCode */
-                if (offset) {
+                if (offBase) {
                     const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
                     const U32 repIndex = (U32)(curr - offset_1);
                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                     const BYTE* const repMatch = repBase + repIndex;
                     if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments  */
-                       & (offset_1 < curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+                       & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
                     if (MEM_read32(ip) == MEM_read32(repMatch)) {
                         /* repcode detected */
                         const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                         size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
                         int const gain2 = (int)(repLength * 4);
-                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
                         if ((repLength >= 4) && (gain2 > gain1))
-                            matchLength = repLength, offset = 0, start = ip;
+                            matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
                 }   }
 
                 /* search match, depth 2 */
-                {   size_t offset2=999999999;
-                    size_t const ml2 = searchMax(ms, ip, iend, &offset2);
-                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                {   size_t ofbCandidate = 999999999;
+                    size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
-                        matchLength = ml2, offset = offset2, start = ip;
+                        matchLength = ml2, offBase = ofbCandidate, start = ip;
                         continue;
             }   }   }
             break;  /* nothing found : store previous solution */
         }
 
         /* catch up */
-        if (offset) {
-            U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+        if (OFFBASE_IS_OFFSET(offBase)) {
+            U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
             const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
             const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
             while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
-            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+            offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
         }
 
         /* store sequence */
 _storeSequence:
-        {   size_t const litLength = start - anchor;
-            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offset, matchLength-MINMATCH);
+        {   size_t const litLength = (size_t)(start - anchor);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
             anchor = ip = start + matchLength;
         }
 
@@ -2106,13 +2050,13 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
             if ( ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow : do not test positions overlapping 2 memory segments  */
-               & (offset_2 < repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
+               & (offset_2 <= repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
             if (MEM_read32(ip) == MEM_read32(repMatch)) {
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                 matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
-                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset history */
-                ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, matchLength-MINMATCH);
+                offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset history */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
                 ip += matchLength;
                 anchor = ip;
                 continue;   /* faster when present ... (?) */
@@ -2178,7 +2122,6 @@ size_t ZSTD_compressBlock_lazy_extDict_row(
 size_t ZSTD_compressBlock_lazy2_extDict_row(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize)
-
 {
     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2);
 }
diff --git a/priv/zstd/lib/compress/zstd_lazy.h b/priv/zstd/lib/compress/zstd_lazy.h
index 150f7b3..3bde673 100644
--- a/priv/zstd/lib/compress/zstd_lazy.h
+++ b/priv/zstd/lib/compress/zstd_lazy.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -25,6 +25,8 @@ extern "C" {
  */
 #define ZSTD_LAZY_DDSS_BUCKET_LOG 2
 
+#define ZSTD_ROW_HASH_TAG_BITS 8        /* nb bits to use for the tag */
+
 U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
 void ZSTD_row_update(ZSTD_matchState_t* const ms, const BYTE* ip);
 
@@ -116,7 +118,7 @@ size_t ZSTD_compressBlock_lazy2_extDict_row(
 size_t ZSTD_compressBlock_btlazy2_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         void const* src, size_t srcSize);
-        
+
 
 #if defined (__cplusplus)
 }
diff --git a/priv/zstd/lib/compress/zstd_ldm.c b/priv/zstd/lib/compress/zstd_ldm.c
index fa4ebea..3d74ff1 100644
--- a/priv/zstd/lib/compress/zstd_ldm.c
+++ b/priv/zstd/lib/compress/zstd_ldm.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -159,12 +159,12 @@ size_t ZSTD_ldm_getTableSize(ldmParams_t params)
     size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
     size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
                            + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
-    return params.enableLdm ? totalSize : 0;
+    return params.enableLdm == ZSTD_ps_enable ? totalSize : 0;
 }
 
 size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
 {
-    return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
+    return params.enableLdm == ZSTD_ps_enable ? (maxChunkSize / params.minMatchLength) : 0;
 }
 
 /** ZSTD_ldm_getBucket() :
@@ -242,11 +242,11 @@ static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
     switch(ms->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
+        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
         break;
 
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
+        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast, ZSTD_tfp_forCCtx);
         break;
 
     case ZSTD_greedy:
@@ -478,7 +478,7 @@ static size_t ZSTD_ldm_generateSequences_internal(
              */
             if (anchor > ip + hashed) {
                 ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
-                /* Continue the outter loop at anchor (ip + hashed == anchor). */
+                /* Continue the outer loop at anchor (ip + hashed == anchor). */
                 ip = anchor - hashed;
                 break;
             }
@@ -549,7 +549,7 @@ size_t ZSTD_ldm_generateSequences(
          * the window through early invalidation.
          * TODO: * Test the chunk size.
          *       * Try invalidation after the sequence generation and test the
-         *         the offset against maxDist directly.
+         *         offset against maxDist directly.
          *
          * NOTE: Because of dictionaries + sequence splitting we MUST make sure
          * that any offset used is valid at the END of the sequence, since it may
@@ -579,7 +579,9 @@ size_t ZSTD_ldm_generateSequences(
     return 0;
 }
 
-void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
+void
+ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch)
+{
     while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
         rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
         if (srcSize <= seq->litLength) {
@@ -657,7 +659,7 @@ void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
 
 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
     ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-    ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+    ZSTD_paramSwitch_e useRowMatchFinder,
     void const* src, size_t srcSize)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
@@ -709,8 +711,8 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
             rep[0] = sequence.offset;
             /* Store the sequence */
             ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
-                          sequence.offset + ZSTD_REP_MOVE,
-                          sequence.matchLength - MINMATCH);
+                          OFFSET_TO_OFFBASE(sequence.offset),
+                          sequence.matchLength);
             ip += sequence.matchLength;
         }
     }
diff --git a/priv/zstd/lib/compress/zstd_ldm.h b/priv/zstd/lib/compress/zstd_ldm.h
index 393466f..f147021 100644
--- a/priv/zstd/lib/compress/zstd_ldm.h
+++ b/priv/zstd/lib/compress/zstd_ldm.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -66,7 +66,7 @@ size_t ZSTD_ldm_generateSequences(
  */
 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
             ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
-            ZSTD_useRowMatchFinderMode_e useRowMatchFinder,
+            ZSTD_paramSwitch_e useRowMatchFinder,
             void const* src, size_t srcSize);
 
 /**
diff --git a/priv/zstd/lib/compress/zstd_ldm_geartab.h b/priv/zstd/lib/compress/zstd_ldm_geartab.h
index e5c24d8..ef34bc5 100644
--- a/priv/zstd/lib/compress/zstd_ldm_geartab.h
+++ b/priv/zstd/lib/compress/zstd_ldm_geartab.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,7 +11,10 @@
 #ifndef ZSTD_LDM_GEARTAB_H
 #define ZSTD_LDM_GEARTAB_H
 
-static U64 ZSTD_ldm_gearTab[256] = {
+#include "../common/compiler.h" /* UNUSED_ATTR */
+#include "../common/mem.h"      /* U64 */
+
+static UNUSED_ATTR const U64 ZSTD_ldm_gearTab[256] = {
     0xf5b8f72c5f77775c, 0x84935f266b7ac412, 0xb647ada9ca730ccc,
     0xb065bb4b114fb1de, 0x34584e7e8c3a9fd0, 0x4e97e17c6ae26b05,
     0x3a03d743bc99a604, 0xcecd042422c4044f, 0x76de76c58524259e,
diff --git a/priv/zstd/lib/compress/zstd_opt.c b/priv/zstd/lib/compress/zstd_opt.c
index 402a7e5..fdd7f9d 100644
--- a/priv/zstd/lib/compress/zstd_opt.c
+++ b/priv/zstd/lib/compress/zstd_opt.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,40 +14,47 @@
 
 
 #define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
-#define ZSTD_FREQ_DIV       4   /* log factor when using previous stats to init next stats */
 #define ZSTD_MAX_PRICE     (1<<30)
 
-#define ZSTD_PREDEF_THRESHOLD 1024   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
+#define ZSTD_PREDEF_THRESHOLD 8   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
 
 
 /*-*************************************
 *  Price functions for optimal parser
 ***************************************/
 
-#if 0    /* approximation at bit level */
+#if 0    /* approximation at bit level (for tests) */
 #  define BITCOST_ACCURACY 0
 #  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
-#  define WEIGHT(stat)  ((void)opt, ZSTD_bitWeight(stat))
-#elif 0  /* fractional bit accuracy */
+#  define WEIGHT(stat, opt) ((void)(opt), ZSTD_bitWeight(stat))
+#elif 0  /* fractional bit accuracy (for tests) */
 #  define BITCOST_ACCURACY 8
 #  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
-#  define WEIGHT(stat,opt) ((void)opt, ZSTD_fracWeight(stat))
+#  define WEIGHT(stat,opt) ((void)(opt), ZSTD_fracWeight(stat))
 #else    /* opt==approx, ultra==accurate */
 #  define BITCOST_ACCURACY 8
 #  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
-#  define WEIGHT(stat,opt) (opt ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
+#  define WEIGHT(stat,opt) ((opt) ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
 #endif
 
+/* ZSTD_bitWeight() :
+ * provide estimated "cost" of a stat in full bits only */
 MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
 {
     return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
 }
 
+/* ZSTD_fracWeight() :
+ * provide fractional-bit "cost" of a stat,
+ * using linear interpolation approximation */
 MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
 {
     U32 const stat = rawStat + 1;
     U32 const hb = ZSTD_highbit32(stat);
     U32 const BWeight = hb * BITCOST_MULTIPLIER;
+    /* Fweight was meant for "Fractional weight"
+     * but it's effectively a value between 1 and 2
+     * using fixed point arithmetic */
     U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
     U32 const weight = BWeight + FWeight;
     assert(hb + BITCOST_ACCURACY < 31);
@@ -58,7 +65,7 @@ MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
 /* debugging function,
  * @return price in bytes as fractional value
  * for debug messages only */
-MEM_STATIC double ZSTD_fCost(U32 price)
+MEM_STATIC double ZSTD_fCost(int price)
 {
     return (double)price / (BITCOST_MULTIPLIER*8);
 }
@@ -66,7 +73,7 @@ MEM_STATIC double ZSTD_fCost(U32 price)
 
 static int ZSTD_compressedLiterals(optState_t const* const optPtr)
 {
-    return optPtr->literalCompressionMode != ZSTD_lcm_uncompressed;
+    return optPtr->literalCompressionMode != ZSTD_ps_disable;
 }
 
 static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
@@ -79,25 +86,52 @@ static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
 }
 
 
-/* ZSTD_downscaleStat() :
- * reduce all elements in table by a factor 2^(ZSTD_FREQ_DIV+malus)
- * return the resulting sum of elements */
-static U32 ZSTD_downscaleStat(unsigned* table, U32 lastEltIndex, int malus)
+static U32 sum_u32(const unsigned table[], size_t nbElts)
+{
+    size_t n;
+    U32 total = 0;
+    for (n=0; n 0 && ZSTD_FREQ_DIV+malus < 31);
+    DEBUGLOG(5, "ZSTD_downscaleStats (nbElts=%u, shift=%u)",
+            (unsigned)lastEltIndex+1, (unsigned)shift );
+    assert(shift < 30);
     for (s=0; s> (ZSTD_FREQ_DIV+malus));
-        sum += table[s];
+        unsigned const base = base1 ? 1 : (table[s]>0);
+        unsigned const newStat = base + (table[s] >> shift);
+        sum += newStat;
+        table[s] = newStat;
     }
     return sum;
 }
 
+/* ZSTD_scaleStats() :
+ * reduce all elt frequencies in table if sum too large
+ * return the resulting sum of elements */
+static U32 ZSTD_scaleStats(unsigned* table, U32 lastEltIndex, U32 logTarget)
+{
+    U32 const prevsum = sum_u32(table, lastEltIndex+1);
+    U32 const factor = prevsum >> logTarget;
+    DEBUGLOG(5, "ZSTD_scaleStats (nbElts=%u, target=%u)", (unsigned)lastEltIndex+1, (unsigned)logTarget);
+    assert(logTarget < 30);
+    if (factor <= 1) return prevsum;
+    return ZSTD_downscaleStats(table, lastEltIndex, ZSTD_highbit32(factor), base_1guaranteed);
+}
+
 /* ZSTD_rescaleFreqs() :
  * if first block (detected by optPtr->litLengthSum == 0) : init statistics
  *    take hints from dictionary if there is one
- *    or init from zero, using src for literals stats, or flat 1 for match symbols
+ *    and init from zero if there is none,
+ *    using src for literals stats, and baseline stats for sequence symbols
  * otherwise downscale existing stats, to be used as seed for next block.
  */
 static void
@@ -109,24 +143,28 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
     DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);
     optPtr->priceType = zop_dynamic;
 
-    if (optPtr->litLengthSum == 0) {  /* first block : init */
-        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {  /* heuristic */
-            DEBUGLOG(5, "(srcSize <= ZSTD_PREDEF_THRESHOLD) => zop_predef");
+    if (optPtr->litLengthSum == 0) {  /* no literals stats collected -> first block assumed -> init */
+
+        /* heuristic: use pre-defined stats for too small inputs */
+        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {
+            DEBUGLOG(5, "srcSize <= %i : use predefined stats", ZSTD_PREDEF_THRESHOLD);
             optPtr->priceType = zop_predef;
         }
 
         assert(optPtr->symbolCosts != NULL);
         if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {
-            /* huffman table presumed generated by dictionary */
+
+            /* huffman stats covering the full value set : table presumed generated by dictionary */
             optPtr->priceType = zop_dynamic;
 
             if (compressedLiterals) {
+                /* generate literals statistics from huffman table */
                 unsigned lit;
                 assert(optPtr->litFreq != NULL);
                 optPtr->litSum = 0;
                 for (lit=0; lit<=MaxLit; lit++) {
                     U32 const scaleLog = 11;   /* scale to 2K */
-                    U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->huf.CTable, lit);
+                    U32 const bitCost = HUF_getNbBitsFromCTable(optPtr->symbolCosts->huf.CTable, lit);
                     assert(bitCost <= scaleLog);
                     optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
                     optPtr->litSum += optPtr->litFreq[lit];
@@ -168,20 +206,26 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
                     optPtr->offCodeSum += optPtr->offCodeFreq[of];
             }   }
 
-        } else {  /* not a dictionary */
+        } else {  /* first block, no dictionary */
 
             assert(optPtr->litFreq != NULL);
             if (compressedLiterals) {
+                /* base initial cost of literals on direct frequency within src */
                 unsigned lit = MaxLit;
                 HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
-                optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
+                optPtr->litSum = ZSTD_downscaleStats(optPtr->litFreq, MaxLit, 8, base_0possible);
             }
 
-            {   unsigned ll;
-                for (ll=0; ll<=MaxLL; ll++)
-                    optPtr->litLengthFreq[ll] = 1;
+            {   unsigned const baseLLfreqs[MaxLL+1] = {
+                    4, 2, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1
+                };
+                ZSTD_memcpy(optPtr->litLengthFreq, baseLLfreqs, sizeof(baseLLfreqs));
+                optPtr->litLengthSum = sum_u32(baseLLfreqs, MaxLL+1);
             }
-            optPtr->litLengthSum = MaxLL+1;
 
             {   unsigned ml;
                 for (ml=0; ml<=MaxML; ml++)
@@ -189,21 +233,25 @@ ZSTD_rescaleFreqs(optState_t* const optPtr,
             }
             optPtr->matchLengthSum = MaxML+1;
 
-            {   unsigned of;
-                for (of=0; of<=MaxOff; of++)
-                    optPtr->offCodeFreq[of] = 1;
+            {   unsigned const baseOFCfreqs[MaxOff+1] = {
+                    6, 2, 1, 1, 2, 3, 4, 4,
+                    4, 3, 2, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1,
+                    1, 1, 1, 1, 1, 1, 1, 1
+                };
+                ZSTD_memcpy(optPtr->offCodeFreq, baseOFCfreqs, sizeof(baseOFCfreqs));
+                optPtr->offCodeSum = sum_u32(baseOFCfreqs, MaxOff+1);
             }
-            optPtr->offCodeSum = MaxOff+1;
 
         }
 
-    } else {   /* new block : re-use previous statistics, scaled down */
+    } else {   /* new block : scale down accumulated statistics */
 
         if (compressedLiterals)
-            optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
-        optPtr->litLengthSum = ZSTD_downscaleStat(optPtr->litLengthFreq, MaxLL, 0);
-        optPtr->matchLengthSum = ZSTD_downscaleStat(optPtr->matchLengthFreq, MaxML, 0);
-        optPtr->offCodeSum = ZSTD_downscaleStat(optPtr->offCodeFreq, MaxOff, 0);
+            optPtr->litSum = ZSTD_scaleStats(optPtr->litFreq, MaxLit, 12);
+        optPtr->litLengthSum = ZSTD_scaleStats(optPtr->litLengthFreq, MaxLL, 11);
+        optPtr->matchLengthSum = ZSTD_scaleStats(optPtr->matchLengthFreq, MaxML, 11);
+        optPtr->offCodeSum = ZSTD_scaleStats(optPtr->offCodeFreq, MaxOff, 11);
     }
 
     ZSTD_setBasePrices(optPtr, optLevel);
@@ -225,11 +273,14 @@ static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
         return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
 
     /* dynamic statistics */
-    {   U32 price = litLength * optPtr->litSumBasePrice;
+    {   U32 price = optPtr->litSumBasePrice * litLength;
+        U32 const litPriceMax = optPtr->litSumBasePrice - BITCOST_MULTIPLIER;
         U32 u;
+        assert(optPtr->litSumBasePrice >= BITCOST_MULTIPLIER);
         for (u=0; u < litLength; u++) {
-            assert(WEIGHT(optPtr->litFreq[literals[u]], optLevel) <= optPtr->litSumBasePrice);   /* literal cost should never be negative */
-            price -= WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+            U32 litPrice = WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+            if (UNLIKELY(litPrice > litPriceMax)) litPrice = litPriceMax;
+            price -= litPrice;
         }
         return price;
     }
@@ -239,7 +290,17 @@ static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
  * cost of literalLength symbol */
 static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)
 {
-    if (optPtr->priceType == zop_predef) return WEIGHT(litLength, optLevel);
+    assert(litLength <= ZSTD_BLOCKSIZE_MAX);
+    if (optPtr->priceType == zop_predef)
+        return WEIGHT(litLength, optLevel);
+
+    /* ZSTD_LLcode() can't compute litLength price for sizes >= ZSTD_BLOCKSIZE_MAX
+     * because it isn't representable in the zstd format.
+     * So instead just pretend it would cost 1 bit more than ZSTD_BLOCKSIZE_MAX - 1.
+     * In such a case, the block would be all literals.
+     */
+    if (litLength == ZSTD_BLOCKSIZE_MAX)
+        return BITCOST_MULTIPLIER + ZSTD_litLengthPrice(ZSTD_BLOCKSIZE_MAX - 1, optPtr, optLevel);
 
     /* dynamic statistics */
     {   U32 const llCode = ZSTD_LLcode(litLength);
@@ -250,22 +311,25 @@ static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optP
 }
 
 /* ZSTD_getMatchPrice() :
- * Provides the cost of the match part (offset + matchLength) of a sequence
+ * Provides the cost of the match part (offset + matchLength) of a sequence.
  * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
- * optLevel: when <2, favors small offset for decompression speed (improved cache efficiency) */
+ * @offBase : sumtype, representing an offset or a repcode, and using numeric representation of ZSTD_storeSeq()
+ * @optLevel: when <2, favors small offset for decompression speed (improved cache efficiency)
+ */
 FORCE_INLINE_TEMPLATE U32
-ZSTD_getMatchPrice(U32 const offset,
+ZSTD_getMatchPrice(U32 const offBase,
                    U32 const matchLength,
              const optState_t* const optPtr,
                    int const optLevel)
 {
     U32 price;
-    U32 const offCode = ZSTD_highbit32(offset+1);
+    U32 const offCode = ZSTD_highbit32(offBase);
     U32 const mlBase = matchLength - MINMATCH;
     assert(matchLength >= MINMATCH);
 
-    if (optPtr->priceType == zop_predef)  /* fixed scheme, do not use statistics */
-        return WEIGHT(mlBase, optLevel) + ((16 + offCode) * BITCOST_MULTIPLIER);
+    if (optPtr->priceType == zop_predef)  /* fixed scheme, does not use statistics */
+        return WEIGHT(mlBase, optLevel)
+             + ((16 + offCode) * BITCOST_MULTIPLIER); /* emulated offset cost */
 
     /* dynamic statistics */
     price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
@@ -284,10 +348,10 @@ ZSTD_getMatchPrice(U32 const offset,
 }
 
 /* ZSTD_updateStats() :
- * assumption : literals + litLengtn <= iend */
+ * assumption : literals + litLength <= iend */
 static void ZSTD_updateStats(optState_t* const optPtr,
                              U32 litLength, const BYTE* literals,
-                             U32 offsetCode, U32 matchLength)
+                             U32 offBase, U32 matchLength)
 {
     /* literals */
     if (ZSTD_compressedLiterals(optPtr)) {
@@ -303,8 +367,8 @@ static void ZSTD_updateStats(optState_t* const optPtr,
         optPtr->litLengthSum++;
     }
 
-    /* match offset code (0-2=>repCode; 3+=>offset+2) */
-    {   U32 const offCode = ZSTD_highbit32(offsetCode+1);
+    /* offset code : follows storeSeq() numeric representation */
+    {   U32 const offCode = ZSTD_highbit32(offBase);
         assert(offCode <= MaxOff);
         optPtr->offCodeFreq[offCode]++;
         optPtr->offCodeSum++;
@@ -338,7 +402,7 @@ MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
 
 /* Update hashTable3 up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
+static U32 ZSTD_insertAndFindFirstIndexHash3 (const ZSTD_matchState_t* ms,
                                               U32* nextToUpdate3,
                                               const BYTE* const ip)
 {
@@ -364,11 +428,13 @@ static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
 *  Binary Tree search
 ***************************************/
 /** ZSTD_insertBt1() : add one or multiple positions to tree.
- *  ip : assumed <= iend-8 .
+ * @param ip assumed <= iend-8 .
+ * @param target The target of ZSTD_updateTree_internal() - we are filling to this position
  * @return : nb of positions added */
 static U32 ZSTD_insertBt1(
-                ZSTD_matchState_t* ms,
+                const ZSTD_matchState_t* ms,
                 const BYTE* const ip, const BYTE* const iend,
+                U32 const target,
                 U32 const mls, const int extDict)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
@@ -391,7 +457,10 @@ static U32 ZSTD_insertBt1(
     U32* smallerPtr = bt + 2*(curr&btMask);
     U32* largerPtr  = smallerPtr + 1;
     U32 dummy32;   /* to be nullified at the end */
-    U32 const windowLow = ms->window.lowLimit;
+    /* windowLow is based on target because
+     * we only need positions that will be in the window at the end of the tree update.
+     */
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, target, cParams->windowLog);
     U32 matchEndIdx = curr+8+1;
     size_t bestLength = 8;
     U32 nbCompares = 1U << cParams->searchLog;
@@ -404,11 +473,12 @@ static U32 ZSTD_insertBt1(
 
     DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
 
+    assert(curr <= target);
     assert(ip <= iend-8);   /* required for h calculation */
     hashTable[h] = curr;   /* Update Hash Table */
 
     assert(windowLow > 0);
-    while (nbCompares-- && (matchIndex >= windowLow)) {
+    for (; nbCompares && (matchIndex >= windowLow); --nbCompares) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         assert(matchIndex < curr);
@@ -492,7 +562,7 @@ void ZSTD_updateTree_internal(
                 idx, target, dictMode);
 
     while(idx < target) {
-        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
+        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, target, mls, dictMode == ZSTD_extDict);
         assert(idx < (U32)(idx + forward));
         idx += forward;
     }
@@ -505,16 +575,17 @@ void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
     ZSTD_updateTree_internal(ms, ip, iend, ms->cParams.minMatch, ZSTD_noDict);
 }
 
-FORCE_INLINE_TEMPLATE
-U32 ZSTD_insertBtAndGetAllMatches (
-                    ZSTD_match_t* matches,   /* store result (found matches) in this table (presumed large enough) */
-                    ZSTD_matchState_t* ms,
-                    U32* nextToUpdate3,
-                    const BYTE* const ip, const BYTE* const iLimit, const ZSTD_dictMode_e dictMode,
-                    const U32 rep[ZSTD_REP_NUM],
-                    U32 const ll0,   /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
-                    const U32 lengthToBeat,
-                    U32 const mls /* template */)
+FORCE_INLINE_TEMPLATE U32
+ZSTD_insertBtAndGetAllMatches (
+                ZSTD_match_t* matches,  /* store result (found matches) in this table (presumed large enough) */
+                ZSTD_matchState_t* ms,
+                U32* nextToUpdate3,
+                const BYTE* const ip, const BYTE* const iLimit,
+                const ZSTD_dictMode_e dictMode,
+                const U32 rep[ZSTD_REP_NUM],
+                const U32 ll0,  /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
+                const U32 lengthToBeat,
+                const U32 mls /* template */)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
@@ -597,7 +668,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
                             repCode, ll0, repOffset, repLen);
                 bestLength = repLen;
-                matches[mnum].off = repCode - ll0;
+                matches[mnum].off = REPCODE_TO_OFFBASE(repCode - ll0 + 1);  /* expect value between 1 and 3 */
                 matches[mnum].len = (U32)repLen;
                 mnum++;
                 if ( (repLen > sufficient_len)
@@ -626,7 +697,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 bestLength = mlen;
                 assert(curr > matchIndex3);
                 assert(mnum==0);  /* no prior solution */
-                matches[0].off = (curr - matchIndex3) + ZSTD_REP_MOVE;
+                matches[0].off = OFFSET_TO_OFFBASE(curr - matchIndex3);
                 matches[0].len = (U32)mlen;
                 mnum = 1;
                 if ( (mlen > sufficient_len) |
@@ -635,11 +706,11 @@ U32 ZSTD_insertBtAndGetAllMatches (
                     return 1;
         }   }   }
         /* no dictMatchState lookup: dicts don't have a populated HC3 table */
-    }
+    }  /* if (mls == 3) */
 
     hashTable[h] = curr;   /* Update Hash Table */
 
-    while (nbCompares-- && (matchIndex >= matchLow)) {
+    for (; nbCompares && (matchIndex >= matchLow); --nbCompares) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         const BYTE* match;
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
@@ -659,21 +730,20 @@ U32 ZSTD_insertBtAndGetAllMatches (
         }
 
         if (matchLength > bestLength) {
-            DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
-                    (U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
+            DEBUGLOG(8, "found match of length %u at distance %u (offBase=%u)",
+                    (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
             assert(matchEndIdx > matchIndex);
             if (matchLength > matchEndIdx - matchIndex)
                 matchEndIdx = matchIndex + (U32)matchLength;
             bestLength = matchLength;
-            matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
+            matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
             matches[mnum].len = (U32)matchLength;
             mnum++;
             if ( (matchLength > ZSTD_OPT_NUM)
                | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
                 if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
                 break; /* drop, to preserve bt consistency (miss a little bit of compression) */
-            }
-        }
+        }   }
 
         if (match[matchLength] < ip[matchLength]) {
             /* match smaller than current */
@@ -692,12 +762,13 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
     *smallerPtr = *largerPtr = 0;
 
+    assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
     if (dictMode == ZSTD_dictMatchState && nbCompares) {
         size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
         U32 dictMatchIndex = dms->hashTable[dmsH];
         const U32* const dmsBt = dms->chainTable;
         commonLengthSmaller = commonLengthLarger = 0;
-        while (nbCompares-- && (dictMatchIndex > dmsLowLimit)) {
+        for (; nbCompares && (dictMatchIndex > dmsLowLimit); --nbCompares) {
             const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
             size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
             const BYTE* match = dmsBase + dictMatchIndex;
@@ -707,19 +778,18 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
             if (matchLength > bestLength) {
                 matchIndex = dictMatchIndex + dmsIndexDelta;
-                DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
-                        (U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
+                DEBUGLOG(8, "found dms match of length %u at distance %u (offBase=%u)",
+                        (U32)matchLength, curr - matchIndex, OFFSET_TO_OFFBASE(curr - matchIndex));
                 if (matchLength > matchEndIdx - matchIndex)
                     matchEndIdx = matchIndex + (U32)matchLength;
                 bestLength = matchLength;
-                matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
+                matches[mnum].off = OFFSET_TO_OFFBASE(curr - matchIndex);
                 matches[mnum].len = (U32)matchLength;
                 mnum++;
                 if ( (matchLength > ZSTD_OPT_NUM)
                    | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
                     break;   /* drop, to guarantee consistency (miss a little bit of compression) */
-                }
-            }
+            }   }
 
             if (dictMatchIndex <= dmsBtLow) { break; }   /* beyond tree size, stop the search */
             if (match[matchLength] < ip[matchLength]) {
@@ -729,39 +799,91 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 /* match is larger than current */
                 commonLengthLarger = matchLength;
                 dictMatchIndex = nextPtr[0];
-            }
-        }
-    }
+    }   }   }  /* if (dictMode == ZSTD_dictMatchState) */
 
     assert(matchEndIdx > curr+8);
     ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
     return mnum;
 }
 
-
-FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
-                        ZSTD_match_t* matches,   /* store result (match found, increasing size) in this table */
-                        ZSTD_matchState_t* ms,
-                        U32* nextToUpdate3,
-                        const BYTE* ip, const BYTE* const iHighLimit, const ZSTD_dictMode_e dictMode,
-                        const U32 rep[ZSTD_REP_NUM],
-                        U32 const ll0,
-                        U32 const lengthToBeat)
+typedef U32 (*ZSTD_getAllMatchesFn)(
+    ZSTD_match_t*,
+    ZSTD_matchState_t*,
+    U32*,
+    const BYTE*,
+    const BYTE*,
+    const U32 rep[ZSTD_REP_NUM],
+    U32 const ll0,
+    U32 const lengthToBeat);
+
+FORCE_INLINE_TEMPLATE U32 ZSTD_btGetAllMatches_internal(
+        ZSTD_match_t* matches,
+        ZSTD_matchState_t* ms,
+        U32* nextToUpdate3,
+        const BYTE* ip,
+        const BYTE* const iHighLimit,
+        const U32 rep[ZSTD_REP_NUM],
+        U32 const ll0,
+        U32 const lengthToBeat,
+        const ZSTD_dictMode_e dictMode,
+        const U32 mls)
 {
-    const ZSTD_compressionParameters* const cParams = &ms->cParams;
-    U32 const matchLengthSearch = cParams->minMatch;
-    DEBUGLOG(8, "ZSTD_BtGetAllMatches");
-    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree_internal(ms, ip, iHighLimit, matchLengthSearch, dictMode);
-    switch(matchLengthSearch)
-    {
-    case 3 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 3);
-    default :
-    case 4 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 4);
-    case 5 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 5);
-    case 7 :
-    case 6 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 6);
+    assert(BOUNDED(3, ms->cParams.minMatch, 6) == mls);
+    DEBUGLOG(8, "ZSTD_BtGetAllMatches(dictMode=%d, mls=%u)", (int)dictMode, mls);
+    if (ip < ms->window.base + ms->nextToUpdate)
+        return 0;   /* skipped area */
+    ZSTD_updateTree_internal(ms, ip, iHighLimit, mls, dictMode);
+    return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, mls);
+}
+
+#define ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls) ZSTD_btGetAllMatches_##dictMode##_##mls
+
+#define GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, mls)            \
+    static U32 ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, mls)(      \
+            ZSTD_match_t* matches,                             \
+            ZSTD_matchState_t* ms,                             \
+            U32* nextToUpdate3,                                \
+            const BYTE* ip,                                    \
+            const BYTE* const iHighLimit,                      \
+            const U32 rep[ZSTD_REP_NUM],                       \
+            U32 const ll0,                                     \
+            U32 const lengthToBeat)                            \
+    {                                                          \
+        return ZSTD_btGetAllMatches_internal(                  \
+                matches, ms, nextToUpdate3, ip, iHighLimit,    \
+                rep, ll0, lengthToBeat, ZSTD_##dictMode, mls); \
     }
+
+#define GEN_ZSTD_BT_GET_ALL_MATCHES(dictMode)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 3)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 4)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 5)  \
+    GEN_ZSTD_BT_GET_ALL_MATCHES_(dictMode, 6)
+
+GEN_ZSTD_BT_GET_ALL_MATCHES(noDict)
+GEN_ZSTD_BT_GET_ALL_MATCHES(extDict)
+GEN_ZSTD_BT_GET_ALL_MATCHES(dictMatchState)
+
+#define ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMode)  \
+    {                                            \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 3), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 4), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 5), \
+        ZSTD_BT_GET_ALL_MATCHES_FN(dictMode, 6)  \
+    }
+
+static ZSTD_getAllMatchesFn
+ZSTD_selectBtGetAllMatches(ZSTD_matchState_t const* ms, ZSTD_dictMode_e const dictMode)
+{
+    ZSTD_getAllMatchesFn const getAllMatchesFns[3][4] = {
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(noDict),
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(extDict),
+        ZSTD_BT_GET_ALL_MATCHES_ARRAY(dictMatchState)
+    };
+    U32 const mls = BOUNDED(3, ms->cParams.minMatch, 6);
+    assert((U32)dictMode < 3);
+    assert(mls - 3 < 4);
+    return getAllMatchesFns[(int)dictMode][mls - 3];
 }
 
 /*************************
@@ -770,16 +892,18 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
 
 /* Struct containing info needed to make decision about ldm inclusion */
 typedef struct {
-    rawSeqStore_t seqStore;         /* External match candidates store for this block */
-    U32 startPosInBlock;            /* Start position of the current match candidate */
-    U32 endPosInBlock;              /* End position of the current match candidate */
-    U32 offset;                     /* Offset of the match candidate */
+    rawSeqStore_t seqStore;   /* External match candidates store for this block */
+    U32 startPosInBlock;      /* Start position of the current match candidate */
+    U32 endPosInBlock;        /* End position of the current match candidate */
+    U32 offset;               /* Offset of the match candidate */
 } ZSTD_optLdm_t;
 
 /* ZSTD_optLdm_skipRawSeqStoreBytes():
- * Moves forward in rawSeqStore by nbBytes, which will update the fields 'pos' and 'posInSequence'.
+ * Moves forward in @rawSeqStore by @nbBytes,
+ * which will update the fields 'pos' and 'posInSequence'.
  */
-static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
+static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes)
+{
     U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
     while (currPos && rawSeqStore->pos < rawSeqStore->size) {
         rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
@@ -800,8 +924,10 @@ static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t
  * Calculates the beginning and end of the next match in the current block.
  * Updates 'pos' and 'posInSequence' of the ldmSeqStore.
  */
-static void ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
-                                                   U32 blockBytesRemaining) {
+static void
+ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
+                                       U32 blockBytesRemaining)
+{
     rawSeq currSeq;
     U32 currBlockEndPos;
     U32 literalsBytesRemaining;
@@ -813,8 +939,8 @@ static void ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 cu
         optLdm->endPosInBlock = UINT_MAX;
         return;
     }
-    /* Calculate appropriate bytes left in matchLength and litLength after adjusting
-       based on ldmSeqStore->posInSequence */
+    /* Calculate appropriate bytes left in matchLength and litLength
+     * after adjusting based on ldmSeqStore->posInSequence */
     currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
     assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
     currBlockEndPos = currPosInBlock + blockBytesRemaining;
@@ -850,15 +976,16 @@ static void ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 cu
 }
 
 /* ZSTD_optLdm_maybeAddMatch():
- * Adds a match if it's long enough, based on it's 'matchStartPosInBlock'
- * and 'matchEndPosInBlock', into 'matches'. Maintains the correct ordering of 'matches'
+ * Adds a match if it's long enough,
+ * based on it's 'matchStartPosInBlock' and 'matchEndPosInBlock',
+ * into 'matches'. Maintains the correct ordering of 'matches'.
  */
 static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
-                                      ZSTD_optLdm_t* optLdm, U32 currPosInBlock) {
-    U32 posDiff = currPosInBlock - optLdm->startPosInBlock;
-    /* Note: ZSTD_match_t actually contains offCode and matchLength (before subtracting MINMATCH) */
-    U32 candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
-    U32 candidateOffCode = optLdm->offset + ZSTD_REP_MOVE;
+                                      const ZSTD_optLdm_t* optLdm, U32 currPosInBlock)
+{
+    U32 const posDiff = currPosInBlock - optLdm->startPosInBlock;
+    /* Note: ZSTD_match_t actually contains offBase and matchLength (before subtracting MINMATCH) */
+    U32 const candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
 
     /* Ensure that current block position is not outside of the match */
     if (currPosInBlock < optLdm->startPosInBlock
@@ -868,10 +995,11 @@ static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
     }
 
     if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
-        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offCode: %u matchLength %u) at block position=%u",
-                 candidateOffCode, candidateMatchLength, currPosInBlock);
+        U32 const candidateOffBase = OFFSET_TO_OFFBASE(optLdm->offset);
+        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offBase: %u matchLength %u) at block position=%u",
+                 candidateOffBase, candidateMatchLength, currPosInBlock);
         matches[*nbMatches].len = candidateMatchLength;
-        matches[*nbMatches].off = candidateOffCode;
+        matches[*nbMatches].off = candidateOffBase;
         (*nbMatches)++;
     }
 }
@@ -879,8 +1007,11 @@ static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
 /* ZSTD_optLdm_processMatchCandidate():
  * Wrapper function to update ldm seq store and call ldm functions as necessary.
  */
-static void ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm, ZSTD_match_t* matches, U32* nbMatches,
-                                              U32 currPosInBlock, U32 remainingBytes) {
+static void
+ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm,
+                                  ZSTD_match_t* matches, U32* nbMatches,
+                                  U32 currPosInBlock, U32 remainingBytes)
+{
     if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
         return;
     }
@@ -891,19 +1022,19 @@ static void ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm, ZSTD_match_
              * at the end of a match from the ldm seq store, and will often be some bytes
              * over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
              */
-            U32 posOvershoot = currPosInBlock - optLdm->endPosInBlock;
+            U32 const posOvershoot = currPosInBlock - optLdm->endPosInBlock;
             ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
-        } 
+        }
         ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
     }
     ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock);
 }
 
+
 /*-*******************************
 *  Optimal parser
 *********************************/
 
-
 static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
 {
     return sol.litlen + sol.mlen;
@@ -944,6 +1075,8 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
     const BYTE* const prefixStart = base + ms->window.dictLimit;
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
 
+    ZSTD_getAllMatchesFn getAllMatches = ZSTD_selectBtGetAllMatches(ms, dictMode);
+
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
     U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
     U32 nextToUpdate3 = ms->nextToUpdate;
@@ -971,7 +1104,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
         /* find first match */
         {   U32 const litlen = (U32)(ip - anchor);
             U32 const ll0 = !litlen;
-            U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
+            U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, ip, iend, rep, ll0, minMatch);
             ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
                                               (U32)(ip-istart), (U32)(iend - ip));
             if (!nbMatches) { ip++; continue; }
@@ -985,18 +1118,18 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
              * in every price. We include the literal length to avoid negative
              * prices when we subtract the previous literal length.
              */
-            opt[0].price = ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
+            opt[0].price = (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
 
             /* large match -> immediate encoding */
             {   U32 const maxML = matches[nbMatches-1].len;
-                U32 const maxOffset = matches[nbMatches-1].off;
-                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffCode=%u at cPos=%u => start new series",
-                            nbMatches, maxML, maxOffset, (U32)(ip-prefixStart));
+                U32 const maxOffBase = matches[nbMatches-1].off;
+                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffBase=%u at cPos=%u => start new series",
+                            nbMatches, maxML, maxOffBase, (U32)(ip-prefixStart));
 
                 if (maxML > sufficient_len) {
                     lastSequence.litlen = litlen;
                     lastSequence.mlen = maxML;
-                    lastSequence.off = maxOffset;
+                    lastSequence.off = maxOffBase;
                     DEBUGLOG(6, "large match (%u>%u), immediate encoding",
                                 maxML, sufficient_len);
                     cur = 0;
@@ -1005,24 +1138,25 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
             }   }
 
             /* set prices for first matches starting position == 0 */
-            {   U32 const literalsPrice = opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+            assert(opt[0].price >= 0);
+            {   U32 const literalsPrice = (U32)opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
                 U32 pos;
                 U32 matchNb;
                 for (pos = 1; pos < minMatch; pos++) {
                     opt[pos].price = ZSTD_MAX_PRICE;   /* mlen, litlen and price will be fixed during forward scanning */
                 }
                 for (matchNb = 0; matchNb < nbMatches; matchNb++) {
-                    U32 const offset = matches[matchNb].off;
+                    U32 const offBase = matches[matchNb].off;
                     U32 const end = matches[matchNb].len;
                     for ( ; pos <= end ; pos++ ) {
-                        U32 const matchPrice = ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
+                        U32 const matchPrice = ZSTD_getMatchPrice(offBase, pos, optStatePtr, optLevel);
                         U32 const sequencePrice = literalsPrice + matchPrice;
                         DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
-                                    pos, ZSTD_fCost(sequencePrice));
+                                    pos, ZSTD_fCost((int)sequencePrice));
                         opt[pos].mlen = pos;
-                        opt[pos].off = offset;
+                        opt[pos].off = offBase;
                         opt[pos].litlen = litlen;
-                        opt[pos].price = sequencePrice;
+                        opt[pos].price = (int)sequencePrice;
                 }   }
                 last_pos = pos-1;
             }
@@ -1037,9 +1171,9 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
             /* Fix current position with one literal if cheaper */
             {   U32 const litlen = (opt[cur-1].mlen == 0) ? opt[cur-1].litlen + 1 : 1;
                 int const price = opt[cur-1].price
-                                + ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
-                                + ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
-                                - ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
+                                + (int)ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
+                                + (int)ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
+                                - (int)ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
                 assert(price < 1000000000); /* overflow check */
                 if (price <= opt[cur].price) {
                     DEBUGLOG(7, "cPos:%zi==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
@@ -1065,7 +1199,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
             assert(cur >= opt[cur].mlen);
             if (opt[cur].mlen != 0) {
                 U32 const prev = cur - opt[cur].mlen;
-                repcodes_t newReps = ZSTD_updateRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
+                repcodes_t const newReps = ZSTD_newRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
                 ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
             } else {
                 ZSTD_memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
@@ -1082,11 +1216,12 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
             }
 
+            assert(opt[cur].price >= 0);
             {   U32 const ll0 = (opt[cur].mlen != 0);
                 U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
-                U32 const previousPrice = opt[cur].price;
+                U32 const previousPrice = (U32)opt[cur].price;
                 U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
-                U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
+                U32 nbMatches = getAllMatches(matches, ms, &nextToUpdate3, inr, iend, opt[cur].rep, ll0, minMatch);
                 U32 matchNb;
 
                 ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
@@ -1119,12 +1254,12 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                     U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
                     U32 mlen;
 
-                    DEBUGLOG(7, "testing match %u => offCode=%4u, mlen=%2u, llen=%2u",
+                    DEBUGLOG(7, "testing match %u => offBase=%4u, mlen=%2u, llen=%2u",
                                 matchNb, matches[matchNb].off, lastML, litlen);
 
                     for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
                         U32 const pos = cur + mlen;
-                        int const price = basePrice + ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+                        int const price = (int)basePrice + (int)ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
 
                         if ((pos > last_pos) || (price < opt[pos].price)) {
                             DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
@@ -1154,7 +1289,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
          * update them while traversing the sequences.
          */
         if (lastSequence.mlen != 0) {
-            repcodes_t reps = ZSTD_updateRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
+            repcodes_t const reps = ZSTD_newRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
             ZSTD_memcpy(rep, &reps, sizeof(reps));
         } else {
             ZSTD_memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
@@ -1185,7 +1320,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 for (storePos=storeStart; storePos <= storeEnd; storePos++) {
                     U32 const llen = opt[storePos].litlen;
                     U32 const mlen = opt[storePos].mlen;
-                    U32 const offCode = opt[storePos].off;
+                    U32 const offBase = opt[storePos].off;
                     U32 const advance = llen + mlen;
                     DEBUGLOG(6, "considering seq starting at %zi, llen=%u, mlen=%u",
                                 anchor - istart, (unsigned)llen, (unsigned)mlen);
@@ -1197,8 +1332,8 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                     }
 
                     assert(anchor + llen <= iend);
-                    ZSTD_updateStats(optStatePtr, llen, anchor, offCode, mlen);
-                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offCode, mlen-MINMATCH);
+                    ZSTD_updateStats(optStatePtr, llen, anchor, offBase, mlen);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offBase, mlen);
                     anchor += advance;
                     ip = anchor;
             }   }
@@ -1210,43 +1345,35 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
     return (size_t)(iend - anchor);
 }
 
+static size_t ZSTD_compressBlock_opt0(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /* optLevel */, dictMode);
+}
+
+static size_t ZSTD_compressBlock_opt2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize, const ZSTD_dictMode_e dictMode)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /* optLevel */, dictMode);
+}
 
 size_t ZSTD_compressBlock_btopt(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_compressBlock_btopt");
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_noDict);
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
 }
 
 
-/* used in 2-pass strategy */
-static U32 ZSTD_upscaleStat(unsigned* table, U32 lastEltIndex, int bonus)
-{
-    U32 s, sum=0;
-    assert(ZSTD_FREQ_DIV+bonus >= 0);
-    for (s=0; slitSum = ZSTD_upscaleStat(optPtr->litFreq, MaxLit, 0);
-    optPtr->litLengthSum = ZSTD_upscaleStat(optPtr->litLengthFreq, MaxLL, 0);
-    optPtr->matchLengthSum = ZSTD_upscaleStat(optPtr->matchLengthFreq, MaxML, 0);
-    optPtr->offCodeSum = ZSTD_upscaleStat(optPtr->offCodeFreq, MaxOff, 0);
-}
 
 /* ZSTD_initStats_ultra():
  * make a first compression pass, just to seed stats with more accurate starting values.
  * only works on first block, with no dictionary and no ldm.
- * this function cannot error, hence its contract must be respected.
+ * this function cannot error out, its narrow contract must be respected.
  */
 static void
 ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
@@ -1263,17 +1390,15 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
     assert(ms->window.dictLimit == ms->window.lowLimit);   /* no dictionary */
     assert(ms->window.dictLimit - ms->nextToUpdate <= 1);  /* no prefix (note: intentional overflow, defined as 2-complement) */
 
-    ZSTD_compressBlock_opt_generic(ms, seqStore, tmpRep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);   /* generate stats into ms->opt*/
+    ZSTD_compressBlock_opt2(ms, seqStore, tmpRep, src, srcSize, ZSTD_noDict);   /* generate stats into ms->opt*/
 
-    /* invalidate first scan from history */
+    /* invalidate first scan from history, only keep entropy stats */
     ZSTD_resetSeqStore(seqStore);
     ms->window.base -= srcSize;
     ms->window.dictLimit += (U32)srcSize;
     ms->window.lowLimit = ms->window.dictLimit;
     ms->nextToUpdate = ms->window.dictLimit;
 
-    /* re-inforce weight of collected statistics */
-    ZSTD_upscaleStats(&ms->opt);
 }
 
 size_t ZSTD_compressBlock_btultra(
@@ -1281,7 +1406,7 @@ size_t ZSTD_compressBlock_btultra(
         const void* src, size_t srcSize)
 {
     DEBUGLOG(5, "ZSTD_compressBlock_btultra (srcSize=%zu)", srcSize);
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
 }
 
 size_t ZSTD_compressBlock_btultra2(
@@ -1291,53 +1416,53 @@ size_t ZSTD_compressBlock_btultra2(
     U32 const curr = (U32)((const BYTE*)src - ms->window.base);
     DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
 
-    /* 2-pass strategy:
+    /* 2-passes strategy:
      * this strategy makes a first pass over first block to collect statistics
-     * and seed next round's statistics with it.
-     * After 1st pass, function forgets everything, and starts a new block.
+     * in order to seed next round's statistics with it.
+     * After 1st pass, function forgets history, and starts a new block.
      * Consequently, this can only work if no data has been previously loaded in tables,
      * aka, no dictionary, no prefix, no ldm preprocessing.
      * The compression ratio gain is generally small (~0.5% on first block),
-     * the cost is 2x cpu time on first block. */
+    ** the cost is 2x cpu time on first block. */
     assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
     if ( (ms->opt.litLengthSum==0)   /* first block */
       && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
       && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
-      && (curr == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
-      && (srcSize > ZSTD_PREDEF_THRESHOLD)
+      && (curr == ms->window.dictLimit)    /* start of frame, nothing already loaded nor skipped */
+      && (srcSize > ZSTD_PREDEF_THRESHOLD) /* input large enough to not employ default stats */
       ) {
         ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
     }
 
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_noDict);
 }
 
 size_t ZSTD_compressBlock_btopt_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
 }
 
 size_t ZSTD_compressBlock_btultra_dictMatchState(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_dictMatchState);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_dictMatchState);
 }
 
 size_t ZSTD_compressBlock_btopt_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_extDict);
+    return ZSTD_compressBlock_opt0(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
 }
 
 size_t ZSTD_compressBlock_btultra_extDict(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_extDict);
+    return ZSTD_compressBlock_opt2(ms, seqStore, rep, src, srcSize, ZSTD_extDict);
 }
 
 /* note : no btultra2 variant for extDict nor dictMatchState,
diff --git a/priv/zstd/lib/compress/zstd_opt.h b/priv/zstd/lib/compress/zstd_opt.h
index 627255f..342e5a3 100644
--- a/priv/zstd/lib/compress/zstd_opt.h
+++ b/priv/zstd/lib/compress/zstd_opt.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/compress/zstdmt_compress.c b/priv/zstd/lib/compress/zstdmt_compress.c
index 22aa3e1..7a2c717 100644
--- a/priv/zstd/lib/compress/zstdmt_compress.c
+++ b/priv/zstd/lib/compress/zstdmt_compress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -102,9 +102,8 @@ typedef struct ZSTDMT_bufferPool_s {
     buffer_t bTable[1];   /* variable size */
 } ZSTDMT_bufferPool;
 
-static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbWorkers, ZSTD_customMem cMem)
+static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned maxNbBuffers, ZSTD_customMem cMem)
 {
-    unsigned const maxNbBuffers = 2*nbWorkers + 3;
     ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_customCalloc(
         sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
     if (bufPool==NULL) return NULL;
@@ -160,9 +159,8 @@ static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const
 }
 
 
-static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, U32 nbWorkers)
+static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, unsigned maxNbBuffers)
 {
-    unsigned const maxNbBuffers = 2*nbWorkers + 3;
     if (srcBufPool==NULL) return NULL;
     if (srcBufPool->totalBuffers >= maxNbBuffers) /* good enough */
         return srcBufPool;
@@ -171,7 +169,7 @@ static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool,
         size_t const bSize = srcBufPool->bufferSize;   /* forward parameters */
         ZSTDMT_bufferPool* newBufPool;
         ZSTDMT_freeBufferPool(srcBufPool);
-        newBufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+        newBufPool = ZSTDMT_createBufferPool(maxNbBuffers, cMem);
         if (newBufPool==NULL) return newBufPool;
         ZSTDMT_setBufferSize(newBufPool, bSize);
         return newBufPool;
@@ -263,6 +261,16 @@ static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
     ZSTD_customFree(buf.start, bufPool->cMem);
 }
 
+/* We need 2 output buffers per worker since each dstBuff must be flushed after it is released.
+ * The 3 additional buffers are as follows:
+ *   1 buffer for input loading
+ *   1 buffer for "next input" when submitting current one
+ *   1 buffer stuck in queue */
+#define BUF_POOL_MAX_NB_BUFFERS(nbWorkers) (2*(nbWorkers) + 3)
+
+/* After a worker releases its rawSeqStore, it is immediately ready for reuse.
+ * So we only need one seq buffer per worker. */
+#define SEQ_POOL_MAX_NB_BUFFERS(nbWorkers) (nbWorkers)
 
 /* =====   Seq Pool Wrapper   ====== */
 
@@ -316,7 +324,7 @@ static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
 
 static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
 {
-    ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(SEQ_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
     if (seqPool == NULL) return NULL;
     ZSTDMT_setNbSeq(seqPool, 0);
     return seqPool;
@@ -329,7 +337,7 @@ static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
 
 static ZSTDMT_seqPool* ZSTDMT_expandSeqPool(ZSTDMT_seqPool* pool, U32 nbWorkers)
 {
-    return ZSTDMT_expandBufferPool(pool, nbWorkers);
+    return ZSTDMT_expandBufferPool(pool, SEQ_POOL_MAX_NB_BUFFERS(nbWorkers));
 }
 
 
@@ -467,7 +475,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
                          ZSTD_dictContentType_e dictContentType)
 {
     /* Adjust parameters */
-    if (params.ldmParams.enableLdm) {
+    if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
         DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
         ZSTD_ldm_adjustParameters(¶ms.ldmParams, ¶ms.cParams);
         assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
@@ -478,7 +486,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
     serialState->nextJobID = 0;
     if (params.fParams.checksumFlag)
         XXH64_reset(&serialState->xxhState, 0);
-    if (params.ldmParams.enableLdm) {
+    if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
         ZSTD_customMem cMem = params.customMem;
         unsigned const hashLog = params.ldmParams.hashLog;
         size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
@@ -564,7 +572,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
     /* A future job may error and skip our job */
     if (serialState->nextJobID == jobID) {
         /* It is now our turn, do any processing necessary */
-        if (serialState->params.ldmParams.enableLdm) {
+        if (serialState->params.ldmParams.enableLdm == ZSTD_ps_enable) {
             size_t error;
             assert(seqStore.seq != NULL && seqStore.pos == 0 &&
                    seqStore.size == 0 && seqStore.capacity > 0);
@@ -594,7 +602,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
     if (seqStore.size > 0) {
         size_t const err = ZSTD_referenceExternalSequences(
             jobCCtx, seqStore.seq, seqStore.size);
-        assert(serialState->params.ldmParams.enableLdm);
+        assert(serialState->params.ldmParams.enableLdm == ZSTD_ps_enable);
         assert(!ZSTD_isError(err));
         (void)err;
     }
@@ -672,7 +680,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
         if (dstBuff.start==NULL) JOB_ERROR(ERROR(memory_allocation));
         job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
     }
-    if (jobParams.ldmParams.enableLdm && rawSeqStore.seq == NULL)
+    if (jobParams.ldmParams.enableLdm == ZSTD_ps_enable && rawSeqStore.seq == NULL)
         JOB_ERROR(ERROR(memory_allocation));
 
     /* Don't compute the checksum for chunks, since we compute it externally,
@@ -680,7 +688,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
      */
     if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
     /* Don't run LDM for the chunks, since we handle it externally */
-    jobParams.ldmParams.enableLdm = 0;
+    jobParams.ldmParams.enableLdm = ZSTD_ps_disable;
     /* Correct nbWorkers to 0. */
     jobParams.nbWorkers = 0;
 
@@ -807,6 +815,15 @@ typedef struct {
 static const roundBuff_t kNullRoundBuff = {NULL, 0, 0};
 
 #define RSYNC_LENGTH 32
+/* Don't create chunks smaller than the zstd block size.
+ * This stops us from regressing compression ratio too much,
+ * and ensures our output fits in ZSTD_compressBound().
+ *
+ * If this is shrunk < ZSTD_BLOCKSIZELOG_MIN then
+ * ZSTD_COMPRESSBOUND() will need to be updated.
+ */
+#define RSYNC_MIN_BLOCK_LOG ZSTD_BLOCKSIZELOG_MAX
+#define RSYNC_MIN_BLOCK_SIZE (1<jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
     assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0);  /* ensure nbJobs is a power of 2 */
     mtctx->jobIDMask = nbJobs - 1;
-    mtctx->bufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    mtctx->bufPool = ZSTDMT_createBufferPool(BUF_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
     mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
     mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
     initError = ZSTDMT_serialState_init(&mtctx->serial);
@@ -1030,7 +1047,7 @@ static size_t ZSTDMT_resize(ZSTDMT_CCtx* mtctx, unsigned nbWorkers)
 {
     if (POOL_resize(mtctx->factory, nbWorkers)) return ERROR(memory_allocation);
     FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbWorkers) , "");
-    mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, nbWorkers);
+    mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, BUF_POOL_MAX_NB_BUFFERS(nbWorkers));
     if (mtctx->bufPool == NULL) return ERROR(memory_allocation);
     mtctx->cctxPool = ZSTDMT_expandCCtxPool(mtctx->cctxPool, nbWorkers);
     if (mtctx->cctxPool == NULL) return ERROR(memory_allocation);
@@ -1135,7 +1152,7 @@ size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
 static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
 {
     unsigned jobLog;
-    if (params->ldmParams.enableLdm) {
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
         /* In Long Range Mode, the windowLog is typically oversized.
          * In which case, it's preferable to determine the jobSize
          * based on cycleLog instead. */
@@ -1179,7 +1196,7 @@ static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
     int const overlapRLog = 9 - ZSTDMT_overlapLog(params->overlapLog, params->cParams.strategy);
     int ovLog = (overlapRLog >= 8) ? 0 : (params->cParams.windowLog - overlapRLog);
     assert(0 <= overlapRLog && overlapRLog <= 8);
-    if (params->ldmParams.enableLdm) {
+    if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
         /* In Long Range Mode, the windowLog is typically oversized.
          * In which case, it's preferable to determine the jobSize
          * based on chainLog instead.
@@ -1252,6 +1269,9 @@ size_t ZSTDMT_initCStream_internal(
         /* Aim for the targetsectionSize as the average job size. */
         U32 const jobSizeKB = (U32)(mtctx->targetSectionSize >> 10);
         U32 const rsyncBits = (assert(jobSizeKB >= 1), ZSTD_highbit32(jobSizeKB) + 10);
+        /* We refuse to create jobs < RSYNC_MIN_BLOCK_SIZE bytes, so make sure our
+         * expected job size is at least 4x larger. */
+        assert(rsyncBits >= RSYNC_MIN_BLOCK_LOG + 2);
         DEBUGLOG(4, "rsyncLog = %u", rsyncBits);
         mtctx->rsync.hash = 0;
         mtctx->rsync.hitMask = (1ULL << rsyncBits) - 1;
@@ -1263,7 +1283,7 @@ size_t ZSTDMT_initCStream_internal(
     ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
     {
         /* If ldm is enabled we need windowSize space. */
-        size_t const windowSize = mtctx->params.ldmParams.enableLdm ? (1U << mtctx->params.cParams.windowLog) : 0;
+        size_t const windowSize = mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable ? (1U << mtctx->params.cParams.windowLog) : 0;
         /* Two buffers of slack, plus extra space for the overlap
          * This is the minimum slack that LDM works with. One extra because
          * flush might waste up to targetSectionSize-1 bytes. Another extra
@@ -1538,17 +1558,21 @@ static range_t ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
 static int ZSTDMT_isOverlapped(buffer_t buffer, range_t range)
 {
     BYTE const* const bufferStart = (BYTE const*)buffer.start;
-    BYTE const* const bufferEnd = bufferStart + buffer.capacity;
     BYTE const* const rangeStart = (BYTE const*)range.start;
-    BYTE const* const rangeEnd = range.size != 0 ? rangeStart + range.size : rangeStart;
 
     if (rangeStart == NULL || bufferStart == NULL)
         return 0;
-    /* Empty ranges cannot overlap */
-    if (bufferStart == bufferEnd || rangeStart == rangeEnd)
-        return 0;
 
-    return bufferStart < rangeEnd && rangeStart < bufferEnd;
+    {
+        BYTE const* const bufferEnd = bufferStart + buffer.capacity;
+        BYTE const* const rangeEnd = rangeStart + range.size;
+
+        /* Empty ranges cannot overlap */
+        if (bufferStart == bufferEnd || rangeStart == rangeEnd)
+            return 0;
+
+        return bufferStart < rangeEnd && rangeStart < bufferEnd;
+    }
 }
 
 static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
@@ -1575,7 +1599,7 @@ static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
 
 static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
 {
-    if (mtctx->params.ldmParams.enableLdm) {
+    if (mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable) {
         ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
         DEBUGLOG(5, "ZSTDMT_waitForLdmComplete");
         DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
@@ -1678,6 +1702,11 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
     if (!mtctx->params.rsyncable)
         /* Rsync is disabled. */
         return syncPoint;
+    if (mtctx->inBuff.filled + input.size - input.pos < RSYNC_MIN_BLOCK_SIZE)
+        /* We don't emit synchronization points if it would produce too small blocks.
+         * We don't have enough input to find a synchronization point, so don't look.
+         */
+        return syncPoint;
     if (mtctx->inBuff.filled + syncPoint.toLoad < RSYNC_LENGTH)
         /* Not enough to compute the hash.
          * We will miss any synchronization points in this RSYNC_LENGTH byte
@@ -1688,10 +1717,28 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
          */
         return syncPoint;
     /* Initialize the loop variables. */
-    if (mtctx->inBuff.filled >= RSYNC_LENGTH) {
-        /* We have enough bytes buffered to initialize the hash.
+    if (mtctx->inBuff.filled < RSYNC_MIN_BLOCK_SIZE) {
+        /* We don't need to scan the first RSYNC_MIN_BLOCK_SIZE positions
+         * because they can't possibly be a sync point. So we can start
+         * part way through the input buffer.
+         */
+        pos = RSYNC_MIN_BLOCK_SIZE - mtctx->inBuff.filled;
+        if (pos >= RSYNC_LENGTH) {
+            prev = istart + pos - RSYNC_LENGTH;
+            hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
+        } else {
+            assert(mtctx->inBuff.filled >= RSYNC_LENGTH);
+            prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
+            hash = ZSTD_rollingHash_compute(prev + pos, (RSYNC_LENGTH - pos));
+            hash = ZSTD_rollingHash_append(hash, istart, pos);
+        }
+    } else {
+        /* We have enough bytes buffered to initialize the hash,
+         * and have processed enough bytes to find a sync point.
          * Start scanning at the beginning of the input.
          */
+        assert(mtctx->inBuff.filled >= RSYNC_MIN_BLOCK_SIZE);
+        assert(RSYNC_MIN_BLOCK_SIZE >= RSYNC_LENGTH);
         pos = 0;
         prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
         hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
@@ -1705,16 +1752,6 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
             syncPoint.flush = 1;
             return syncPoint;
         }
-    } else {
-        /* We don't have enough bytes buffered to initialize the hash, but
-         * we know we have at least RSYNC_LENGTH bytes total.
-         * Start scanning after the first RSYNC_LENGTH bytes less the bytes
-         * already buffered.
-         */
-        pos = RSYNC_LENGTH - mtctx->inBuff.filled;
-        prev = (BYTE const*)mtctx->inBuff.buffer.start - pos;
-        hash = ZSTD_rollingHash_compute(mtctx->inBuff.buffer.start, mtctx->inBuff.filled);
-        hash = ZSTD_rollingHash_append(hash, istart, pos);
     }
     /* Starting with the hash of the previous RSYNC_LENGTH bytes, roll
      * through the input. If we hit a synchronization point, then cut the
@@ -1724,16 +1761,24 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
      * then a block will be emitted anyways, but this is okay, since if we
      * are already synchronized we will remain synchronized.
      */
+    assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
     for (; pos < syncPoint.toLoad; ++pos) {
         BYTE const toRemove = pos < RSYNC_LENGTH ? prev[pos] : istart[pos - RSYNC_LENGTH];
-        /* if (pos >= RSYNC_LENGTH) assert(ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash); */
+        /* This assert is very expensive, and Debian compiles with asserts enabled.
+         * So disable it for now. We can get similar coverage by checking it at the
+         * beginning & end of the loop.
+         * assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
+         */
         hash = ZSTD_rollingHash_rotate(hash, toRemove, istart[pos], primePower);
+        assert(mtctx->inBuff.filled + pos >= RSYNC_MIN_BLOCK_SIZE);
         if ((hash & hitMask) == hitMask) {
             syncPoint.toLoad = pos + 1;
             syncPoint.flush = 1;
+            ++pos; /* for assert */
             break;
         }
     }
+    assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
     return syncPoint;
 }
 
diff --git a/priv/zstd/lib/compress/zstdmt_compress.h b/priv/zstd/lib/compress/zstdmt_compress.h
index 2fee2ec..ed4dc0e 100644
--- a/priv/zstd/lib/compress/zstdmt_compress.h
+++ b/priv/zstd/lib/compress/zstdmt_compress.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -65,8 +65,11 @@ size_t ZSTDMT_nextInputSizeHint(const ZSTDMT_CCtx* mtctx);
  *  Private use only. Init streaming operation.
  *  expects params to be valid.
  *  must receive dict, or cdict, or none, but not both.
+ *  mtctx can be freshly constructed or reused from a prior compression.
+ *  If mtctx is reused, memory allocations from the prior compression may not be freed,
+ *  even if they are not needed for the current compression.
  *  @return : 0, or an error code */
-size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
+size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* mtctx,
                     const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
                     const ZSTD_CDict* cdict,
                     ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
diff --git a/priv/zstd/lib/decompress/huf_decompress.c b/priv/zstd/lib/decompress/huf_decompress.c
index b93c9a0..c2d1f63 100644
--- a/priv/zstd/lib/decompress/huf_decompress.c
+++ b/priv/zstd/lib/decompress/huf_decompress.c
@@ -1,7 +1,7 @@
 /* ******************************************************************
  * huff0 huffman decoder,
  * part of Finite State Entropy library
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  *
  *  You can contact the author at :
  *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
@@ -19,9 +19,16 @@
 #include "../common/compiler.h"
 #include "../common/bitstream.h"  /* BIT_* */
 #include "../common/fse.h"        /* to compress headers */
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/error_private.h"
+#include "../common/zstd_internal.h"
+#include "../common/bits.h"       /* ZSTD_highbit32, ZSTD_countTrailingZeros64 */
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+
+#define HUF_DECODER_FAST_TABLELOG 11
 
 /* **************************************************************
 *  Macros
@@ -36,6 +43,28 @@
 #error "Cannot force the use of the X1 and X2 decoders at the same time!"
 #endif
 
+/* When DYNAMIC_BMI2 is enabled, fast decoders are only called when bmi2 is
+ * supported at runtime, so we can add the BMI2 target attribute.
+ * When it is disabled, we will still get BMI2 if it is enabled statically.
+ */
+#if DYNAMIC_BMI2
+# define HUF_FAST_BMI2_ATTRS BMI2_TARGET_ATTRIBUTE
+#else
+# define HUF_FAST_BMI2_ATTRS
+#endif
+
+#ifdef __cplusplus
+# define HUF_EXTERN_C extern "C"
+#else
+# define HUF_EXTERN_C
+#endif
+#define HUF_ASM_DECL HUF_EXTERN_C
+
+#if DYNAMIC_BMI2
+# define HUF_NEED_BMI2_FUNCTION 1
+#else
+# define HUF_NEED_BMI2_FUNCTION 0
+#endif
 
 /* **************************************************************
 *  Error Management
@@ -53,6 +82,11 @@
 /* **************************************************************
 *  BMI2 Variant Wrappers
 ****************************************************************/
+typedef size_t (*HUF_DecompressUsingDTableFn)(void *dst, size_t dstSize,
+                                              const void *cSrc,
+                                              size_t cSrcSize,
+                                              const HUF_DTable *DTable);
+
 #if DYNAMIC_BMI2
 
 #define HUF_DGEN(fn)                                                        \
@@ -65,7 +99,7 @@
         return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
     }                                                                       \
                                                                             \
-    static TARGET_ATTRIBUTE("bmi2") size_t fn##_bmi2(                       \
+    static BMI2_TARGET_ATTRIBUTE size_t fn##_bmi2(                          \
                   void* dst,  size_t dstSize,                               \
             const void* cSrc, size_t cSrcSize,                              \
             const HUF_DTable* DTable)                                       \
@@ -74,9 +108,9 @@
     }                                                                       \
                                                                             \
     static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
-                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
     {                                                                       \
-        if (bmi2) {                                                         \
+        if (flags & HUF_flags_bmi2) {                                       \
             return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
         }                                                                   \
         return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
@@ -86,9 +120,9 @@
 
 #define HUF_DGEN(fn)                                                        \
     static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
-                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+                     size_t cSrcSize, HUF_DTable const* DTable, int flags)  \
     {                                                                       \
-        (void)bmi2;                                                         \
+        (void)flags;                                                        \
         return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
     }
 
@@ -107,13 +141,164 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
     return dtd;
 }
 
+static size_t HUF_initFastDStream(BYTE const* ip) {
+    BYTE const lastByte = ip[7];
+    size_t const bitsConsumed = lastByte ? 8 - ZSTD_highbit32(lastByte) : 0;
+    size_t const value = MEM_readLEST(ip) | 1;
+    assert(bitsConsumed <= 8);
+    assert(sizeof(size_t) == 8);
+    return value << bitsConsumed;
+}
+
+
+/**
+ * The input/output arguments to the Huffman fast decoding loop:
+ *
+ * ip [in/out] - The input pointers, must be updated to reflect what is consumed.
+ * op [in/out] - The output pointers, must be updated to reflect what is written.
+ * bits [in/out] - The bitstream containers, must be updated to reflect the current state.
+ * dt [in] - The decoding table.
+ * ilimit [in] - The input limit, stop when any input pointer is below ilimit.
+ * oend [in] - The end of the output stream. op[3] must not cross oend.
+ * iend [in] - The end of each input stream. ip[i] may cross iend[i],
+ *             as long as it is above ilimit, but that indicates corruption.
+ */
+typedef struct {
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U64 bits[4];
+    void const* dt;
+    BYTE const* ilimit;
+    BYTE* oend;
+    BYTE const* iend[4];
+} HUF_DecompressFastArgs;
+
+typedef void (*HUF_DecompressFastLoopFn)(HUF_DecompressFastArgs*);
+
+/**
+ * Initializes args for the fast decoding loop.
+ * @returns 1 on success
+ *          0 if the fallback implementation should be used.
+ *          Or an error code on failure.
+ */
+static size_t HUF_DecompressFastArgs_init(HUF_DecompressFastArgs* args, void* dst, size_t dstSize, void const* src, size_t srcSize, const HUF_DTable* DTable)
+{
+    void const* dt = DTable + 1;
+    U32 const dtLog = HUF_getDTableDesc(DTable).tableLog;
+
+    const BYTE* const ilimit = (const BYTE*)src + 6 + 8;
+
+    BYTE* const oend = (BYTE*)dst + dstSize;
+
+    /* The fast decoding loop assumes 64-bit little-endian.
+     * This condition is false on x32.
+     */
+    if (!MEM_isLittleEndian() || MEM_32bits())
+        return 0;
+
+    /* strict minimum : jump table + 1 byte per stream */
+    if (srcSize < 10)
+        return ERROR(corruption_detected);
+
+    /* Must have at least 8 bytes per stream because we don't handle initializing smaller bit containers.
+     * If table log is not correct at this point, fallback to the old decoder.
+     * On small inputs we don't have enough data to trigger the fast loop, so use the old decoder.
+     */
+    if (dtLog != HUF_DECODER_FAST_TABLELOG)
+        return 0;
+
+    /* Read the jump table. */
+    {
+        const BYTE* const istart = (const BYTE*)src;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = srcSize - (length1 + length2 + length3 + 6);
+        args->iend[0] = istart + 6;  /* jumpTable */
+        args->iend[1] = args->iend[0] + length1;
+        args->iend[2] = args->iend[1] + length2;
+        args->iend[3] = args->iend[2] + length3;
+
+        /* HUF_initFastDStream() requires this, and this small of an input
+         * won't benefit from the ASM loop anyways.
+         * length1 must be >= 16 so that ip[0] >= ilimit before the loop
+         * starts.
+         */
+        if (length1 < 16 || length2 < 8 || length3 < 8 || length4 < 8)
+            return 0;
+        if (length4 > srcSize) return ERROR(corruption_detected);   /* overflow */
+    }
+    /* ip[] contains the position that is currently loaded into bits[]. */
+    args->ip[0] = args->iend[1] - sizeof(U64);
+    args->ip[1] = args->iend[2] - sizeof(U64);
+    args->ip[2] = args->iend[3] - sizeof(U64);
+    args->ip[3] = (BYTE const*)src + srcSize - sizeof(U64);
+
+    /* op[] contains the output pointers. */
+    args->op[0] = (BYTE*)dst;
+    args->op[1] = args->op[0] + (dstSize+3)/4;
+    args->op[2] = args->op[1] + (dstSize+3)/4;
+    args->op[3] = args->op[2] + (dstSize+3)/4;
+
+    /* No point to call the ASM loop for tiny outputs. */
+    if (args->op[3] >= oend)
+        return 0;
+
+    /* bits[] is the bit container.
+        * It is read from the MSB down to the LSB.
+        * It is shifted left as it is read, and zeros are
+        * shifted in. After the lowest valid bit a 1 is
+        * set, so that CountTrailingZeros(bits[]) can be used
+        * to count how many bits we've consumed.
+        */
+    args->bits[0] = HUF_initFastDStream(args->ip[0]);
+    args->bits[1] = HUF_initFastDStream(args->ip[1]);
+    args->bits[2] = HUF_initFastDStream(args->ip[2]);
+    args->bits[3] = HUF_initFastDStream(args->ip[3]);
+
+    /* If ip[] >= ilimit, it is guaranteed to be safe to
+        * reload bits[]. It may be beyond its section, but is
+        * guaranteed to be valid (>= istart).
+        */
+    args->ilimit = ilimit;
+
+    args->oend = oend;
+    args->dt = dt;
+
+    return 1;
+}
+
+static size_t HUF_initRemainingDStream(BIT_DStream_t* bit, HUF_DecompressFastArgs const* args, int stream, BYTE* segmentEnd)
+{
+    /* Validate that we haven't overwritten. */
+    if (args->op[stream] > segmentEnd)
+        return ERROR(corruption_detected);
+    /* Validate that we haven't read beyond iend[].
+        * Note that ip[] may be < iend[] because the MSB is
+        * the next bit to read, and we may have consumed 100%
+        * of the stream, so down to iend[i] - 8 is valid.
+        */
+    if (args->ip[stream] < args->iend[stream] - 8)
+        return ERROR(corruption_detected);
+
+    /* Construct the BIT_DStream_t. */
+    assert(sizeof(size_t) == 8);
+    bit->bitContainer = MEM_readLEST(args->ip[stream]);
+    bit->bitsConsumed = ZSTD_countTrailingZeros64(args->bits[stream]);
+    bit->start = (const char*)args->iend[0];
+    bit->limitPtr = bit->start + sizeof(size_t);
+    bit->ptr = (const char*)args->ip[stream];
+
+    return 0;
+}
+
 
 #ifndef HUF_FORCE_DECOMPRESS_X2
 
 /*-***************************/
 /*  single-symbol decoding   */
 /*-***************************/
-typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1;   /* single-symbol decoding */
+typedef struct { BYTE nbBits; BYTE byte; } HUF_DEltX1;   /* single-symbol decoding */
 
 /**
  * Packs 4 HUF_DEltX1 structs into a U64. This is used to lay down 4 entries at
@@ -122,14 +307,45 @@ typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1;   /* single-symbol decodi
 static U64 HUF_DEltX1_set4(BYTE symbol, BYTE nbBits) {
     U64 D4;
     if (MEM_isLittleEndian()) {
-        D4 = symbol + (nbBits << 8);
+        D4 = (U64)((symbol << 8) + nbBits);
     } else {
-        D4 = (symbol << 8) + nbBits;
+        D4 = (U64)(symbol + (nbBits << 8));
     }
+    assert(D4 < (1U << 16));
     D4 *= 0x0001000100010001ULL;
     return D4;
 }
 
+/**
+ * Increase the tableLog to targetTableLog and rescales the stats.
+ * If tableLog > targetTableLog this is a no-op.
+ * @returns New tableLog
+ */
+static U32 HUF_rescaleStats(BYTE* huffWeight, U32* rankVal, U32 nbSymbols, U32 tableLog, U32 targetTableLog)
+{
+    if (tableLog > targetTableLog)
+        return tableLog;
+    if (tableLog < targetTableLog) {
+        U32 const scale = targetTableLog - tableLog;
+        U32 s;
+        /* Increase the weight for all non-zero probability symbols by scale. */
+        for (s = 0; s < nbSymbols; ++s) {
+            huffWeight[s] += (BYTE)((huffWeight[s] == 0) ? 0 : scale);
+        }
+        /* Update rankVal to reflect the new weights.
+         * All weights except 0 get moved to weight + scale.
+         * Weights [1, scale] are empty.
+         */
+        for (s = targetTableLog; s > scale; --s) {
+            rankVal[s] = rankVal[s - scale];
+        }
+        for (s = scale; s > 0; --s) {
+            rankVal[s] = 0;
+        }
+    }
+    return targetTableLog;
+}
+
 typedef struct {
         U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
         U32 rankStart[HUF_TABLELOG_ABSOLUTEMAX + 1];
@@ -138,13 +354,7 @@ typedef struct {
         BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];
 } HUF_ReadDTableX1_Workspace;
 
-
-size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
-{
-    return HUF_readDTableX1_wksp_bmi2(DTable, src, srcSize, workSpace, wkspSize, /* bmi2 */ 0);
-}
-
-size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize, int flags)
 {
     U32 tableLog = 0;
     U32 nbSymbols = 0;
@@ -159,11 +369,15 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
     DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
     /* ZSTD_memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
-    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), bmi2);
+    iSize = HUF_readStats_wksp(wksp->huffWeight, HUF_SYMBOLVALUE_MAX + 1, wksp->rankVal, &nbSymbols, &tableLog, src, srcSize, wksp->statsWksp, sizeof(wksp->statsWksp), flags);
     if (HUF_isError(iSize)) return iSize;
 
+
     /* Table header */
     {   DTableDesc dtd = HUF_getDTableDesc(DTable);
+        U32 const maxTableLog = dtd.maxTableLog + 1;
+        U32 const targetTableLog = MIN(maxTableLog, HUF_DECODER_FAST_TABLELOG);
+        tableLog = HUF_rescaleStats(wksp->huffWeight, wksp->rankVal, nbSymbols, tableLog, targetTableLog);
         if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, Huffman tree cannot fit in */
         dtd.tableType = 0;
         dtd.tableLog = (BYTE)tableLog;
@@ -182,9 +396,8 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
      * rankStart[0] is not filled because there are no entries in the table for
      * weight 0.
      */
-    {
-        int n;
-        int nextRankStart = 0;
+    {   int n;
+        U32 nextRankStart = 0;
         int const unroll = 4;
         int const nLimit = (int)nbSymbols - unroll + 1;
         for (n=0; n<(int)tableLog+1; n++) {
@@ -207,14 +420,13 @@ size_t HUF_readDTableX1_wksp_bmi2(HUF_DTable* DTable, const void* src, size_t sr
 
     /* fill DTable
      * We fill all entries of each weight in order.
-     * That way length is a constant for each iteration of the outter loop.
+     * That way length is a constant for each iteration of the outer loop.
      * We can switch based on the length to a different inner loop which is
      * optimized for that particular case.
      */
-    {
-        U32 w;
-        int symbol=wksp->rankVal[0];
-        int rankStart=0;
+    {   U32 w;
+        int symbol = wksp->rankVal[0];
+        int rankStart = 0;
         for (w=1; wrankVal[w];
             int const length = (1 << w) >> 1;
@@ -304,11 +516,15 @@ HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, cons
     BYTE* const pStart = p;
 
     /* up to 4 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
-        HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+    if ((pEnd - p) > 3) {
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
+            HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
+        }
+    } else {
+        BIT_reloadDStream(bitDPtr);
     }
 
     /* [0-3] symbols remaining */
@@ -320,7 +536,7 @@ HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, cons
     while (p < pEnd)
         HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
 
-    return pEnd-pStart;
+    return (size_t)(pEnd-pStart);
 }
 
 FORCE_INLINE_TEMPLATE size_t
@@ -346,6 +562,10 @@ HUF_decompress1X1_usingDTable_internal_body(
     return dstSize;
 }
 
+/* HUF_decompress4X1_usingDTable_internal_body():
+ * Conditions :
+ * @dstSize >= 6
+ */
 FORCE_INLINE_TEMPLATE size_t
 HUF_decompress4X1_usingDTable_internal_body(
           void* dst,  size_t dstSize,
@@ -388,33 +608,37 @@ HUF_decompress4X1_usingDTable_internal_body(
         U32 endSignal = 1;
 
         if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);      /* overflow */
+        if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
         CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
         CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
         CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
         CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
 
         /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
-        for ( ; (endSignal) & (op4 < olimit) ; ) {
-            HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
-            endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
-            endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
-            endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
-            endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+        if ((size_t)(oend - op4) >= sizeof(size_t)) {
+            for ( ; (endSignal) & (op4 < olimit) ; ) {
+                HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
+                HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
+                endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+            }
         }
 
         /* check corruption */
@@ -440,74 +664,232 @@ HUF_decompress4X1_usingDTable_internal_body(
     }
 }
 
+#if HUF_NEED_BMI2_FUNCTION
+static BMI2_TARGET_ATTRIBUTE
+size_t HUF_decompress4X1_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+#endif
 
-typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
-                                               const void *cSrc,
-                                               size_t cSrcSize,
-                                               const HUF_DTable *DTable);
+static
+size_t HUF_decompress4X1_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X1_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
 
-HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
-HUF_DGEN(HUF_decompress4X1_usingDTable_internal)
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
 
+HUF_ASM_DECL void HUF_decompress4X1_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
 
+#endif
 
-size_t HUF_decompress1X1_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X1_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
 {
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    U16 const* const dtable = (U16 const*)args->dt;
+    BYTE* const oend = args->oend;
+    BYTE const* const ilimit = args->ilimit;
+
+    /* Copy the arguments to local variables */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy(&ip, &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+        int symbol;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= (stream == 3 ? oend : op[stream + 1]));
+            assert(ip[stream] >= ilimit);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each iteration produces 5 output symbols per stream */
+            size_t const oiters = (size_t)(oend - op[3]) / 5;
+            /* Each iteration consumes up to 11 bits * 5 = 55 bits < 7 bytes
+             * per stream.
+             */
+            size_t const iiters = (size_t)(ip[0] - ilimit) / 7;
+            /* We can safely run iters iterations before running bounds checks */
+            size_t const iters = MIN(oiters, iiters);
+            size_t const symbols = iters * 5;
+
+            /* We can simply check that op[3] < olimit, instead of checking all
+             * of our bounds, since we can't hit the other bounds until we've run
+             * iters iterations, which only happens when op[3] == olimit.
+             */
+            olimit = op[3] + symbols;
+
+            /* Exit fast decoding loop once we get close to the end. */
+            if (op[3] + 20 > olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+        do {
+            /* Decode 5 symbols in each of the 4 streams */
+            for (symbol = 0; symbol < 5; ++symbol) {
+                for (stream = 0; stream < 4; ++stream) {
+                    int const index = (int)(bits[stream] >> 53);
+                    int const entry = (int)dtable[index];
+                    bits[stream] <<= (entry & 63);
+                    op[stream][symbol] = (BYTE)((entry >> 8) & 0xFF);
+                }
+            }
+            /* Reload the bitstreams */
+            for (stream = 0; stream < 4; ++stream) {
+                int const ctz = ZSTD_countTrailingZeros64(bits[stream]);
+                int const nbBits = ctz & 7;
+                int const nbBytes = ctz >> 3;
+                op[stream] += 5;
+                ip[stream] -= nbBytes;
+                bits[stream] = MEM_read64(ip[stream]) | 1;
+                bits[stream] <<= nbBits;
+            }
+        } while (op[3] < olimit);
+    }
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy(&args->ip, &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
 }
 
-size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
+/**
+ * @returns @p dstSize on success (>= 6)
+ *          0 if the fallback implementation should be used
+ *          An error if an error occurred
+ */
+static HUF_FAST_BMI2_ATTRS
+size_t
+HUF_decompress4X1_usingDTable_internal_fast(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn)
 {
-    const BYTE* ip = (const BYTE*) cSrc;
+    void const* dt = DTable + 1;
+    const BYTE* const iend = (const BYTE*)cSrc + 6;
+    BYTE* const oend = (BYTE*)dst + dstSize;
+    HUF_DecompressFastArgs args;
+    {   size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init fast loop args");
+        if (ret == 0)
+            return 0;
+    }
 
-    size_t const hSize = HUF_readDTableX1_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
-    if (HUF_isError(hSize)) return hSize;
-    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
-    ip += hSize; cSrcSize -= hSize;
+    assert(args.ip[0] >= args.ilimit);
+    loopFn(&args);
+
+    /* Our loop guarantees that ip[] >= ilimit and that we haven't
+    * overwritten any op[].
+    */
+    assert(args.ip[0] >= iend);
+    assert(args.ip[1] >= iend);
+    assert(args.ip[2] >= iend);
+    assert(args.ip[3] >= iend);
+    assert(args.op[3] <= oend);
+    (void)iend;
+
+    /* finish bit streams one by one. */
+    {   size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* segmentEnd = (BYTE*)dst;
+        int i;
+        for (i = 0; i < 4; ++i) {
+            BIT_DStream_t bit;
+            if (segmentSize <= (size_t)(oend - segmentEnd))
+                segmentEnd += segmentSize;
+            else
+                segmentEnd = oend;
+            FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
+            /* Decompress and validate that we've produced exactly the expected length. */
+            args.op[i] += HUF_decodeStreamX1(args.op[i], &bit, segmentEnd, (HUF_DEltX1 const*)dt, HUF_DECODER_FAST_TABLELOG);
+            if (args.op[i] != segmentEnd) return ERROR(corruption_detected);
+        }
+    }
 
-    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+    /* decoded size */
+    assert(dstSize != 0);
+    return dstSize;
 }
 
+HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
 
-size_t HUF_decompress4X1_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
+static size_t HUF_decompress4X1_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
 {
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X1_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X1_usingDTable_internal_fast_c_loop;
+
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X1_usingDTable_internal_bmi2;
+# if ZSTD_ENABLE_ASM_X86_64_BMI2
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+        }
+# endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+    }
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X1_usingDTable_internal_fast_asm_loop;
+    }
+#endif
+
+    if (!(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X1_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
 }
 
-static size_t HUF_decompress4X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+static size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize, int bmi2)
+                                   void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
 }
 
-size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
-}
-
-
 #endif /* HUF_FORCE_DECOMPRESS_X2 */
 
 
@@ -518,106 +900,226 @@ size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
 /* *************************/
 
 typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX2;  /* double-symbols decoding */
-typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+typedef struct { BYTE symbol; } sortedSymbol_t;
 typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
 typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
 
+/**
+ * Constructs a HUF_DEltX2 in a U32.
+ */
+static U32 HUF_buildDEltX2U32(U32 symbol, U32 nbBits, U32 baseSeq, int level)
+{
+    U32 seq;
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, sequence) == 0);
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, nbBits) == 2);
+    DEBUG_STATIC_ASSERT(offsetof(HUF_DEltX2, length) == 3);
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U32));
+    if (MEM_isLittleEndian()) {
+        seq = level == 1 ? symbol : (baseSeq + (symbol << 8));
+        return seq + (nbBits << 16) + ((U32)level << 24);
+    } else {
+        seq = level == 1 ? (symbol << 8) : ((baseSeq << 8) + symbol);
+        return (seq << 16) + (nbBits << 8) + (U32)level;
+    }
+}
 
-/* HUF_fillDTableX2Level2() :
- * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
-static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 consumed,
-                           const U32* rankValOrigin, const int minWeight,
-                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
-                           U32 nbBitsBaseline, U16 baseSeq, U32* wksp, size_t wkspSize)
+/**
+ * Constructs a HUF_DEltX2.
+ */
+static HUF_DEltX2 HUF_buildDEltX2(U32 symbol, U32 nbBits, U32 baseSeq, int level)
 {
     HUF_DEltX2 DElt;
-    U32* rankVal = wksp;
+    U32 const val = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
+    DEBUG_STATIC_ASSERT(sizeof(DElt) == sizeof(val));
+    ZSTD_memcpy(&DElt, &val, sizeof(val));
+    return DElt;
+}
 
-    assert(wkspSize >= HUF_TABLELOG_MAX + 1);
-    (void)wkspSize;
-    /* get pre-calculated rankVal */
-    ZSTD_memcpy(rankVal, rankValOrigin, sizeof(U32) * (HUF_TABLELOG_MAX + 1));
+/**
+ * Constructs 2 HUF_DEltX2s and packs them into a U64.
+ */
+static U64 HUF_buildDEltX2U64(U32 symbol, U32 nbBits, U16 baseSeq, int level)
+{
+    U32 DElt = HUF_buildDEltX2U32(symbol, nbBits, baseSeq, level);
+    return (U64)DElt + ((U64)DElt << 32);
+}
 
-    /* fill skipped values */
+/**
+ * Fills the DTable rank with all the symbols from [begin, end) that are each
+ * nbBits long.
+ *
+ * @param DTableRank The start of the rank in the DTable.
+ * @param begin The first symbol to fill (inclusive).
+ * @param end The last symbol to fill (exclusive).
+ * @param nbBits Each symbol is nbBits long.
+ * @param tableLog The table log.
+ * @param baseSeq If level == 1 { 0 } else { the first level symbol }
+ * @param level The level in the table. Must be 1 or 2.
+ */
+static void HUF_fillDTableX2ForWeight(
+    HUF_DEltX2* DTableRank,
+    sortedSymbol_t const* begin, sortedSymbol_t const* end,
+    U32 nbBits, U32 tableLog,
+    U16 baseSeq, int const level)
+{
+    U32 const length = 1U << ((tableLog - nbBits) & 0x1F /* quiet static-analyzer */);
+    const sortedSymbol_t* ptr;
+    assert(level >= 1 && level <= 2);
+    switch (length) {
+    case 1:
+        for (ptr = begin; ptr != end; ++ptr) {
+            HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
+            *DTableRank++ = DElt;
+        }
+        break;
+    case 2:
+        for (ptr = begin; ptr != end; ++ptr) {
+            HUF_DEltX2 const DElt = HUF_buildDEltX2(ptr->symbol, nbBits, baseSeq, level);
+            DTableRank[0] = DElt;
+            DTableRank[1] = DElt;
+            DTableRank += 2;
+        }
+        break;
+    case 4:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+            DTableRank += 4;
+        }
+        break;
+    case 8:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
+            DTableRank += 8;
+        }
+        break;
+    default:
+        for (ptr = begin; ptr != end; ++ptr) {
+            U64 const DEltX2 = HUF_buildDEltX2U64(ptr->symbol, nbBits, baseSeq, level);
+            HUF_DEltX2* const DTableRankEnd = DTableRank + length;
+            for (; DTableRank != DTableRankEnd; DTableRank += 8) {
+                ZSTD_memcpy(DTableRank + 0, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 2, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 4, &DEltX2, sizeof(DEltX2));
+                ZSTD_memcpy(DTableRank + 6, &DEltX2, sizeof(DEltX2));
+            }
+        }
+        break;
+    }
+}
+
+/* HUF_fillDTableX2Level2() :
+ * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
+static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 targetLog, const U32 consumedBits,
+                           const U32* rankVal, const int minWeight, const int maxWeight1,
+                           const sortedSymbol_t* sortedSymbols, U32 const* rankStart,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    /* Fill skipped values (all positions up to rankVal[minWeight]).
+     * These are positions only get a single symbol because the combined weight
+     * is too large.
+     */
     if (minWeight>1) {
-        U32 i, skipSize = rankVal[minWeight];
-        MEM_writeLE16(&(DElt.sequence), baseSeq);
-        DElt.nbBits   = (BYTE)(consumed);
-        DElt.length   = 1;
-        for (i = 0; i < skipSize; i++)
-            DTable[i] = DElt;
+        U32 const length = 1U << ((targetLog - consumedBits) & 0x1F /* quiet static-analyzer */);
+        U64 const DEltX2 = HUF_buildDEltX2U64(baseSeq, consumedBits, /* baseSeq */ 0, /* level */ 1);
+        int const skipSize = rankVal[minWeight];
+        assert(length > 1);
+        assert((U32)skipSize < length);
+        switch (length) {
+        case 2:
+            assert(skipSize == 1);
+            ZSTD_memcpy(DTable, &DEltX2, sizeof(DEltX2));
+            break;
+        case 4:
+            assert(skipSize <= 4);
+            ZSTD_memcpy(DTable + 0, &DEltX2, sizeof(DEltX2));
+            ZSTD_memcpy(DTable + 2, &DEltX2, sizeof(DEltX2));
+            break;
+        default:
+            {
+                int i;
+                for (i = 0; i < skipSize; i += 8) {
+                    ZSTD_memcpy(DTable + i + 0, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 2, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 4, &DEltX2, sizeof(DEltX2));
+                    ZSTD_memcpy(DTable + i + 6, &DEltX2, sizeof(DEltX2));
+                }
+            }
+        }
     }
 
-    /* fill DTable */
-    {   U32 s; for (s=0; s= 1 */
-
-            rankVal[weight] += length;
-    }   }
+    /* Fill each of the second level symbols by weight. */
+    {
+        int w;
+        for (w = minWeight; w < maxWeight1; ++w) {
+            int const begin = rankStart[w];
+            int const end = rankStart[w+1];
+            U32 const nbBits = nbBitsBaseline - w;
+            U32 const totalBits = nbBits + consumedBits;
+            HUF_fillDTableX2ForWeight(
+                DTable + rankVal[w],
+                sortedSymbols + begin, sortedSymbols + end,
+                totalBits, targetLog,
+                baseSeq, /* level */ 2);
+        }
+    }
 }
 
-
 static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
-                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
-                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
-                           const U32 nbBitsBaseline, U32* wksp, size_t wkspSize)
+                           const sortedSymbol_t* sortedList,
+                           const U32* rankStart, rankValCol_t* rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
 {
-    U32* rankVal = wksp;
+    U32* const rankVal = rankValOrigin[0];
     const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
     const U32 minBits  = nbBitsBaseline - maxWeight;
-    U32 s;
-
-    assert(wkspSize >= HUF_TABLELOG_MAX + 1);
-    wksp += HUF_TABLELOG_MAX + 1;
-    wkspSize -= HUF_TABLELOG_MAX + 1;
-
-    ZSTD_memcpy(rankVal, rankValOrigin, sizeof(U32) * (HUF_TABLELOG_MAX + 1));
-
-    /* fill DTable */
-    for (s=0; s= minBits) {   /* enough room for a second symbol */
-            U32 sortedRank;
+    int w;
+    int const wEnd = (int)maxWeight + 1;
+
+    /* Fill DTable in order of weight. */
+    for (w = 1; w < wEnd; ++w) {
+        int const begin = (int)rankStart[w];
+        int const end = (int)rankStart[w+1];
+        U32 const nbBits = nbBitsBaseline - w;
+
+        if (targetLog-nbBits >= minBits) {
+            /* Enough room for a second symbol. */
+            int start = rankVal[w];
+            U32 const length = 1U << ((targetLog - nbBits) & 0x1F /* quiet static-analyzer */);
             int minWeight = nbBits + scaleLog;
+            int s;
             if (minWeight < 1) minWeight = 1;
-            sortedRank = rankStart[minWeight];
-            HUF_fillDTableX2Level2(DTable+start, targetLog-nbBits, nbBits,
-                           rankValOrigin[nbBits], minWeight,
-                           sortedList+sortedRank, sortedListSize-sortedRank,
-                           nbBitsBaseline, symbol, wksp, wkspSize);
+            /* Fill the DTable for every symbol of weight w.
+             * These symbols get at least 1 second symbol.
+             */
+            for (s = begin; s != end; ++s) {
+                HUF_fillDTableX2Level2(
+                    DTable + start, targetLog, nbBits,
+                    rankValOrigin[nbBits], minWeight, wEnd,
+                    sortedList, rankStart,
+                    nbBitsBaseline, sortedList[s].symbol);
+                start += length;
+            }
         } else {
-            HUF_DEltX2 DElt;
-            MEM_writeLE16(&(DElt.sequence), symbol);
-            DElt.nbBits = (BYTE)(nbBits);
-            DElt.length = 1;
-            {   U32 const end = start + length;
-                U32 u;
-                for (u = start; u < end; u++) DTable[u] = DElt;
-        }   }
-        rankVal[weight] += length;
+            /* Only a single symbol. */
+            HUF_fillDTableX2ForWeight(
+                DTable + rankVal[w],
+                sortedList + begin, sortedList + end,
+                nbBits, targetLog,
+                /* baseSeq */ 0, /* level */ 1);
+        }
     }
 }
 
 typedef struct {
     rankValCol_t rankVal[HUF_TABLELOG_MAX];
     U32 rankStats[HUF_TABLELOG_MAX + 1];
-    U32 rankStart0[HUF_TABLELOG_MAX + 2];
+    U32 rankStart0[HUF_TABLELOG_MAX + 3];
     sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
     BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
     U32 calleeWksp[HUF_READ_STATS_WORKSPACE_SIZE_U32];
@@ -625,11 +1127,11 @@ typedef struct {
 
 size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
                        const void* src, size_t srcSize,
-                             void* workSpace, size_t wkspSize)
+                             void* workSpace, size_t wkspSize, int flags)
 {
-    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    U32 tableLog, maxW, nbSymbols;
     DTableDesc dtd = HUF_getDTableDesc(DTable);
-    U32 const maxTableLog = dtd.maxTableLog;
+    U32 maxTableLog = dtd.maxTableLog;
     size_t iSize;
     void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
     HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
@@ -647,11 +1149,12 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
     if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     /* ZSTD_memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
-    iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), /* bmi2 */ 0);
+    iSize = HUF_readStats_wksp(wksp->weightList, HUF_SYMBOLVALUE_MAX + 1, wksp->rankStats, &nbSymbols, &tableLog, src, srcSize, wksp->calleeWksp, sizeof(wksp->calleeWksp), flags);
     if (HUF_isError(iSize)) return iSize;
 
     /* check result */
     if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+    if (tableLog <= HUF_DECODER_FAST_TABLELOG && maxTableLog > HUF_DECODER_FAST_TABLELOG) maxTableLog = HUF_DECODER_FAST_TABLELOG;
 
     /* find maxWeight */
     for (maxW = tableLog; wksp->rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
@@ -664,7 +1167,7 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
             rankStart[w] = curr;
         }
         rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
-        sizeOfSort = nextRankStart;
+        rankStart[maxW+1] = nextRankStart;
     }
 
     /* sort symbols by weight */
@@ -673,7 +1176,6 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
             U32 const w = wksp->weightList[s];
             U32 const r = rankStart[w]++;
             wksp->sortedSymbol[r].symbol = (BYTE)s;
-            wksp->sortedSymbol[r].weight = (BYTE)w;
         }
         rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
     }
@@ -698,10 +1200,9 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
     }   }   }   }
 
     HUF_fillDTableX2(dt, maxTableLog,
-                   wksp->sortedSymbol, sizeOfSort,
+                   wksp->sortedSymbol,
                    wksp->rankStart0, wksp->rankVal, maxW,
-                   tableLog+1,
-                   wksp->calleeWksp, sizeof(wksp->calleeWksp) / sizeof(U32));
+                   tableLog+1);
 
     dtd.tableLog = (BYTE)maxTableLog;
     dtd.tableType = 1;
@@ -714,7 +1215,7 @@ FORCE_INLINE_TEMPLATE U32
 HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    ZSTD_memcpy(op, dt+val, 2);
+    ZSTD_memcpy(op, &dt[val].sequence, 2);
     BIT_skipBits(DStream, dt[val].nbBits);
     return dt[val].length;
 }
@@ -723,15 +1224,17 @@ FORCE_INLINE_TEMPLATE U32
 HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
-    ZSTD_memcpy(op, dt+val, 1);
-    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
-    else {
+    ZSTD_memcpy(op, &dt[val].sequence, 1);
+    if (dt[val].length==1) {
+        BIT_skipBits(DStream, dt[val].nbBits);
+    } else {
         if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
             BIT_skipBits(DStream, dt[val].nbBits);
             if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
                 /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
                 DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
-    }   }
+        }
+    }
     return 1;
 }
 
@@ -753,19 +1256,37 @@ HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
     BYTE* const pStart = p;
 
     /* up to 8 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
-        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    if ((size_t)(pEnd - p) >= sizeof(bitDPtr->bitContainer)) {
+        if (dtLog <= 11 && MEM_64bits()) {
+            /* up to 10 symbols at a time */
+            while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-9)) {
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            }
+        } else {
+            /* up to 8 symbols at a time */
+            while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
+                HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+                HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+            }
+        }
+    } else {
+        BIT_reloadDStream(bitDPtr);
     }
 
     /* closer to end : up to 2 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    if ((size_t)(pEnd - p) >= 2) {
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
 
-    while (p <= pEnd-2)
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+        while (p <= pEnd-2)
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+    }
 
     if (p < pEnd)
         p += HUF_decodeLastSymbolX2(p, bitDPtr, dt, dtLog);
@@ -800,6 +1321,10 @@ HUF_decompress1X2_usingDTable_internal_body(
     return dstSize;
 }
 
+/* HUF_decompress4X2_usingDTable_internal_body():
+ * Conditions:
+ * @dstSize >= 6
+ */
 FORCE_INLINE_TEMPLATE size_t
 HUF_decompress4X2_usingDTable_internal_body(
           void* dst,  size_t dstSize,
@@ -840,58 +1365,62 @@ HUF_decompress4X2_usingDTable_internal_body(
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
         U32 const dtLog = dtd.tableLog;
 
-        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        if (length4 > cSrcSize) return ERROR(corruption_detected);  /* overflow */
+        if (opStart4 > oend) return ERROR(corruption_detected);     /* overflow */
+        if (dstSize < 6) return ERROR(corruption_detected);         /* stream 4-split doesn't work */
         CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
         CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
         CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
         CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
 
         /* 16-32 symbols per loop (4-8 symbols per stream) */
-        for ( ; (endSignal) & (op4 < olimit); ) {
+        if ((size_t)(oend - op4) >= sizeof(size_t)) {
+            for ( ; (endSignal) & (op4 < olimit); ) {
 #if defined(__clang__) && (defined(__x86_64__) || defined(__i386__))
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
-            endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
-            endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
-            endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
-            endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+                endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+                endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+                endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
 #else
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
-            endSignal = (U32)LIKELY(
-                        (BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
-                      & (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
-                      & (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
-                      & (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+                HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+                HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+                HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+                HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+                endSignal = (U32)LIKELY((U32)
+                            (BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
+                        & (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
 #endif
+            }
         }
 
         /* check corruption */
@@ -915,68 +1444,281 @@ HUF_decompress4X2_usingDTable_internal_body(
     }
 }
 
-HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
-HUF_DGEN(HUF_decompress4X2_usingDTable_internal)
+#if HUF_NEED_BMI2_FUNCTION
+static BMI2_TARGET_ATTRIBUTE
+size_t HUF_decompress4X2_usingDTable_internal_bmi2(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+#endif
+
+static
+size_t HUF_decompress4X2_usingDTable_internal_default(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable) {
+    return HUF_decompress4X2_usingDTable_internal_body(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+HUF_ASM_DECL void HUF_decompress4X2_usingDTable_internal_fast_asm_loop(HUF_DecompressFastArgs* args) ZSTDLIB_HIDDEN;
 
-size_t HUF_decompress1X2_usingDTable(
+#endif
+
+static HUF_FAST_BMI2_ATTRS
+void HUF_decompress4X2_usingDTable_internal_fast_c_loop(HUF_DecompressFastArgs* args)
+{
+    U64 bits[4];
+    BYTE const* ip[4];
+    BYTE* op[4];
+    BYTE* oend[4];
+    HUF_DEltX2 const* const dtable = (HUF_DEltX2 const*)args->dt;
+    BYTE const* const ilimit = args->ilimit;
+
+    /* Copy the arguments to local registers. */
+    ZSTD_memcpy(&bits, &args->bits, sizeof(bits));
+    ZSTD_memcpy(&ip, &args->ip, sizeof(ip));
+    ZSTD_memcpy(&op, &args->op, sizeof(op));
+
+    oend[0] = op[1];
+    oend[1] = op[2];
+    oend[2] = op[3];
+    oend[3] = args->oend;
+
+    assert(MEM_isLittleEndian());
+    assert(!MEM_32bits());
+
+    for (;;) {
+        BYTE* olimit;
+        int stream;
+        int symbol;
+
+        /* Assert loop preconditions */
+#ifndef NDEBUG
+        for (stream = 0; stream < 4; ++stream) {
+            assert(op[stream] <= oend[stream]);
+            assert(ip[stream] >= ilimit);
+        }
+#endif
+        /* Compute olimit */
+        {
+            /* Each loop does 5 table lookups for each of the 4 streams.
+             * Each table lookup consumes up to 11 bits of input, and produces
+             * up to 2 bytes of output.
+             */
+            /* We can consume up to 7 bytes of input per iteration per stream.
+             * We also know that each input pointer is >= ip[0]. So we can run
+             * iters loops before running out of input.
+             */
+            size_t iters = (size_t)(ip[0] - ilimit) / 7;
+            /* Each iteration can produce up to 10 bytes of output per stream.
+             * Each output stream my advance at different rates. So take the
+             * minimum number of safe iterations among all the output streams.
+             */
+            for (stream = 0; stream < 4; ++stream) {
+                size_t const oiters = (size_t)(oend[stream] - op[stream]) / 10;
+                iters = MIN(iters, oiters);
+            }
+
+            /* Each iteration produces at least 5 output symbols. So until
+             * op[3] crosses olimit, we know we haven't executed iters
+             * iterations yet. This saves us maintaining an iters counter,
+             * at the expense of computing the remaining # of iterations
+             * more frequently.
+             */
+            olimit = op[3] + (iters * 5);
+
+            /* Exit the fast decoding loop if we are too close to the end. */
+            if (op[3] + 10 > olimit)
+                break;
+
+            /* Exit the decoding loop if any input pointer has crossed the
+             * previous one. This indicates corruption, and a precondition
+             * to our loop is that ip[i] >= ip[0].
+             */
+            for (stream = 1; stream < 4; ++stream) {
+                if (ip[stream] < ip[stream - 1])
+                    goto _out;
+            }
+        }
+
+#ifndef NDEBUG
+        for (stream = 1; stream < 4; ++stream) {
+            assert(ip[stream] >= ip[stream - 1]);
+        }
+#endif
+
+        do {
+            /* Do 5 table lookups for each of the first 3 streams */
+            for (symbol = 0; symbol < 5; ++symbol) {
+                for (stream = 0; stream < 3; ++stream) {
+                    int const index = (int)(bits[stream] >> 53);
+                    HUF_DEltX2 const entry = dtable[index];
+                    MEM_write16(op[stream], entry.sequence);
+                    bits[stream] <<= (entry.nbBits);
+                    op[stream] += (entry.length);
+                }
+            }
+            /* Do 1 table lookup from the final stream */
+            {
+                int const index = (int)(bits[3] >> 53);
+                HUF_DEltX2 const entry = dtable[index];
+                MEM_write16(op[3], entry.sequence);
+                bits[3] <<= (entry.nbBits);
+                op[3] += (entry.length);
+            }
+            /* Do 4 table lookups from the final stream & reload bitstreams */
+            for (stream = 0; stream < 4; ++stream) {
+                /* Do a table lookup from the final stream.
+                 * This is interleaved with the reloading to reduce register
+                 * pressure. This shouldn't be necessary, but compilers can
+                 * struggle with codegen with high register pressure.
+                 */
+                {
+                    int const index = (int)(bits[3] >> 53);
+                    HUF_DEltX2 const entry = dtable[index];
+                    MEM_write16(op[3], entry.sequence);
+                    bits[3] <<= (entry.nbBits);
+                    op[3] += (entry.length);
+                }
+                /* Reload the bistreams. The final bitstream must be reloaded
+                 * after the 5th symbol was decoded.
+                 */
+                {
+                    int const ctz = ZSTD_countTrailingZeros64(bits[stream]);
+                    int const nbBits = ctz & 7;
+                    int const nbBytes = ctz >> 3;
+                    ip[stream] -= nbBytes;
+                    bits[stream] = MEM_read64(ip[stream]) | 1;
+                    bits[stream] <<= nbBits;
+                }
+            }
+        } while (op[3] < olimit);
+    }
+
+_out:
+
+    /* Save the final values of each of the state variables back to args. */
+    ZSTD_memcpy(&args->bits, &bits, sizeof(bits));
+    ZSTD_memcpy(&args->ip, &ip, sizeof(ip));
+    ZSTD_memcpy(&args->op, &op, sizeof(op));
+}
+
+
+static HUF_FAST_BMI2_ATTRS size_t
+HUF_decompress4X2_usingDTable_internal_fast(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
+    const HUF_DTable* DTable,
+    HUF_DecompressFastLoopFn loopFn) {
+    void const* dt = DTable + 1;
+    const BYTE* const iend = (const BYTE*)cSrc + 6;
+    BYTE* const oend = (BYTE*)dst + dstSize;
+    HUF_DecompressFastArgs args;
+    {
+        size_t const ret = HUF_DecompressFastArgs_init(&args, dst, dstSize, cSrc, cSrcSize, DTable);
+        FORWARD_IF_ERROR(ret, "Failed to init asm args");
+        if (ret == 0)
+            return 0;
+    }
+
+    assert(args.ip[0] >= args.ilimit);
+    loopFn(&args);
+
+    /* note : op4 already verified within main loop */
+    assert(args.ip[0] >= iend);
+    assert(args.ip[1] >= iend);
+    assert(args.ip[2] >= iend);
+    assert(args.ip[3] >= iend);
+    assert(args.op[3] <= oend);
+    (void)iend;
+
+    /* finish bitStreams one by one */
+    {
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* segmentEnd = (BYTE*)dst;
+        int i;
+        for (i = 0; i < 4; ++i) {
+            BIT_DStream_t bit;
+            if (segmentSize <= (size_t)(oend - segmentEnd))
+                segmentEnd += segmentSize;
+            else
+                segmentEnd = oend;
+            FORWARD_IF_ERROR(HUF_initRemainingDStream(&bit, &args, i, segmentEnd), "corruption");
+            args.op[i] += HUF_decodeStreamX2(args.op[i], &bit, segmentEnd, (HUF_DEltX2 const*)dt, HUF_DECODER_FAST_TABLELOG);
+            if (args.op[i] != segmentEnd)
+                return ERROR(corruption_detected);
+        }
+    }
+
+    /* decoded size */
+    return dstSize;
+}
+
+static size_t HUF_decompress4X2_usingDTable_internal(void* dst, size_t dstSize, void const* cSrc,
+                    size_t cSrcSize, HUF_DTable const* DTable, int flags)
 {
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+    HUF_DecompressUsingDTableFn fallbackFn = HUF_decompress4X2_usingDTable_internal_default;
+    HUF_DecompressFastLoopFn loopFn = HUF_decompress4X2_usingDTable_internal_fast_c_loop;
+
+#if DYNAMIC_BMI2
+    if (flags & HUF_flags_bmi2) {
+        fallbackFn = HUF_decompress4X2_usingDTable_internal_bmi2;
+# if ZSTD_ENABLE_ASM_X86_64_BMI2
+        if (!(flags & HUF_flags_disableAsm)) {
+            loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+        }
+# endif
+    } else {
+        return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
+    }
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2 && defined(__BMI2__)
+    if (!(flags & HUF_flags_disableAsm)) {
+        loopFn = HUF_decompress4X2_usingDTable_internal_fast_asm_loop;
+    }
+#endif
+
+    if (!(flags & HUF_flags_disableFast)) {
+        size_t const ret = HUF_decompress4X2_usingDTable_internal_fast(dst, dstSize, cSrc, cSrcSize, DTable, loopFn);
+        if (ret != 0)
+            return ret;
+    }
+    return fallbackFn(dst, dstSize, cSrc, cSrcSize, DTable);
 }
 
+HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
+
 size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
+                                   void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
     size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize,
-                                               workSpace, wkspSize);
+                                               workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, flags);
 }
 
-
-size_t HUF_decompress4X2_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-}
-
-static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+static size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize, int bmi2)
+                                   void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
     size_t hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize,
-                                         workSpace, wkspSize);
+                                         workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
-}
-
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
 }
 
-
 #endif /* HUF_FORCE_DECOMPRESS_X1 */
 
 
@@ -984,66 +1726,28 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
 /* Universal decompression selectors */
 /* ***********************************/
 
-size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize,
-                                    const void* cSrc, size_t cSrcSize,
-                                    const HUF_DTable* DTable)
-{
-    DTableDesc const dtd = HUF_getDTableDesc(DTable);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-    (void)dtd;
-    assert(dtd.tableType == 0);
-    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-    (void)dtd;
-    assert(dtd.tableType == 1);
-    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#else
-    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
-                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#endif
-}
-
-size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
-                                    const void* cSrc, size_t cSrcSize,
-                                    const HUF_DTable* DTable)
-{
-    DTableDesc const dtd = HUF_getDTableDesc(DTable);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-    (void)dtd;
-    assert(dtd.tableType == 0);
-    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-    (void)dtd;
-    assert(dtd.tableType == 1);
-    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#else
-    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
-                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
-#endif
-}
-
 
 #if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
 typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
-static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+static const algo_time_t algoTime[16 /* Quantization */][2 /* single, double */] =
 {
     /* single, double, quad */
-    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
-    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
-    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
-    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
-    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
-    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
-    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
-    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
-    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
-    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
-    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
-    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
-    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
-    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
-    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
-    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+    {{0,0}, {1,1}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}},  /* Q==1 : impossible */
+    {{ 150,216}, { 381,119}},   /* Q == 2 : 12-18% */
+    {{ 170,205}, { 514,112}},   /* Q == 3 : 18-25% */
+    {{ 177,199}, { 539,110}},   /* Q == 4 : 25-32% */
+    {{ 197,194}, { 644,107}},   /* Q == 5 : 32-38% */
+    {{ 221,192}, { 735,107}},   /* Q == 6 : 38-44% */
+    {{ 256,189}, { 881,106}},   /* Q == 7 : 44-50% */
+    {{ 359,188}, {1167,109}},   /* Q == 8 : 50-56% */
+    {{ 582,187}, {1570,114}},   /* Q == 9 : 56-62% */
+    {{ 688,187}, {1712,122}},   /* Q ==10 : 62-69% */
+    {{ 825,186}, {1965,136}},   /* Q ==11 : 69-75% */
+    {{ 976,185}, {2131,150}},   /* Q ==12 : 75-81% */
+    {{1180,186}, {2070,175}},   /* Q ==13 : 81-87% */
+    {{1377,185}, {1731,202}},   /* Q ==14 : 87-93% */
+    {{1412,185}, {1695,202}},   /* Q ==15 : 93-99% */
 };
 #endif
 
@@ -1070,42 +1774,15 @@ U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
         U32 const D256 = (U32)(dstSize >> 8);
         U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
         U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
-        DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, to reduce cache eviction */
+        DTime1 += DTime1 >> 5;  /* small advantage to algorithm using less memory, to reduce cache eviction */
         return DTime1 < DTime0;
     }
 #endif
 }
 
-
-size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
-                                     size_t dstSize, const void* cSrc,
-                                     size_t cSrcSize, void* workSpace,
-                                     size_t wkspSize)
-{
-    /* validation checks */
-    if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if (cSrcSize == 0) return ERROR(corruption_detected);
-
-    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-        (void)algoNb;
-        assert(algoNb == 0);
-        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-        (void)algoNb;
-        assert(algoNb == 1);
-        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
-#else
-        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                            cSrcSize, workSpace, wkspSize):
-                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
-#endif
-    }
-}
-
 size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
                                   const void* cSrc, size_t cSrcSize,
-                                  void* workSpace, size_t wkspSize)
+                                  void* workSpace, size_t wkspSize, int flags)
 {
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -1118,71 +1795,71 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
         (void)algoNb;
         assert(algoNb == 0);
         return HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize);
+                                cSrcSize, workSpace, wkspSize, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
         (void)algoNb;
         assert(algoNb == 1);
         return HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize);
+                                cSrcSize, workSpace, wkspSize, flags);
 #else
         return algoNb ? HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize):
+                                cSrcSize, workSpace, wkspSize, flags):
                         HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
-                                cSrcSize, workSpace, wkspSize);
+                                cSrcSize, workSpace, wkspSize, flags);
 #endif
     }
 }
 
 
-size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
 #if defined(HUF_FORCE_DECOMPRESS_X1)
     (void)dtd;
     assert(dtd.tableType == 0);
-    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
     (void)dtd;
     assert(dtd.tableType == 1);
-    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #else
-    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
-                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #endif
 }
 
 #ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX1_wksp_bmi2(dctx, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize, flags);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, flags);
 }
 #endif
 
-size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int flags)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
 #if defined(HUF_FORCE_DECOMPRESS_X1)
     (void)dtd;
     assert(dtd.tableType == 0);
-    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
     (void)dtd;
     assert(dtd.tableType == 1);
-    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #else
-    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
-                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, flags);
 #endif
 }
 
-size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int flags)
 {
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -1192,160 +1869,14 @@ size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t ds
 #if defined(HUF_FORCE_DECOMPRESS_X1)
         (void)algoNb;
         assert(algoNb == 0);
-        return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
 #elif defined(HUF_FORCE_DECOMPRESS_X2)
         (void)algoNb;
         assert(algoNb == 1);
-        return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
 #else
-        return algoNb ? HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
-                        HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags) :
+                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, flags);
 #endif
     }
 }
-
-#ifndef ZSTD_NO_UNUSED_FUNCTIONS
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_readDTableX1_wksp(DTable, src, srcSize,
-                                 workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
-}
-#endif
-
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
-{
-  U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-  return HUF_readDTableX2_wksp(DTable, src, srcSize,
-                               workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
-}
-#endif
-
-#ifndef HUF_FORCE_DECOMPRESS_X2
-size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
-}
-#endif
-
-#ifndef HUF_FORCE_DECOMPRESS_X1
-size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
-}
-#endif
-
-typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
-
-size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
-    static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
-#endif
-
-    /* validation checks */
-    if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
-    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
-    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
-
-    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-        (void)algoNb;
-        assert(algoNb == 0);
-        return HUF_decompress4X1(dst, dstSize, cSrc, cSrcSize);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-        (void)algoNb;
-        assert(algoNb == 1);
-        return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);
-#else
-        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
-#endif
-    }
-}
-
-size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    /* validation checks */
-    if (dstSize == 0) return ERROR(dstSize_tooSmall);
-    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
-    if (cSrcSize == dstSize) { ZSTD_memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
-    if (cSrcSize == 1) { ZSTD_memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
-
-    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-#if defined(HUF_FORCE_DECOMPRESS_X1)
-        (void)algoNb;
-        assert(algoNb == 0);
-        return HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
-#elif defined(HUF_FORCE_DECOMPRESS_X2)
-        (void)algoNb;
-        assert(algoNb == 1);
-        return HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
-#else
-        return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
-                        HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
-#endif
-    }
-}
-
-size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X_hufOnly_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
-                                         workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
-                             const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
-                                      workSpace, sizeof(workSpace));
-}
-#endif
diff --git a/priv/zstd/lib/decompress/huf_decompress_amd64.S b/priv/zstd/lib/decompress/huf_decompress_amd64.S
new file mode 100644
index 0000000..671624f
--- /dev/null
+++ b/priv/zstd/lib/decompress/huf_decompress_amd64.S
@@ -0,0 +1,576 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "../common/portability_macros.h"
+
+/* Stack marking
+ * ref: https://wiki.gentoo.org/wiki/Hardened/GNU_stack_quickstart
+ */
+#if defined(__ELF__) && defined(__GNUC__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+#if ZSTD_ENABLE_ASM_X86_64_BMI2
+
+/* Calling convention:
+ *
+ * %rdi contains the first argument: HUF_DecompressAsmArgs*.
+ * %rbp isn't maintained (no frame pointer).
+ * %rsp contains the stack pointer that grows down.
+ *      No red-zone is assumed, only addresses >= %rsp are used.
+ * All register contents are preserved.
+ *
+ * TODO: Support Windows calling convention.
+ */
+
+ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
+ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
+ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
+ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
+.global HUF_decompress4X1_usingDTable_internal_fast_asm_loop
+.global HUF_decompress4X2_usingDTable_internal_fast_asm_loop
+.global _HUF_decompress4X1_usingDTable_internal_fast_asm_loop
+.global _HUF_decompress4X2_usingDTable_internal_fast_asm_loop
+.text
+
+/* Sets up register mappings for clarity.
+ * op[], bits[], dtable & ip[0] each get their own register.
+ * ip[1,2,3] & olimit alias var[].
+ * %rax is a scratch register.
+ */
+
+#define op0    rsi
+#define op1    rbx
+#define op2    rcx
+#define op3    rdi
+
+#define ip0    r8
+#define ip1    r9
+#define ip2    r10
+#define ip3    r11
+
+#define bits0  rbp
+#define bits1  rdx
+#define bits2  r12
+#define bits3  r13
+#define dtable r14
+#define olimit r15
+
+/* var[] aliases ip[1,2,3] & olimit
+ * ip[1,2,3] are saved every iteration.
+ * olimit is only used in compute_olimit.
+ */
+#define var0   r15
+#define var1   r9
+#define var2   r10
+#define var3   r11
+
+/* 32-bit var registers */
+#define vard0  r15d
+#define vard1  r9d
+#define vard2  r10d
+#define vard3  r11d
+
+/* Calls X(N) for each stream 0, 1, 2, 3. */
+#define FOR_EACH_STREAM(X) \
+    X(0);                  \
+    X(1);                  \
+    X(2);                  \
+    X(3)
+
+/* Calls X(N, idx) for each stream 0, 1, 2, 3. */
+#define FOR_EACH_STREAM_WITH_INDEX(X, idx) \
+    X(0, idx);                             \
+    X(1, idx);                             \
+    X(2, idx);                             \
+    X(3, idx)
+
+/* Define both _HUF_* & HUF_* symbols because MacOS
+ * C symbols are prefixed with '_' & Linux symbols aren't.
+ */
+_HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
+HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
+    ZSTD_CET_ENDBRANCH
+    /* Save all registers - even if they are callee saved for simplicity. */
+    push %rax
+    push %rbx
+    push %rcx
+    push %rdx
+    push %rbp
+    push %rsi
+    push %rdi
+    push %r8
+    push %r9
+    push %r10
+    push %r11
+    push %r12
+    push %r13
+    push %r14
+    push %r15
+
+    /* Read HUF_DecompressAsmArgs* args from %rax */
+    movq %rdi, %rax
+    movq  0(%rax), %ip0
+    movq  8(%rax), %ip1
+    movq 16(%rax), %ip2
+    movq 24(%rax), %ip3
+    movq 32(%rax), %op0
+    movq 40(%rax), %op1
+    movq 48(%rax), %op2
+    movq 56(%rax), %op3
+    movq 64(%rax), %bits0
+    movq 72(%rax), %bits1
+    movq 80(%rax), %bits2
+    movq 88(%rax), %bits3
+    movq 96(%rax), %dtable
+    push %rax      /* argument */
+    push 104(%rax) /* ilimit */
+    push 112(%rax) /* oend */
+    push %olimit   /* olimit space */
+
+    subq $24, %rsp
+
+.L_4X1_compute_olimit:
+    /* Computes how many iterations we can do safely
+     * %r15, %rax may be clobbered
+     * rbx, rdx must be saved
+     * op3 & ip0 mustn't be clobbered
+     */
+    movq %rbx, 0(%rsp)
+    movq %rdx, 8(%rsp)
+
+    movq 32(%rsp), %rax /* rax = oend */
+    subq %op3,    %rax  /* rax = oend - op3 */
+
+    /* r15 = (oend - op3) / 5 */
+    movabsq $-3689348814741910323, %rdx
+    mulq %rdx
+    movq %rdx, %r15
+    shrq $2, %r15
+
+    movq %ip0,     %rax /* rax = ip0 */
+    movq 40(%rsp), %rdx /* rdx = ilimit */
+    subq %rdx,     %rax /* rax = ip0 - ilimit */
+    movq %rax,     %rbx /* rbx = ip0 - ilimit */
+
+    /* rdx = (ip0 - ilimit) / 7 */
+    movabsq $2635249153387078803, %rdx
+    mulq %rdx
+    subq %rdx, %rbx
+    shrq %rbx
+    addq %rbx, %rdx
+    shrq $2, %rdx
+
+    /* r15 = min(%rdx, %r15) */
+    cmpq %rdx, %r15
+    cmova %rdx, %r15
+
+    /* r15 = r15 * 5 */
+    leaq (%r15, %r15, 4), %r15
+
+    /* olimit = op3 + r15 */
+    addq %op3, %olimit
+
+    movq 8(%rsp), %rdx
+    movq 0(%rsp), %rbx
+
+    /* If (op3 + 20 > olimit) */
+    movq %op3, %rax    /* rax = op3 */
+    addq $20,  %rax    /* rax = op3 + 20 */
+    cmpq %rax, %olimit /* op3 + 20 > olimit */
+    jb .L_4X1_exit
+
+    /* If (ip1 < ip0) go to exit */
+    cmpq %ip0, %ip1
+    jb .L_4X1_exit
+
+    /* If (ip2 < ip1) go to exit */
+    cmpq %ip1, %ip2
+    jb .L_4X1_exit
+
+    /* If (ip3 < ip2) go to exit */
+    cmpq %ip2, %ip3
+    jb .L_4X1_exit
+
+/* Reads top 11 bits from bits[n]
+ * Loads dt[bits[n]] into var[n]
+ */
+#define GET_NEXT_DELT(n)                \
+    movq $53, %var##n;                  \
+    shrxq %var##n, %bits##n, %var##n;   \
+    movzwl (%dtable,%var##n,2),%vard##n
+
+/* var[n] must contain the DTable entry computed with GET_NEXT_DELT
+ * Moves var[n] to %rax
+ * bits[n] <<= var[n] & 63
+ * op[n][idx] = %rax >> 8
+ * %ah is a way to access bits [8, 16) of %rax
+ */
+#define DECODE_FROM_DELT(n, idx)       \
+    movq %var##n, %rax;                \
+    shlxq %var##n, %bits##n, %bits##n; \
+    movb %ah, idx(%op##n)
+
+/* Assumes GET_NEXT_DELT has been called.
+ * Calls DECODE_FROM_DELT then GET_NEXT_DELT
+ */
+#define DECODE_AND_GET_NEXT(n, idx) \
+    DECODE_FROM_DELT(n, idx);       \
+    GET_NEXT_DELT(n)                \
+
+/* // ctz & nbBytes is stored in bits[n]
+ * // nbBits is stored in %rax
+ * ctz  = CTZ[bits[n]]
+ * nbBits  = ctz & 7
+ * nbBytes = ctz >> 3
+ * op[n]  += 5
+ * ip[n]  -= nbBytes
+ * // Note: x86-64 is little-endian ==> no bswap
+ * bits[n] = MEM_readST(ip[n]) | 1
+ * bits[n] <<= nbBits
+ */
+#define RELOAD_BITS(n)             \
+    bsfq %bits##n, %bits##n;       \
+    movq %bits##n, %rax;           \
+    andq $7, %rax;                 \
+    shrq $3, %bits##n;             \
+    leaq 5(%op##n), %op##n;        \
+    subq %bits##n, %ip##n;         \
+    movq (%ip##n), %bits##n;       \
+    orq $1, %bits##n;              \
+    shlx %rax, %bits##n, %bits##n
+
+    /* Store clobbered variables on the stack */
+    movq %olimit, 24(%rsp)
+    movq %ip1, 0(%rsp)
+    movq %ip2, 8(%rsp)
+    movq %ip3, 16(%rsp)
+
+    /* Call GET_NEXT_DELT for each stream */
+    FOR_EACH_STREAM(GET_NEXT_DELT)
+
+    .p2align 6
+
+.L_4X1_loop_body:
+    /* Decode 5 symbols in each of the 4 streams (20 total)
+     * Must have called GET_NEXT_DELT for each stream
+     */
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 0)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 1)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 2)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 3)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE_FROM_DELT, 4)
+
+    /* Load ip[1,2,3] from stack (var[] aliases them)
+     * ip[] is needed for RELOAD_BITS
+     * Each will be stored back to the stack after RELOAD
+     */
+    movq 0(%rsp), %ip1
+    movq 8(%rsp), %ip2
+    movq 16(%rsp), %ip3
+
+    /* Reload each stream & fetch the next table entry
+     * to prepare for the next iteration
+     */
+    RELOAD_BITS(0)
+    GET_NEXT_DELT(0)
+
+    RELOAD_BITS(1)
+    movq %ip1, 0(%rsp)
+    GET_NEXT_DELT(1)
+
+    RELOAD_BITS(2)
+    movq %ip2, 8(%rsp)
+    GET_NEXT_DELT(2)
+
+    RELOAD_BITS(3)
+    movq %ip3, 16(%rsp)
+    GET_NEXT_DELT(3)
+
+    /* If op3 < olimit: continue the loop */
+    cmp %op3, 24(%rsp)
+    ja .L_4X1_loop_body
+
+    /* Reload ip[1,2,3] from stack */
+    movq 0(%rsp), %ip1
+    movq 8(%rsp), %ip2
+    movq 16(%rsp), %ip3
+
+    /* Re-compute olimit */
+    jmp .L_4X1_compute_olimit
+
+#undef GET_NEXT_DELT
+#undef DECODE_FROM_DELT
+#undef DECODE
+#undef RELOAD_BITS
+.L_4X1_exit:
+    addq $24, %rsp
+
+    /* Restore stack (oend & olimit) */
+    pop %rax /* olimit */
+    pop %rax /* oend */
+    pop %rax /* ilimit */
+    pop %rax /* arg */
+
+    /* Save ip / op / bits */
+    movq %ip0,  0(%rax)
+    movq %ip1,  8(%rax)
+    movq %ip2, 16(%rax)
+    movq %ip3, 24(%rax)
+    movq %op0, 32(%rax)
+    movq %op1, 40(%rax)
+    movq %op2, 48(%rax)
+    movq %op3, 56(%rax)
+    movq %bits0, 64(%rax)
+    movq %bits1, 72(%rax)
+    movq %bits2, 80(%rax)
+    movq %bits3, 88(%rax)
+
+    /* Restore registers */
+    pop %r15
+    pop %r14
+    pop %r13
+    pop %r12
+    pop %r11
+    pop %r10
+    pop %r9
+    pop %r8
+    pop %rdi
+    pop %rsi
+    pop %rbp
+    pop %rdx
+    pop %rcx
+    pop %rbx
+    pop %rax
+    ret
+
+_HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
+HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
+    ZSTD_CET_ENDBRANCH
+    /* Save all registers - even if they are callee saved for simplicity. */
+    push %rax
+    push %rbx
+    push %rcx
+    push %rdx
+    push %rbp
+    push %rsi
+    push %rdi
+    push %r8
+    push %r9
+    push %r10
+    push %r11
+    push %r12
+    push %r13
+    push %r14
+    push %r15
+
+    movq %rdi, %rax
+    movq  0(%rax), %ip0
+    movq  8(%rax), %ip1
+    movq 16(%rax), %ip2
+    movq 24(%rax), %ip3
+    movq 32(%rax), %op0
+    movq 40(%rax), %op1
+    movq 48(%rax), %op2
+    movq 56(%rax), %op3
+    movq 64(%rax), %bits0
+    movq 72(%rax), %bits1
+    movq 80(%rax), %bits2
+    movq 88(%rax), %bits3
+    movq 96(%rax), %dtable
+    push %rax      /* argument */
+    push %rax      /* olimit */
+    push 104(%rax) /* ilimit */
+
+    movq 112(%rax), %rax
+    push %rax /* oend3 */
+
+    movq %op3, %rax
+    push %rax /* oend2 */
+
+    movq %op2, %rax
+    push %rax /* oend1 */
+
+    movq %op1, %rax
+    push %rax /* oend0 */
+
+    /* Scratch space */
+    subq $8, %rsp
+
+.L_4X2_compute_olimit:
+    /* Computes how many iterations we can do safely
+     * %r15, %rax may be clobbered
+     * rdx must be saved
+     * op[1,2,3,4] & ip0 mustn't be clobbered
+     */
+    movq %rdx, 0(%rsp)
+
+    /* We can consume up to 7 input bytes each iteration. */
+    movq %ip0,     %rax  /* rax = ip0 */
+    movq 40(%rsp), %rdx  /* rdx = ilimit */
+    subq %rdx,     %rax  /* rax = ip0 - ilimit */
+    movq %rax,    %r15   /* r15 = ip0 - ilimit */
+
+    /* rdx = rax / 7 */
+    movabsq $2635249153387078803, %rdx
+    mulq %rdx
+    subq %rdx, %r15
+    shrq %r15
+    addq %r15, %rdx
+    shrq $2, %rdx
+
+    /* r15 = (ip0 - ilimit) / 7 */
+    movq %rdx, %r15
+
+    /* r15 = min(r15, min(oend0 - op0, oend1 - op1, oend2 - op2, oend3 - op3) / 10) */
+    movq 8(%rsp),  %rax /* rax = oend0 */
+    subq %op0,     %rax /* rax = oend0 - op0 */
+    movq 16(%rsp), %rdx /* rdx = oend1 */
+    subq %op1,     %rdx /* rdx = oend1 - op1 */
+
+    cmpq  %rax,    %rdx
+    cmova %rax,    %rdx /* rdx = min(%rdx, %rax) */
+
+    movq 24(%rsp), %rax /* rax = oend2 */
+    subq %op2,     %rax /* rax = oend2 - op2 */
+
+    cmpq  %rax,    %rdx
+    cmova %rax,    %rdx /* rdx = min(%rdx, %rax) */
+
+    movq 32(%rsp), %rax /* rax = oend3 */
+    subq %op3,     %rax /* rax = oend3 - op3 */
+
+    cmpq  %rax,    %rdx
+    cmova %rax,    %rdx /* rdx = min(%rdx, %rax) */
+
+    movabsq $-3689348814741910323, %rax
+    mulq %rdx
+    shrq $3,       %rdx /* rdx = rdx / 10 */
+
+    /* r15 = min(%rdx, %r15) */
+    cmpq  %rdx, %r15
+    cmova %rdx, %r15
+
+    /* olimit = op3 + 5 * r15 */
+    movq %r15, %rax
+    leaq (%op3, %rax, 4), %olimit
+    addq %rax, %olimit
+
+    movq 0(%rsp), %rdx
+
+    /* If (op3 + 10 > olimit) */
+    movq %op3, %rax    /* rax = op3 */
+    addq $10,  %rax    /* rax = op3 + 10 */
+    cmpq %rax, %olimit /* op3 + 10 > olimit */
+    jb .L_4X2_exit
+
+    /* If (ip1 < ip0) go to exit */
+    cmpq %ip0, %ip1
+    jb .L_4X2_exit
+
+    /* If (ip2 < ip1) go to exit */
+    cmpq %ip1, %ip2
+    jb .L_4X2_exit
+
+    /* If (ip3 < ip2) go to exit */
+    cmpq %ip2, %ip3
+    jb .L_4X2_exit
+
+#define DECODE(n, idx)              \
+    movq %bits##n, %rax;            \
+    shrq $53, %rax;                 \
+    movzwl 0(%dtable,%rax,4),%r8d;  \
+    movzbl 2(%dtable,%rax,4),%r15d; \
+    movzbl 3(%dtable,%rax,4),%eax;  \
+    movw %r8w, (%op##n);            \
+    shlxq %r15, %bits##n, %bits##n; \
+    addq %rax, %op##n
+
+#define RELOAD_BITS(n)              \
+    bsfq %bits##n, %bits##n;        \
+    movq %bits##n, %rax;            \
+    shrq $3, %bits##n;              \
+    andq $7, %rax;                  \
+    subq %bits##n, %ip##n;          \
+    movq (%ip##n), %bits##n;        \
+    orq $1, %bits##n;               \
+    shlxq %rax, %bits##n, %bits##n
+
+
+    movq %olimit, 48(%rsp)
+
+    .p2align 6
+
+.L_4X2_loop_body:
+    /* We clobber r8, so store it on the stack */
+    movq %r8, 0(%rsp)
+
+    /* Decode 5 symbols from each of the 4 streams (20 symbols total). */
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 0)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 1)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 2)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 3)
+    FOR_EACH_STREAM_WITH_INDEX(DECODE, 4)
+
+    /* Reload r8 */
+    movq 0(%rsp), %r8
+
+    FOR_EACH_STREAM(RELOAD_BITS)
+
+    cmp %op3, 48(%rsp)
+    ja .L_4X2_loop_body
+    jmp .L_4X2_compute_olimit
+
+#undef DECODE
+#undef RELOAD_BITS
+.L_4X2_exit:
+    addq $8, %rsp
+    /* Restore stack (oend & olimit) */
+    pop %rax /* oend0 */
+    pop %rax /* oend1 */
+    pop %rax /* oend2 */
+    pop %rax /* oend3 */
+    pop %rax /* ilimit */
+    pop %rax /* olimit */
+    pop %rax /* arg */
+
+    /* Save ip / op / bits */
+    movq %ip0,  0(%rax)
+    movq %ip1,  8(%rax)
+    movq %ip2, 16(%rax)
+    movq %ip3, 24(%rax)
+    movq %op0, 32(%rax)
+    movq %op1, 40(%rax)
+    movq %op2, 48(%rax)
+    movq %op3, 56(%rax)
+    movq %bits0, 64(%rax)
+    movq %bits1, 72(%rax)
+    movq %bits2, 80(%rax)
+    movq %bits3, 88(%rax)
+
+    /* Restore registers */
+    pop %r15
+    pop %r14
+    pop %r13
+    pop %r12
+    pop %r11
+    pop %r10
+    pop %r9
+    pop %r8
+    pop %rdi
+    pop %rsi
+    pop %rbp
+    pop %rdx
+    pop %rcx
+    pop %rbx
+    pop %rax
+    ret
+
+#endif
diff --git a/priv/zstd/lib/decompress/zstd_ddict.c b/priv/zstd/lib/decompress/zstd_ddict.c
index ce33547..ad5c34a 100644
--- a/priv/zstd/lib/decompress/zstd_ddict.c
+++ b/priv/zstd/lib/decompress/zstd_ddict.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -19,7 +19,6 @@
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "zstd_decompress_internal.h"
 #include "zstd_ddict.h"
@@ -134,7 +133,7 @@ static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
         ZSTD_memcpy(internalBuffer, dict, dictSize);
     }
     ddict->dictSize = dictSize;
-    ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    ddict->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
 
     /* parse dictionary content */
     FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) , "");
@@ -240,5 +239,5 @@ size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
 unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;
-    return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
+    return ddict->dictID;
 }
diff --git a/priv/zstd/lib/decompress/zstd_ddict.h b/priv/zstd/lib/decompress/zstd_ddict.h
index bd03268..c4ca887 100644
--- a/priv/zstd/lib/decompress/zstd_ddict.h
+++ b/priv/zstd/lib/decompress/zstd_ddict.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/decompress/zstd_decompress.c b/priv/zstd/lib/decompress/zstd_decompress.c
index 910bc03..d487966 100644
--- a/priv/zstd/lib/decompress/zstd_decompress.c
+++ b/priv/zstd/lib/decompress/zstd_decompress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -56,17 +56,16 @@
 *  Dependencies
 *********************************************************/
 #include "../common/zstd_deps.h"   /* ZSTD_memcpy, ZSTD_memmove, ZSTD_memset */
-#include "../common/cpu.h"         /* bmi2 */
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/xxhash.h" /* XXH64_reset, XXH64_update, XXH64_digest, XXH64 */
 #include "../common/zstd_internal.h"  /* blockProperties_t */
 #include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
 #include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
 #include "zstd_decompress_block.h"   /* ZSTD_decompressBlock_internal */
+#include "../common/bits.h"  /* ZSTD_highbit32 */
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
 #  include "../legacy/zstd_legacy.h"
@@ -79,11 +78,11 @@
  *************************************/
 
 #define DDICT_HASHSET_MAX_LOAD_FACTOR_COUNT_MULT 4
-#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3   /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
-                                                     * Currently, that means a 0.75 load factor.
-                                                     * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
-                                                     * the load factor of the ddict hash set.
-                                                     */
+#define DDICT_HASHSET_MAX_LOAD_FACTOR_SIZE_MULT 3  /* These two constants represent SIZE_MULT/COUNT_MULT load factor without using a float.
+                                                    * Currently, that means a 0.75 load factor.
+                                                    * So, if count * COUNT_MULT / size * SIZE_MULT != 0, then we've exceeded
+                                                    * the load factor of the ddict hash set.
+                                                    */
 
 #define DDICT_HASHSET_TABLE_BASE_SIZE 64
 #define DDICT_HASHSET_RESIZE_FACTOR 2
@@ -177,12 +176,15 @@ static const ZSTD_DDict* ZSTD_DDictHashSet_getDDict(ZSTD_DDictHashSet* hashSet,
 static ZSTD_DDictHashSet* ZSTD_createDDictHashSet(ZSTD_customMem customMem) {
     ZSTD_DDictHashSet* ret = (ZSTD_DDictHashSet*)ZSTD_customMalloc(sizeof(ZSTD_DDictHashSet), customMem);
     DEBUGLOG(4, "Allocating new hash set");
+    if (!ret)
+        return NULL;
     ret->ddictPtrTable = (const ZSTD_DDict**)ZSTD_customCalloc(DDICT_HASHSET_TABLE_BASE_SIZE * sizeof(ZSTD_DDict*), customMem);
-    ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
-    ret->ddictPtrCount = 0;
-    if (!ret || !ret->ddictPtrTable) {
+    if (!ret->ddictPtrTable) {
+        ZSTD_customFree(ret, customMem);
         return NULL;
     }
+    ret->ddictPtrTableSize = DDICT_HASHSET_TABLE_BASE_SIZE;
+    ret->ddictPtrCount = 0;
     return ret;
 }
 
@@ -241,6 +243,7 @@ static void ZSTD_DCtx_resetParameters(ZSTD_DCtx* dctx)
     dctx->outBufferMode = ZSTD_bm_buffered;
     dctx->forceIgnoreChecksum = ZSTD_d_validateChecksum;
     dctx->refMultipleDDicts = ZSTD_rmd_refSingleDDict;
+    dctx->disableHufAsm = 0;
 }
 
 static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
@@ -255,11 +258,15 @@ static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
     dctx->inBuffSize  = 0;
     dctx->outBuffSize = 0;
     dctx->streamStage = zdss_init;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
     dctx->legacyContext = NULL;
     dctx->previousLegacyVersion = 0;
+#endif
     dctx->noForwardProgress = 0;
     dctx->oversizedDuration = 0;
-    dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+#if DYNAMIC_BMI2
+    dctx->bmi2 = ZSTD_cpuSupportsBmi2();
+#endif
     dctx->ddictSet = NULL;
     ZSTD_DCtx_resetParameters(dctx);
 #ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
@@ -280,8 +287,7 @@ ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
     return dctx;
 }
 
-ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
-{
+static ZSTD_DCtx* ZSTD_createDCtx_internal(ZSTD_customMem customMem) {
     if ((!customMem.customAlloc) ^ (!customMem.customFree)) return NULL;
 
     {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_customMalloc(sizeof(*dctx), customMem);
@@ -292,10 +298,15 @@ ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
     }
 }
 
+ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDCtx_internal(customMem);
+}
+
 ZSTD_DCtx* ZSTD_createDCtx(void)
 {
     DEBUGLOG(3, "ZSTD_createDCtx");
-    return ZSTD_createDCtx_advanced(ZSTD_defaultCMem);
+    return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
 }
 
 static void ZSTD_clearDict(ZSTD_DCtx* dctx)
@@ -380,6 +391,19 @@ unsigned ZSTD_isFrame(const void* buffer, size_t size)
     return 0;
 }
 
+/*! ZSTD_isSkippableFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ */
+unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size)
+{
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+    return 0;
+}
+
 /** ZSTD_frameHeaderSize_internal() :
  *  srcSize must be large enough to reach header size fields.
  *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
@@ -415,16 +439,40 @@ size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
  *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
  * @return : 0, `zfhPtr` is correctly filled,
  *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
- *           or an error code, which can be tested using ZSTD_isError() */
+**           or an error code, which can be tested using ZSTD_isError() */
 size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
 {
     const BYTE* ip = (const BYTE*)src;
     size_t const minInputSize = ZSTD_startingInputLength(format);
 
-    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
-    if (srcSize < minInputSize) return minInputSize;
-    RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter");
+    DEBUGLOG(5, "ZSTD_getFrameHeader_advanced: minInputSize = %zu, srcSize = %zu", minInputSize, srcSize);
 
+    if (srcSize > 0) {
+        /* note : technically could be considered an assert(), since it's an invalid entry */
+        RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter : src==NULL, but srcSize>0");
+    }
+    if (srcSize < minInputSize) {
+        if (srcSize > 0 && format != ZSTD_f_zstd1_magicless) {
+            /* when receiving less than @minInputSize bytes,
+             * control these bytes at least correspond to a supported magic number
+             * in order to error out early if they don't.
+            **/
+            size_t const toCopy = MIN(4, srcSize);
+            unsigned char hbuf[4]; MEM_writeLE32(hbuf, ZSTD_MAGICNUMBER);
+            assert(src != NULL);
+            ZSTD_memcpy(hbuf, src, toCopy);
+            if ( MEM_readLE32(hbuf) != ZSTD_MAGICNUMBER ) {
+                /* not a zstd frame : let's check if it's a skippable frame */
+                MEM_writeLE32(hbuf, ZSTD_MAGIC_SKIPPABLE_START);
+                ZSTD_memcpy(hbuf, src, toCopy);
+                if ((MEM_readLE32(hbuf) & ZSTD_MAGIC_SKIPPABLE_MASK) != ZSTD_MAGIC_SKIPPABLE_START) {
+                    RETURN_ERROR(prefix_unknown,
+                                "first bytes don't correspond to any supported magic number");
+        }   }   }
+        return minInputSize;
+    }
+
+    ZSTD_memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzers may not understand that zfhPtr will be read only if return value is zero, since they are 2 different signals */
     if ( (format != ZSTD_f_zstd1_magicless)
       && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
         if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
@@ -466,7 +514,9 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
         }
         switch(dictIDSizeCode)
         {
-            default: assert(0);  /* impossible */
+            default:
+                assert(0);  /* impossible */
+                ZSTD_FALLTHROUGH;
             case 0 : break;
             case 1 : dictID = ip[pos]; pos++; break;
             case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
@@ -474,7 +524,9 @@ size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, s
         }
         switch(fcsID)
         {
-            default: assert(0);  /* impossible */
+            default:
+                assert(0);  /* impossible */
+                ZSTD_FALLTHROUGH;
             case 0 : if (singleSegment) frameContentSize = ip[pos]; break;
             case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
             case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
@@ -503,7 +555,6 @@ size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t src
     return ZSTD_getFrameHeader_advanced(zfhPtr, src, srcSize, ZSTD_f_zstd1);
 }
 
-
 /** ZSTD_getFrameContentSize() :
  *  compatible with legacy mode
  * @return : decompressed size of the single frame pointed to be `src` if known, otherwise
@@ -544,6 +595,37 @@ static size_t readSkippableFrameSize(void const* src, size_t srcSize)
     }
 }
 
+/*! ZSTD_readSkippableFrame() :
+ * Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
+ *
+ * The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
+ * i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
+ * in the magicVariant.
+ *
+ * Returns an error if destination buffer is not large enough, or if the frame is not skippable.
+ *
+ * @return : number of bytes written or a ZSTD error.
+ */
+ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
+                                            const void* src, size_t srcSize)
+{
+    U32 const magicNumber = MEM_readLE32(src);
+    size_t skippableFrameSize = readSkippableFrameSize(src, srcSize);
+    size_t skippableContentSize = skippableFrameSize - ZSTD_SKIPPABLEHEADERSIZE;
+
+    /* check input validity */
+    RETURN_ERROR_IF(!ZSTD_isSkippableFrame(src, srcSize), frameParameter_unsupported, "");
+    RETURN_ERROR_IF(skippableFrameSize < ZSTD_SKIPPABLEHEADERSIZE || skippableFrameSize > srcSize, srcSize_wrong, "");
+    RETURN_ERROR_IF(skippableContentSize > dstCapacity, dstSize_tooSmall, "");
+
+    /* deliver payload */
+    if (skippableContentSize > 0  && dst != NULL)
+        ZSTD_memcpy(dst, (const BYTE *)src + ZSTD_SKIPPABLEHEADERSIZE, skippableContentSize);
+    if (magicVariant != NULL)
+        *magicVariant = magicNumber - ZSTD_MAGIC_SKIPPABLE_START;
+    return skippableContentSize;
+}
+
 /** ZSTD_findDecompressedSize() :
  *  compatible with legacy mode
  *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
@@ -700,10 +782,11 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
             ip += 4;
         }
 
+        frameSizeInfo.nbBlocks = nbBlocks;
         frameSizeInfo.compressedSize = (size_t)(ip - ipstart);
         frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
                                         ? zfh.frameContentSize
-                                        : nbBlocks * zfh.blockSizeMax;
+                                        : (unsigned long long)nbBlocks * zfh.blockSizeMax;
         return frameSizeInfo;
     }
 }
@@ -743,6 +826,48 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
     return bound;
 }
 
+size_t ZSTD_decompressionMargin(void const* src, size_t srcSize)
+{
+    size_t margin = 0;
+    unsigned maxBlockSize = 0;
+
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        ZSTD_frameHeader zfh;
+
+        FORWARD_IF_ERROR(ZSTD_getFrameHeader(&zfh, src, srcSize), "");
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ERROR(corruption_detected);
+
+        if (zfh.frameType == ZSTD_frame) {
+            /* Add the frame header to our margin */
+            margin += zfh.headerSize;
+            /* Add the checksum to our margin */
+            margin += zfh.checksumFlag ? 4 : 0;
+            /* Add 3 bytes per block */
+            margin += 3 * frameSizeInfo.nbBlocks;
+
+            /* Compute the max block size */
+            maxBlockSize = MAX(maxBlockSize, zfh.blockSizeMax);
+        } else {
+            assert(zfh.frameType == ZSTD_skippableFrame);
+            /* Add the entire skippable frame size to our margin. */
+            margin += compressedSize;
+        }
+
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+    }
+
+    /* Add the max block size back to the margin. */
+    margin += maxBlockSize;
+
+    return margin;
+}
 
 /*-*************************************************************
  *   Frame decoding
@@ -768,7 +893,7 @@ static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
         if (srcSize == 0) return 0;
         RETURN_ERROR(dstBuffer_null, "");
     }
-    ZSTD_memcpy(dst, src, srcSize);
+    ZSTD_memmove(dst, src, srcSize);
     return srcSize;
 }
 
@@ -846,6 +971,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
 
     /* Loop on each block */
     while (1) {
+        BYTE* oBlockEnd = oend;
         size_t decodedSize;
         blockProperties_t blockProperties;
         size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSrcSize, &blockProperties);
@@ -855,16 +981,34 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         remainingSrcSize -= ZSTD_blockHeaderSize;
         RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong, "");
 
+        if (ip >= op && ip < oBlockEnd) {
+            /* We are decompressing in-place. Limit the output pointer so that we
+             * don't overwrite the block that we are currently reading. This will
+             * fail decompression if the input & output pointers aren't spaced
+             * far enough apart.
+             *
+             * This is important to set, even when the pointers are far enough
+             * apart, because ZSTD_decompressBlock_internal() can decide to store
+             * literals in the output buffer, after the block it is decompressing.
+             * Since we don't want anything to overwrite our input, we have to tell
+             * ZSTD_decompressBlock_internal to never write past ip.
+             *
+             * See ZSTD_allocateLiteralsBuffer() for reference.
+             */
+            oBlockEnd = op + (ip - op);
+        }
+
         switch(blockProperties.blockType)
         {
         case bt_compressed:
-            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oend-op), ip, cBlockSize, /* frame */ 1);
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, (size_t)(oBlockEnd-op), ip, cBlockSize, /* frame */ 1, not_streaming);
             break;
         case bt_raw :
+            /* Use oend instead of oBlockEnd because this function is safe to overlap. It uses memmove. */
             decodedSize = ZSTD_copyRawBlock(op, (size_t)(oend-op), ip, cBlockSize);
             break;
         case bt_rle :
-            decodedSize = ZSTD_setRleBlock(op, (size_t)(oend-op), *ip, blockProperties.origSize);
+            decodedSize = ZSTD_setRleBlock(op, (size_t)(oBlockEnd-op), *ip, blockProperties.origSize);
             break;
         case bt_reserved :
         default:
@@ -899,6 +1043,7 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
     }
     ZSTD_DCtx_trace_end(dctx, (U64)(op-ostart), (U64)(ip-istart), /* streaming */ 0);
     /* Allow caller to get size read */
+    DEBUGLOG(4, "ZSTD_decompressFrame: decompressed frame of size %zi, consuming %zi bytes of input", op-ostart, ip - (const BYTE*)*srcPtr);
     *srcPtr = ip;
     *srcSizePtr = remainingSrcSize;
     return (size_t)(op-ostart);
@@ -1009,7 +1154,7 @@ static ZSTD_DDict const* ZSTD_getDDict(ZSTD_DCtx* dctx)
     switch (dctx->dictUses) {
     default:
         assert(0 /* Impossible */);
-        /* fall-through */
+        ZSTD_FALLTHROUGH;
     case ZSTD_dont_use:
         ZSTD_clearDict(dctx);
         return NULL;
@@ -1031,7 +1176,7 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 {
 #if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
     size_t regenSize;
-    ZSTD_DCtx* const dctx = ZSTD_createDCtx();
+    ZSTD_DCtx* const dctx =  ZSTD_createDCtx_internal(ZSTD_defaultCMem);
     RETURN_ERROR_IF(dctx==NULL, memory_allocation, "NULL pointer!");
     regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
     ZSTD_freeDCtx(dctx);
@@ -1051,8 +1196,8 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
 
 /**
- * Similar to ZSTD_nextSrcSizeToDecompress(), but when when a block input can be streamed,
- * we allow taking a partial block as the input. Currently only raw uncompressed blocks can
+ * Similar to ZSTD_nextSrcSizeToDecompress(), but when a block input can be streamed, we
+ * allow taking a partial block as the input. Currently only raw uncompressed blocks can
  * be streamed.
  *
  * For blocks that can be streamed, this allows us to reduce the latency until we produce
@@ -1065,7 +1210,7 @@ static size_t ZSTD_nextSrcSizeToDecompressWithInputSize(ZSTD_DCtx* dctx, size_t
         return dctx->expected;
     if (dctx->bType != bt_raw)
         return dctx->expected;
-    return MIN(MAX(inputSize, 1), dctx->expected);
+    return BOUNDED(1, inputSize, dctx->expected);
 }
 
 ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
@@ -1073,7 +1218,9 @@ ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
     {
     default:   /* should not happen */
         assert(0);
+        ZSTD_FALLTHROUGH;
     case ZSTDds_getFrameHeaderSize:
+        ZSTD_FALLTHROUGH;
     case ZSTDds_decodeFrameHeader:
         return ZSTDnit_frameHeader;
     case ZSTDds_decodeBlockHeader:
@@ -1085,6 +1232,7 @@ ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
     case ZSTDds_checkChecksum:
         return ZSTDnit_checksum;
     case ZSTDds_decodeSkippableHeader:
+        ZSTD_FALLTHROUGH;
     case ZSTDds_skipFrame:
         return ZSTDnit_skippableFrame;
     }
@@ -1168,7 +1316,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             {
             case bt_compressed:
                 DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
-                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1, is_streaming);
                 dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_raw :
@@ -1249,7 +1397,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 
     default:
         assert(0);   /* impossible */
-        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
+        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
     }
 }
 
@@ -1290,11 +1438,11 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
         /* in minimal huffman, we always use X1 variants */
         size_t const hSize = HUF_readDTableX1_wksp(entropy->hufTable,
                                                 dictPtr, dictEnd - dictPtr,
-                                                workspace, workspaceSize);
+                                                workspace, workspaceSize, /* flags */ 0);
 #else
         size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
                                                 dictPtr, (size_t)(dictEnd - dictPtr),
-                                                workspace, workspaceSize);
+                                                workspace, workspaceSize, /* flags */ 0);
 #endif
         RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
         dictPtr += hSize;
@@ -1393,7 +1541,7 @@ size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
     dctx->prefixStart = NULL;
     dctx->virtualStart = NULL;
     dctx->dictEnd = NULL;
-    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    dctx->entropy.hufTable[0] = (HUF_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);  /* cover both little and big endian */
     dctx->litEntropy = dctx->fseEntropy = 0;
     dctx->dictID = 0;
     dctx->bType = bt_reserved;
@@ -1455,7 +1603,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
  *  This could for one of the following reasons :
  *  - The frame does not require a dictionary (most common case).
  *  - The frame was built with dictID intentionally removed.
- *    Needed dictionary is a hidden information.
+ *    Needed dictionary is a hidden piece of information.
  *    Note : this use case also happens when using a non-conformant dictionary.
  *  - `srcSize` is too small, and as a result, frame header could not be decoded.
  *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
@@ -1464,7 +1612,7 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
  *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
 unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
 {
-    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0 };
+    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0, 0, 0 };
     size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
     if (ZSTD_isError(hError)) return 0;
     return zfp.dictID;
@@ -1493,7 +1641,7 @@ size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
 ZSTD_DStream* ZSTD_createDStream(void)
 {
     DEBUGLOG(3, "ZSTD_createDStream");
-    return ZSTD_createDStream_advanced(ZSTD_defaultCMem);
+    return ZSTD_createDCtx_internal(ZSTD_defaultCMem);
 }
 
 ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
@@ -1503,7 +1651,7 @@ ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
 
 ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
 {
-    return ZSTD_createDCtx_advanced(customMem);
+    return ZSTD_createDCtx_internal(customMem);
 }
 
 size_t ZSTD_freeDStream(ZSTD_DStream* zds)
@@ -1571,7 +1719,9 @@ size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t di
 size_t ZSTD_initDStream(ZSTD_DStream* zds)
 {
     DEBUGLOG(4, "ZSTD_initDStream");
-    return ZSTD_initDStream_usingDDict(zds, NULL);
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(zds, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_DCtx_refDDict(zds, NULL), "");
+    return ZSTD_startingInputLength(zds->format);
 }
 
 /* ZSTD_initDStream_usingDDict() :
@@ -1579,6 +1729,7 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
  * this function cannot fail */
 size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
 {
+    DEBUGLOG(4, "ZSTD_initDStream_usingDDict");
     FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
     FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
     return ZSTD_startingInputLength(dctx->format);
@@ -1589,6 +1740,7 @@ size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
  * this function cannot fail */
 size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
 {
+    DEBUGLOG(4, "ZSTD_resetDStream");
     FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
     return ZSTD_startingInputLength(dctx->format);
 }
@@ -1660,6 +1812,11 @@ ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
             bounds.lowerBound = (int)ZSTD_rmd_refSingleDDict;
             bounds.upperBound = (int)ZSTD_rmd_refMultipleDDicts;
             return bounds;
+        case ZSTD_d_disableHuffmanAssembly:
+            bounds.lowerBound = 0;
+            bounds.upperBound = 1;
+            return bounds;
+
         default:;
     }
     bounds.error = ERROR(parameter_unsupported);
@@ -1700,6 +1857,9 @@ size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value
         case ZSTD_d_refMultipleDDicts:
             *value = (int)dctx->refMultipleDDicts;
             return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            *value = (int)dctx->disableHufAsm;
+            return 0;
         default:;
     }
     RETURN_ERROR(parameter_unsupported, "");
@@ -1733,6 +1893,10 @@ size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value
             }
             dctx->refMultipleDDicts = (ZSTD_refMultipleDDicts_e)value;
             return 0;
+        case ZSTD_d_disableHuffmanAssembly:
+            CHECK_DBOUNDS(ZSTD_d_disableHuffmanAssembly, value);
+            dctx->disableHufAsm = value != 0;
+            return 0;
         default:;
     }
     RETURN_ERROR(parameter_unsupported, "");
@@ -1763,7 +1927,8 @@ size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
 size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
 {
     size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
-    unsigned long long const neededRBSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
+    /* space is needed to store the litbuffer after the output of a given block without stomping the extDict of a previous run, as well as to cover both windows against wildcopy*/
+    unsigned long long const neededRBSize = windowSize + blockSize + ZSTD_BLOCKSIZE_MAX + (WILDCOPY_OVERLENGTH * 2);
     unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
     size_t const minRBSize = (size_t) neededSize;
     RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
@@ -1897,10 +2062,12 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             DEBUGLOG(5, "stage zdss_init => transparent reset ");
             zds->streamStage = zdss_loadHeader;
             zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
             zds->legacyVersion = 0;
+#endif
             zds->hostageByte = 0;
             zds->expectedOutBuffer = *output;
-            /* fall-through */
+            ZSTD_FALLTHROUGH;
 
         case zdss_loadHeader :
             DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
@@ -1917,7 +2084,6 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 if (zds->refMultipleDDicts && zds->ddictSet) {
                     ZSTD_DCtx_selectFrameDDict(zds);
                 }
-                DEBUGLOG(5, "header size : %u", (U32)hSize);
                 if (ZSTD_isError(hSize)) {
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
                     U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
@@ -1949,6 +2115,11 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                             zds->lhSize += remainingInput;
                         }
                         input->pos = input->size;
+                        /* check first few bytes */
+                        FORWARD_IF_ERROR(
+                            ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format),
+                            "First few bytes detected incorrect" );
+                        /* return hint input size */
                         return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
                     assert(ip != NULL);
@@ -1966,8 +2137,9 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, (size_t)(oend-op), istart, cSize, ZSTD_getDDict(zds));
                     if (ZSTD_isError(decompressedSize)) return decompressedSize;
                     DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
+                    assert(istart != NULL);
                     ip = istart + cSize;
-                    op += decompressedSize;
+                    op = op ? op + decompressedSize : op; /* can occur if frameContentSize = 0 (empty frame) */
                     zds->expected = 0;
                     zds->streamStage = zdss_init;
                     someMoreWork = 0;
@@ -2038,7 +2210,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                         zds->outBuffSize = neededOutBuffSize;
             }   }   }
             zds->streamStage = zdss_read;
-            /* fall-through */
+            ZSTD_FALLTHROUGH;
 
         case zdss_read:
             DEBUGLOG(5, "stage zdss_read");
@@ -2051,13 +2223,14 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 }
                 if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
                     FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
+                    assert(ip != NULL);
                     ip += neededInSize;
                     /* Function modifies the stage so we must break */
                     break;
             }   }
             if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
             zds->streamStage = zdss_load;
-            /* fall-through */
+            ZSTD_FALLTHROUGH;
 
         case zdss_load:
             {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
@@ -2065,7 +2238,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 int const isSkipFrame = ZSTD_isSkipFrame(zds);
                 size_t loadedSize;
                 /* At this point we shouldn't be decompressing a block that we can stream. */
-                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip));
+                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, (size_t)(iend - ip)));
                 if (isSkipFrame) {
                     loadedSize = MIN(toLoad, (size_t)(iend-ip));
                 } else {
@@ -2074,8 +2247,11 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                                     "should never happen");
                     loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, (size_t)(iend-ip));
                 }
-                ip += loadedSize;
-                zds->inPos += loadedSize;
+                if (loadedSize != 0) {
+                    /* ip may be NULL */
+                    ip += loadedSize;
+                    zds->inPos += loadedSize;
+                }
                 if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
 
                 /* decode loaded input */
@@ -2085,14 +2261,17 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 break;
             }
         case zdss_flush:
-            {   size_t const toFlushSize = zds->outEnd - zds->outStart;
+            {
+                size_t const toFlushSize = zds->outEnd - zds->outStart;
                 size_t const flushedSize = ZSTD_limitCopy(op, (size_t)(oend-op), zds->outBuff + zds->outStart, toFlushSize);
-                op += flushedSize;
+
+                op = op ? op + flushedSize : op;
+
                 zds->outStart += flushedSize;
                 if (flushedSize == toFlushSize) {  /* flush completed */
                     zds->streamStage = zdss_read;
                     if ( (zds->outBuffSize < zds->fParams.frameContentSize)
-                      && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
+                        && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
                         DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
                                 (int)(zds->outBuffSize - zds->outStart),
                                 (U32)zds->fParams.blockSizeMax);
@@ -2106,7 +2285,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
 
         default:
             assert(0);    /* impossible */
-            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
+            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compilers require default to do something */
     }   }
 
     /* result */
@@ -2119,8 +2298,8 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
     if ((ip==istart) && (op==ostart)) {  /* no forward progress */
         zds->noForwardProgress ++;
         if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
-            RETURN_ERROR_IF(op==oend, dstSize_tooSmall, "");
-            RETURN_ERROR_IF(ip==iend, srcSize_wrong, "");
+            RETURN_ERROR_IF(op==oend, noForwardProgress_destFull, "");
+            RETURN_ERROR_IF(ip==iend, noForwardProgress_inputEmpty, "");
             assert(0);
         }
     } else {
@@ -2157,11 +2336,17 @@ size_t ZSTD_decompressStream_simpleArgs (
                             void* dst, size_t dstCapacity, size_t* dstPos,
                       const void* src, size_t srcSize, size_t* srcPos)
 {
-    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
-    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
-    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
-    size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
-    *dstPos = output.pos;
-    *srcPos = input.pos;
-    return cErr;
+    ZSTD_outBuffer output;
+    ZSTD_inBuffer  input;
+    output.dst = dst;
+    output.size = dstCapacity;
+    output.pos = *dstPos;
+    input.src = src;
+    input.size = srcSize;
+    input.pos = *srcPos;
+    {   size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
+        *dstPos = output.pos;
+        *srcPos = input.pos;
+        return cErr;
+    }
 }
diff --git a/priv/zstd/lib/decompress/zstd_decompress_block.c b/priv/zstd/lib/decompress/zstd_decompress_block.c
index 349dcdc..0a06a02 100644
--- a/priv/zstd/lib/decompress/zstd_decompress_block.c
+++ b/priv/zstd/lib/decompress/zstd_decompress_block.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -20,12 +20,12 @@
 #include "../common/mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "../common/fse.h"
-#define HUF_STATIC_LINKING_ONLY
 #include "../common/huf.h"
 #include "../common/zstd_internal.h"
 #include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
 #include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
 #include "zstd_decompress_block.h"
+#include "../common/bits.h"  /* ZSTD_highbit32 */
 
 /*_*******************************************************
 *  Macros
@@ -69,15 +69,56 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
     }
 }
 
+/* Allocate buffer for literals, either overlapping current dst, or split between dst and litExtraBuffer, or stored entirely within litExtraBuffer */
+static void ZSTD_allocateLiteralsBuffer(ZSTD_DCtx* dctx, void* const dst, const size_t dstCapacity, const size_t litSize,
+    const streaming_operation streaming, const size_t expectedWriteSize, const unsigned splitImmediately)
+{
+    if (streaming == not_streaming && dstCapacity > ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH + litSize + WILDCOPY_OVERLENGTH)
+    {
+        /* room for litbuffer to fit without read faulting */
+        dctx->litBuffer = (BYTE*)dst + ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_in_dst;
+    }
+    else if (litSize > ZSTD_LITBUFFEREXTRASIZE)
+    {
+        /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
+        if (splitImmediately) {
+            /* won't fit in litExtraBuffer, so it will be split between end of dst and extra buffer */
+            dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
+            dctx->litBufferEnd = dctx->litBuffer + litSize - ZSTD_LITBUFFEREXTRASIZE;
+        }
+        else {
+            /* initially this will be stored entirely in dst during huffman decoding, it will partially be shifted to litExtraBuffer after */
+            dctx->litBuffer = (BYTE*)dst + expectedWriteSize - litSize;
+            dctx->litBufferEnd = (BYTE*)dst + expectedWriteSize;
+        }
+        dctx->litBufferLocation = ZSTD_split;
+    }
+    else
+    {
+        /* fits entirely within litExtraBuffer, so no split is necessary */
+        dctx->litBuffer = dctx->litExtraBuffer;
+        dctx->litBufferEnd = dctx->litBuffer + litSize;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    }
+}
 
 /* Hidden declaration for fullbench */
 size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
-                          const void* src, size_t srcSize);
+                          const void* src, size_t srcSize,
+                          void* dst, size_t dstCapacity, const streaming_operation streaming);
 /*! ZSTD_decodeLiteralsBlock() :
+ * Where it is possible to do so without being stomped by the output during decompression, the literals block will be stored
+ * in the dstBuffer.  If there is room to do so, it will be stored in full in the excess dst space after where the current
+ * block will be output.  Otherwise it will be stored at the end of the current dst blockspace, with a small portion being
+ * stored in dctx->litExtraBuffer to help keep it "ahead" of the current output write.
+ *
  * @return : nb of bytes read from src (< srcSize )
  *  note : symbol not declared but exposed for fullbench */
 size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
-                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+                          const void* src, size_t srcSize,   /* note : srcSize < BLOCKSIZE */
+                          void* dst, size_t dstCapacity, const streaming_operation streaming)
 {
     DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
     RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected, "");
@@ -90,15 +131,19 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
         case set_repeat:
             DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
             RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted, "");
-            /* fall-through */
+            ZSTD_FALLTHROUGH;
 
         case set_compressed:
-            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3");
+            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need up to 5 for case 3");
             {   size_t lhSize, litSize, litCSize;
                 U32 singleStream=0;
                 U32 const lhlCode = (istart[0] >> 2) & 3;
                 U32 const lhc = MEM_readLE32(istart);
                 size_t hufSuccess;
+                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
+                int const flags = 0
+                    | (ZSTD_DCtx_get_bmi2(dctx) ? HUF_flags_bmi2 : 0)
+                    | (dctx->disableHufAsm ? HUF_flags_disableAsm : 0);
                 switch(lhlCode)
                 {
                 case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -121,8 +166,15 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
                     break;
                 }
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                if (!singleStream)
+                    RETURN_ERROR_IF(litSize < MIN_LITERALS_FOR_4_STREAMS, literals_headerWrong,
+                        "Not enough literals (%zu) for the 4-streams mode (min %u)",
+                        litSize, MIN_LITERALS_FOR_4_STREAMS);
                 RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
+                RETURN_ERROR_IF(expectedWriteSize < litSize , dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 0);
 
                 /* prefetch huffman table if cold */
                 if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
@@ -131,13 +183,14 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
 
                 if (litEncType==set_repeat) {
                     if (singleStream) {
-                        hufSuccess = HUF_decompress1X_usingDTable_bmi2(
+                        hufSuccess = HUF_decompress1X_usingDTable(
                             dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                            dctx->HUFptr, dctx->bmi2);
+                            dctx->HUFptr, flags);
                     } else {
-                        hufSuccess = HUF_decompress4X_usingDTable_bmi2(
+                        assert(litSize >= MIN_LITERALS_FOR_4_STREAMS);
+                        hufSuccess = HUF_decompress4X_usingDTable(
                             dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                            dctx->HUFptr, dctx->bmi2);
+                            dctx->HUFptr, flags);
                     }
                 } else {
                     if (singleStream) {
@@ -145,20 +198,27 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                         hufSuccess = HUF_decompress1X_DCtx_wksp(
                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
                             istart+lhSize, litCSize, dctx->workspace,
-                            sizeof(dctx->workspace));
+                            sizeof(dctx->workspace), flags);
 #else
-                        hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2(
+                        hufSuccess = HUF_decompress1X1_DCtx_wksp(
                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
                             istart+lhSize, litCSize, dctx->workspace,
-                            sizeof(dctx->workspace), dctx->bmi2);
+                            sizeof(dctx->workspace), flags);
 #endif
                     } else {
-                        hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2(
+                        hufSuccess = HUF_decompress4X_hufOnly_wksp(
                             dctx->entropy.hufTable, dctx->litBuffer, litSize,
                             istart+lhSize, litCSize, dctx->workspace,
-                            sizeof(dctx->workspace), dctx->bmi2);
+                            sizeof(dctx->workspace), flags);
                     }
                 }
+                if (dctx->litBufferLocation == ZSTD_split)
+                {
+                    ZSTD_memcpy(dctx->litExtraBuffer, dctx->litBufferEnd - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memmove(dctx->litBuffer + ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH, dctx->litBuffer, litSize - ZSTD_LITBUFFEREXTRASIZE);
+                    dctx->litBuffer += ZSTD_LITBUFFEREXTRASIZE - WILDCOPY_OVERLENGTH;
+                    dctx->litBufferEnd -= WILDCOPY_OVERLENGTH;
+                }
 
                 RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
 
@@ -166,13 +226,13 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                 dctx->litSize = litSize;
                 dctx->litEntropy = 1;
                 if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
-                ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                 return litCSize + lhSize;
             }
 
         case set_basic:
             {   size_t litSize, lhSize;
                 U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
                 switch(lhlCode)
                 {
                 case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -185,27 +245,41 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 case 3:
                     lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize = 3");
                     litSize = MEM_readLE24(istart) >> 4;
                     break;
                 }
 
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
+                RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
                 if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
                     RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
-                    ZSTD_memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                    if (dctx->litBufferLocation == ZSTD_split)
+                    {
+                        ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize - ZSTD_LITBUFFEREXTRASIZE);
+                        ZSTD_memcpy(dctx->litExtraBuffer, istart + lhSize + litSize - ZSTD_LITBUFFEREXTRASIZE, ZSTD_LITBUFFEREXTRASIZE);
+                    }
+                    else
+                    {
+                        ZSTD_memcpy(dctx->litBuffer, istart + lhSize, litSize);
+                    }
                     dctx->litPtr = dctx->litBuffer;
                     dctx->litSize = litSize;
-                    ZSTD_memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                     return lhSize+litSize;
                 }
                 /* direct reference into compressed stream */
                 dctx->litPtr = istart+lhSize;
                 dctx->litSize = litSize;
+                dctx->litBufferEnd = dctx->litPtr + litSize;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
                 return lhSize+litSize;
             }
 
         case set_rle:
             {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
                 size_t litSize, lhSize;
+                size_t expectedWriteSize = MIN(ZSTD_BLOCKSIZE_MAX, dstCapacity);
                 switch(lhlCode)
                 {
                 case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -214,16 +288,28 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 case 1:
                     lhSize = 2;
+                    RETURN_ERROR_IF(srcSize<3, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 3");
                     litSize = MEM_readLE16(istart) >> 4;
                     break;
                 case 3:
                     lhSize = 3;
+                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 2; here we need lhSize+1 = 4");
                     litSize = MEM_readLE24(istart) >> 4;
-                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4");
                     break;
                 }
+                RETURN_ERROR_IF(litSize > 0 && dst == NULL, dstSize_tooSmall, "NULL not handled");
                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
-                ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+                RETURN_ERROR_IF(expectedWriteSize < litSize, dstSize_tooSmall, "");
+                ZSTD_allocateLiteralsBuffer(dctx, dst, dstCapacity, litSize, streaming, expectedWriteSize, 1);
+                if (dctx->litBufferLocation == ZSTD_split)
+                {
+                    ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize - ZSTD_LITBUFFEREXTRASIZE);
+                    ZSTD_memset(dctx->litExtraBuffer, istart[lhSize], ZSTD_LITBUFFEREXTRASIZE);
+                }
+                else
+                {
+                    ZSTD_memset(dctx->litBuffer, istart[lhSize], litSize);
+                }
                 dctx->litPtr = dctx->litBuffer;
                 dctx->litSize = litSize;
                 return lhSize+1;
@@ -343,7 +429,7 @@ static const ZSTD_seqSymbol ML_defaultDTable[(1<nbBits = 0;
     cell->nextState = 0;
     assert(nbAddBits < 255);
-    cell->nbAdditionalBits = (BYTE)nbAddBits;
+    cell->nbAdditionalBits = nbAddBits;
     cell->baseValue = baseValue;
 }
 
@@ -367,7 +453,7 @@ static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddB
 FORCE_INLINE_TEMPLATE
 void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
             const short* normalizedCounter, unsigned maxSymbolValue,
-            const U32* baseValue, const U32* nbAdditionalBits,
+            const U32* baseValue, const U8* nbAdditionalBits,
             unsigned tableLog, void* wksp, size_t wkspSize)
 {
     ZSTD_seqSymbol* const tableDecode = dt+1;
@@ -430,14 +516,15 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
                 for (i = 8; i < n; i += 8) {
                     MEM_write64(spread + pos + i, sv);
                 }
-                pos += n;
+                assert(n>=0);
+                pos += (size_t)n;
             }
         }
         /* Now we spread those positions across the table.
-         * The benefit of doing it in two stages is that we avoid the the
+         * The benefit of doing it in two stages is that we avoid the
          * variable size inner loop, which caused lots of branch misses.
          * Now we can run through all the positions without any branch misses.
-         * We unroll the loop twice, since that is what emperically worked best.
+         * We unroll the loop twice, since that is what empirically worked best.
          */
         {
             size_t position = 0;
@@ -464,7 +551,7 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
             for (i=0; i highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+                while (UNLIKELY(position > highThreshold)) position = (position + step) & tableMask;   /* lowprob area */
         }   }
         assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
     }
@@ -475,10 +562,10 @@ void ZSTD_buildFSETable_body(ZSTD_seqSymbol* dt,
         for (u=0; u max, corruption_detected, "");
         {   U32 const symbol = *(const BYTE*)src;
             U32 const baseline = baseValue[symbol];
-            U32 const nbBits = nbAdditionalBits[symbol];
+            U8 const nbBits = nbAdditionalBits[symbol];
             ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
         }
         *DTablePtr = DTableSpace;
@@ -620,7 +707,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       LL_defaultDTable, dctx->fseEntropy,
                                                       dctx->ddictIsCold, nbSeq,
                                                       dctx->workspace, sizeof(dctx->workspace),
-                                                      dctx->bmi2);
+                                                      ZSTD_DCtx_get_bmi2(dctx));
             RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
             ip += llhSize;
         }
@@ -632,7 +719,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       OF_defaultDTable, dctx->fseEntropy,
                                                       dctx->ddictIsCold, nbSeq,
                                                       dctx->workspace, sizeof(dctx->workspace),
-                                                      dctx->bmi2);
+                                                      ZSTD_DCtx_get_bmi2(dctx));
             RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
             ip += ofhSize;
         }
@@ -644,7 +731,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                                                       ML_defaultDTable, dctx->fseEntropy,
                                                       dctx->ddictIsCold, nbSeq,
                                                       dctx->workspace, sizeof(dctx->workspace),
-                                                      dctx->bmi2);
+                                                      ZSTD_DCtx_get_bmi2(dctx));
             RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
             ip += mlhSize;
         }
@@ -713,7 +800,7 @@ HINT_INLINE void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset) {
  *         - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
  *           The src buffer must be before the dst buffer.
  */
-static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
+static void ZSTD_safecopy(BYTE* op, const BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
     ptrdiff_t const diff = op - ip;
     BYTE* const oend = op + length;
 
@@ -729,6 +816,7 @@ static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_
         /* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
         assert(length >= 8);
         ZSTD_overlapCopy8(&op, &ip, diff);
+        length -= 8;
         assert(op - ip >= 8);
         assert(op <= oend);
     }
@@ -743,12 +831,35 @@ static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_
         assert(oend > oend_w);
         ZSTD_wildcopy(op, ip, oend_w - op, ovtype);
         ip += oend_w - op;
-        op = oend_w;
+        op += oend_w - op;
     }
     /* Handle the leftovers. */
     while (op < oend) *op++ = *ip++;
 }
 
+/* ZSTD_safecopyDstBeforeSrc():
+ * This version allows overlap with dst before src, or handles the non-overlap case with dst after src
+ * Kept separate from more common ZSTD_safecopy case to avoid performance impact to the safecopy common case */
+static void ZSTD_safecopyDstBeforeSrc(BYTE* op, BYTE const* ip, ptrdiff_t length) {
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
+
+    if (length < 8 || diff > -8) {
+        /* Handle short lengths, close overlaps, and dst not before src. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+
+    if (op <= oend - WILDCOPY_OVERLENGTH && diff < -WILDCOPY_VECLEN) {
+        ZSTD_wildcopy(op, ip, oend - WILDCOPY_OVERLENGTH - op, ZSTD_no_overlap);
+        ip += oend - WILDCOPY_OVERLENGTH - op;
+        op += oend - WILDCOPY_OVERLENGTH - op;
+    }
+
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
 /* ZSTD_execSequenceEnd():
  * This version handles cases that are near the end of the output buffer. It requires
  * more careful checks to make sure there is no overflow. By separating out these hard
@@ -759,9 +870,9 @@ static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_
  */
 FORCE_NOINLINE
 size_t ZSTD_execSequenceEnd(BYTE* op,
-                            BYTE* const oend, seq_t sequence,
-                            const BYTE** litPtr, const BYTE* const litLimit,
-                            const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+    BYTE* const oend, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
 {
     BYTE* const oLitEnd = op + sequence.litLength;
     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -784,27 +895,76 @@ size_t ZSTD_execSequenceEnd(BYTE* op,
     if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
         /* offset beyond prefix */
         RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
-        match = dictEnd - (prefixStart-match);
+        match = dictEnd - (prefixStart - match);
         if (match + sequence.matchLength <= dictEnd) {
             ZSTD_memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
         }
         /* span extDict & currentPrefixSegment */
         {   size_t const length1 = dictEnd - match;
-            ZSTD_memmove(oLitEnd, match, length1);
-            op = oLitEnd + length1;
-            sequence.matchLength -= length1;
-            match = prefixStart;
-    }   }
+        ZSTD_memmove(oLitEnd, match, length1);
+        op = oLitEnd + length1;
+        sequence.matchLength -= length1;
+        match = prefixStart;
+        }
+    }
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
+    return sequenceLength;
+}
+
+/* ZSTD_execSequenceEndSplitLitBuffer():
+ * This version is intended to be used during instances where the litBuffer is still split.  It is kept separate to avoid performance impact for the good case.
+ */
+FORCE_NOINLINE
+size_t ZSTD_execSequenceEndSplitLitBuffer(BYTE* op,
+    BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+
+    /* bounds checks : careful of address space overflow in 32-bit mode */
+    RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
+    RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
+    assert(op < op + sequenceLength);
+    assert(oLitEnd < op + sequenceLength);
+
+    /* copy literals */
+    RETURN_ERROR_IF(op > *litPtr && op < *litPtr + sequence.litLength, dstSize_tooSmall, "output should not catch up to and overwrite literal buffer");
+    ZSTD_safecopyDstBeforeSrc(op, *litPtr, sequence.litLength);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
+        match = dictEnd - (prefixStart - match);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+        ZSTD_memmove(oLitEnd, match, length1);
+        op = oLitEnd + length1;
+        sequence.matchLength -= length1;
+        match = prefixStart;
+        }
+    }
     ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
     return sequenceLength;
 }
 
 HINT_INLINE
 size_t ZSTD_execSequence(BYTE* op,
-                         BYTE* const oend, seq_t sequence,
-                         const BYTE** litPtr, const BYTE* const litLimit,
-                         const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+    BYTE* const oend, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
 {
     BYTE* const oLitEnd = op + sequence.litLength;
     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -815,6 +975,103 @@ size_t ZSTD_execSequence(BYTE* op,
 
     assert(op != NULL /* Precondition */);
     assert(oend_w < oend /* No underflow */);
+
+#if defined(__aarch64__)
+    /* prefetch sequence starting from match that will be used for copy later */
+    PREFETCH_L1(match);
+#endif
+    /* Handle edge cases in a slow path:
+     *   - Read beyond end of literals
+     *   - Match end is within WILDCOPY_OVERLIMIT of oend
+     *   - 32-bit mode and the match length overflows
+     */
+    if (UNLIKELY(
+        iLitEnd > litLimit ||
+        oMatchEnd > oend_w ||
+        (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
+        return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(op <= oLitEnd /* No overflow */);
+    assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
+    assert(oMatchEnd <= oend /* No underflow */);
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (UNLIKELY(sequence.litLength > 16)) {
+        ZSTD_wildcopy(op + 16, (*litPtr) + 16, sequence.litLength - 16, ZSTD_no_overlap);
+    }
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* Copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix -> go into extDict */
+        RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
+        match = dictEnd + (match - prefixStart);
+        if (match + sequence.matchLength <= dictEnd) {
+            ZSTD_memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+        ZSTD_memmove(oLitEnd, match, length1);
+        op = oLitEnd + length1;
+        sequence.matchLength -= length1;
+        match = prefixStart;
+        }
+    }
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
+    }
+    assert(sequence.offset < WILDCOPY_VECLEN);
+
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
+
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength - 8, ZSTD_overlap_src_before_dst);
+    }
+    return sequenceLength;
+}
+
+HINT_INLINE
+size_t ZSTD_execSequenceSplitLitBuffer(BYTE* op,
+    BYTE* const oend, const BYTE* const oend_w, seq_t sequence,
+    const BYTE** litPtr, const BYTE* const litLimit,
+    const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    assert(op != NULL /* Precondition */);
+    assert(oend_w < oend /* No underflow */);
     /* Handle edge cases in a slow path:
      *   - Read beyond end of literals
      *   - Match end is within WILDCOPY_OVERLIMIT of oend
@@ -824,7 +1081,7 @@ size_t ZSTD_execSequence(BYTE* op,
             iLitEnd > litLimit ||
             oMatchEnd > oend_w ||
             (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
-        return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+        return ZSTD_execSequenceEndSplitLitBuffer(op, oend, oend_w, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
 
     /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
     assert(op <= oLitEnd /* No overflow */);
@@ -892,6 +1149,7 @@ size_t ZSTD_execSequence(BYTE* op,
     return sequenceLength;
 }
 
+
 static void
 ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
 {
@@ -905,24 +1163,14 @@ ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqS
 }
 
 FORCE_INLINE_TEMPLATE void
-ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD)
+ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, U16 nextState, U32 nbBits)
 {
-    ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state];
-    U32 const nbBits = DInfo.nbBits;
     size_t const lowBits = BIT_readBits(bitD, nbBits);
-    DStatePtr->state = DInfo.nextState + lowBits;
-}
-
-FORCE_INLINE_TEMPLATE void
-ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, ZSTD_seqSymbol const DInfo)
-{
-    U32 const nbBits = DInfo.nbBits;
-    size_t const lowBits = BIT_readBits(bitD, nbBits);
-    DStatePtr->state = DInfo.nextState + lowBits;
+    DStatePtr->state = nextState + lowBits;
 }
 
 /* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
- * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
+ * offset bits. But we can only read at most STREAM_ACCUMULATOR_MIN_32
  * bits before reloading. This value is the maximum number of bytes we read
  * after reloading when we are decoding long offsets.
  */
@@ -937,102 +1185,125 @@ FORCE_INLINE_TEMPLATE seq_t
 ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
 {
     seq_t seq;
-    ZSTD_seqSymbol const llDInfo = seqState->stateLL.table[seqState->stateLL.state];
-    ZSTD_seqSymbol const mlDInfo = seqState->stateML.table[seqState->stateML.state];
-    ZSTD_seqSymbol const ofDInfo = seqState->stateOffb.table[seqState->stateOffb.state];
-    U32 const llBase = llDInfo.baseValue;
-    U32 const mlBase = mlDInfo.baseValue;
-    U32 const ofBase = ofDInfo.baseValue;
-    BYTE const llBits = llDInfo.nbAdditionalBits;
-    BYTE const mlBits = mlDInfo.nbAdditionalBits;
-    BYTE const ofBits = ofDInfo.nbAdditionalBits;
-    BYTE const totalBits = llBits+mlBits+ofBits;
-
-    /* sequence */
-    {   size_t offset;
-        if (ofBits > 1) {
-            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
-            ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
-            assert(ofBits <= MaxOff);
-            if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
-                U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
-                offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
-                BIT_reloadDStream(&seqState->DStream);
-                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
-                assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32);   /* to avoid another reload */
-            } else {
-                offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
-                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
-            }
-            seqState->prevOffset[2] = seqState->prevOffset[1];
-            seqState->prevOffset[1] = seqState->prevOffset[0];
-            seqState->prevOffset[0] = offset;
-        } else {
-            U32 const ll0 = (llBase == 0);
-            if (LIKELY((ofBits == 0))) {
-                if (LIKELY(!ll0))
-                    offset = seqState->prevOffset[0];
-                else {
-                    offset = seqState->prevOffset[1];
-                    seqState->prevOffset[1] = seqState->prevOffset[0];
-                    seqState->prevOffset[0] = offset;
-                }
-            } else {
-                offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
-                {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
-                    temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
-                    if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
-                    seqState->prevOffset[1] = seqState->prevOffset[0];
-                    seqState->prevOffset[0] = offset = temp;
-        }   }   }
-        seq.offset = offset;
-    }
-
-    seq.matchLength = mlBase;
-    if (mlBits > 0)
-        seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
-
-    if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
-        BIT_reloadDStream(&seqState->DStream);
-    if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
-        BIT_reloadDStream(&seqState->DStream);
-    /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
-    ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
-
-    seq.litLength = llBase;
-    if (llBits > 0)
-        seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
-
-    if (MEM_32bits())
-        BIT_reloadDStream(&seqState->DStream);
-
-    DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
-                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
-
-    /* ANS state update
-     * gcc-9.0.0 does 2.5% worse with ZSTD_updateFseStateWithDInfo().
-     * clang-9.2.0 does 7% worse with ZSTD_updateFseState().
-     * Naturally it seems like ZSTD_updateFseStateWithDInfo() should be the
-     * better option, so it is the default for other compilers. But, if you
-     * measure that it is worse, please put up a pull request.
+    /*
+     * ZSTD_seqSymbol is a structure with a total of 64 bits wide. So it can be
+     * loaded in one operation and extracted its fields by simply shifting or
+     * bit-extracting on aarch64.
+     * GCC doesn't recognize this and generates more unnecessary ldr/ldrb/ldrh
+     * operations that cause performance drop. This can be avoided by using this
+     * ZSTD_memcpy hack.
      */
-    {
-#if defined(__GNUC__) && !defined(__clang__)
-        const int kUseUpdateFseState = 1;
+#if defined(__aarch64__) && (defined(__GNUC__) && !defined(__clang__))
+    ZSTD_seqSymbol llDInfoS, mlDInfoS, ofDInfoS;
+    ZSTD_seqSymbol* const llDInfo = &llDInfoS;
+    ZSTD_seqSymbol* const mlDInfo = &mlDInfoS;
+    ZSTD_seqSymbol* const ofDInfo = &ofDInfoS;
+    ZSTD_memcpy(llDInfo, seqState->stateLL.table + seqState->stateLL.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(mlDInfo, seqState->stateML.table + seqState->stateML.state, sizeof(ZSTD_seqSymbol));
+    ZSTD_memcpy(ofDInfo, seqState->stateOffb.table + seqState->stateOffb.state, sizeof(ZSTD_seqSymbol));
 #else
-        const int kUseUpdateFseState = 0;
+    const ZSTD_seqSymbol* const llDInfo = seqState->stateLL.table + seqState->stateLL.state;
+    const ZSTD_seqSymbol* const mlDInfo = seqState->stateML.table + seqState->stateML.state;
+    const ZSTD_seqSymbol* const ofDInfo = seqState->stateOffb.table + seqState->stateOffb.state;
 #endif
-        if (kUseUpdateFseState) {
-            ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
-            ZSTD_updateFseState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
-            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-            ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
-        } else {
-            ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llDInfo);    /* <=  9 bits */
-            ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlDInfo);    /* <=  9 bits */
-            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-            ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofDInfo);  /* <=  8 bits */
+    seq.matchLength = mlDInfo->baseValue;
+    seq.litLength = llDInfo->baseValue;
+    {   U32 const ofBase = ofDInfo->baseValue;
+        BYTE const llBits = llDInfo->nbAdditionalBits;
+        BYTE const mlBits = mlDInfo->nbAdditionalBits;
+        BYTE const ofBits = ofDInfo->nbAdditionalBits;
+        BYTE const totalBits = llBits+mlBits+ofBits;
+
+        U16 const llNext = llDInfo->nextState;
+        U16 const mlNext = mlDInfo->nextState;
+        U16 const ofNext = ofDInfo->nextState;
+        U32 const llnbBits = llDInfo->nbBits;
+        U32 const mlnbBits = mlDInfo->nbBits;
+        U32 const ofnbBits = ofDInfo->nbBits;
+
+        assert(llBits <= MaxLLBits);
+        assert(mlBits <= MaxMLBits);
+        assert(ofBits <= MaxOff);
+        /*
+         * As gcc has better branch and block analyzers, sometimes it is only
+         * valuable to mark likeliness for clang, it gives around 3-4% of
+         * performance.
+         */
+
+        /* sequence */
+        {   size_t offset;
+    #if defined(__clang__)
+            if (LIKELY(ofBits > 1)) {
+    #else
+            if (ofBits > 1) {
+    #endif
+                ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+                ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 > LONG_OFFSETS_MAX_EXTRA_BITS_32);
+                ZSTD_STATIC_ASSERT(STREAM_ACCUMULATOR_MIN_32 - LONG_OFFSETS_MAX_EXTRA_BITS_32 >= MaxMLBits);
+                if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
+                    /* Always read extra bits, this keeps the logic simple,
+                     * avoids branches, and avoids accidentally reading 0 bits.
+                     */
+                    U32 const extraBits = LONG_OFFSETS_MAX_EXTRA_BITS_32;
+                    offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                    BIT_reloadDStream(&seqState->DStream);
+                    offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+                } else {
+                    offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+                }
+                seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset;
+            } else {
+                U32 const ll0 = (llDInfo->baseValue == 0);
+                if (LIKELY((ofBits == 0))) {
+                    offset = seqState->prevOffset[ll0];
+                    seqState->prevOffset[1] = seqState->prevOffset[!ll0];
+                    seqState->prevOffset[0] = offset;
+                } else {
+                    offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
+                    {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                        temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                        if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                        seqState->prevOffset[1] = seqState->prevOffset[0];
+                        seqState->prevOffset[0] = offset = temp;
+            }   }   }
+            seq.offset = offset;
         }
+
+    #if defined(__clang__)
+        if (UNLIKELY(mlBits > 0))
+    #else
+        if (mlBits > 0)
+    #endif
+            seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
+
+        if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+            BIT_reloadDStream(&seqState->DStream);
+        if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+            BIT_reloadDStream(&seqState->DStream);
+        /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
+        ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+    #if defined(__clang__)
+        if (UNLIKELY(llBits > 0))
+    #else
+        if (llBits > 0)
+    #endif
+            seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
+
+        if (MEM_32bits())
+            BIT_reloadDStream(&seqState->DStream);
+
+        DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
+                    (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+
+        ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llNext, llnbBits);    /* <=  9 bits */
+        ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlNext, mlnbBits);    /* <=  9 bits */
+        if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+        ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofNext, ofnbBits);  /* <=  8 bits */
     }
 
     return seq;
@@ -1085,9 +1356,11 @@ MEM_STATIC void ZSTD_assertValidSequence(
 #endif
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+
+
 FORCE_INLINE_TEMPLATE size_t
 DONT_VECTORIZE
-ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
+ZSTD_decompressSequences_bodySplitLitBuffer( ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
                          const void* seqStart, size_t seqSize, int nbSeq,
                          const ZSTD_longOffset_e isLongOffset,
@@ -1099,11 +1372,11 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
     BYTE* const oend = ostart + maxDstSize;
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
-    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* litBufferEnd = dctx->litBufferEnd;
     const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
     const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    DEBUGLOG(5, "ZSTD_decompressSequences_body");
+    DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer");
     (void)frame;
 
     /* Regen sequences */
@@ -1124,55 +1397,237 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
                 BIT_DStream_endOfBuffer < BIT_DStream_completed &&
                 BIT_DStream_completed < BIT_DStream_overflow);
 
+        /* decompress without overrunning litPtr begins */
+        {
+            seq_t sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            /* Align the decompression loop to 32 + 16 bytes.
+                *
+                * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
+                * speed swings based on the alignment of the decompression loop. This
+                * performance swing is caused by parts of the decompression loop falling
+                * out of the DSB. The entire decompression loop should fit in the DSB,
+                * when it can't we get much worse performance. You can measure if you've
+                * hit the good case or the bad case with this perf command for some
+                * compressed file test.zst:
+                *
+                *   perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
+                *             -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
+                *
+                * If you see most cycles served out of the MITE you've hit the bad case.
+                * If you see most cycles served out of the DSB you've hit the good case.
+                * If it is pretty even then you may be in an okay case.
+                *
+                * This issue has been reproduced on the following CPUs:
+                *   - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
+                *               Use Instruments->Counters to get DSB/MITE cycles.
+                *               I never got performance swings, but I was able to
+                *               go from the good case of mostly DSB to half of the
+                *               cycles served from MITE.
+                *   - Coffeelake: Intel i9-9900k
+                *   - Coffeelake: Intel i7-9700k
+                *
+                * I haven't been able to reproduce the instability or DSB misses on any
+                * of the following CPUS:
+                *   - Haswell
+                *   - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
+                *   - Skylake
+                *
+                * Alignment is done for each of the three major decompression loops:
+                *   - ZSTD_decompressSequences_bodySplitLitBuffer - presplit section of the literal buffer
+                *   - ZSTD_decompressSequences_bodySplitLitBuffer - postsplit section of the literal buffer
+                *   - ZSTD_decompressSequences_body
+                * Alignment choices are made to minimize large swings on bad cases and influence on performance
+                * from changes external to this code, rather than to overoptimize on the current commit.
+                *
+                * If you are seeing performance stability this script can help test.
+                * It tests on 4 commits in zstd where I saw performance change.
+                *
+                *   https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
+                */
 #if defined(__GNUC__) && defined(__x86_64__)
-        /* Align the decompression loop to 32 + 16 bytes.
-         *
-         * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
-         * speed swings based on the alignment of the decompression loop. This
-         * performance swing is caused by parts of the decompression loop falling
-         * out of the DSB. The entire decompression loop should fit in the DSB,
-         * when it can't we get much worse performance. You can measure if you've
-         * hit the good case or the bad case with this perf command for some
-         * compressed file test.zst:
-         *
-         *   perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
-         *             -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
-         *
-         * If you see most cycles served out of the MITE you've hit the bad case.
-         * If you see most cycles served out of the DSB you've hit the good case.
-         * If it is pretty even then you may be in an okay case.
-         *
-         * This issue has been reproduced on the following CPUs:
-         *   - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
-         *               Use Instruments->Counters to get DSB/MITE cycles.
-         *               I never got performance swings, but I was able to
-         *               go from the good case of mostly DSB to half of the
-         *               cycles served from MITE.
-         *   - Coffeelake: Intel i9-9900k
-         *   - Coffeelake: Intel i7-9700k
-         *
-         * I haven't been able to reproduce the instability or DSB misses on any
-         * of the following CPUS:
-         *   - Haswell
-         *   - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
-         *   - Skylake
-         *
-         * If you are seeing performance stability this script can help test.
-         * It tests on 4 commits in zstd where I saw performance change.
-         *
-         *   https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
-         */
-        __asm__(".p2align 6");
-        __asm__("nop");
-        __asm__(".p2align 5");
-        __asm__("nop");
-#  if __GNUC__ >= 9
-        /* better for gcc-9 and gcc-10, worse for clang and gcc-8 */
-        __asm__(".p2align 3");
+            __asm__(".p2align 6");
+#  if __GNUC__ >= 7
+	    /* good for gcc-7, gcc-9, and gcc-11 */
+            __asm__("nop");
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 4");
+#    if __GNUC__ == 8 || __GNUC__ == 10
+	    /* good for gcc-8 and gcc-10 */
+            __asm__("nop");
+            __asm__(".p2align 3");
+#    endif
+#  endif
+#endif
+
+            /* Handle the initial state where litBuffer is currently split between dst and litExtraBuffer */
+            for (; litPtr + sequence.litLength <= dctx->litBufferEnd; ) {
+                size_t const oneSeqSize = ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence.litLength - WILDCOPY_OVERLENGTH, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                    return oneSeqSize;
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                op += oneSeqSize;
+                if (UNLIKELY(!--nbSeq))
+                    break;
+                BIT_reloadDStream(&(seqState.DStream));
+                sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+            }
+
+            /* If there are more sequences, they will need to read literals from litExtraBuffer; copy over the remainder from dst and update litPtr and litEnd */
+            if (nbSeq > 0) {
+                const size_t leftoverLit = dctx->litBufferEnd - litPtr;
+                if (leftoverLit)
+                {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequence.litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {
+                    size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                    assert(!ZSTD_isError(oneSeqSize));
+                    if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                    if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                        return oneSeqSize;
+                    DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                    op += oneSeqSize;
+                    if (--nbSeq)
+                        BIT_reloadDStream(&(seqState.DStream));
+                }
+            }
+        }
+
+        if (nbSeq > 0) /* there is remaining lit from extra buffer */
+        {
+
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+            __asm__("nop");
+#  if __GNUC__ != 7
+            /* worse for gcc-7 better for gcc-8, gcc-9, and gcc-10 and clang */
+            __asm__(".p2align 4");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  elif __GNUC__ >= 11
+            __asm__(".p2align 3");
+#  else
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 3");
+#  endif
+#endif
+
+            for (; ; ) {
+                seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litBufferEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+                if (UNLIKELY(ZSTD_isError(oneSeqSize)))
+                    return oneSeqSize;
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                op += oneSeqSize;
+                if (UNLIKELY(!--nbSeq))
+                    break;
+                BIT_reloadDStream(&(seqState.DStream));
+            }
+        }
+
+        /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences_bodySplitLitBuffer: after decode loop, remaining nbSeq : %i", nbSeq);
+        RETURN_ERROR_IF(nbSeq, corruption_detected, "");
+        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
+        /* save reps for next block */
+        { U32 i; for (i=0; ientropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    if (dctx->litBufferLocation == ZSTD_split)  /* split hasn't been reached yet, first get dst then copy litExtraBuffer */
+    {
+        size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+        litPtr = dctx->litExtraBuffer;
+        litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+        dctx->litBufferLocation = ZSTD_not_in_dst;
+    }
+    {   size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_body(ZSTD_DCtx* dctx,
+    void* dst, size_t maxDstSize,
+    const void* seqStart, size_t seqSize, int nbSeq,
+    const ZSTD_longOffset_e isLongOffset,
+    const int frame)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = dctx->litBufferLocation == ZSTD_not_in_dst ? ostart + maxDstSize : dctx->litBuffer;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const prefixStart = (const BYTE*)(dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*)(dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*)(dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences_body: nbSeq = %d", nbSeq);
+    (void)frame;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i = 0; i < ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend - ip)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        assert(dst != NULL);
+
+        ZSTD_STATIC_ASSERT(
+            BIT_DStream_unfinished < BIT_DStream_completed &&
+            BIT_DStream_endOfBuffer < BIT_DStream_completed &&
+            BIT_DStream_completed < BIT_DStream_overflow);
+
+#if defined(__GNUC__) && defined(__x86_64__)
+            __asm__(".p2align 6");
+            __asm__("nop");
+#  if __GNUC__ >= 7
+            __asm__(".p2align 5");
+            __asm__("nop");
+            __asm__(".p2align 3");
 #  else
-        __asm__(".p2align 4");
+            __asm__(".p2align 4");
+            __asm__("nop");
+            __asm__(".p2align 3");
 #  endif
 #endif
+
         for ( ; ; ) {
             seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
             size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
@@ -1218,6 +1673,16 @@ ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
 {
     return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
 }
+
+static size_t
+ZSTD_decompressSequencesSplitLitBuffer_default(ZSTD_DCtx* dctx,
+                                               void* dst, size_t maxDstSize,
+                                         const void* seqStart, size_t seqSize, int nbSeq,
+                                         const ZSTD_longOffset_e isLongOffset,
+                                         const int frame)
+{
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
@@ -1250,10 +1715,10 @@ ZSTD_decompressSequencesLong_body(
     const BYTE* ip = (const BYTE*)seqStart;
     const BYTE* const iend = ip + seqSize;
     BYTE* const ostart = (BYTE*)dst;
-    BYTE* const oend = ostart + maxDstSize;
+    BYTE* const oend = dctx->litBufferLocation == ZSTD_in_dst ? dctx->litBuffer : ostart + maxDstSize;
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
-    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* litBufferEnd = dctx->litBufferEnd;
     const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
     const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
@@ -1289,32 +1754,94 @@ ZSTD_decompressSequencesLong_body(
         }
         RETURN_ERROR_IF(seqNblitBufferLocation == ZSTD_split && litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength > dctx->litBufferEnd)
+            {
+                /* lit buffer is reaching split point, empty out the first buffer and transition to litExtraBuffer */
+                const size_t leftoverLit = dctx->litBufferEnd - litPtr;
+                if (leftoverLit)
+                {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
-            assert(!ZSTD_isError(oneSeqSize));
-            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
 #endif
-            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
 
-            prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
-            sequences[seqNb & STORED_SEQS_MASK] = sequence;
-            op += oneSeqSize;
+                prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                op += oneSeqSize;
+            }
+            else
+            {
+                /* lit buffer is either wholly contained in first or second split, or not split at all*/
+                oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                    ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK].litLength - WILDCOPY_OVERLENGTH, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
+                    ZSTD_execSequence(op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb - ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+
+                prefetchPos = ZSTD_prefetchMatch(prefetchPos, sequence, prefixStart, dictEnd);
+                sequences[seqNb & STORED_SEQS_MASK] = sequence;
+                op += oneSeqSize;
+            }
         }
         RETURN_ERROR_IF(seqNblitBufferLocation == ZSTD_split && litPtr + sequence->litLength > dctx->litBufferEnd)
+            {
+                const size_t leftoverLit = dctx->litBufferEnd - litPtr;
+                if (leftoverLit)
+                {
+                    RETURN_ERROR_IF(leftoverLit > (size_t)(oend - op), dstSize_tooSmall, "remaining lit must fit within dstBuffer");
+                    ZSTD_safecopyDstBeforeSrc(op, litPtr, leftoverLit);
+                    sequence->litLength -= leftoverLit;
+                    op += leftoverLit;
+                }
+                litPtr = dctx->litExtraBuffer;
+                litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+                dctx->litBufferLocation = ZSTD_not_in_dst;
+                {
+                    size_t const oneSeqSize = ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
 #if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
-            assert(!ZSTD_isError(oneSeqSize));
-            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+                    assert(!ZSTD_isError(oneSeqSize));
+                    if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
 #endif
-            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
-            op += oneSeqSize;
+                    if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                    op += oneSeqSize;
+                }
+            }
+            else
+            {
+                size_t const oneSeqSize = dctx->litBufferLocation == ZSTD_split ?
+                    ZSTD_execSequenceSplitLitBuffer(op, oend, litPtr + sequence->litLength - WILDCOPY_OVERLENGTH, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd) :
+                    ZSTD_execSequence(op, oend, *sequence, &litPtr, litBufferEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+                assert(!ZSTD_isError(oneSeqSize));
+                if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+            }
         }
 
         /* save reps for next block */
@@ -1322,10 +1849,21 @@ ZSTD_decompressSequencesLong_body(
     }
 
     /* last literal segment */
-    {   size_t const lastLLSize = litEnd - litPtr;
+    if (dctx->litBufferLocation == ZSTD_split)  /* first deplete literal buffer in dst, then copy litExtraBuffer */
+    {
+        size_t const lastLLSize = litBufferEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend - op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            ZSTD_memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+        litPtr = dctx->litExtraBuffer;
+        litBufferEnd = dctx->litExtraBuffer + ZSTD_LITBUFFEREXTRASIZE;
+    }
+    {   size_t const lastLLSize = litBufferEnd - litPtr;
         RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
         if (op != NULL) {
-            ZSTD_memcpy(op, litPtr, lastLLSize);
+            ZSTD_memmove(op, litPtr, lastLLSize);
             op += lastLLSize;
         }
     }
@@ -1349,7 +1887,7 @@ ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
 #if DYNAMIC_BMI2
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
-static TARGET_ATTRIBUTE("bmi2") size_t
+static BMI2_TARGET_ATTRIBUTE size_t
 DONT_VECTORIZE
 ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
@@ -1359,10 +1897,20 @@ ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
 {
     return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
 }
+static BMI2_TARGET_ATTRIBUTE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequencesSplitLitBuffer_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequences_bodySplitLitBuffer(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
-static TARGET_ATTRIBUTE("bmi2") size_t
+static BMI2_TARGET_ATTRIBUTE size_t
 ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
@@ -1391,11 +1939,25 @@ ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
 {
     DEBUGLOG(5, "ZSTD_decompressSequences");
 #if DYNAMIC_BMI2
-    if (dctx->bmi2) {
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
         return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
     }
 #endif
-  return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+static size_t
+ZSTD_decompressSequencesSplitLitBuffer(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                                 const void* seqStart, size_t seqSize, int nbSeq,
+                                 const ZSTD_longOffset_e isLongOffset,
+                                 const int frame)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesSplitLitBuffer");
+#if DYNAMIC_BMI2
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
+        return ZSTD_decompressSequencesSplitLitBuffer_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    }
+#endif
+    return ZSTD_decompressSequencesSplitLitBuffer_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
 }
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
 
@@ -1415,7 +1977,7 @@ ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
 {
     DEBUGLOG(5, "ZSTD_decompressSequencesLong");
 #if DYNAMIC_BMI2
-    if (dctx->bmi2) {
+    if (ZSTD_DCtx_get_bmi2(dctx)) {
         return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
     }
 #endif
@@ -1424,55 +1986,101 @@ ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
 #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
 
 
+/**
+ * @returns The total size of the history referencable by zstd, including
+ * both the prefix and the extDict. At @p op any offset larger than this
+ * is invalid.
+ */
+static size_t ZSTD_totalHistorySize(BYTE* op, BYTE const* virtualStart)
+{
+    return (size_t)(op - virtualStart);
+}
 
-#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
-    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
-/* ZSTD_getLongOffsetsShare() :
+typedef struct {
+    unsigned longOffsetShare;
+    unsigned maxNbAdditionalBits;
+} ZSTD_OffsetInfo;
+
+/* ZSTD_getOffsetInfo() :
  * condition : offTable must be valid
  * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
- *           compared to maximum possible of (1< 22) total += 1;
+    ZSTD_OffsetInfo info = {0, 0};
+    /* If nbSeq == 0, then the offTable is uninitialized, but we have
+     * no sequences, so both values should be 0.
+     */
+    if (nbSeq != 0) {
+        const void* ptr = offTable;
+        U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+        const ZSTD_seqSymbol* table = offTable + 1;
+        U32 const max = 1 << tableLog;
+        U32 u;
+        DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+        assert(max <= (1 << OffFSELog));  /* max not too large */
+        for (u=0; u 22) info.longOffsetShare += 1;
+        }
+
+        assert(tableLog <= OffFSELog);
+        info.longOffsetShare <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
     }
 
-    assert(tableLog <= OffFSELog);
-    total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+    return info;
+}
 
-    return total;
+/**
+ * @returns The maximum offset we can decode in one read of our bitstream, without
+ * reloading more bits in the middle of the offset bits read. Any offsets larger
+ * than this must use the long offset decoder.
+ */
+static size_t ZSTD_maxShortOffset(void)
+{
+    if (MEM_64bits()) {
+        /* We can decode any offset without reloading bits.
+         * This might change if the max window size grows.
+         */
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        return (size_t)-1;
+    } else {
+        /* The maximum offBase is (1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1.
+         * This offBase would require STREAM_ACCUMULATOR_MIN extra bits.
+         * Then we have to subtract ZSTD_REP_NUM to get the maximum possible offset.
+         */
+        size_t const maxOffbase = ((size_t)1 << (STREAM_ACCUMULATOR_MIN + 1)) - 1;
+        size_t const maxOffset = maxOffbase - ZSTD_REP_NUM;
+        assert(ZSTD_highbit32((U32)maxOffbase) == STREAM_ACCUMULATOR_MIN);
+        return maxOffset;
+    }
 }
-#endif
 
 size_t
 ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
                               void* dst, size_t dstCapacity,
-                        const void* src, size_t srcSize, const int frame)
+                        const void* src, size_t srcSize, const int frame, const streaming_operation streaming)
 {   /* blockType == blockCompressed */
     const BYTE* ip = (const BYTE*)src;
-    /* isLongOffset must be true if there are long offsets.
-     * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
-     * We don't expect that to be the case in 64-bit mode.
-     * In block mode, window size is not known, so we have to be conservative.
-     * (note: but it could be evaluated from current-lowLimit)
-     */
-    ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN))));
     DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
 
-    RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
+    /* Note : the wording of the specification
+     * allows compressed block to be sized exactly ZSTD_BLOCKSIZE_MAX.
+     * This generally does not happen, as it makes little sense,
+     * since an uncompressed block would feature same size and have no decompression cost.
+     * Also, note that decoder from reference libzstd before < v1.5.4
+     * would consider this edge case as an error.
+     * As a consequence, avoid generating compressed blocks of size ZSTD_BLOCKSIZE_MAX
+     * for broader compatibility with the deployed ecosystem of zstd decoders */
+    RETURN_ERROR_IF(srcSize > ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
 
     /* Decode literals section */
-    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
-        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
+    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize, dst, dstCapacity, streaming);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : cSize=%u, nbLiterals=%zu", (U32)litCSize, dctx->litSize);
         if (ZSTD_isError(litCSize)) return litCSize;
         ip += litCSize;
         srcSize -= litCSize;
@@ -1480,6 +2088,23 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
 
     /* Build Decoding Tables */
     {
+        /* Compute the maximum block size, which must also work when !frame and fParams are unset.
+         * Additionally, take the min with dstCapacity to ensure that the totalHistorySize fits in a size_t.
+         */
+        size_t const blockSizeMax = MIN(dstCapacity, (frame ? dctx->fParams.blockSizeMax : ZSTD_BLOCKSIZE_MAX));
+        size_t const totalHistorySize = ZSTD_totalHistorySize((BYTE*)dst + blockSizeMax, (BYTE const*)dctx->virtualStart);
+        /* isLongOffset must be true if there are long offsets.
+         * Offsets are long if they are larger than ZSTD_maxShortOffset().
+         * We don't expect that to be the case in 64-bit mode.
+         *
+         * We check here to see if our history is large enough to allow long offsets.
+         * If it isn't, then we can't possible have (valid) long offsets. If the offset
+         * is invalid, then it is okay to read it incorrectly.
+         *
+         * If isLongOffsets is true, then we will later check our decoding table to see
+         * if it is even possible to generate long offsets.
+         */
+        ZSTD_longOffset_e isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (totalHistorySize > ZSTD_maxShortOffset()));
         /* These macros control at build-time which decompressor implementation
          * we use. If neither is defined, we do some inspection and dispatch at
          * runtime.
@@ -1487,6 +2112,11 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
         int usePrefetchDecoder = dctx->ddictIsCold;
+#else
+        /* Set to 1 to avoid computing offset info if we don't need to.
+         * Otherwise this value is ignored.
+         */
+        int usePrefetchDecoder = 1;
 #endif
         int nbSeq;
         size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
@@ -1496,30 +2126,45 @@ ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
 
         RETURN_ERROR_IF(dst == NULL && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
 
-#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
-    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
-        if ( !usePrefetchDecoder
-          && (!frame || (dctx->fParams.windowSize > (1<<24)))
-          && (nbSeq>ADVANCED_SEQS) ) {  /* could probably use a larger nbSeq limit */
-            U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
-            U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
-            usePrefetchDecoder = (shareLongOffsets >= minShare);
+        /* If we could potentially have long offsets, or we might want to use the prefetch decoder,
+         * compute information about the share of long offsets, and the maximum nbAdditionalBits.
+         * NOTE: could probably use a larger nbSeq limit
+         */
+        if (isLongOffset || (!usePrefetchDecoder && (totalHistorySize > (1u << 24)) && (nbSeq > 8))) {
+            ZSTD_OffsetInfo const info = ZSTD_getOffsetInfo(dctx->OFTptr, nbSeq);
+            if (isLongOffset && info.maxNbAdditionalBits <= STREAM_ACCUMULATOR_MIN) {
+                /* If isLongOffset, but the maximum number of additional bits that we see in our table is small
+                 * enough, then we know it is impossible to have too long an offset in this block, so we can
+                 * use the regular offset decoder.
+                 */
+                isLongOffset = ZSTD_lo_isRegularOffset;
+            }
+            if (!usePrefetchDecoder) {
+                U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+                usePrefetchDecoder = (info.longOffsetShare >= minShare);
+            }
         }
-#endif
 
         dctx->ddictIsCold = 0;
 
 #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
     !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
-        if (usePrefetchDecoder)
+        if (usePrefetchDecoder) {
+#else
+        (void)usePrefetchDecoder;
+        {
 #endif
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
             return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
 #endif
+        }
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
         /* else */
-        return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+        if (dctx->litBufferLocation == ZSTD_split)
+            return ZSTD_decompressSequencesSplitLitBuffer(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+        else
+            return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
 #endif
     }
 }
@@ -1542,7 +2187,7 @@ size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
 {
     size_t dSize;
     ZSTD_checkContinuity(dctx, dst, dstCapacity);
-    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0, not_streaming);
     dctx->previousDstEnd = (char*)dst + dSize;
     return dSize;
 }
diff --git a/priv/zstd/lib/decompress/zstd_decompress_block.h b/priv/zstd/lib/decompress/zstd_decompress_block.h
index 049a0cd..67791db 100644
--- a/priv/zstd/lib/decompress/zstd_decompress_block.h
+++ b/priv/zstd/lib/decompress/zstd_decompress_block.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -33,6 +33,12 @@
  */
 
 
+ /* Streaming state is used to inform allocation of the literal buffer */
+typedef enum {
+    not_streaming = 0,
+    is_streaming = 1
+} streaming_operation;
+
 /* ZSTD_decompressBlock_internal() :
  * decompress block, starting at `src`,
  * into destination buffer `dst`.
@@ -41,7 +47,7 @@
  */
 size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
                                void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize, const int frame);
+                         const void* src, size_t srcSize, const int frame, const streaming_operation streaming);
 
 /* ZSTD_buildFSETable() :
  * generate FSE decoding table for one symbol (ll, ml or off)
@@ -54,7 +60,7 @@ size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
  */
 void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
              const short* normalizedCounter, unsigned maxSymbolValue,
-             const U32* baseValue, const U32* nbAdditionalBits,
+             const U32* baseValue, const U8* nbAdditionalBits,
                    unsigned tableLog, void* wksp, size_t wkspSize,
                    int bmi2);
 
diff --git a/priv/zstd/lib/decompress/zstd_decompress_internal.h b/priv/zstd/lib/decompress/zstd_decompress_internal.h
index ebda0c9..c2ec5d9 100644
--- a/priv/zstd/lib/decompress/zstd_decompress_internal.h
+++ b/priv/zstd/lib/decompress/zstd_decompress_internal.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -20,7 +20,7 @@
  *  Dependencies
  *********************************************************/
 #include "../common/mem.h"             /* BYTE, U16, U32 */
-#include "../common/zstd_internal.h"   /* ZSTD_seqSymbol */
+#include "../common/zstd_internal.h"   /* constants : MaxLL, MaxML, MaxOff, LLFSELog, etc. */
 
 
 
@@ -40,7 +40,7 @@ static UNUSED_ATTR const U32 OF_base[MaxOff+1] = {
                  0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
                  0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD, 0x1FFFFFFD, 0x3FFFFFFD, 0x7FFFFFFD };
 
-static UNUSED_ATTR const U32 OF_bits[MaxOff+1] = {
+static UNUSED_ATTR const U8 OF_bits[MaxOff+1] = {
                      0,  1,  2,  3,  4,  5,  6,  7,
                      8,  9, 10, 11, 12, 13, 14, 15,
                     16, 17, 18, 19, 20, 21, 22, 23,
@@ -75,12 +75,13 @@ static UNUSED_ATTR const U32 ML_base[MaxML+1] = {
 
 #define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE (sizeof(S16) * (MaxSeq + 1) + (1u << MaxFSELog) + sizeof(U64))
 #define ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32 ((ZSTD_BUILD_FSE_TABLE_WKSP_SIZE + sizeof(U32) - 1) / sizeof(U32))
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
 
 typedef struct {
     ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
     ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
     ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
-    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
+    HUF_DTable hufTable[HUF_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];  /* can accommodate HUF_decompress4X */
     U32 rep[ZSTD_REP_NUM];
     U32 workspace[ZSTD_BUILD_FSE_TABLE_WKSP_SIZE_U32];
 } ZSTD_entropyDTables_t;
@@ -106,6 +107,22 @@ typedef struct {
     size_t ddictPtrCount;
 } ZSTD_DDictHashSet;
 
+#ifndef ZSTD_DECODER_INTERNAL_BUFFER
+#  define ZSTD_DECODER_INTERNAL_BUFFER  (1 << 16)
+#endif
+
+#define ZSTD_LBMIN 64
+#define ZSTD_LBMAX (128 << 10)
+
+/* extra buffer, compensates when dst is not large enough to store litBuffer */
+#define ZSTD_LITBUFFEREXTRASIZE  BOUNDED(ZSTD_LBMIN, ZSTD_DECODER_INTERNAL_BUFFER, ZSTD_LBMAX)
+
+typedef enum {
+    ZSTD_not_in_dst = 0,  /* Stored entirely within litExtraBuffer */
+    ZSTD_in_dst = 1,           /* Stored entirely within dst (in memory after current output write) */
+    ZSTD_split = 2            /* Split between litExtraBuffer and dst */
+} ZSTD_litLocation_e;
+
 struct ZSTD_DCtx_s
 {
     const ZSTD_seqSymbol* LLTptr;
@@ -136,7 +153,9 @@ struct ZSTD_DCtx_s
     size_t litSize;
     size_t rleSize;
     size_t staticSize;
+#if DYNAMIC_BMI2 != 0
     int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+#endif
 
     /* dictionary */
     ZSTD_DDict* ddictLocal;
@@ -146,6 +165,7 @@ struct ZSTD_DCtx_s
     ZSTD_dictUses_e dictUses;
     ZSTD_DDictHashSet* ddictSet;                    /* Hash set for multiple ddicts */
     ZSTD_refMultipleDDicts_e refMultipleDDicts;     /* User specified: if == 1, will allow references to multiple DDicts. Default == 0 (disabled) */
+    int disableHufAsm;
 
     /* streaming */
     ZSTD_dStreamStage streamStage;
@@ -158,16 +178,21 @@ struct ZSTD_DCtx_s
     size_t outStart;
     size_t outEnd;
     size_t lhSize;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
     void* legacyContext;
     U32 previousLegacyVersion;
     U32 legacyVersion;
+#endif
     U32 hostageByte;
     int noForwardProgress;
     ZSTD_bufferMode_e outBufferMode;
     ZSTD_outBuffer expectedOutBuffer;
 
     /* workspace */
-    BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE* litBuffer;
+    const BYTE* litBufferEnd;
+    ZSTD_litLocation_e litBufferLocation;
+    BYTE litExtraBuffer[ZSTD_LITBUFFEREXTRASIZE + WILDCOPY_OVERLENGTH]; /* literal buffer can be split between storage within dst and within this scratch buffer */
     BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
 
     size_t oversizedDuration;
@@ -183,6 +208,14 @@ struct ZSTD_DCtx_s
 #endif
 };  /* typedef'd to ZSTD_DCtx within "zstd.h" */
 
+MEM_STATIC int ZSTD_DCtx_get_bmi2(const struct ZSTD_DCtx_s *dctx) {
+#if DYNAMIC_BMI2 != 0
+	return dctx->bmi2;
+#else
+    (void)dctx;
+	return 0;
+#endif
+}
 
 /*-*******************************************************
  *  Shared internal functions
diff --git a/priv/zstd/lib/deprecated/zbuff.h b/priv/zstd/lib/deprecated/zbuff.h
index b83ea0f..a968245 100644
--- a/priv/zstd/lib/deprecated/zbuff.h
+++ b/priv/zstd/lib/deprecated/zbuff.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/deprecated/zbuff_common.c b/priv/zstd/lib/deprecated/zbuff_common.c
index e7d01a0..5a2f2db 100644
--- a/priv/zstd/lib/deprecated/zbuff_common.c
+++ b/priv/zstd/lib/deprecated/zbuff_common.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/deprecated/zbuff_compress.c b/priv/zstd/lib/deprecated/zbuff_compress.c
index 2e72267..1d86821 100644
--- a/priv/zstd/lib/deprecated/zbuff_compress.c
+++ b/priv/zstd/lib/deprecated/zbuff_compress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,6 +15,7 @@
 ***************************************/
 #define ZBUFF_STATIC_LINKING_ONLY
 #include "zbuff.h"
+#include "../common/error_private.h"
 
 
 /*-***********************************************************
@@ -73,13 +74,32 @@ size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
                                    ZSTD_parameters params, unsigned long long pledgedSrcSize)
 {
     if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* preserve "0 == unknown" behavior */
-    return ZSTD_initCStream_advanced(zbc, dict, dictSize, params, pledgedSrcSize);
+    FORWARD_IF_ERROR(ZSTD_CCtx_reset(zbc, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setPledgedSrcSize(zbc, pledgedSrcSize), "");
+
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_windowLog, params.cParams.windowLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_hashLog, params.cParams.hashLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_chainLog, params.cParams.chainLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_searchLog, params.cParams.searchLog), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_minMatch, params.cParams.minMatch), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_targetLength, params.cParams.targetLength), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_strategy, params.cParams.strategy), "");
+
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_contentSizeFlag, params.fParams.contentSizeFlag), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_checksumFlag, params.fParams.checksumFlag), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_dictIDFlag, params.fParams.noDictIDFlag), "");
+
+    FORWARD_IF_ERROR(ZSTD_CCtx_loadDictionary(zbc, dict, dictSize), "");
+    return 0;
 }
 
-
 size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* zbc, const void* dict, size_t dictSize, int compressionLevel)
 {
-    return ZSTD_initCStream_usingDict(zbc, dict, dictSize, compressionLevel);
+    FORWARD_IF_ERROR(ZSTD_CCtx_reset(zbc, ZSTD_reset_session_only), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_setParameter(zbc, ZSTD_c_compressionLevel, compressionLevel), "");
+    FORWARD_IF_ERROR(ZSTD_CCtx_loadDictionary(zbc, dict, dictSize), "");
+    return 0;
 }
 
 size_t ZBUFF_compressInit(ZBUFF_CCtx* zbc, int compressionLevel)
diff --git a/priv/zstd/lib/deprecated/zbuff_decompress.c b/priv/zstd/lib/deprecated/zbuff_decompress.c
index d73c0f3..12a66af 100644
--- a/priv/zstd/lib/deprecated/zbuff_decompress.c
+++ b/priv/zstd/lib/deprecated/zbuff_decompress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -13,6 +13,8 @@
 /* *************************************
 *  Dependencies
 ***************************************/
+#define ZSTD_DISABLE_DEPRECATE_WARNINGS  /* suppress warning on ZSTD_initDStream_usingDict */
+#include "../zstd.h"        /* ZSTD_CStream, ZSTD_DStream, ZSTDLIB_API */
 #define ZBUFF_STATIC_LINKING_ONLY
 #include "zbuff.h"
 
diff --git a/priv/zstd/lib/dictBuilder/cover.c b/priv/zstd/lib/dictBuilder/cover.c
index 8364444..9e5e7d5 100644
--- a/priv/zstd/lib/dictBuilder/cover.c
+++ b/priv/zstd/lib/dictBuilder/cover.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -34,12 +34,20 @@
 #include "../common/pool.h"
 #include "../common/threading.h"
 #include "../common/zstd_internal.h" /* includes zstd.h */
+#include "../common/bits.h" /* ZSTD_highbit32 */
 #include "../zdict.h"
 #include "cover.h"
 
 /*-*************************************
 *  Constants
 ***************************************/
+/**
+* There are 32bit indexes used to ref samples, so limit samples size to 4GB
+* on 64bit builds.
+* For 32bit builds we choose 1 GB.
+* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
+* contiguous buffer, so 1GB is already a high limit.
+*/
 #define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
 #define COVER_DEFAULT_SPLITPOINT 1.0
 
@@ -47,7 +55,7 @@
 *  Console display
 ***************************************/
 #ifndef LOCALDISPLAYLEVEL
-static int g_displayLevel = 2;
+static int g_displayLevel = 0;
 #endif
 #undef  DISPLAY
 #define DISPLAY(...)                                                           \
@@ -534,7 +542,7 @@ static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
 
 /**
  * Prepare a context for dictionary building.
- * The context is only dependent on the parameter `d` and can used multiple
+ * The context is only dependent on the parameter `d` and can be used multiple
  * times.
  * Returns 0 on success or error code on error.
  * The context must be destroyed with `COVER_ctx_destroy()`.
@@ -639,7 +647,7 @@ static size_t COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
 
 void COVER_warnOnSmallCorpus(size_t maxDictSize, size_t nbDmers, int displayLevel)
 {
-  const double ratio = (double)nbDmers / maxDictSize;
+  const double ratio = (double)nbDmers / (double)maxDictSize;
   if (ratio >= 10) {
       return;
   }
@@ -735,7 +743,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
   COVER_map_t activeDmers;
   parameters.splitPoint = 1.0;
   /* Initialize global data */
-  g_displayLevel = parameters.zParams.notificationLevel;
+  g_displayLevel = (int)parameters.zParams.notificationLevel;
   /* Checks */
   if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
     DISPLAYLEVEL(1, "Cover parameters incorrect\n");
@@ -943,9 +951,17 @@ void COVER_best_finish(COVER_best_t *best, ZDICT_cover_params_t parameters,
   }
 }
 
+static COVER_dictSelection_t setDictSelection(BYTE* buf, size_t s, size_t csz)
+{
+    COVER_dictSelection_t ds;
+    ds.dictContent = buf;
+    ds.dictSize = s;
+    ds.totalCompressedSize = csz;
+    return ds;
+}
+
 COVER_dictSelection_t COVER_dictSelectionError(size_t error) {
-    COVER_dictSelection_t selection = { NULL, 0, error };
-    return selection;
+    return setDictSelection(NULL, 0, error);
 }
 
 unsigned COVER_dictSelectionIsError(COVER_dictSelection_t selection) {
@@ -998,9 +1014,8 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBuffe
   }
 
   if (params.shrinkDict == 0) {
-    COVER_dictSelection_t selection = { largestDictbuffer, dictContentSize, totalCompressedSize };
     free(candidateDictBuffer);
-    return selection;
+    return setDictSelection(largestDictbuffer, dictContentSize, totalCompressedSize);
   }
 
   largestDict = dictContentSize;
@@ -1032,20 +1047,16 @@ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent, size_t dictBuffe
       return COVER_dictSelectionError(totalCompressedSize);
     }
 
-    if (totalCompressedSize <= largestCompressed * regressionTolerance) {
-      COVER_dictSelection_t selection = { candidateDictBuffer, dictContentSize, totalCompressedSize };
+    if ((double)totalCompressedSize <= (double)largestCompressed * regressionTolerance) {
       free(largestDictbuffer);
-      return selection;
+      return setDictSelection( candidateDictBuffer, dictContentSize, totalCompressedSize );
     }
     dictContentSize *= 2;
   }
   dictContentSize = largestDict;
   totalCompressedSize = largestCompressed;
-  {
-    COVER_dictSelection_t selection = { largestDictbuffer, dictContentSize, totalCompressedSize };
-    free(candidateDictBuffer);
-    return selection;
-  }
+  free(candidateDictBuffer);
+  return setDictSelection( largestDictbuffer, dictContentSize, totalCompressedSize );
 }
 
 /**
diff --git a/priv/zstd/lib/dictBuilder/cover.h b/priv/zstd/lib/dictBuilder/cover.h
index 1aacddd..252624b 100644
--- a/priv/zstd/lib/dictBuilder/cover.h
+++ b/priv/zstd/lib/dictBuilder/cover.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/dictBuilder/fastcover.c b/priv/zstd/lib/dictBuilder/fastcover.c
index ed789f9..46bba01 100644
--- a/priv/zstd/lib/dictBuilder/fastcover.c
+++ b/priv/zstd/lib/dictBuilder/fastcover.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -32,6 +32,13 @@
 /*-*************************************
 *  Constants
 ***************************************/
+/**
+* There are 32bit indexes used to ref samples, so limit samples size to 4GB
+* on 64bit builds.
+* For 32bit builds we choose 1 GB.
+* Most 32bit platforms have 2GB user-mode addressable space and we allocate a large
+* contiguous buffer, so 1GB is already a high limit.
+*/
 #define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
 #define FASTCOVER_MAX_F 31
 #define FASTCOVER_MAX_ACCEL 10
@@ -44,7 +51,7 @@
 *  Console display
 ***************************************/
 #ifndef LOCALDISPLAYLEVEL
-static int g_displayLevel = 2;
+static int g_displayLevel = 0;
 #endif
 #undef  DISPLAY
 #define DISPLAY(...)                                                           \
@@ -297,7 +304,7 @@ FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
 
 /**
  * Prepare a context for dictionary building.
- * The context is only dependent on the parameter `d` and can used multiple
+ * The context is only dependent on the parameter `d` and can be used multiple
  * times.
  * Returns 0 on success or error code on error.
  * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
@@ -549,7 +556,7 @@ ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
     ZDICT_cover_params_t coverParams;
     FASTCOVER_accel_t accelParams;
     /* Initialize global data */
-    g_displayLevel = parameters.zParams.notificationLevel;
+    g_displayLevel = (int)parameters.zParams.notificationLevel;
     /* Assign splitPoint and f if not provided */
     parameters.splitPoint = 1.0;
     parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
@@ -632,7 +639,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
     const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
     const unsigned shrinkDict = 0;
     /* Local variables */
-    const int displayLevel = parameters->zParams.notificationLevel;
+    const int displayLevel = (int)parameters->zParams.notificationLevel;
     unsigned iteration = 1;
     unsigned d;
     unsigned k;
@@ -716,7 +723,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
         data->parameters.splitPoint = splitPoint;
         data->parameters.steps = kSteps;
         data->parameters.shrinkDict = shrinkDict;
-        data->parameters.zParams.notificationLevel = g_displayLevel;
+        data->parameters.zParams.notificationLevel = (unsigned)g_displayLevel;
         /* Check the parameters */
         if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
                                        data->ctx->f, accel)) {
diff --git a/priv/zstd/lib/dictBuilder/zdict.c b/priv/zstd/lib/dictBuilder/zdict.c
index 459cbe4..f22e04d 100644
--- a/priv/zstd/lib/dictBuilder/zdict.c
+++ b/priv/zstd/lib/dictBuilder/zdict.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -44,7 +44,6 @@
 #ifndef ZDICT_STATIC_LINKING_ONLY
 #  define ZDICT_STATIC_LINKING_ONLY
 #endif
-#define HUF_STATIC_LINKING_ONLY
 
 #include "../common/mem.h"           /* read */
 #include "../common/fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
@@ -54,6 +53,7 @@
 #include "../compress/zstd_compress_internal.h" /* ZSTD_loadCEntropy() */
 #include "../zdict.h"
 #include "divsufsort.h"
+#include "../common/bits.h"          /* ZSTD_NbCommonBytes */
 
 
 /*-*************************************
@@ -130,65 +130,6 @@ size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictSize)
 /*-********************************************************
 *  Dictionary training functions
 **********************************************************/
-static unsigned ZDICT_NbCommonBytes (size_t val)
-{
-    if (MEM_isLittleEndian()) {
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            _BitScanForward64( &r, (U64)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctzll((U64)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r=0;
-            _BitScanForward( &r, (U32)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctz((U32)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
-        }
-    } else {  /* Big Endian CPU */
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            _BitScanReverse64( &r, val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clzll(val) >> 3);
-#       else
-            unsigned r;
-            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
-            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
-            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r = 0;
-            _BitScanReverse( &r, (unsigned long)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clz((U32)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-    }   }
-}
-
-
 /*! ZDICT_count() :
     Count the nb of common bytes between 2 pointers.
     Note : this function presumes end of buffer followed by noisy guard band.
@@ -203,7 +144,7 @@ static size_t ZDICT_count(const void* pIn, const void* pMatch)
             pMatch = (const char*)pMatch+sizeof(size_t);
             continue;
         }
-        pIn = (const char*)pIn+ZDICT_NbCommonBytes(diff);
+        pIn = (const char*)pIn+ZSTD_NbCommonBytes(diff);
         return (size_t)((const char*)pIn - pStart);
     }
 }
@@ -235,7 +176,7 @@ static dictItem ZDICT_analyzePos(
     U32 savings[LLIMIT] = {0};
     const BYTE* b = (const BYTE*)buffer;
     size_t maxLength = LLIMIT;
-    size_t pos = suffix[start];
+    size_t pos = (size_t)suffix[start];
     U32 end = start;
     dictItem solution;
 
@@ -369,7 +310,7 @@ static dictItem ZDICT_analyzePos(
             savings[i] = savings[i-1] + (lengthList[i] * (i-3));
 
         DISPLAYLEVEL(4, "Selected dict at position %u, of length %u : saves %u (ratio: %.2f)  \n",
-                     (unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / maxLength);
+                     (unsigned)pos, (unsigned)maxLength, (unsigned)savings[maxLength], (double)savings[maxLength] / (double)maxLength);
 
         solution.pos = (U32)pos;
         solution.length = (U32)maxLength;
@@ -379,7 +320,7 @@ static dictItem ZDICT_analyzePos(
         {   U32 id;
             for (id=start; id1) && (table[u-1].savings < elt.savings))
-            table[u] = table[u-1], u--;
+                table[u] = table[u-1], u--;
             table[u] = elt;
             return u;
     }   }
@@ -442,7 +383,7 @@ static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const
 
         if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
             /* append */
-            int const addedLength = (int)eltEnd - (table[u].pos + table[u].length);
+            int const addedLength = (int)eltEnd - (int)(table[u].pos + table[u].length);
             table[u].savings += elt.length / 8;    /* rough approx bonus */
             if (addedLength > 0) {   /* otherwise, elt fully included into existing */
                 table[u].length += addedLength;
@@ -582,7 +523,7 @@ static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
             if (solution.length==0) { cursor++; continue; }
             ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
             cursor += solution.length;
-            DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / bufferSize * 100);
+            DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / (double)bufferSize * 100.0);
     }   }
 
 _cleanup:
@@ -662,8 +603,8 @@ static void ZDICT_countEStats(EStats_ress_t esr, const ZSTD_parameters* params,
 
             if (nbSeq >= 2) { /* rep offsets */
                 const seqDef* const seq = seqStorePtr->sequencesStart;
-                U32 offset1 = seq[0].offset - 3;
-                U32 offset2 = seq[1].offset - 3;
+                U32 offset1 = seq[0].offBase - ZSTD_REP_NUM;
+                U32 offset2 = seq[1].offBase - ZSTD_REP_NUM;
                 if (offset1 >= MAXREPOFFSET) offset1 = 0;
                 if (offset2 >= MAXREPOFFSET) offset2 = 0;
                 repOffsets[offset1] += 3;
@@ -734,6 +675,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles);
     size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles);
     BYTE* dstPtr = (BYTE*)dstBuffer;
+    U32 wksp[HUF_CTABLE_WORKSPACE_SIZE_U32];
 
     /* init */
     DEBUGLOG(4, "ZDICT_analyzeEntropy");
@@ -766,8 +708,15 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
         pos += fileSizes[u];
     }
 
+    if (notificationLevel >= 4) {
+        /* writeStats */
+        DISPLAYLEVEL(4, "Offset Code Frequencies : \n");
+        for (u=0; u<=offcodeMax; u++) {
+            DISPLAYLEVEL(4, "%2u :%7u \n", u, offcodeCount[u]);
+    }   }
+
     /* analyze, build stats, starting with literals */
-    {   size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
+    {   size_t maxNbBits = HUF_buildCTable_wksp(hufTable, countLit, 255, huffLog, wksp, sizeof(wksp));
         if (HUF_isError(maxNbBits)) {
             eSize = maxNbBits;
             DISPLAYLEVEL(1, " HUF_buildCTable error \n");
@@ -776,7 +725,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
         if (maxNbBits==8) {  /* not compressible : will fail on HUF_writeCTable() */
             DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
             ZDICT_flatLit(countLit);  /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
-            maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
+            maxNbBits = HUF_buildCTable_wksp(hufTable, countLit, 255, huffLog, wksp, sizeof(wksp));
             assert(maxNbBits==9);
         }
         huffLog = (U32)maxNbBits;
@@ -817,7 +766,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     llLog = (U32)errorCode;
 
     /* write result to buffer */
-    {   size_t const hhSize = HUF_writeCTable(dstPtr, maxDstSize, hufTable, 255, huffLog);
+    {   size_t const hhSize = HUF_writeCTable_wksp(dstPtr, maxDstSize, hufTable, 255, huffLog, wksp, sizeof(wksp));
         if (HUF_isError(hhSize)) {
             eSize = hhSize;
             DISPLAYLEVEL(1, "HUF_writeCTable error \n");
@@ -872,7 +821,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
 #else
     /* at this stage, we don't use the result of "most common first offset",
-       as the impact of statistics is not properly evaluated */
+     * as the impact of statistics is not properly evaluated */
     MEM_writeLE32(dstPtr+0, repStartValue[0]);
     MEM_writeLE32(dstPtr+4, repStartValue[1]);
     MEM_writeLE32(dstPtr+8, repStartValue[2]);
@@ -888,6 +837,17 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
 }
 
 
+/**
+ * @returns the maximum repcode value
+ */
+static U32 ZDICT_maxRep(U32 const reps[ZSTD_REP_NUM])
+{
+    U32 maxRep = reps[0];
+    int r;
+    for (r = 1; r < ZSTD_REP_NUM; ++r)
+        maxRep = MAX(maxRep, reps[r]);
+    return maxRep;
+}
 
 size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
                           const void* customDictContent, size_t dictContentSize,
@@ -899,11 +859,13 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
     BYTE header[HBUFFSIZE];
     int const compressionLevel = (params.compressionLevel == 0) ? ZSTD_CLEVEL_DEFAULT : params.compressionLevel;
     U32 const notificationLevel = params.notificationLevel;
+    /* The final dictionary content must be at least as large as the largest repcode */
+    size_t const minContentSize = (size_t)ZDICT_maxRep(repStartValue);
+    size_t paddingSize;
 
     /* check conditions */
     DEBUGLOG(4, "ZDICT_finalizeDictionary");
     if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
-    if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
 
     /* dictionary header */
@@ -927,12 +889,43 @@ size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
         hSize += eSize;
     }
 
-    /* copy elements in final buffer ; note : src and dst buffer can overlap */
-    if (hSize + dictContentSize > dictBufferCapacity) dictContentSize = dictBufferCapacity - hSize;
-    {   size_t const dictSize = hSize + dictContentSize;
-        char* dictEnd = (char*)dictBuffer + dictSize;
-        memmove(dictEnd - dictContentSize, customDictContent, dictContentSize);
-        memcpy(dictBuffer, header, hSize);
+    /* Shrink the content size if it doesn't fit in the buffer */
+    if (hSize + dictContentSize > dictBufferCapacity) {
+        dictContentSize = dictBufferCapacity - hSize;
+    }
+
+    /* Pad the dictionary content with zeros if it is too small */
+    if (dictContentSize < minContentSize) {
+        RETURN_ERROR_IF(hSize + minContentSize > dictBufferCapacity, dstSize_tooSmall,
+                        "dictBufferCapacity too small to fit max repcode");
+        paddingSize = minContentSize - dictContentSize;
+    } else {
+        paddingSize = 0;
+    }
+
+    {
+        size_t const dictSize = hSize + paddingSize + dictContentSize;
+
+        /* The dictionary consists of the header, optional padding, and the content.
+         * The padding comes before the content because the "best" position in the
+         * dictionary is the last byte.
+         */
+        BYTE* const outDictHeader = (BYTE*)dictBuffer;
+        BYTE* const outDictPadding = outDictHeader + hSize;
+        BYTE* const outDictContent = outDictPadding + paddingSize;
+
+        assert(dictSize <= dictBufferCapacity);
+        assert(outDictContent + dictContentSize == (BYTE*)dictBuffer + dictSize);
+
+        /* First copy the customDictContent into its final location.
+         * `customDictContent` and `dictBuffer` may overlap, so we must
+         * do this before any other writes into the output buffer.
+         * Then copy the header & padding into the output buffer.
+         */
+        memmove(outDictContent, customDictContent, dictContentSize);
+        memcpy(outDictHeader, header, hSize);
+        memset(outDictPadding, 0, paddingSize);
+
         return dictSize;
     }
 }
diff --git a/priv/zstd/lib/dll/example/Makefile b/priv/zstd/lib/dll/example/Makefile
index 03b034d..86cf690 100644
--- a/priv/zstd/lib/dll/example/Makefile
+++ b/priv/zstd/lib/dll/example/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/dll/example/README.md b/priv/zstd/lib/dll/example/README.md
index 9e30fd5..46aec79 100644
--- a/priv/zstd/lib/dll/example/README.md
+++ b/priv/zstd/lib/dll/example/README.md
@@ -46,7 +46,7 @@ The compiled executable will require ZSTD DLL which is available at `dll\libzstd
 Open `example\fullbench-dll.sln` to compile `fullbench-dll` that uses a
 dynamic ZSTD library from the `dll` directory. The solution works with Visual C++
 2010 or newer. When one will open the solution with Visual C++ newer than 2010
-then the solution will upgraded to the current version.
+then the solution will be upgraded to the current version.
 
 ## Using ZSTD DLL with Visual C++
 
diff --git a/priv/zstd/lib/legacy/zstd_legacy.h b/priv/zstd/lib/legacy/zstd_legacy.h
index a6f1174..dd17325 100644
--- a/priv/zstd/lib/legacy/zstd_legacy.h
+++ b/priv/zstd/lib/legacy/zstd_legacy.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -242,6 +242,13 @@ MEM_STATIC ZSTD_frameSizeInfo ZSTD_findFrameSizeInfoLegacy(const void *src, size
         frameSizeInfo.compressedSize = ERROR(srcSize_wrong);
         frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
     }
+    /* In all cases, decompressedBound == nbBlocks * ZSTD_BLOCKSIZE_MAX.
+     * So we can compute nbBlocks without having to change every function.
+     */
+    if (frameSizeInfo.decompressedBound != ZSTD_CONTENTSIZE_ERROR) {
+        assert((frameSizeInfo.decompressedBound & (ZSTD_BLOCKSIZE_MAX - 1)) == 0);
+        frameSizeInfo.nbBlocks = (size_t)(frameSizeInfo.decompressedBound / ZSTD_BLOCKSIZE_MAX);
+    }
     return frameSizeInfo;
 }
 
diff --git a/priv/zstd/lib/legacy/zstd_v01.c b/priv/zstd/lib/legacy/zstd_v01.c
index 7ab5547..1a3aad0 100644
--- a/priv/zstd/lib/legacy/zstd_v01.c
+++ b/priv/zstd/lib/legacy/zstd_v01.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -190,28 +190,6 @@ typedef   signed long long  S64;
 /****************************************************************
 *  Memory I/O
 *****************************************************************/
-/* FSE_FORCE_MEMORY_ACCESS
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets generating assembly depending on alignment.
- *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef FSE_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define FSE_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define FSE_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
-
 
 static unsigned FSE_32bits(void)
 {
@@ -224,24 +202,6 @@ static unsigned FSE_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(FSE_FORCE_MEMORY_ACCESS) && (FSE_FORCE_MEMORY_ACCESS==2)
-
-static U16 FSE_read16(const void* memPtr) { return *(const U16*) memPtr; }
-static U32 FSE_read32(const void* memPtr) { return *(const U32*) memPtr; }
-static U64 FSE_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-#elif defined(FSE_FORCE_MEMORY_ACCESS) && (FSE_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
-
-static U16 FSE_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-static U32 FSE_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-static U64 FSE_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-#else
-
 static U16 FSE_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -257,8 +217,6 @@ static U64 FSE_read64(const void* memPtr)
     U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
 }
 
-#endif /* FSE_FORCE_MEMORY_ACCESS */
-
 static U16 FSE_readLE16(const void* memPtr)
 {
     if (FSE_isLittleEndian())
@@ -343,8 +301,7 @@ FORCE_INLINE unsigned FSE_highbit32 (U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
     unsigned long r;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (GCC_VERSION >= 304)   /* GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -1194,7 +1151,7 @@ static size_t HUF_decompress (void* dst, size_t maxDstSize, const void* cSrc, si
     zstd - standard compression library
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -1763,20 +1720,26 @@ static size_t ZSTD_execSequence(BYTE* op,
     static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
     static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* subtracted */
     const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
     const size_t litLength = sequence.litLength;
     BYTE* const endMatch = op + litLength + sequence.matchLength;    /* risk : address space overflow (32-bits) */
     const BYTE* const litEnd = *litPtr + litLength;
 
-    /* check */
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
     if (endMatch > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
-    if (litEnd > litLimit) return ERROR(corruption_detected);
-    if (sequence.matchLength > (size_t)(*litPtr-op))  return ERROR(dstSize_tooSmall);    /* overwrite literal segment */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+    if (sequence.matchLength > (size_t)(*litPtr-op)) return ERROR(dstSize_tooSmall);  /* overwrite literal segment */
 
     /* copy Literals */
-    if (((size_t)(*litPtr - op) < 8) || ((size_t)(oend-litEnd) < 8) || (op+litLength > oend-8))
-        memmove(op, *litPtr, litLength);   /* overwrite risk */
-    else
-        ZSTD_wildcopy(op, *litPtr, litLength);
+    ZSTD_memmove(op, *litPtr, sequence.litLength);   /* note : v0.1 seems to allow scenarios where output or input are close to end of buffer */
+
     op += litLength;
     *litPtr = litEnd;   /* update for next sequence */
 
diff --git a/priv/zstd/lib/legacy/zstd_v01.h b/priv/zstd/lib/legacy/zstd_v01.h
index f777eb6..6ac8769 100644
--- a/priv/zstd/lib/legacy/zstd_v01.h
+++ b/priv/zstd/lib/legacy/zstd_v01.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/legacy/zstd_v02.c b/priv/zstd/lib/legacy/zstd_v02.c
index 89fdc71..e09bb4a 100644
--- a/priv/zstd/lib/legacy/zstd_v02.c
+++ b/priv/zstd/lib/legacy/zstd_v02.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -28,7 +28,7 @@
    low-level memory access routines
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -115,27 +115,6 @@ extern "C" {
 /****************************************************************
 *  Memory I/O
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets generating assembly depending on alignment.
- *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define MEM_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
 
 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
@@ -146,33 +125,6 @@ MEM_STATIC unsigned MEM_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
-
-/* violates C standard on structure alignment.
-Only use if no other choice to achieve best performance on target platform */
-MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
-MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
-MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
-
-#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
-
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-
-#else
-
-/* default method, safe and standard.
-   can sometimes prove slower */
-
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -193,9 +145,6 @@ MEM_STATIC void MEM_write16(void* memPtr, U16 value)
     memcpy(memPtr, &value, sizeof(value));
 }
 
-#endif /* MEM_FORCE_MEMORY_ACCESS */
-
-
 MEM_STATIC U16 MEM_readLE16(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -272,7 +221,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
    header file (to include)
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -353,9 +302,8 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
 MEM_STATIC unsigned BIT_highbit32 (U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -437,7 +385,7 @@ MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
 }
 
 /*! BIT_lookBitsFast :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
@@ -457,7 +405,7 @@ MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 }
 
 /*!BIT_readBitsFast :
-*  unsafe version; only works only if nbBits >= 1 */
+*  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t value = BIT_lookBitsFast(bitD, nbBits);
@@ -514,7 +462,7 @@ MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
    Error codes and messages
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -613,7 +561,7 @@ typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be mor
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -757,7 +705,7 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -826,7 +774,7 @@ static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, si
     Header File
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -886,7 +834,7 @@ typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
     Header File for static linking only
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -950,7 +898,7 @@ typedef struct ZSTD_DCtx_s ZSTD_DCtx;
    FSE : Finite State Entropy coder
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1454,7 +1402,7 @@ static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, siz
    Huff0 : Huffman coder, part of New Generation Entropy library
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -2613,7 +2561,7 @@ static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_
     zstd - standard compression library
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -3118,12 +3066,19 @@ static size_t ZSTD_execSequence(BYTE* op,
     const BYTE* const litEnd = *litPtr + sequence.litLength;
 
     /* checks */
-    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use the pointer check */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
     if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
     if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
 
     /* copy Literals */
-    ZSTD_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
     op = oLitEnd;
     *litPtr = litEnd;   /* update for next sequence */
 
diff --git a/priv/zstd/lib/legacy/zstd_v02.h b/priv/zstd/lib/legacy/zstd_v02.h
index 1b37195..dab0260 100644
--- a/priv/zstd/lib/legacy/zstd_v02.h
+++ b/priv/zstd/lib/legacy/zstd_v02.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/legacy/zstd_v03.c b/priv/zstd/lib/legacy/zstd_v03.c
index 5262d51..b0d7f52 100644
--- a/priv/zstd/lib/legacy/zstd_v03.c
+++ b/priv/zstd/lib/legacy/zstd_v03.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -29,7 +29,7 @@
    low-level memory access routines
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -116,27 +116,6 @@ extern "C" {
 /****************************************************************
 *  Memory I/O
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets generating assembly depending on alignment.
- *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define MEM_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
 
 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
@@ -147,33 +126,6 @@ MEM_STATIC unsigned MEM_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
-
-/* violates C standard on structure alignment.
-Only use if no other choice to achieve best performance on target platform */
-MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
-MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
-MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
-
-#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
-
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-
-#else
-
-/* default method, safe and standard.
-   can sometimes prove slower */
-
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -194,10 +146,6 @@ MEM_STATIC void MEM_write16(void* memPtr, U16 value)
     memcpy(memPtr, &value, sizeof(value));
 }
 
-
-#endif /* MEM_FORCE_MEMORY_ACCESS */
-
-
 MEM_STATIC U16 MEM_readLE16(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -274,7 +222,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
    header file (to include)
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -356,9 +304,8 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
 MEM_STATIC unsigned BIT_highbit32 (U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -439,7 +386,7 @@ MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
 }
 
 /*! BIT_lookBitsFast :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
@@ -459,7 +406,7 @@ MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 }
 
 /*!BIT_readBitsFast :
-*  unsafe version; only works only if nbBits >= 1 */
+*  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t value = BIT_lookBitsFast(bitD, nbBits);
@@ -516,7 +463,7 @@ MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
    Error codes and messages
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -615,7 +562,7 @@ typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be mor
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -759,7 +706,7 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -827,7 +774,7 @@ static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, si
     Header File
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -887,7 +834,7 @@ typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
     Header File for static linking only
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -951,7 +898,7 @@ typedef struct ZSTD_DCtx_s ZSTD_DCtx;
    FSE : Finite State Entropy coder
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1455,7 +1402,7 @@ static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, siz
    Huff0 : Huffman coder, part of New Generation Entropy library
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -2252,7 +2199,7 @@ static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_
     zstd - standard compression library
     Copyright (C) 2014-2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2759,18 +2706,24 @@ static size_t ZSTD_execSequence(BYTE* op,
     const BYTE* const litEnd = *litPtr + sequence.litLength;
 
     /* checks */
-    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+    if (sequence.offset > (U32)(oLitEnd - base)) return ERROR(corruption_detected);
+
     if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
     if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
 
     /* copy Literals */
-    ZSTD_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
     op = oLitEnd;
     *litPtr = litEnd;   /* update for next sequence */
 
     /* copy Match */
-    {
-        const BYTE* match = op - sequence.offset;
+    {   const BYTE* match = op - sequence.offset;
 
         /* check */
         if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
diff --git a/priv/zstd/lib/legacy/zstd_v03.h b/priv/zstd/lib/legacy/zstd_v03.h
index 7a00d43..9bf3cce 100644
--- a/priv/zstd/lib/legacy/zstd_v03.h
+++ b/priv/zstd/lib/legacy/zstd_v03.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/legacy/zstd_v04.c b/priv/zstd/lib/legacy/zstd_v04.c
index bee1b99..57be832 100644
--- a/priv/zstd/lib/legacy/zstd_v04.c
+++ b/priv/zstd/lib/legacy/zstd_v04.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -87,27 +87,6 @@ extern "C" {
 /****************************************************************
 *  Memory I/O
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets generating assembly depending on alignment.
- *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define MEM_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
 
 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
@@ -118,33 +97,6 @@ MEM_STATIC unsigned MEM_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
-
-/* violates C standard on structure alignment.
-Only use if no other choice to achieve best performance on target platform */
-MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
-MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
-MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
-
-#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
-
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-
-#else
-
-/* default method, safe and standard.
-   can sometimes prove slower */
-
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -165,9 +117,6 @@ MEM_STATIC void MEM_write16(void* memPtr, U16 value)
     memcpy(memPtr, &value, sizeof(value));
 }
 
-#endif /* MEM_FORCE_MEMORY_ACCESS */
-
-
 MEM_STATIC U16 MEM_readLE16(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -545,7 +494,7 @@ If there is an error, the function will return an error code, which can be teste
    header file (to include)
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -627,9 +576,8 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
 MEM_STATIC unsigned BIT_highbit32 (U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -704,7 +652,7 @@ MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
 }
 
 /*! BIT_lookBitsFast :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
@@ -724,7 +672,7 @@ MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 }
 
 /*!BIT_readBitsFast :
-*  unsafe version; only works only if nbBits >= 1 */
+*  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t value = BIT_lookBitsFast(bitD, nbBits);
@@ -785,7 +733,7 @@ MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -934,7 +882,7 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
    FSE : Finite State Entropy coder
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1440,7 +1388,7 @@ static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, siz
    header file
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1518,7 +1466,7 @@ static unsigned    HUF_isError(size_t code);        /* tells if a return value i
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1605,7 +1553,7 @@ static size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const
    Huff0 : Huffman coder, part of New Generation Entropy library
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -2405,7 +2353,7 @@ static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_
     zstd - decompression module fo v0.4 legacy format
     Copyright (C) 2015-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2880,13 +2828,19 @@ static size_t ZSTD_execSequence(BYTE* op,
     const BYTE* const litEnd = *litPtr + sequence.litLength;
     const BYTE* match = oLitEnd - sequence.offset;
 
-    /* check */
-    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
     if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
-    if (litEnd > litLimit) return ERROR(corruption_detected);   /* risk read beyond lit buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
 
     /* copy Literals */
-    ZSTD_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    ZSTD_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
     op = oLitEnd;
     *litPtr = litEnd;   /* update for next sequence */
 
@@ -3287,7 +3241,7 @@ static void ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* dict, s
     Buffered version of Zstd compression library
     Copyright (C) 2015, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
diff --git a/priv/zstd/lib/legacy/zstd_v04.h b/priv/zstd/lib/legacy/zstd_v04.h
index 66b97ab..640240d 100644
--- a/priv/zstd/lib/legacy/zstd_v04.h
+++ b/priv/zstd/lib/legacy/zstd_v04.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/legacy/zstd_v05.c b/priv/zstd/lib/legacy/zstd_v05.c
index eb8966b..93a1169 100644
--- a/priv/zstd/lib/legacy/zstd_v05.c
+++ b/priv/zstd/lib/legacy/zstd_v05.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -19,7 +19,7 @@
    low-level memory access routines
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -106,27 +106,6 @@ extern "C" {
 /*-**************************************************************
 *  Memory I/O
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS :
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets depending on alignment.
- *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define MEM_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
 
 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
@@ -137,37 +116,6 @@ MEM_STATIC unsigned MEM_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
-
-/* violates C standard, by lying on structure alignment.
-Only use if no other choice to achieve best performance on target platform */
-MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
-MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
-MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
-MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
-MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
-
-#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
-
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign*)memPtr)->u32 = value; }
-MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign*)memPtr)->u64 = value; }
-
-#else
-
-/* default method, safe and standard.
-   can sometimes prove slower */
-
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -198,9 +146,6 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value)
     memcpy(memPtr, &value, sizeof(value));
 }
 
-#endif /* MEM_FORCE_MEMORY_ACCESS */
-
-
 MEM_STATIC U16 MEM_readLE16(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -265,7 +210,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
     Header File for static linking only
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -289,7 +234,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net
+    - zstd homepage : https://facebook.github.io/zstd
 */
 #ifndef ZSTD_STATIC_H
 #define ZSTD_STATIC_H
@@ -401,7 +346,7 @@ size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx, void* dst, size_t dstCapacity
     Header File for include
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -488,7 +433,7 @@ static const size_t ZSTDv05_frameHeaderSize_min = 5;
 #define FSEv05_ENCODING_DYNAMIC 3
 
 
-#define HufLog 12
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
 
 #define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
 #define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
@@ -556,7 +501,7 @@ typedef struct {
    header file
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -675,7 +620,7 @@ size_t FSEv05_decompress_usingDTable(void* dst, size_t dstCapacity, const void*
    header file (to include)
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -756,9 +701,8 @@ MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits);
 MEM_STATIC unsigned BITv05_highbit32 (U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -830,7 +774,7 @@ MEM_STATIC size_t BITv05_lookBits(BITv05_DStream_t* bitD, U32 nbBits)
 }
 
 /*! BITv05_lookBitsFast :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BITv05_lookBitsFast(BITv05_DStream_t* bitD, U32 nbBits)
 {
     const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
@@ -850,7 +794,7 @@ MEM_STATIC size_t BITv05_readBits(BITv05_DStream_t* bitD, unsigned nbBits)
 }
 
 /*!BITv05_readBitsFast :
-*  unsafe version; only works only if nbBits >= 1 */
+*  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits)
 {
     size_t value = BITv05_lookBitsFast(bitD, nbBits);
@@ -905,7 +849,7 @@ MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* DStream)
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1055,7 +999,7 @@ MEM_STATIC unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr)
    FSEv05 : Finite State Entropy coder
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1541,7 +1485,7 @@ size_t FSEv05_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t
    header file
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1614,7 +1558,7 @@ const char* HUFv05_getErrorName(size_t code);   /* provides error code string (u
    header file, for static linking only
    Copyright (C) 2013-2016, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1706,7 +1650,7 @@ size_t HUFv05_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void
    Huff0 : Huffman coder, part of New Generation Entropy library
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -2551,7 +2495,7 @@ size_t HUFv05_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cS
     zstd - standard compression library
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2649,7 +2593,7 @@ struct ZSTDv05_DCtx_s
     FSEv05_DTable LLTable[FSEv05_DTABLE_SIZE_U32(LLFSEv05Log)];
     FSEv05_DTable OffTable[FSEv05_DTABLE_SIZE_U32(OffFSEv05Log)];
     FSEv05_DTable MLTable[FSEv05_DTABLE_SIZE_U32(MLFSEv05Log)];
-    unsigned   hufTableX4[HUFv05_DTABLE_SIZE(HufLog)];
+    unsigned   hufTableX4[HUFv05_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];
     const void* previousDstEnd;
     const void* base;
     const void* vBase;
@@ -2677,7 +2621,7 @@ size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx)
     dctx->base = NULL;
     dctx->vBase = NULL;
     dctx->dictEnd = NULL;
-    dctx->hufTableX4[0] = HufLog;
+    dctx->hufTableX4[0] = ZSTD_HUFFDTABLE_CAPACITY_LOG;
     dctx->flagStaticTables = 0;
     return 0;
 }
@@ -3238,13 +3182,19 @@ static size_t ZSTDv05_execSequence(BYTE* op,
     const BYTE* const litEnd = *litPtr + sequence.litLength;
     const BYTE* match = oLitEnd - sequence.offset;
 
-    /* check */
-    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
     if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
-    if (litEnd > litLimit) return ERROR(corruption_detected);   /* risk read beyond lit buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
 
     /* copy Literals */
-    ZSTDv05_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    ZSTDv05_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
     op = oLitEnd;
     *litPtr = litEnd;   /* update for next sequence */
 
@@ -3750,7 +3700,7 @@ size_t ZSTDv05_decompressBegin_usingDict(ZSTDv05_DCtx* dctx, const void* dict, s
     Buffered version of Zstd compression library
     Copyright (C) 2015-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
diff --git a/priv/zstd/lib/legacy/zstd_v05.h b/priv/zstd/lib/legacy/zstd_v05.h
index bd423bf..2dcffc9 100644
--- a/priv/zstd/lib/legacy/zstd_v05.h
+++ b/priv/zstd/lib/legacy/zstd_v05.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/legacy/zstd_v06.c b/priv/zstd/lib/legacy/zstd_v06.c
index fcb16d4..175f7cc 100644
--- a/priv/zstd/lib/legacy/zstd_v06.c
+++ b/priv/zstd/lib/legacy/zstd_v06.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -23,7 +23,7 @@
    low-level memory access routines
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -108,27 +108,6 @@ extern "C" {
 /*-**************************************************************
 *  Memory I/O
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS :
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets depending on alignment.
- *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define MEM_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
 
 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
@@ -139,33 +118,6 @@ MEM_STATIC unsigned MEM_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
-
-/* violates C standard, by lying on structure alignment.
-Only use if no other choice to achieve best performance on target platform */
-MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
-MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
-MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
-
-#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
-
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-
-#else
-
-/* default method, safe and standard.
-   can sometimes prove slower */
-
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -186,9 +138,6 @@ MEM_STATIC void MEM_write16(void* memPtr, U16 value)
     memcpy(memPtr, &value, sizeof(value));
 }
 
-
-#endif /* MEM_FORCE_MEMORY_ACCESS */
-
 MEM_STATIC U32 MEM_swap32(U32 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
@@ -284,7 +233,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
     Header File for static linking only
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -308,7 +257,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net
+    - zstd homepage : https://facebook.github.io/zstd
 */
 #ifndef ZSTDv06_STATIC_H
 #define ZSTDv06_STATIC_H
@@ -415,7 +364,7 @@ ZSTDLIBv06_API size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx, void* dst, siz
     Header File for include
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -482,7 +431,7 @@ typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
 #define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
 #define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
 
-#define HufLog 12
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
 
 #define IS_HUF 0
 #define IS_PCH 1
@@ -622,7 +571,7 @@ void ZSTDv06_seqToCodes(const seqStore_t* seqStorePtr, size_t const nbSeq);
    Public Prototypes declaration
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -770,7 +719,7 @@ If there is an error, the function will return an error code, which can be teste
    header file (to include)
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -860,9 +809,8 @@ MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, unsigned nbBits);
 MEM_STATIC unsigned BITv06_highbit32 ( U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -932,7 +880,7 @@ MEM_STATIC size_t BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuff
 }
 
 /*! BITv06_lookBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BITv06_lookBitsFast(const BITv06_DStream_t* bitD, U32 nbBits)
 {
     U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
@@ -952,7 +900,7 @@ MEM_STATIC size_t BITv06_readBits(BITv06_DStream_t* bitD, U32 nbBits)
 }
 
 /*! BITv06_readBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, U32 nbBits)
 {
     size_t const value = BITv06_lookBitsFast(bitD, nbBits);
@@ -1006,7 +954,7 @@ MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* DStream)
    header file for static linking (only)
    Copyright (C) 2013-2015, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1214,7 +1162,7 @@ MEM_STATIC BYTE FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStre
    Common functions of New Generation Entropy library
    Copyright (C) 2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1359,7 +1307,7 @@ size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned
    FSE : Finite State Entropy decoder
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1683,7 +1631,7 @@ size_t FSEv06_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t
    header file
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1753,7 +1701,7 @@ size_t HUFv06_compressBound(size_t size);       /**< maximum compressed size */
    header file, for static linking only
    Copyright (C) 2013-2016, Yann Collet
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1935,7 +1883,7 @@ MEM_STATIC size_t HUFv06_readStats(BYTE* huffWeight, size_t hwSize, U32* rankSta
    Huffman decoder, part of New Generation Entropy library
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -2680,7 +2628,7 @@ size_t HUFv06_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cS
     Common functions of Zstd compression library
     Copyright (C) 2015-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2704,7 +2652,7 @@ size_t HUFv06_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cS
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net/
+    - zstd homepage : https://facebook.github.io/zstd/
 */
 
 
@@ -2734,7 +2682,7 @@ const char* ZBUFFv06_getErrorName(size_t errorCode) { return ERR_getErrorName(er
     zstd - standard compression library
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2758,7 +2706,7 @@ const char* ZBUFFv06_getErrorName(size_t errorCode) { return ERR_getErrorName(er
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net
+    - zstd homepage : https://facebook.github.io/zstd
 */
 
 /* ***************************************************************
@@ -2810,7 +2758,7 @@ struct ZSTDv06_DCtx_s
     FSEv06_DTable LLTable[FSEv06_DTABLE_SIZE_U32(LLFSELog)];
     FSEv06_DTable OffTable[FSEv06_DTABLE_SIZE_U32(OffFSELog)];
     FSEv06_DTable MLTable[FSEv06_DTABLE_SIZE_U32(MLFSELog)];
-    unsigned   hufTableX4[HUFv06_DTABLE_SIZE(HufLog)];
+    unsigned   hufTableX4[HUFv06_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];
     const void* previousDstEnd;
     const void* base;
     const void* vBase;
@@ -2838,7 +2786,7 @@ size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx)
     dctx->base = NULL;
     dctx->vBase = NULL;
     dctx->dictEnd = NULL;
-    dctx->hufTableX4[0] = HufLog;
+    dctx->hufTableX4[0] = ZSTD_HUFFDTABLE_CAPACITY_LOG;
     dctx->flagRepeatTable = 0;
     return 0;
 }
@@ -3374,13 +3322,19 @@ static size_t ZSTDv06_execSequence(BYTE* op,
     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
     const BYTE* match = oLitEnd - sequence.offset;
 
-    /* check */
-    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    /* checks */
+    size_t const seqLength = sequence.litLength + sequence.matchLength;
+
+    if (seqLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);
+    /* Now we know there are no overflow in literal nor match lengths, can use pointer checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);
+
     if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
-    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
 
     /* copy Literals */
-    ZSTDv06_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    ZSTDv06_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
     op = oLitEnd;
     *litPtr = iLitEnd;   /* update for next sequence */
 
@@ -3893,7 +3847,7 @@ size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, s
     Buffered version of Zstd compression library
     Copyright (C) 2015-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -3917,7 +3871,7 @@ size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, s
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net/
+    - zstd homepage : https://facebook.github.io/zstd/
 */
 
 
@@ -4039,7 +3993,8 @@ size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* zbd,
                     size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
                     if (ZSTDv06_isError(hSize)) return hSize;
                     if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
-                        memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        if (ip != NULL)
+                            memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
                         zbd->lhSize += iend-ip;
                         *dstCapacityPtr = 0;
                         return (hSize - zbd->lhSize) + ZSTDv06_blockHeaderSize;   /* remaining header bytes + next block header */
diff --git a/priv/zstd/lib/legacy/zstd_v06.h b/priv/zstd/lib/legacy/zstd_v06.h
index 9e32b76..6338910 100644
--- a/priv/zstd/lib/legacy/zstd_v06.h
+++ b/priv/zstd/lib/legacy/zstd_v06.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/legacy/zstd_v07.c b/priv/zstd/lib/legacy/zstd_v07.c
index 0d0e466..15dc3ef 100644
--- a/priv/zstd/lib/legacy/zstd_v07.c
+++ b/priv/zstd/lib/legacy/zstd_v07.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -184,7 +184,7 @@ ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockS
    low-level memory access routines
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -268,27 +268,6 @@ extern "C" {
 /*-**************************************************************
 *  Memory I/O
 *****************************************************************/
-/* MEM_FORCE_MEMORY_ACCESS :
- * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
- * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
- * The below switch allow to select different access method for improved performance.
- * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
- *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
- * Method 2 : direct access. This method is portable but violate C standard.
- *            It can generate buggy code on targets depending on alignment.
- *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
- * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
- * Prefer these methods in priority order (0 > 1 > 2)
- */
-#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
-#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
-#    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
-#    define MEM_FORCE_MEMORY_ACCESS 1
-#  endif
-#endif
 
 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
@@ -299,33 +278,6 @@ MEM_STATIC unsigned MEM_isLittleEndian(void)
     return one.c[0];
 }
 
-#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
-
-/* violates C standard, by lying on structure alignment.
-Only use if no other choice to achieve best performance on target platform */
-MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
-MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
-MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
-
-#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
-
-/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
-/* currently only defined for gcc and icc */
-typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
-
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-
-#else
-
-/* default method, safe and standard.
-   can sometimes prove slower */
-
 MEM_STATIC U16 MEM_read16(const void* memPtr)
 {
     U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
@@ -346,8 +298,6 @@ MEM_STATIC void MEM_write16(void* memPtr, U16 value)
     memcpy(memPtr, &value, sizeof(value));
 }
 
-#endif /* MEM_FORCE_MEMORY_ACCESS */
-
 MEM_STATIC U32 MEM_swap32(U32 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
@@ -442,7 +392,7 @@ MEM_STATIC size_t MEM_readLEST(const void* memPtr)
    header file (to include)
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -530,9 +480,8 @@ MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, unsigned nbBits);
 MEM_STATIC unsigned BITv07_highbit32 (U32 val)
 {
 #   if defined(_MSC_VER)   /* Visual */
-    unsigned long r=0;
-    _BitScanReverse ( &r, val );
-    return (unsigned) r;
+    unsigned long r;
+    return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return __builtin_clz (val) ^ 31;
 #   else   /* Software version */
@@ -600,7 +549,7 @@ MEM_STATIC size_t BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuff
 }
 
 /*! BITv07_lookBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BITv07_lookBitsFast(const BITv07_DStream_t* bitD, U32 nbBits)
 {
     U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
@@ -620,7 +569,7 @@ MEM_STATIC size_t BITv07_readBits(BITv07_DStream_t* bitD, U32 nbBits)
 }
 
 /*! BITv07_readBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+*   unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, U32 nbBits)
 {
     size_t const value = BITv07_lookBitsFast(bitD, nbBits);
@@ -674,7 +623,7 @@ MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* DStream)
    Public Prototypes declaration
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -982,7 +931,7 @@ MEM_STATIC BYTE FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStre
    header file
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1155,7 +1104,7 @@ size_t HUFv07_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void
    Common functions of New Generation Entropy library
    Copyright (C) 2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1379,7 +1328,7 @@ size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
    FSE : Finite State Entropy decoder
    Copyright (C) 2013-2015, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -1703,7 +1652,7 @@ size_t FSEv07_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t
    Huffman decoder, part of New Generation Entropy library
    Copyright (C) 2013-2016, Yann Collet.
 
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
@@ -2581,7 +2530,7 @@ size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize,
     Common functions of Zstd compression library
     Copyright (C) 2015-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2605,7 +2554,7 @@ size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize,
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net/
+    - zstd homepage : https://facebook.github.io/zstd/
 */
 
 
@@ -2651,7 +2600,7 @@ static void ZSTDv07_defaultFreeFunction(void* opaque, void* address)
     Header File for include
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2721,7 +2670,7 @@ typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
 #define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
 #define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
 
-#define HufLog 12
+#define ZSTD_HUFFDTABLE_CAPACITY_LOG 12
 typedef enum { lbt_huffman, lbt_repeat, lbt_raw, lbt_rle } litBlockType_t;
 
 #define LONGNBSEQ 0x7F00
@@ -2858,7 +2807,7 @@ static const ZSTDv07_customMem defaultCustomMem = { ZSTDv07_defaultAllocFunction
     zstd - standard compression library
     Copyright (C) 2014-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -2882,7 +2831,7 @@ static const ZSTDv07_customMem defaultCustomMem = { ZSTDv07_defaultAllocFunction
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net
+    - zstd homepage : https://facebook.github.io/zstd
 */
 
 /* ***************************************************************
@@ -2935,7 +2884,7 @@ struct ZSTDv07_DCtx_s
     FSEv07_DTable LLTable[FSEv07_DTABLE_SIZE_U32(LLFSELog)];
     FSEv07_DTable OffTable[FSEv07_DTABLE_SIZE_U32(OffFSELog)];
     FSEv07_DTable MLTable[FSEv07_DTABLE_SIZE_U32(MLFSELog)];
-    HUFv07_DTable hufTable[HUFv07_DTABLE_SIZE(HufLog)];  /* can accommodate HUFv07_decompress4X */
+    HUFv07_DTable hufTable[HUFv07_DTABLE_SIZE(ZSTD_HUFFDTABLE_CAPACITY_LOG)];  /* can accommodate HUFv07_decompress4X */
     const void* previousDstEnd;
     const void* base;
     const void* vBase;
@@ -2971,7 +2920,7 @@ size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx)
     dctx->base = NULL;
     dctx->vBase = NULL;
     dctx->dictEnd = NULL;
-    dctx->hufTable[0] = (HUFv07_DTable)((HufLog)*0x1000001);
+    dctx->hufTable[0] = (HUFv07_DTable)((ZSTD_HUFFDTABLE_CAPACITY_LOG)*0x1000001);
     dctx->litEntropy = dctx->fseEntropy = 0;
     dctx->dictID = 0;
     { int i; for (i=0; irep[i] = repStartValue[i]; }
@@ -3603,11 +3552,14 @@ size_t ZSTDv07_execSequence(BYTE* op,
     const BYTE* match = oLitEnd - sequence.offset;
 
     /* check */
-    if ((oLitEnd>oend_w) | (oMatchEnd>oend)) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
-    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    assert(oend >= op);
+    if (sequence.litLength + WILDCOPY_OVERLENGTH > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    if (sequenceLength > (size_t)(oend - op)) return ERROR(dstSize_tooSmall);
+    assert(litLimit >= *litPtr);
+    if (sequence.litLength > (size_t)(litLimit - *litPtr)) return ERROR(corruption_detected);;
 
     /* copy Literals */
-    ZSTDv07_wildcopy(op, *litPtr, sequence.litLength);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    ZSTDv07_wildcopy(op, *litPtr, (ptrdiff_t)sequence.litLength);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
     op = oLitEnd;
     *litPtr = iLitEnd;   /* update for next sequence */
 
@@ -3621,7 +3573,7 @@ size_t ZSTDv07_execSequence(BYTE* op,
             return sequenceLength;
         }
         /* span extDict & currentPrefixSegment */
-        {   size_t const length1 = dictEnd - match;
+        {   size_t const length1 = (size_t)(dictEnd - match);
             memmove(oLitEnd, match, length1);
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
@@ -4257,7 +4209,7 @@ ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
     Buffered version of Zstd compression library
     Copyright (C) 2015-2016, Yann Collet.
 
-    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+    BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
     Redistribution and use in source and binary forms, with or without
     modification, are permitted provided that the following conditions are
@@ -4281,7 +4233,7 @@ ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
     OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     You can contact the author at :
-    - zstd homepage : http://www.zstd.net/
+    - zstd homepage : https://facebook.github.io/zstd/
 */
 
 
@@ -4421,7 +4373,8 @@ size_t ZBUFFv07_decompressContinue(ZBUFFv07_DCtx* zbd,
                 if (hSize != 0) {
                     size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
                     if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
-                        memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        if (ip != NULL)
+                            memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
                         zbd->lhSize += iend-ip;
                         *dstCapacityPtr = 0;
                         return (hSize - zbd->lhSize) + ZSTDv07_blockHeaderSize;   /* remaining header bytes + next block header */
diff --git a/priv/zstd/lib/legacy/zstd_v07.h b/priv/zstd/lib/legacy/zstd_v07.h
index bc35cfa..1ff3904 100644
--- a/priv/zstd/lib/legacy/zstd_v07.h
+++ b/priv/zstd/lib/legacy/zstd_v07.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Yann Collet, Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/lib/libzstd.mk b/priv/zstd/lib/libzstd.mk
new file mode 100644
index 0000000..5e11d5d
--- /dev/null
+++ b/priv/zstd/lib/libzstd.mk
@@ -0,0 +1,214 @@
+# ################################################################
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under both the BSD-style license (found in the
+# LICENSE file in the root directory of this source tree) and the GPLv2 (found
+# in the COPYING file in the root directory of this source tree).
+# You may select, at your option, one of the above-listed licenses.
+# ################################################################
+
+##################################################################
+# Input Variables
+##################################################################
+
+# Zstd lib directory
+LIBZSTD ?= ./
+
+# ZSTD_LIB_MINIFY is a helper variable that
+# configures a bunch of other variables to space-optimized defaults.
+ZSTD_LIB_MINIFY ?= 0
+
+# Legacy support
+ifneq ($(ZSTD_LIB_MINIFY), 0)
+  ZSTD_LEGACY_SUPPORT ?= 0
+else
+  ZSTD_LEGACY_SUPPORT ?= 5
+endif
+ZSTD_LEGACY_MULTITHREADED_API ?= 0
+
+# Build size optimizations
+ifneq ($(ZSTD_LIB_MINIFY), 0)
+  HUF_FORCE_DECOMPRESS_X1 ?= 1
+  HUF_FORCE_DECOMPRESS_X2 ?= 0
+  ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT ?= 1
+  ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG ?= 0
+  ZSTD_NO_INLINE ?= 1
+  ZSTD_STRIP_ERROR_STRINGS ?= 1
+else
+  HUF_FORCE_DECOMPRESS_X1 ?= 0
+  HUF_FORCE_DECOMPRESS_X2 ?= 0
+  ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT ?= 0
+  ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG ?= 0
+  ZSTD_NO_INLINE ?= 0
+  ZSTD_STRIP_ERROR_STRINGS ?= 0
+endif
+
+# Assembly support
+ZSTD_NO_ASM ?= 0
+
+##################################################################
+# libzstd helpers
+##################################################################
+
+VOID ?= /dev/null
+
+# Make 4.3 doesn't support '\#' anymore (https://lwn.net/Articles/810071/)
+NUM_SYMBOL := \#
+
+# define silent mode as default (verbose mode with V=1 or VERBOSE=1)
+$(V)$(VERBOSE).SILENT:
+
+# When cross-compiling from linux to windows,
+# one might need to specify TARGET_SYSTEM as "Windows."
+# Building from Fedora fails without it.
+# (but Ubuntu and Debian don't need to set anything)
+TARGET_SYSTEM ?= $(OS)
+
+# Version numbers
+LIBVER_SRC := $(LIBZSTD)/zstd.h
+LIBVER_MAJOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MAJOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
+LIBVER_MINOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MINOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
+LIBVER_PATCH_SCRIPT:=`sed -n '/define ZSTD_VERSION_RELEASE/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
+LIBVER_SCRIPT:= $(LIBVER_MAJOR_SCRIPT).$(LIBVER_MINOR_SCRIPT).$(LIBVER_PATCH_SCRIPT)
+LIBVER_MAJOR := $(shell echo $(LIBVER_MAJOR_SCRIPT))
+LIBVER_MINOR := $(shell echo $(LIBVER_MINOR_SCRIPT))
+LIBVER_PATCH := $(shell echo $(LIBVER_PATCH_SCRIPT))
+LIBVER := $(shell echo $(LIBVER_SCRIPT))
+CCVER := $(shell $(CC) --version)
+ZSTD_VERSION?= $(LIBVER)
+
+ifneq ($(ZSTD_LIB_MINIFY), 0)
+  HAVE_CC_OZ ?= $(shell echo "" | $(CC) -Oz -x c -c - -o /dev/null 2> /dev/null && echo 1 || echo 0)
+ifneq ($(HAVE_CC_OZ), 0)
+    # Some compilers (clang) support an even more space-optimized setting.
+    CFLAGS += -Oz
+else
+    CFLAGS += -Os
+endif
+  CFLAGS += -fno-stack-protector -fomit-frame-pointer -fno-ident \
+            -DDYNAMIC_BMI2=0 -DNDEBUG
+else
+  CFLAGS ?= -O3
+endif
+
+DEBUGLEVEL ?= 0
+CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -DDEBUGLEVEL=$(DEBUGLEVEL)
+ifeq ($(TARGET_SYSTEM),Windows_NT)   # MinGW assumed
+  CPPFLAGS += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
+endif
+DEBUGFLAGS= -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
+            -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
+            -Wstrict-prototypes -Wundef -Wpointer-arith \
+            -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
+            -Wredundant-decls -Wmissing-prototypes -Wc++-compat
+CFLAGS   += $(DEBUGFLAGS) $(MOREFLAGS)
+ASFLAGS  += $(DEBUGFLAGS) $(MOREFLAGS) $(CFLAGS)
+LDFLAGS  += $(MOREFLAGS)
+FLAGS     = $(CPPFLAGS) $(CFLAGS) $(ASFLAGS) $(LDFLAGS)
+
+ifndef ALREADY_APPENDED_NOEXECSTACK
+export ALREADY_APPENDED_NOEXECSTACK := 1
+ifeq ($(shell echo "int main(int argc, char* argv[]) { (void)argc; (void)argv; return 0; }" | $(CC) $(FLAGS) -z noexecstack -x c -Werror - -o $(VOID) 2>$(VOID) && echo 1 || echo 0),1)
+LDFLAGS += -z noexecstack
+endif
+ifeq ($(shell echo | $(CC) $(FLAGS) -Wa,--noexecstack -x assembler -Werror -c - -o $(VOID) 2>$(VOID) && echo 1 || echo 0),1)
+CFLAGS  += -Wa,--noexecstack
+# CFLAGS are also added to ASFLAGS
+else ifeq ($(shell echo | $(CC) $(FLAGS) -Qunused-arguments -Wa,--noexecstack -x assembler -Werror -c - -o $(VOID) 2>$(VOID) && echo 1 || echo 0),1)
+# See e.g.: https://github.com/android/ndk/issues/171
+CFLAGS  += -Qunused-arguments -Wa,--noexecstack
+# CFLAGS are also added to ASFLAGS
+endif
+endif
+
+ifeq ($(shell echo "int main(int argc, char* argv[]) { (void)argc; (void)argv; return 0; }" | $(CC) $(FLAGS) -z cet-report=error -x c -Werror - -o $(VOID) 2>$(VOID) && echo 1 || echo 0),1)
+LDFLAGS += -z cet-report=error
+endif
+
+HAVE_COLORNEVER = $(shell echo a | grep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
+GREP_OPTIONS ?=
+ifeq ($(HAVE_COLORNEVER), 1)
+  GREP_OPTIONS += --color=never
+endif
+GREP = grep $(GREP_OPTIONS)
+
+ZSTD_COMMON_FILES := $(sort $(wildcard $(LIBZSTD)/common/*.c))
+ZSTD_COMPRESS_FILES := $(sort $(wildcard $(LIBZSTD)/compress/*.c))
+ZSTD_DECOMPRESS_FILES := $(sort $(wildcard $(LIBZSTD)/decompress/*.c))
+ZSTD_DICTBUILDER_FILES := $(sort $(wildcard $(LIBZSTD)/dictBuilder/*.c))
+ZSTD_DEPRECATED_FILES := $(sort $(wildcard $(LIBZSTD)/deprecated/*.c))
+ZSTD_LEGACY_FILES :=
+
+ZSTD_DECOMPRESS_AMD64_ASM_FILES := $(sort $(wildcard $(LIBZSTD)/decompress/*_amd64.S))
+
+ifneq ($(ZSTD_NO_ASM), 0)
+  CPPFLAGS += -DZSTD_DISABLE_ASM
+else
+  # Unconditionally add the ASM files they are disabled by
+  # macros in the .S file.
+  ZSTD_DECOMPRESS_FILES += $(ZSTD_DECOMPRESS_AMD64_ASM_FILES)
+endif
+
+ifneq ($(HUF_FORCE_DECOMPRESS_X1), 0)
+  CFLAGS += -DHUF_FORCE_DECOMPRESS_X1
+endif
+
+ifneq ($(HUF_FORCE_DECOMPRESS_X2), 0)
+  CFLAGS += -DHUF_FORCE_DECOMPRESS_X2
+endif
+
+ifneq ($(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT), 0)
+  CFLAGS += -DZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+endif
+
+ifneq ($(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG), 0)
+  CFLAGS += -DZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+endif
+
+ifneq ($(ZSTD_NO_INLINE), 0)
+  CFLAGS += -DZSTD_NO_INLINE
+endif
+
+ifneq ($(ZSTD_STRIP_ERROR_STRINGS), 0)
+  CFLAGS += -DZSTD_STRIP_ERROR_STRINGS
+endif
+
+ifneq ($(ZSTD_LEGACY_MULTITHREADED_API), 0)
+  CFLAGS += -DZSTD_LEGACY_MULTITHREADED_API
+endif
+
+ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
+ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
+  ZSTD_LEGACY_FILES += $(shell ls $(LIBZSTD)/legacy/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
+endif
+endif
+CPPFLAGS  += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
+
+UNAME := $(shell uname)
+
+ifndef BUILD_DIR
+ifeq ($(UNAME), Darwin)
+  ifeq ($(shell md5 < /dev/null > /dev/null; echo $$?), 0)
+    HASH ?= md5
+  endif
+else ifeq ($(UNAME), FreeBSD)
+  HASH ?= gmd5sum
+else ifeq ($(UNAME), NetBSD)
+  HASH ?= md5 -n
+else ifeq ($(UNAME), OpenBSD)
+  HASH ?= md5
+endif
+HASH ?= md5sum
+
+HASH_DIR = conf_$(shell echo $(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $(ZSTD_FILES) | $(HASH) | cut -f 1 -d " " )
+HAVE_HASH :=$(shell echo 1 | $(HASH) > /dev/null && echo 1 || echo 0)
+ifeq ($(HAVE_HASH),0)
+  $(info warning : could not find HASH ($(HASH)), needed to differentiate builds using different flags)
+  BUILD_DIR := obj/generic_noconf
+endif
+endif # BUILD_DIR
+
+ZSTD_SUBDIR := $(LIBZSTD)/common $(LIBZSTD)/compress $(LIBZSTD)/decompress $(LIBZSTD)/dictBuilder $(LIBZSTD)/legacy $(LIBZSTD)/deprecated
+vpath %.c $(ZSTD_SUBDIR)
+vpath %.S $(ZSTD_SUBDIR)
diff --git a/priv/zstd/lib/libzstd.pc.in b/priv/zstd/lib/libzstd.pc.in
index 8465c97..d5cc027 100644
--- a/priv/zstd/lib/libzstd.pc.in
+++ b/priv/zstd/lib/libzstd.pc.in
@@ -1,6 +1,6 @@
 #   ZSTD - standard compression algorithm
-#   Copyright (C) 2014-2016, Yann Collet, Facebook
-#   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+#   Copyright (c) Meta Platforms, Inc. and affiliates.
+#   BSD 2-Clause License (https://opensource.org/licenses/bsd-license.php)
 
 prefix=@PREFIX@
 exec_prefix=@EXEC_PREFIX@
@@ -9,7 +9,8 @@ libdir=@LIBDIR@
 
 Name: zstd
 Description: fast lossless compression algorithm library
-URL: http://www.zstd.net/
+URL: https://facebook.github.io/zstd/
 Version: @VERSION@
 Libs: -L${libdir} -lzstd
+Libs.private: @LIBS_PRIVATE@
 Cflags: -I${includedir}
diff --git a/priv/zstd/lib/module.modulemap b/priv/zstd/lib/module.modulemap
new file mode 100644
index 0000000..eff98df
--- /dev/null
+++ b/priv/zstd/lib/module.modulemap
@@ -0,0 +1,35 @@
+module libzstd [extern_c] {
+    header "zstd.h"
+    export *
+    config_macros [exhaustive] \
+        /* zstd.h */ \
+        ZSTD_STATIC_LINKING_ONLY, \
+        ZSTDLIB_VISIBILITY, \
+        ZSTDLIB_VISIBLE, \
+        ZSTDLIB_HIDDEN, \
+        ZSTD_DLL_EXPORT, \
+        ZSTDLIB_STATIC_API, \
+        ZSTD_DISABLE_DEPRECATE_WARNINGS, \
+        ZSTD_CLEVEL_DEFAULT, \
+        /* zdict.h */ \
+        ZDICT_STATIC_LINKING_ONLY, \
+        ZDICTLIB_VISIBLE, \
+        ZDICTLIB_HIDDEN, \
+        ZDICTLIB_VISIBILITY, \
+        ZDICTLIB_STATIC_API, \
+        ZDICT_DISABLE_DEPRECATE_WARNINGS, \
+        /* zstd_errors.h */ \
+        ZSTDERRORLIB_VISIBLE, \
+        ZSTDERRORLIB_HIDDEN, \
+        ZSTDERRORLIB_VISIBILITY
+
+    module dictbuilder [extern_c] {
+        header "zdict.h"
+        export *
+    }
+
+    module errors [extern_c] {
+        header "zstd_errors.h"
+        export *
+    }
+}
diff --git a/priv/zstd/lib/zdict.h b/priv/zstd/lib/zdict.h
index 75b05db..2268f94 100644
--- a/priv/zstd/lib/zdict.h
+++ b/priv/zstd/lib/zdict.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -8,32 +8,43 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
-#ifndef DICTBUILDER_H_001
-#define DICTBUILDER_H_001
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
+#ifndef ZSTD_ZDICT_H
+#define ZSTD_ZDICT_H
 
 /*======  Dependencies  ======*/
 #include   /* size_t */
 
 
 /* =====   ZDICTLIB_API : control library symbols visibility   ===== */
-#ifndef ZDICTLIB_VISIBILITY
-#  if defined(__GNUC__) && (__GNUC__ >= 4)
-#    define ZDICTLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#ifndef ZDICTLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZDICTLIB_VISIBILITY
+#    define ZDICTLIB_VISIBLE ZDICTLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZDICTLIB_VISIBLE __attribute__ ((visibility ("default")))
+#  else
+#    define ZDICTLIB_VISIBLE
+#  endif
+#endif
+
+#ifndef ZDICTLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZDICTLIB_HIDDEN __attribute__ ((visibility ("hidden")))
 #  else
-#    define ZDICTLIB_VISIBILITY
+#    define ZDICTLIB_HIDDEN
 #  endif
 #endif
+
 #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBILITY
+#  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBLE
 #elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
-#  define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#  define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
 #else
-#  define ZDICTLIB_API ZDICTLIB_VISIBILITY
+#  define ZDICTLIB_API ZDICTLIB_VISIBLE
 #endif
 
 /*******************************************************************************
@@ -46,7 +57,7 @@ extern "C" {
  *
  * Zstd can use dictionaries to improve compression ratio of small data.
  * Traditionally small files don't compress well because there is very little
- * repetion in a single sample, since it is small. But, if you are compressing
+ * repetition in a single sample, since it is small. But, if you are compressing
  * many similar files, like a bunch of JSON records that share the same
  * structure, you can train a dictionary on ahead of time on some samples of
  * these files. Then, zstd can use the dictionary to find repetitions that are
@@ -110,8 +121,8 @@ extern "C" {
  * The zstd CLI defaults to a 110KB dictionary. You likely don't need a
  * dictionary larger than that. But, most use cases can get away with a
  * smaller dictionary. The advanced dictionary builders can automatically
- * shrink the dictionary for you, and select a the smallest size that
- * doesn't hurt compression ratio too much. See the `shrinkDict` parameter.
+ * shrink the dictionary for you, and select the smallest size that doesn't
+ * hurt compression ratio too much. See the `shrinkDict` parameter.
  * A smaller dictionary can save memory, and potentially speed up
  * compression.
  *
@@ -132,7 +143,7 @@ extern "C" {
  *
  *   # Benchmark levels 1-3 without a dictionary
  *   zstd -b1e3 -r /path/to/my/files
- *   # Benchmark levels 1-3 with a dictioanry
+ *   # Benchmark levels 1-3 with a dictionary
  *   zstd -b1e3 -r /path/to/my/files -D /path/to/my/dictionary
  *
  * When should I retrain a dictionary?
@@ -201,9 +212,9 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCap
                                     const size_t* samplesSizes, unsigned nbSamples);
 
 typedef struct {
-    int      compressionLevel;   /*< optimize for a specific zstd compression level; 0 means default */
-    unsigned notificationLevel;  /*< Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
-    unsigned dictID;             /*< force dictID value; 0 means auto mode (32-bits random value)
+    int      compressionLevel;   /**< optimize for a specific zstd compression level; 0 means default */
+    unsigned notificationLevel;  /**< Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
+    unsigned dictID;             /**< force dictID value; 0 means auto mode (32-bits random value)
                                   *   NOTE: The zstd format reserves some dictionary IDs for future use.
                                   *         You may use them in private settings, but be warned that they
                                   *         may be used by zstd in a public dictionary registry in the future.
@@ -237,7 +248,6 @@ typedef struct {
  * is presumed that the most profitable content is at the end of the dictionary,
  * since that is the cheapest to reference.
  *
- * `dictContentSize` must be >= ZDICT_CONTENTSIZE_MIN bytes.
  * `maxDictSize` must be >= max(dictContentSize, ZSTD_DICTSIZE_MIN).
  *
  * @return: size of dictionary stored into `dstDictBuffer` (<= `maxDictSize`),
@@ -261,9 +271,21 @@ ZDICTLIB_API size_t ZDICT_getDictHeaderSize(const void* dictBuffer, size_t dictS
 ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode);
 ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
 
+#endif   /* ZSTD_ZDICT_H */
 
+#if defined(ZDICT_STATIC_LINKING_ONLY) && !defined(ZSTD_ZDICT_H_STATIC)
+#define ZSTD_ZDICT_H_STATIC
 
-#ifdef ZDICT_STATIC_LINKING_ONLY
+/* This can be overridden externally to hide static symbols. */
+#ifndef ZDICTLIB_STATIC_API
+#  if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#    define ZDICTLIB_STATIC_API __declspec(dllexport) ZDICTLIB_VISIBLE
+#  elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#    define ZDICTLIB_STATIC_API __declspec(dllimport) ZDICTLIB_VISIBLE
+#  else
+#    define ZDICTLIB_STATIC_API ZDICTLIB_VISIBLE
+#  endif
+#endif
 
 /* ====================================================================================
  * The definitions in this section are considered experimental.
@@ -272,8 +294,9 @@ ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
  * Use them only in association with static linking.
  * ==================================================================================== */
 
-#define ZDICT_CONTENTSIZE_MIN 128
 #define ZDICT_DICTSIZE_MIN    256
+/* Deprecated: Remove in v1.6.0 */
+#define ZDICT_CONTENTSIZE_MIN 128
 
 /*! ZDICT_cover_params_t:
  *  k and d are the only required parameters.
@@ -318,7 +341,7 @@ typedef struct {
  *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
  */
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
+ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_cover(
           void *dictBuffer, size_t dictBufferCapacity,
     const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
           ZDICT_cover_params_t parameters);
@@ -340,7 +363,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
  *          See ZDICT_trainFromBuffer() for details on failure modes.
  * Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
  */
-ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_cover(
           void* dictBuffer, size_t dictBufferCapacity,
     const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
           ZDICT_cover_params_t* parameters);
@@ -361,7 +384,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
  *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
  */
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
+ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
                     size_t dictBufferCapacity, const void *samplesBuffer,
                     const size_t *samplesSizes, unsigned nbSamples,
                     ZDICT_fastCover_params_t parameters);
@@ -384,7 +407,7 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_fastCover(void *dictBuffer,
  *          See ZDICT_trainFromBuffer() for details on failure modes.
  * Note: ZDICT_optimizeTrainFromBuffer_fastCover() requires about 6 * 2^f bytes of memory for each thread.
  */
-ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
+ZDICTLIB_STATIC_API size_t ZDICT_optimizeTrainFromBuffer_fastCover(void* dictBuffer,
                     size_t dictBufferCapacity, const void* samplesBuffer,
                     const size_t* samplesSizes, unsigned nbSamples,
                     ZDICT_fastCover_params_t* parameters);
@@ -409,7 +432,7 @@ typedef struct {
  *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
  *  Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
  */
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
+ZDICTLIB_STATIC_API size_t ZDICT_trainFromBuffer_legacy(
     void* dictBuffer, size_t dictBufferCapacity,
     const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
     ZDICT_legacy_params_t parameters);
@@ -421,32 +444,31 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
    or _CRT_SECURE_NO_WARNINGS in Visual.
    Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
 #ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
-#  define ZDICT_DEPRECATED(message) ZDICTLIB_API   /* disable deprecation warnings */
+#  define ZDICT_DEPRECATED(message) /* disable deprecation warnings */
 #else
 #  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
 #  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
-#    define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
+#    define ZDICT_DEPRECATED(message) [[deprecated(message)]]
 #  elif defined(__clang__) || (ZDICT_GCC_VERSION >= 405)
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
+#    define ZDICT_DEPRECATED(message) __attribute__((deprecated(message)))
 #  elif (ZDICT_GCC_VERSION >= 301)
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated))
+#    define ZDICT_DEPRECATED(message) __attribute__((deprecated))
 #  elif defined(_MSC_VER)
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __declspec(deprecated(message))
+#    define ZDICT_DEPRECATED(message) __declspec(deprecated(message))
 #  else
 #    pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API
+#    define ZDICT_DEPRECATED(message)
 #  endif
 #endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
 
 ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
+ZDICTLIB_STATIC_API
 size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
                                   const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
 
 
-#endif   /* ZDICT_STATIC_LINKING_ONLY */
+#endif   /* ZSTD_ZDICT_H_STATIC */
 
 #if defined (__cplusplus)
 }
 #endif
-
-#endif   /* DICTBUILDER_H_001 */
diff --git a/priv/zstd/lib/zstd.h b/priv/zstd/lib/zstd.h
index 4651e6c..95aac07 100644
--- a/priv/zstd/lib/zstd.h
+++ b/priv/zstd/lib/zstd.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,27 +14,61 @@ extern "C" {
 #ifndef ZSTD_H_235446
 #define ZSTD_H_235446
 
-/* ======   Dependency   ======*/
+/* ======   Dependencies   ======*/
 #include    /* INT_MAX */
 #include    /* size_t */
 
 
 /* =====   ZSTDLIB_API : control library symbols visibility   ===== */
-#ifndef ZSTDLIB_VISIBILITY
-#  if defined(__GNUC__) && (__GNUC__ >= 4)
-#    define ZSTDLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#ifndef ZSTDLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZSTDLIB_VISIBILITY
+#    define ZSTDLIB_VISIBLE ZSTDLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDLIB_VISIBLE __attribute__ ((visibility ("default")))
 #  else
-#    define ZSTDLIB_VISIBILITY
+#    define ZSTDLIB_VISIBLE
 #  endif
 #endif
+
+#ifndef ZSTDLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZSTDLIB_HIDDEN
+#  endif
+#endif
+
 #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBILITY
+#  define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBLE
 #elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
-#  define ZSTDLIB_API __declspec(dllimport) ZSTDLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#  define ZSTDLIB_API __declspec(dllimport) ZSTDLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
 #else
-#  define ZSTDLIB_API ZSTDLIB_VISIBILITY
+#  define ZSTDLIB_API ZSTDLIB_VISIBLE
 #endif
 
+/* Deprecation warnings :
+ * Should these warnings be a problem, it is generally possible to disable them,
+ * typically with -Wno-deprecated-declarations for gcc or _CRT_SECURE_NO_WARNINGS in Visual.
+ * Otherwise, it's also possible to define ZSTD_DISABLE_DEPRECATE_WARNINGS.
+ */
+#ifdef ZSTD_DISABLE_DEPRECATE_WARNINGS
+#  define ZSTD_DEPRECATED(message) /* disable deprecation warnings */
+#else
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZSTD_DEPRECATED(message) [[deprecated(message)]]
+#  elif (defined(GNUC) && (GNUC > 4 || (GNUC == 4 && GNUC_MINOR >= 5))) || defined(__clang__)
+#    define ZSTD_DEPRECATED(message) __attribute__((deprecated(message)))
+#  elif defined(__GNUC__) && (__GNUC__ >= 3)
+#    define ZSTD_DEPRECATED(message) __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZSTD_DEPRECATED(message) __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZSTD_DEPRECATED for this compiler")
+#    define ZSTD_DEPRECATED(message)
+#  endif
+#endif /* ZSTD_DISABLE_DEPRECATE_WARNINGS */
+
 
 /*******************************************************************************
   Introduction
@@ -72,7 +106,7 @@ extern "C" {
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
 #define ZSTD_VERSION_MINOR    5
-#define ZSTD_VERSION_RELEASE  0
+#define ZSTD_VERSION_RELEASE  4
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
 
 /*! ZSTD_versionNumber() :
@@ -163,7 +197,9 @@ ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t
  *  "empty", "unknown" and "error" results to the same return value (0),
  *  while ZSTD_getFrameContentSize() gives them separate return values.
  * @return : decompressed size of `src` frame content _if known and not empty_, 0 otherwise. */
-ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
+ZSTD_DEPRECATED("Replaced by ZSTD_getFrameContentSize")
+ZSTDLIB_API
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
 
 /*! ZSTD_findFrameCompressedSize() : Requires v1.4.0+
  * `src` should point to the start of a ZSTD frame or skippable frame.
@@ -175,8 +211,30 @@ ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize)
 
 
 /*======  Helper functions  ======*/
-#define ZSTD_COMPRESSBOUND(srcSize)   ((srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
-ZSTDLIB_API size_t      ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+/* ZSTD_compressBound() :
+ * maximum compressed size in worst case single-pass scenario.
+ * When invoking `ZSTD_compress()` or any other one-pass compression function,
+ * it's recommended to provide @dstCapacity >= ZSTD_compressBound(srcSize)
+ * as it eliminates one potential failure scenario,
+ * aka not enough room in dst buffer to write the compressed frame.
+ * Note : ZSTD_compressBound() itself can fail, if @srcSize > ZSTD_MAX_INPUT_SIZE .
+ *        In which case, ZSTD_compressBound() will return an error code
+ *        which can be tested using ZSTD_isError().
+ *
+ * ZSTD_COMPRESSBOUND() :
+ * same as ZSTD_compressBound(), but as a macro.
+ * It can be used to produce constants, which can be useful for static allocation,
+ * for example to size a static array on stack.
+ * Will produce constant value 0 if srcSize too large.
+ */
+#define ZSTD_MAX_INPUT_SIZE ((sizeof(size_t)==8) ? 0xFF00FF00FF00FF00LLU : 0xFF00FF00U)
+#define ZSTD_COMPRESSBOUND(srcSize)   (((size_t)(srcSize) >= ZSTD_MAX_INPUT_SIZE) ? 0 : (srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
+ZSTDLIB_API size_t ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+/* ZSTD_isError() :
+ * Most ZSTD_* functions returning a size_t value can be tested for error,
+ * using ZSTD_isError().
+ * @return 1 if error, 0 otherwise
+ */
 ZSTDLIB_API unsigned    ZSTD_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
 ZSTDLIB_API const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
 ZSTDLIB_API int         ZSTD_minCLevel(void);               /*!< minimum negative compression level allowed, requires v1.4.0+ */
@@ -247,7 +305,7 @@ ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx,
  *
  *   It's possible to reset all parameters to "default" using ZSTD_CCtx_reset().
  *
- *   This API supercedes all other "advanced" API entry points in the experimental section.
+ *   This API supersedes all other "advanced" API entry points in the experimental section.
  *   In the future, we expect to remove from experimental API entry points which are redundant with this API.
  */
 
@@ -417,8 +475,11 @@ typedef enum {
      * ZSTD_c_stableOutBuffer
      * ZSTD_c_blockDelimiters
      * ZSTD_c_validateSequences
-     * ZSTD_c_splitBlocks
+     * ZSTD_c_useBlockSplitter
      * ZSTD_c_useRowMatchFinder
+     * ZSTD_c_prefetchCDictTables
+     * ZSTD_c_enableSeqProducerFallback
+     * ZSTD_c_maxBlockSize
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -437,7 +498,11 @@ typedef enum {
      ZSTD_c_experimentalParam12=1009,
      ZSTD_c_experimentalParam13=1010,
      ZSTD_c_experimentalParam14=1011,
-     ZSTD_c_experimentalParam15=1012
+     ZSTD_c_experimentalParam15=1012,
+     ZSTD_c_experimentalParam16=1013,
+     ZSTD_c_experimentalParam17=1014,
+     ZSTD_c_experimentalParam18=1015,
+     ZSTD_c_experimentalParam19=1016
 } ZSTD_cParameter;
 
 typedef struct {
@@ -500,7 +565,7 @@ typedef enum {
  *                  They will be used to compress next frame.
  *                  Resetting session never fails.
  *  - The parameters : changes all parameters back to "default".
- *                  This removes any reference to any dictionary too.
+ *                  This also removes any reference to any dictionary or external sequence producer.
  *                  Parameters can only be changed between 2 sessions (i.e. no compression is currently ongoing)
  *                  otherwise the reset fails, and function returns an error value (which can be tested using ZSTD_isError())
  *  - Both : similar to resetting the session, followed by resetting parameters.
@@ -550,13 +615,15 @@ typedef enum {
      * ZSTD_d_stableOutBuffer
      * ZSTD_d_forceIgnoreChecksum
      * ZSTD_d_refMultipleDDicts
+     * ZSTD_d_disableHuffmanAssembly
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly
      */
      ZSTD_d_experimentalParam1=1000,
      ZSTD_d_experimentalParam2=1001,
      ZSTD_d_experimentalParam3=1002,
-     ZSTD_d_experimentalParam4=1003
+     ZSTD_d_experimentalParam4=1003,
+     ZSTD_d_experimentalParam5=1004
 
 } ZSTD_dParameter;
 
@@ -735,8 +802,6 @@ ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output
  * This following is a legacy streaming API, available since v1.0+ .
  * It can be replaced by ZSTD_CCtx_reset() and ZSTD_compressStream2().
  * It is redundant, but remains fully supported.
- * Streaming in combination with advanced parameters and dictionary compression
- * can only be used through the new API.
  ******************************************************************************/
 
 /*!
@@ -745,6 +810,9 @@ ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     ZSTD_CCtx_refCDict(zcs, NULL); // clear the dictionary (if any)
  *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
+ *
+ * Note that ZSTD_initCStream() clears any previously set dictionary. Use the new API
+ * to compress with a dictionary.
  */
 ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
 /*!
@@ -795,13 +863,31 @@ ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);  /* accept NULL pointer
 
 /*===== Streaming decompression functions =====*/
 
-/* This function is redundant with the advanced API and equivalent to:
+/*! ZSTD_initDStream() :
+ * Initialize/reset DStream state for new decompression operation.
+ * Call before new decompression operation using same DStream.
  *
+ * Note : This function is redundant with the advanced API and equivalent to:
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
  *     ZSTD_DCtx_refDDict(zds, NULL);
  */
 ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
 
+/*! ZSTD_decompressStream() :
+ * Streaming decompression function.
+ * Call repetitively to consume full input updating it as necessary.
+ * Function will update both input and output `pos` fields exposing current state via these fields:
+ * - `input.pos < input.size`, some input remaining and caller should provide remaining input
+ *   on the next call.
+ * - `output.pos < output.size`, decoder finished and flushed all remaining buffers.
+ * - `output.pos == output.size`, potentially uncflushed data present in the internal buffers,
+ *   call ZSTD_decompressStream() again to flush remaining data to output.
+ * Note : with no additional input, amount of data flushed <= ZSTD_BLOCKSIZE_MAX.
+ *
+ * @return : 0 when a frame is completely decoded and fully flushed,
+ *           or an error code, which can be tested using ZSTD_isError(),
+ *           or any other value > 0, which means there is some decoding or flushing to do to complete current frame.
+ */
 ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
 
 ZSTDLIB_API size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
@@ -920,7 +1006,7 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict);
  *  If @return == 0, the dictID could not be decoded.
  *  This could for one of the following reasons :
  *  - The frame does not require a dictionary to be decoded (most common case).
- *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden information.
+ *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden piece of information.
  *    Note : this use case also happens when using a non-conformant dictionary.
  *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
  *  - This is not a Zstandard frame.
@@ -932,7 +1018,7 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
  * Advanced dictionary and prefix API (Requires v1.4.0+)
  *
  * This API allows dictionaries to be used with ZSTD_compress2(),
- * ZSTD_compressStream2(), and ZSTD_decompress(). Dictionaries are sticky, and
+ * ZSTD_compressStream2(), and ZSTD_decompressDCtx(). Dictionaries are sticky, and
  * only reset with the context is reset with ZSTD_reset_parameters or
  * ZSTD_reset_session_and_parameters. Prefixes are single-use.
  ******************************************************************************/
@@ -944,8 +1030,9 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special: Loading a NULL (or 0-size) dictionary invalidates previous dictionary,
  *           meaning "return to no-dictionary mode".
- *  Note 1 : Dictionary is sticky, it will be used for all future compressed frames.
- *           To return to "no-dictionary" situation, load a NULL dictionary (or reset parameters).
+ *  Note 1 : Dictionary is sticky, it will be used for all future compressed frames,
+ *           until parameters are reset, a new dictionary is loaded, or the dictionary
+ *           is explicitly invalidated by loading a NULL dictionary.
  *  Note 2 : Loading a dictionary involves building tables.
  *           It's also a CPU consuming operation, with non-negligible impact on latency.
  *           Tables are dependent on compression parameters, and for this reason,
@@ -958,7 +1045,7 @@ ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
 ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
 
 /*! ZSTD_CCtx_refCDict() : Requires v1.4.0+
- *  Reference a prepared dictionary, to be used for all next compressed frames.
+ *  Reference a prepared dictionary, to be used for all future compressed frames.
  *  Note that compression parameters are enforced from within CDict,
  *  and supersede any compression parameter previously set within CCtx.
  *  The parameters ignored are labelled as "superseded-by-cdict" in the ZSTD_cParameter enum docs.
@@ -993,9 +1080,9 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx,
                                  const void* prefix, size_t prefixSize);
 
 /*! ZSTD_DCtx_loadDictionary() : Requires v1.4.0+
- *  Create an internal DDict from dict buffer,
- *  to be used to decompress next frames.
- *  The dictionary remains valid for all future frames, until explicitly invalidated.
+ *  Create an internal DDict from dict buffer, to be used to decompress all future frames.
+ *  The dictionary remains valid for all future frames, until explicitly invalidated, or
+ *  a new dictionary is loaded.
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  *  Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
  *            meaning "return to no-dictionary mode".
@@ -1019,9 +1106,10 @@ ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, s
  *  The memory for the table is allocated on the first call to refDDict, and can be
  *  freed with ZSTD_freeDCtx().
  *
+ *  If called with ZSTD_d_refMultipleDDicts disabled (the default), only one dictionary
+ *  will be managed, and referencing a dictionary effectively "discards" any previous one.
+ *
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
- *  Note 1 : Currently, only one dictionary can be managed.
- *           Referencing a new dictionary effectively "discards" any previous one.
  *  Special: referencing a NULL DDict means "return to no-dictionary mode".
  *  Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
  */
@@ -1073,27 +1161,16 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
 #define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
 
-/* Deprecation warnings :
- * Should these warnings be a problem, it is generally possible to disable them,
- * typically with -Wno-deprecated-declarations for gcc or _CRT_SECURE_NO_WARNINGS in Visual.
- * Otherwise, it's also possible to define ZSTD_DISABLE_DEPRECATE_WARNINGS.
- */
-#ifdef ZSTD_DISABLE_DEPRECATE_WARNINGS
-#  define ZSTD_DEPRECATED(message) ZSTDLIB_API  /* disable deprecation warnings */
-#else
-#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
-#    define ZSTD_DEPRECATED(message) [[deprecated(message)]] ZSTDLIB_API
-#  elif (defined(GNUC) && (GNUC > 4 || (GNUC == 4 && GNUC_MINOR >= 5))) || defined(__clang__)
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_API __attribute__((deprecated(message)))
-#  elif defined(__GNUC__) && (__GNUC__ >= 3)
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_API __attribute__((deprecated))
-#  elif defined(_MSC_VER)
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_API __declspec(deprecated(message))
+/* This can be overridden externally to hide static symbols. */
+#ifndef ZSTDLIB_STATIC_API
+#  if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#    define ZSTDLIB_STATIC_API __declspec(dllexport) ZSTDLIB_VISIBLE
+#  elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#    define ZSTDLIB_STATIC_API __declspec(dllimport) ZSTDLIB_VISIBLE
 #  else
-#    pragma message("WARNING: You need to implement ZSTD_DEPRECATED for this compiler")
-#    define ZSTD_DEPRECATED(message) ZSTDLIB_API
+#    define ZSTDLIB_STATIC_API ZSTDLIB_VISIBLE
 #  endif
-#endif /* ZSTD_DISABLE_DEPRECATE_WARNINGS */
+#endif
 
 /****************************************************************************************
  *   experimental API (static linking only)
@@ -1129,6 +1206,7 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #define ZSTD_TARGETLENGTH_MIN     0   /* note : comparing this constant to an unsigned results in a tautological test */
 #define ZSTD_STRATEGY_MIN        ZSTD_fast
 #define ZSTD_STRATEGY_MAX        ZSTD_btultra2
+#define ZSTD_BLOCKSIZE_MAX_MIN (1 << 10) /* The minimum valid max blocksize. Maximum blocksizes smaller than this make compressBound() inaccurate. */
 
 
 #define ZSTD_OVERLAPLOG_MIN       0
@@ -1157,9 +1235,6 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #define ZSTD_SRCSIZEHINT_MIN        0
 #define ZSTD_SRCSIZEHINT_MAX        INT_MAX
 
-/* internal */
-#define ZSTD_HASHLOG3_MAX           17
-
 
 /* ---  Advanced types  --- */
 
@@ -1302,10 +1377,14 @@ typedef enum {
 } ZSTD_literalCompressionMode_e;
 
 typedef enum {
-  ZSTD_urm_auto = 0,                   /* Automatically determine whether or not we use row matchfinder */
-  ZSTD_urm_disableRowMatchFinder = 1,  /* Never use row matchfinder */
-  ZSTD_urm_enableRowMatchFinder = 2    /* Always use row matchfinder when applicable */
-} ZSTD_useRowMatchFinderMode_e;
+  /* Note: This enum controls features which are conditionally beneficial. Zstd typically will make a final
+   * decision on whether or not to enable the feature (ZSTD_ps_auto), but setting the switch to ZSTD_ps_enable
+   * or ZSTD_ps_disable allow for a force enable/disable the feature.
+   */
+  ZSTD_ps_auto = 0,         /* Let the library automatically determine whether the feature shall be enabled */
+  ZSTD_ps_enable = 1,       /* Force-enable the feature */
+  ZSTD_ps_disable = 2       /* Do not use the feature */
+} ZSTD_paramSwitch_e;
 
 /***************************************
 *  Frame size functions
@@ -1332,7 +1411,7 @@ typedef enum {
  *   note 5 : ZSTD_findDecompressedSize handles multiple frames, and so it must traverse the input to
  *            read each contained frame header.  This is fast as most of the data is skipped,
  *            however it does mean that all frame data must be present and valid. */
-ZSTDLIB_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
 
 /*! ZSTD_decompressBound() :
  *  `src` should point to the start of a series of ZSTD encoded and/or skippable frames
@@ -1347,41 +1426,97 @@ ZSTDLIB_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t
  *  note 3  : when the decompressed size field isn't available, the upper-bound for that frame is calculated by:
  *              upper-bound = # blocks * min(128 KB, Window_Size)
  */
-ZSTDLIB_API unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize);
 
 /*! ZSTD_frameHeaderSize() :
  *  srcSize must be >= ZSTD_FRAMEHEADERSIZE_PREFIX.
  * @return : size of the Frame Header,
  *           or an error code (if srcSize is too small) */
-ZSTDLIB_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
+
+/*! ZSTD_decompressionMargin() :
+ * Zstd supports in-place decompression, where the input and output buffers overlap.
+ * In this case, the output buffer must be at least (Margin + Output_Size) bytes large,
+ * and the input buffer must be at the end of the output buffer.
+ *
+ *  _______________________ Output Buffer ________________________
+ * |                                                              |
+ * |                                        ____ Input Buffer ____|
+ * |                                       |                      |
+ * v                                       v                      v
+ * |---------------------------------------|-----------|----------|
+ * ^                                                   ^          ^
+ * |___________________ Output_Size ___________________|_ Margin _|
+ *
+ * NOTE: See also ZSTD_DECOMPRESSION_MARGIN().
+ * NOTE: This applies only to single-pass decompression through ZSTD_decompress() or
+ * ZSTD_decompressDCtx().
+ * NOTE: This function supports multi-frame input.
+ *
+ * @param src The compressed frame(s)
+ * @param srcSize The size of the compressed frame(s)
+ * @returns The decompression margin or an error that can be checked with ZSTD_isError().
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_decompressionMargin(const void* src, size_t srcSize);
+
+/*! ZSTD_DECOMPRESS_MARGIN() :
+ * Similar to ZSTD_decompressionMargin(), but instead of computing the margin from
+ * the compressed frame, compute it from the original size and the blockSizeLog.
+ * See ZSTD_decompressionMargin() for details.
+ *
+ * WARNING: This macro does not support multi-frame input, the input must be a single
+ * zstd frame. If you need that support use the function, or implement it yourself.
+ *
+ * @param originalSize The original uncompressed size of the data.
+ * @param blockSize    The block size == MIN(windowSize, ZSTD_BLOCKSIZE_MAX).
+ *                     Unless you explicitly set the windowLog smaller than
+ *                     ZSTD_BLOCKSIZELOG_MAX you can just use ZSTD_BLOCKSIZE_MAX.
+ */
+#define ZSTD_DECOMPRESSION_MARGIN(originalSize, blockSize) ((size_t)(                                              \
+        ZSTD_FRAMEHEADERSIZE_MAX                                                              /* Frame header */ + \
+        4                                                                                         /* checksum */ + \
+        ((originalSize) == 0 ? 0 : 3 * (((originalSize) + (blockSize) - 1) / blockSize)) /* 3 bytes per block */ + \
+        (blockSize)                                                                    /* One block of margin */   \
+    ))
 
 typedef enum {
   ZSTD_sf_noBlockDelimiters = 0,         /* Representation of ZSTD_Sequence has no block delimiters, sequences only */
   ZSTD_sf_explicitBlockDelimiters = 1    /* Representation of ZSTD_Sequence contains explicit block delimiters */
 } ZSTD_sequenceFormat_e;
 
+/*! ZSTD_sequenceBound() :
+ * `srcSize` : size of the input buffer
+ *  @return : upper-bound for the number of sequences that can be generated
+ *            from a buffer of srcSize bytes
+ *
+ *  note : returns number of sequences - to get bytes, multiply by sizeof(ZSTD_Sequence).
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_sequenceBound(size_t srcSize);
+
 /*! ZSTD_generateSequences() :
- * Generate sequences using ZSTD_compress2, given a source buffer.
+ * Generate sequences using ZSTD_compress2(), given a source buffer.
  *
  * Each block will end with a dummy sequence
  * with offset == 0, matchLength == 0, and litLength == length of last literals.
  * litLength may be == 0, and if so, then the sequence of (of: 0 ml: 0 ll: 0)
  * simply acts as a block delimiter.
  *
- * zc can be used to insert custom compression params.
- * This function invokes ZSTD_compress2
+ * @zc can be used to insert custom compression params.
+ * This function invokes ZSTD_compress2().
  *
  * The output of this function can be fed into ZSTD_compressSequences() with CCtx
  * setting of ZSTD_c_blockDelimiters as ZSTD_sf_explicitBlockDelimiters
  * @return : number of sequences generated
  */
 
-ZSTDLIB_API size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
-                                          size_t outSeqsSize, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t
+ZSTD_generateSequences( ZSTD_CCtx* zc,
+                        ZSTD_Sequence* outSeqs, size_t outSeqsSize,
+                        const void* src, size_t srcSize);
 
 /*! ZSTD_mergeBlockDelimiters() :
  * Given an array of ZSTD_Sequence, remove all sequences that represent block delimiters/last literals
- * by merging them into into the literals of the next sequence.
+ * by merging them into the literals of the next sequence.
  *
  * As such, the final generated result has no explicit representation of block boundaries,
  * and the final last literals segment is not represented in the sequences.
@@ -1390,10 +1525,12 @@ ZSTDLIB_API size_t ZSTD_generateSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
  * setting of ZSTD_c_blockDelimiters as ZSTD_sf_noBlockDelimiters
  * @return : number of sequences left after merging
  */
-ZSTDLIB_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
+ZSTDLIB_STATIC_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize);
 
 /*! ZSTD_compressSequences() :
- * Compress an array of ZSTD_Sequence, generated from the original source buffer, into dst.
+ * Compress an array of ZSTD_Sequence, associated with @src buffer, into dst.
+ * @src contains the entire input (not just the literals).
+ * If @srcSize > sum(sequence.length), the remaining bytes are considered all literals
  * If a dictionary is included, then the cctx should reference the dict. (see: ZSTD_CCtx_refCDict(), ZSTD_CCtx_loadDictionary(), etc.)
  * The entire source is compressed into a single frame.
  *
@@ -1418,17 +1555,18 @@ ZSTDLIB_API size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t se
  * Note: Repcodes are, as of now, always re-calculated within this function, so ZSTD_Sequence::rep is unused.
  * Note 2: Once we integrate ability to ingest repcodes, the explicit block delims mode must respect those repcodes exactly,
  *         and cannot emit an RLE block that disagrees with the repcode history
- * @return : final compressed size or a ZSTD error.
+ * @return : final compressed size, or a ZSTD error code.
  */
-ZSTDLIB_API size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size_t dstSize,
-                                  const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
-                                  const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t
+ZSTD_compressSequences( ZSTD_CCtx* cctx, void* dst, size_t dstSize,
+                        const ZSTD_Sequence* inSeqs, size_t inSeqsSize,
+                        const void* src, size_t srcSize);
 
 
 /*! ZSTD_writeSkippableFrame() :
  * Generates a zstd skippable frame containing data given by src, and writes it to dst buffer.
  *
- * Skippable frames begin with a a 4-byte magic number. There are 16 possible choices of magic number,
+ * Skippable frames begin with a 4-byte magic number. There are 16 possible choices of magic number,
  * ranging from ZSTD_MAGIC_SKIPPABLE_START to ZSTD_MAGIC_SKIPPABLE_START+15.
  * As such, the parameter magicVariant controls the exact skippable frame magic number variant used, so
  * the magic number used will be ZSTD_MAGIC_SKIPPABLE_START + magicVariant.
@@ -1438,9 +1576,29 @@ ZSTDLIB_API size_t ZSTD_compressSequences(ZSTD_CCtx* const cctx, void* dst, size
  *
  * @return : number of bytes written or a ZSTD error.
  */
-ZSTDLIB_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
+ZSTDLIB_STATIC_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
                                             const void* src, size_t srcSize, unsigned magicVariant);
 
+/*! ZSTD_readSkippableFrame() :
+ * Retrieves a zstd skippable frame containing data given by src, and writes it to dst buffer.
+ *
+ * The parameter magicVariant will receive the magicVariant that was supplied when the frame was written,
+ * i.e. magicNumber - ZSTD_MAGIC_SKIPPABLE_START.  This can be NULL if the caller is not interested
+ * in the magicVariant.
+ *
+ * Returns an error if destination buffer is not large enough, or if the frame is not skippable.
+ *
+ * @return : number of bytes written or a ZSTD error.
+ */
+ZSTDLIB_API size_t ZSTD_readSkippableFrame(void* dst, size_t dstCapacity, unsigned* magicVariant,
+                                            const void* src, size_t srcSize);
+
+/*! ZSTD_isSkippableFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier for a skippable frame.
+ */
+ZSTDLIB_API unsigned ZSTD_isSkippableFrame(const void* buffer, size_t size);
+
+
 
 /***************************************
 *  Memory management
@@ -1466,13 +1624,16 @@ ZSTDLIB_API size_t ZSTD_writeSkippableFrame(void* dst, size_t dstCapacity,
  *  and ZSTD_estimateCCtxSize_usingCCtxParams(), which can be used in tandem with ZSTD_CCtxParams_setParameter().
  *  Both can be used to estimate memory using custom compression parameters and arbitrary srcSize limits.
  *
- *  Note 2 : only single-threaded compression is supported.
+ *  Note : only single-threaded compression is supported.
  *  ZSTD_estimateCCtxSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+ *
+ *  Note 2 : ZSTD_estimateCCtxSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+ *  Size estimates assume that no external sequence producer is registered.
  */
-ZSTDLIB_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
-ZSTDLIB_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
-ZSTDLIB_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
-ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDCtxSize(void);
 
 /*! ZSTD_estimateCStreamSize() :
  *  ZSTD_estimateCStreamSize() will provide a budget large enough for any compression level up to selected one.
@@ -1486,21 +1647,26 @@ ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
  *  or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
  *  Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
  *         an internal ?Dict will be created, which additional size is not estimated here.
- *         In this case, get total size by adding ZSTD_estimate?DictSize */
-ZSTDLIB_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
-ZSTDLIB_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
-ZSTDLIB_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
-ZSTDLIB_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
-ZSTDLIB_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
+ *         In this case, get total size by adding ZSTD_estimate?DictSize
+ *  Note 2 : only single-threaded compression is supported.
+ *  ZSTD_estimateCStreamSize_usingCCtxParams() will return an error code if ZSTD_c_nbWorkers is >= 1.
+ *  Note 3 : ZSTD_estimateCStreamSize* functions are not compatible with the Block-Level Sequence Producer API at this time.
+ *  Size estimates assume that no external sequence producer is registered.
+ */
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
 
 /*! ZSTD_estimate?DictSize() :
  *  ZSTD_estimateCDictSize() will bet that src size is relatively "small", and content is copied, like ZSTD_createCDict().
  *  ZSTD_estimateCDictSize_advanced() makes it possible to control compression parameters precisely, like ZSTD_createCDict_advanced().
  *  Note : dictionaries created by reference (`ZSTD_dlm_byRef`) are logically smaller.
  */
-ZSTDLIB_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
-ZSTDLIB_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
+ZSTDLIB_STATIC_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
 
 /*! ZSTD_initStatic*() :
  *  Initialize an object using a pre-allocated fixed-size buffer.
@@ -1523,20 +1689,20 @@ ZSTDLIB_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e
  *  Limitation 2 : static cctx currently not compatible with multi-threading.
  *  Limitation 3 : static dctx is incompatible with legacy support.
  */
-ZSTDLIB_API ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
-ZSTDLIB_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
+ZSTDLIB_STATIC_API ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
 
-ZSTDLIB_API ZSTD_DCtx*    ZSTD_initStaticDCtx(void* workspace, size_t workspaceSize);
-ZSTDLIB_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
+ZSTDLIB_STATIC_API ZSTD_DCtx*    ZSTD_initStaticDCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
 
-ZSTDLIB_API const ZSTD_CDict* ZSTD_initStaticCDict(
+ZSTDLIB_STATIC_API const ZSTD_CDict* ZSTD_initStaticCDict(
                                         void* workspace, size_t workspaceSize,
                                         const void* dict, size_t dictSize,
                                         ZSTD_dictLoadMethod_e dictLoadMethod,
                                         ZSTD_dictContentType_e dictContentType,
                                         ZSTD_compressionParameters cParams);
 
-ZSTDLIB_API const ZSTD_DDict* ZSTD_initStaticDDict(
+ZSTDLIB_STATIC_API const ZSTD_DDict* ZSTD_initStaticDDict(
                                         void* workspace, size_t workspaceSize,
                                         const void* dict, size_t dictSize,
                                         ZSTD_dictLoadMethod_e dictLoadMethod,
@@ -1557,44 +1723,44 @@ __attribute__((__unused__))
 #endif
 ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL };  /**< this constant defers to stdlib's functions */
 
-ZSTDLIB_API ZSTD_CCtx*    ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
-ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
-ZSTDLIB_API ZSTD_DCtx*    ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
-ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_CCtx*    ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_DCtx*    ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_STATIC_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
 
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
+ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
                                                   ZSTD_dictLoadMethod_e dictLoadMethod,
                                                   ZSTD_dictContentType_e dictContentType,
                                                   ZSTD_compressionParameters cParams,
                                                   ZSTD_customMem customMem);
 
-/* ! Thread pool :
- * These prototypes make it possible to share a thread pool among multiple compression contexts.
- * This can limit resources for applications with multiple threads where each one uses
- * a threaded compression mode (via ZSTD_c_nbWorkers parameter).
- * ZSTD_createThreadPool creates a new thread pool with a given number of threads.
- * Note that the lifetime of such pool must exist while being used.
- * ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
- * to use an internal thread pool).
- * ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
+/*! Thread pool :
+ *  These prototypes make it possible to share a thread pool among multiple compression contexts.
+ *  This can limit resources for applications with multiple threads where each one uses
+ *  a threaded compression mode (via ZSTD_c_nbWorkers parameter).
+ *  ZSTD_createThreadPool creates a new thread pool with a given number of threads.
+ *  Note that the lifetime of such pool must exist while being used.
+ *  ZSTD_CCtx_refThreadPool assigns a thread pool to a context (use NULL argument value
+ *  to use an internal thread pool).
+ *  ZSTD_freeThreadPool frees a thread pool, accepts NULL pointer.
  */
 typedef struct POOL_ctx_s ZSTD_threadPool;
-ZSTDLIB_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
-ZSTDLIB_API void ZSTD_freeThreadPool (ZSTD_threadPool* pool);  /* accept NULL pointer */
-ZSTDLIB_API size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool);
+ZSTDLIB_STATIC_API ZSTD_threadPool* ZSTD_createThreadPool(size_t numThreads);
+ZSTDLIB_STATIC_API void ZSTD_freeThreadPool (ZSTD_threadPool* pool);  /* accept NULL pointer */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refThreadPool(ZSTD_CCtx* cctx, ZSTD_threadPool* pool);
 
 
 /*
  * This API is temporary and is expected to change or disappear in the future!
  */
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced2(
+ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_advanced2(
     const void* dict, size_t dictSize,
     ZSTD_dictLoadMethod_e dictLoadMethod,
     ZSTD_dictContentType_e dictContentType,
     const ZSTD_CCtx_params* cctxParams,
     ZSTD_customMem customMem);
 
-ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(
+ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_advanced(
     const void* dict, size_t dictSize,
     ZSTD_dictLoadMethod_e dictLoadMethod,
     ZSTD_dictContentType_e dictContentType,
@@ -1611,22 +1777,22 @@ ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(
  *  As a consequence, `dictBuffer` **must** outlive CDict,
  *  and its content must remain unmodified throughout the lifetime of CDict.
  *  note: equivalent to ZSTD_createCDict_advanced(), with dictLoadMethod==ZSTD_dlm_byRef */
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
+ZSTDLIB_STATIC_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
 
 /*! ZSTD_getCParams() :
  * @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
  * `estimatedSrcSize` value is optional, select 0 if not known */
-ZSTDLIB_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
 /*! ZSTD_getParams() :
  *  same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
  *  All fields of `ZSTD_frameParameters` are set to default : contentSize=1, checksum=0, noDictID=0 */
-ZSTDLIB_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+ZSTDLIB_STATIC_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
 /*! ZSTD_checkCParams() :
  *  Ensure param values remain within authorized range.
  * @return 0 on success, or an error code (can be checked with ZSTD_isError()) */
-ZSTDLIB_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
+ZSTDLIB_STATIC_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
 
 /*! ZSTD_adjustCParams() :
  *  optimize params for a given `srcSize` and `dictSize`.
@@ -1634,24 +1800,33 @@ ZSTDLIB_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
  * `dictSize` must be `0` when there is no dictionary.
  *  cPar can be invalid : all parameters will be clamped within valid range in the @return struct.
  *  This function never fails (wide contract) */
-ZSTDLIB_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
+ZSTDLIB_STATIC_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
+
+/*! ZSTD_CCtx_setCParams() :
+ *  Set all parameters provided within @cparams into the working @cctx.
+ *  Note : if modifying parameters during compression (MT mode only),
+ *         note that changes to the .windowLog parameter will be ignored.
+ * @return 0 on success, or an error code (can be checked with ZSTD_isError()) */
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setCParams(ZSTD_CCtx* cctx, ZSTD_compressionParameters cparams);
 
 /*! ZSTD_compress_advanced() :
  *  Note : this function is now DEPRECATED.
  *         It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_setParameter() and other parameter setters.
  *  This prototype will generate compilation warnings. */
 ZSTD_DEPRECATED("use ZSTD_compress2")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compress_advanced(ZSTD_CCtx* cctx,
-                                          void* dst, size_t dstCapacity,
-                                    const void* src, size_t srcSize,
-                                    const void* dict,size_t dictSize,
-                                          ZSTD_parameters params);
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                        const void* dict,size_t dictSize,
+                              ZSTD_parameters params);
 
 /*! ZSTD_compress_usingCDict_advanced() :
  *  Note : this function is now DEPRECATED.
  *         It can be replaced by ZSTD_compress2(), in combination with ZSTD_CCtx_loadDictionary() and other parameter setters.
  *  This prototype will generate compilation warnings. */
 ZSTD_DEPRECATED("use ZSTD_compress2 with ZSTD_CCtx_loadDictionary")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
                                               void* dst, size_t dstCapacity,
                                         const void* src, size_t srcSize,
@@ -1662,18 +1837,18 @@ size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
 /*! ZSTD_CCtx_loadDictionary_byReference() :
  *  Same as ZSTD_CCtx_loadDictionary(), but dictionary content is referenced, instead of being copied into CCtx.
  *  It saves some memory, but also requires that `dict` outlives its usage within `cctx` */
-ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
 
 /*! ZSTD_CCtx_loadDictionary_advanced() :
  *  Same as ZSTD_CCtx_loadDictionary(), but gives finer control over
  *  how to load the dictionary (by copy ? by reference ?)
  *  and how to interpret it (automatic ? force raw mode ? full mode only ?) */
-ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
 
 /*! ZSTD_CCtx_refPrefix_advanced() :
  *  Same as ZSTD_CCtx_refPrefix(), but gives finer control over
  *  how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) */
-ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
 
 /* ===   experimental parameters   === */
 /* these parameters can be used with ZSTD_setParameter()
@@ -1712,9 +1887,15 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  * See the comments on that enum for an explanation of the feature. */
 #define ZSTD_c_forceAttachDict ZSTD_c_experimentalParam4
 
-/* Controls how the literals are compressed (default is auto).
- * The value must be of type ZSTD_literalCompressionMode_e.
- * See ZSTD_literalCompressionMode_e enum definition for details.
+/* Controlled with ZSTD_paramSwitch_e enum.
+ * Default is ZSTD_ps_auto.
+ * Set to ZSTD_ps_disable to never compress literals.
+ * Set to ZSTD_ps_enable to always compress literals. (Note: uncompressed literals
+ * may still be emitted if huffman is not beneficial to use.)
+ *
+ * By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
+ * literals compression based on the compression parameters - specifically,
+ * negative compression levels do not use literal compression.
  */
 #define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
 
@@ -1777,7 +1958,7 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  *
  * Note that this means that the CDict tables can no longer be copied into the
  * CCtx, so the dict attachment mode ZSTD_dictForceCopy will no longer be
- * useable. The dictionary can only be attached or reloaded.
+ * usable. The dictionary can only be attached or reloaded.
  *
  * In general, you should expect compression to be faster--sometimes very much
  * so--and CDict creation to be slightly slower. Eventually, we will probably
@@ -1789,13 +1970,16 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  * Experimental parameter.
  * Default is 0 == disabled. Set to 1 to enable.
  *
- * Tells the compressor that the ZSTD_inBuffer will ALWAYS be the same
- * between calls, except for the modifications that zstd makes to pos (the
- * caller must not modify pos). This is checked by the compressor, and
- * compression will fail if it ever changes. This means the only flush
- * mode that makes sense is ZSTD_e_end, so zstd will error if ZSTD_e_end
- * is not used. The data in the ZSTD_inBuffer in the range [src, src + pos)
- * MUST not be modified during compression or you will get data corruption.
+ * Tells the compressor that input data presented with ZSTD_inBuffer
+ * will ALWAYS be the same between calls.
+ * Technically, the @src pointer must never be changed,
+ * and the @pos field can only be updated by zstd.
+ * However, it's possible to increase the @size field,
+ * allowing scenarios where more data can be appended after compressions starts.
+ * These conditions are checked by the compressor,
+ * and compression will fail if they are not respected.
+ * Also, data in the ZSTD_inBuffer within the range [src, src + pos)
+ * MUST not be modified during compression or it will result in data corruption.
  *
  * When this flag is enabled zstd won't allocate an input window buffer,
  * because the user guarantees it can reference the ZSTD_inBuffer until
@@ -1803,18 +1987,15 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  * large enough to fit a block (see ZSTD_c_stableOutBuffer). This will also
  * avoid the memcpy() from the input buffer to the input window buffer.
  *
- * NOTE: ZSTD_compressStream2() will error if ZSTD_e_end is not used.
- * That means this flag cannot be used with ZSTD_compressStream().
- *
  * NOTE: So long as the ZSTD_inBuffer always points to valid memory, using
  * this flag is ALWAYS memory safe, and will never access out-of-bounds
- * memory. However, compression WILL fail if you violate the preconditions.
+ * memory. However, compression WILL fail if conditions are not respected.
  *
- * WARNING: The data in the ZSTD_inBuffer in the range [dst, dst + pos) MUST
- * not be modified during compression or you will get data corruption. This
- * is because zstd needs to reference data in the ZSTD_inBuffer to find
+ * WARNING: The data in the ZSTD_inBuffer in the range [src, src + pos) MUST
+ * not be modified during compression or it will result in data corruption.
+ * This is because zstd needs to reference data in the ZSTD_inBuffer to find
  * matches. Normally zstd maintains its own window buffer for this purpose,
- * but passing this flag tells zstd to use the user provided buffer.
+ * but passing this flag tells zstd to rely on user provided buffer instead.
  */
 #define ZSTD_c_stableInBuffer ZSTD_c_experimentalParam9
 
@@ -1859,30 +2040,33 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  * Without validation, providing a sequence that does not conform to the zstd spec will cause
  * undefined behavior, and may produce a corrupted block.
  *
- * With validation enabled, a if sequence is invalid (see doc/zstd_compression_format.md for
+ * With validation enabled, if sequence is invalid (see doc/zstd_compression_format.md for
  * specifics regarding offset/matchlength requirements) then the function will bail out and
  * return an error.
  *
  */
 #define ZSTD_c_validateSequences ZSTD_c_experimentalParam12
 
-/* ZSTD_c_splitBlocks
- * Default is 0 == disabled. Set to 1 to enable block splitting.
+/* ZSTD_c_useBlockSplitter
+ * Controlled with ZSTD_paramSwitch_e enum.
+ * Default is ZSTD_ps_auto.
+ * Set to ZSTD_ps_disable to never use block splitter.
+ * Set to ZSTD_ps_enable to always use block splitter.
  *
- * Will attempt to split blocks in order to improve compression ratio at the cost of speed.
+ * By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
+ * block splitting based on the compression parameters.
  */
-#define ZSTD_c_splitBlocks ZSTD_c_experimentalParam13
+#define ZSTD_c_useBlockSplitter ZSTD_c_experimentalParam13
 
 /* ZSTD_c_useRowMatchFinder
- * Default is ZSTD_urm_auto.
- * Controlled with ZSTD_useRowMatchFinderMode_e enum.
- *
- * By default, in ZSTD_urm_auto, when finalizing the compression parameters, the library
- * will decide at runtime whether to use the row-based matchfinder based on support for SIMD
- * instructions as well as the windowLog.
+ * Controlled with ZSTD_paramSwitch_e enum.
+ * Default is ZSTD_ps_auto.
+ * Set to ZSTD_ps_disable to never use row-based matchfinder.
+ * Set to ZSTD_ps_enable to force usage of row-based matchfinder.
  *
- * Set to ZSTD_urm_disableRowMatchFinder to never use row-based matchfinder.
- * Set to ZSTD_urm_enableRowMatchFinder to force usage of row-based matchfinder.
+ * By default, in ZSTD_ps_auto, the library will decide at runtime whether to use
+ * the row-based matchfinder based on support for SIMD instructions and the window log.
+ * Note that this only pertains to compression strategies: greedy, lazy, and lazy2
  */
 #define ZSTD_c_useRowMatchFinder ZSTD_c_experimentalParam14
 
@@ -1906,12 +2090,85 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  */
 #define ZSTD_c_deterministicRefPrefix ZSTD_c_experimentalParam15
 
+/* ZSTD_c_prefetchCDictTables
+ * Controlled with ZSTD_paramSwitch_e enum. Default is ZSTD_ps_auto.
+ *
+ * In some situations, zstd uses CDict tables in-place rather than copying them
+ * into the working context. (See docs on ZSTD_dictAttachPref_e above for details).
+ * In such situations, compression speed is seriously impacted when CDict tables are
+ * "cold" (outside CPU cache). This parameter instructs zstd to prefetch CDict tables
+ * when they are used in-place.
+ *
+ * For sufficiently small inputs, the cost of the prefetch will outweigh the benefit.
+ * For sufficiently large inputs, zstd will by default memcpy() CDict tables
+ * into the working context, so there is no need to prefetch. This parameter is
+ * targeted at a middle range of input sizes, where a prefetch is cheap enough to be
+ * useful but memcpy() is too expensive. The exact range of input sizes where this
+ * makes sense is best determined by careful experimentation.
+ *
+ * Note: for this parameter, ZSTD_ps_auto is currently equivalent to ZSTD_ps_disable,
+ * but in the future zstd may conditionally enable this feature via an auto-detection
+ * heuristic for cold CDicts.
+ * Use ZSTD_ps_disable to opt out of prefetching under any circumstances.
+ */
+#define ZSTD_c_prefetchCDictTables ZSTD_c_experimentalParam16
+
+/* ZSTD_c_enableSeqProducerFallback
+ * Allowed values are 0 (disable) and 1 (enable). The default setting is 0.
+ *
+ * Controls whether zstd will fall back to an internal sequence producer if an
+ * external sequence producer is registered and returns an error code. This fallback
+ * is block-by-block: the internal sequence producer will only be called for blocks
+ * where the external sequence producer returns an error code. Fallback parsing will
+ * follow any other cParam settings, such as compression level, the same as in a
+ * normal (fully-internal) compression operation.
+ *
+ * The user is strongly encouraged to read the full Block-Level Sequence Producer API
+ * documentation (below) before setting this parameter. */
+#define ZSTD_c_enableSeqProducerFallback ZSTD_c_experimentalParam17
+
+/* ZSTD_c_maxBlockSize
+ * Allowed values are between 1KB and ZSTD_BLOCKSIZE_MAX (128KB).
+ * The default is ZSTD_BLOCKSIZE_MAX, and setting to 0 will set to the default.
+ *
+ * This parameter can be used to set an upper bound on the blocksize
+ * that overrides the default ZSTD_BLOCKSIZE_MAX. It cannot be used to set upper
+ * bounds greater than ZSTD_BLOCKSIZE_MAX or bounds lower than 1KB (will make
+ * compressBound() innacurate). Only currently meant to be used for testing.
+ *
+ */
+#define ZSTD_c_maxBlockSize ZSTD_c_experimentalParam18
+
+/* ZSTD_c_searchForExternalRepcodes
+ * This parameter affects how zstd parses external sequences, such as sequences
+ * provided through the compressSequences() API or from an external block-level
+ * sequence producer.
+ *
+ * If set to ZSTD_ps_enable, the library will check for repeated offsets in
+ * external sequences, even if those repcodes are not explicitly indicated in
+ * the "rep" field. Note that this is the only way to exploit repcode matches
+ * while using compressSequences() or an external sequence producer, since zstd
+ * currently ignores the "rep" field of external sequences.
+ *
+ * If set to ZSTD_ps_disable, the library will not exploit repeated offsets in
+ * external sequences, regardless of whether the "rep" field has been set. This
+ * reduces sequence compression overhead by about 25% while sacrificing some
+ * compression ratio.
+ *
+ * The default value is ZSTD_ps_auto, for which the library will enable/disable
+ * based on compression level.
+ *
+ * Note: for now, this param only has an effect if ZSTD_c_blockDelimiters is
+ * set to ZSTD_sf_explicitBlockDelimiters. That may change in the future.
+ */
+#define ZSTD_c_searchForExternalRepcodes ZSTD_c_experimentalParam19
+
 /*! ZSTD_CCtx_getParameter() :
  *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
  *  and store it into int* value.
  * @return : 0, or an error code (which can be tested with ZSTD_isError()).
  */
-ZSTDLIB_API size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value);
 
 
 /*! ZSTD_CCtx_params :
@@ -1931,25 +2188,25 @@ ZSTDLIB_API size_t ZSTD_CCtx_getParameter(const ZSTD_CCtx* cctx, ZSTD_cParameter
  *  This can be used with ZSTD_estimateCCtxSize_advanced_usingCCtxParams()
  *  for static allocation of CCtx for single-threaded compression.
  */
-ZSTDLIB_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
-ZSTDLIB_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);  /* accept NULL pointer */
+ZSTDLIB_STATIC_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
+ZSTDLIB_STATIC_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);  /* accept NULL pointer */
 
 /*! ZSTD_CCtxParams_reset() :
  *  Reset params to default values.
  */
-ZSTDLIB_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
 
 /*! ZSTD_CCtxParams_init() :
  *  Initializes the compression parameters of cctxParams according to
  *  compression level. All other parameters are reset to their default values.
  */
-ZSTDLIB_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
 
 /*! ZSTD_CCtxParams_init_advanced() :
  *  Initializes the compression and frame parameters of cctxParams according to
  *  params. All other parameters are reset to their default values.
  */
-ZSTDLIB_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
 
 /*! ZSTD_CCtxParams_setParameter() : Requires v1.4.0+
  *  Similar to ZSTD_CCtx_setParameter.
@@ -1959,14 +2216,14 @@ ZSTDLIB_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, Z
  * @result : a code representing success or failure (which can be tested with
  *           ZSTD_isError()).
  */
-ZSTDLIB_API size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, int value);
 
 /*! ZSTD_CCtxParams_getParameter() :
  * Similar to ZSTD_CCtx_getParameter.
  * Get the requested value of one compression parameter, selected by enum ZSTD_cParameter.
  * @result : 0, or an error code (which can be tested with ZSTD_isError()).
  */
-ZSTDLIB_API size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params, ZSTD_cParameter param, int* value);
+ZSTDLIB_STATIC_API size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params, ZSTD_cParameter param, int* value);
 
 /*! ZSTD_CCtx_setParametersUsingCCtxParams() :
  *  Apply a set of ZSTD_CCtx_params to the compression context.
@@ -1975,7 +2232,7 @@ ZSTDLIB_API size_t ZSTD_CCtxParams_getParameter(const ZSTD_CCtx_params* params,
  *    if nbWorkers>=1, new parameters will be picked up at next job,
  *       with a few restrictions (windowLog, pledgedSrcSize, nbWorkers, jobSize, and overlapLog are not updated).
  */
-ZSTDLIB_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+ZSTDLIB_STATIC_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
         ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
 
 /*! ZSTD_compressStream2_simpleArgs() :
@@ -1984,7 +2241,7 @@ ZSTDLIB_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
  *  This variant might be helpful for binders from dynamic languages
  *  which have troubles handling structures containing memory pointers.
  */
-ZSTDLIB_API size_t ZSTD_compressStream2_simpleArgs (
+ZSTDLIB_STATIC_API size_t ZSTD_compressStream2_simpleArgs (
                             ZSTD_CCtx* cctx,
                             void* dst, size_t dstCapacity, size_t* dstPos,
                       const void* src, size_t srcSize, size_t* srcPos,
@@ -2000,33 +2257,33 @@ ZSTDLIB_API size_t ZSTD_compressStream2_simpleArgs (
  *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
  *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
  *  Note 3 : Skippable Frame Identifiers are considered valid. */
-ZSTDLIB_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
+ZSTDLIB_STATIC_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
 
 /*! ZSTD_createDDict_byReference() :
  *  Create a digested dictionary, ready to start decompression operation without startup delay.
  *  Dictionary content is referenced, and therefore stays in dictBuffer.
  *  It is important that dictBuffer outlives DDict,
  *  it must remain read accessible throughout the lifetime of DDict */
-ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
+ZSTDLIB_STATIC_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
 
 /*! ZSTD_DCtx_loadDictionary_byReference() :
  *  Same as ZSTD_DCtx_loadDictionary(),
  *  but references `dict` content instead of copying it into `dctx`.
  *  This saves memory if `dict` remains around.,
  *  However, it's imperative that `dict` remains accessible (and unmodified) while being used, so it must outlive decompression. */
-ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
 
 /*! ZSTD_DCtx_loadDictionary_advanced() :
  *  Same as ZSTD_DCtx_loadDictionary(),
  *  but gives direct control over
  *  how to load the dictionary (by copy ? by reference ?)
  *  and how to interpret it (automatic ? force raw mode ? full mode only ?). */
-ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
 
 /*! ZSTD_DCtx_refPrefix_advanced() :
  *  Same as ZSTD_DCtx_refPrefix(), but gives finer control over
  *  how to interpret prefix content (automatic ? force raw mode (default) ? full mode only ?) */
-ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
 
 /*! ZSTD_DCtx_setMaxWindowSize() :
  *  Refuses allocating internal buffers for frames requiring a window size larger than provided limit.
@@ -2035,14 +2292,14 @@ ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* pre
  *  By default, a decompression context accepts all window sizes <= (1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT)
  * @return : 0, or an error code (which can be tested using ZSTD_isError()).
  */
-ZSTDLIB_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
 
 /*! ZSTD_DCtx_getParameter() :
  *  Get the requested decompression parameter value, selected by enum ZSTD_dParameter,
  *  and store it into int* value.
  * @return : 0, or an error code (which can be tested with ZSTD_isError()).
  */
-ZSTDLIB_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
+ZSTDLIB_STATIC_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param, int* value);
 
 /* ZSTD_d_format
  * experimental parameter,
@@ -2062,7 +2319,7 @@ ZSTDLIB_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param
  * in the range [dst, dst + pos) MUST not be modified during decompression
  * or you will get data corruption.
  *
- * When this flags is enabled zstd won't allocate an output buffer, because
+ * When this flag is enabled zstd won't allocate an output buffer, because
  * it can write directly to the ZSTD_outBuffer, but it will still allocate
  * an input buffer large enough to fit any compressed block. This will also
  * avoid the memcpy() from the internal output buffer to the ZSTD_outBuffer.
@@ -2115,6 +2372,17 @@ ZSTDLIB_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param
  */
 #define ZSTD_d_refMultipleDDicts ZSTD_d_experimentalParam4
 
+/* ZSTD_d_disableHuffmanAssembly
+ * Set to 1 to disable the Huffman assembly implementation.
+ * The default value is 0, which allows zstd to use the Huffman assembly
+ * implementation if available.
+ *
+ * This parameter can be used to disable Huffman assembly at runtime.
+ * If you want to disable it at compile time you can define the macro
+ * ZSTD_DISABLE_ASM.
+ */
+#define ZSTD_d_disableHuffmanAssembly ZSTD_d_experimentalParam5
+
 
 /*! ZSTD_DCtx_setFormat() :
  *  This function is REDUNDANT. Prefer ZSTD_DCtx_setParameter().
@@ -2123,6 +2391,7 @@ ZSTDLIB_API size_t ZSTD_DCtx_getParameter(ZSTD_DCtx* dctx, ZSTD_dParameter param
  *  such ZSTD_f_zstd1_magicless for example.
  * @return : 0, or an error code (which can be tested using ZSTD_isError()). */
 ZSTD_DEPRECATED("use ZSTD_DCtx_setParameter() instead")
+ZSTDLIB_STATIC_API
 size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
 
 /*! ZSTD_decompressStream_simpleArgs() :
@@ -2131,7 +2400,7 @@ size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
  *  This can be helpful for binders from dynamic languages
  *  which have troubles handling structures containing memory pointers.
  */
-ZSTDLIB_API size_t ZSTD_decompressStream_simpleArgs (
+ZSTDLIB_STATIC_API size_t ZSTD_decompressStream_simpleArgs (
                             ZSTD_DCtx* dctx,
                             void* dst, size_t dstCapacity, size_t* dstPos,
                       const void* src, size_t srcSize, size_t* srcPos);
@@ -2159,6 +2428,7 @@ ZSTDLIB_API size_t ZSTD_decompressStream_simpleArgs (
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
                          int compressionLevel,
                          unsigned long long pledgedSrcSize);
@@ -2176,6 +2446,7 @@ size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs,
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
                      const void* dict, size_t dictSize,
                            int compressionLevel);
@@ -2196,6 +2467,7 @@ size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs,
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
                     const void* dict, size_t dictSize,
                           ZSTD_parameters params,
@@ -2205,11 +2477,12 @@ size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
  * This function is DEPRECATED, and equivalent to:
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
  *     ZSTD_CCtx_refCDict(zcs, cdict);
- * 
+ *
  * note : cdict will just be referenced, and must outlive compression session
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
 
 /*! ZSTD_initCStream_usingCDict_advanced() :
@@ -2228,6 +2501,7 @@ size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);
  * This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset and ZSTD_CCtx_refCDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
                                const ZSTD_CDict* cdict,
                                      ZSTD_frameParameters fParams,
@@ -2252,6 +2526,7 @@ size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
  *  This prototype will generate compilation warnings.
  */
 ZSTD_DEPRECATED("use ZSTD_CCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API
 size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
 
 
@@ -2270,7 +2545,7 @@ typedef struct {
  * Note : (ingested - consumed) is amount of input data buffered internally, not yet compressed.
  * Aggregates progression inside active worker threads.
  */
-ZSTDLIB_API ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
+ZSTDLIB_STATIC_API ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
 
 /*! ZSTD_toFlushNow() :
  *  Tell how many bytes are ready to be flushed immediately.
@@ -2285,7 +2560,7 @@ ZSTDLIB_API ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx
  *    therefore flush speed is limited by production speed of oldest job
  *    irrespective of the speed of concurrent (and newer) jobs.
  */
-ZSTDLIB_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
+ZSTDLIB_STATIC_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
 
 
 /*=====   Advanced Streaming decompression functions  =====*/
@@ -2297,9 +2572,9 @@ ZSTDLIB_API size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx);
  *     ZSTD_DCtx_loadDictionary(zds, dict, dictSize);
  *
  * note: no dictionary will be used if dict == NULL or dictSize < 8
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  */
-ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_loadDictionary, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
 
 /*!
  * This function is deprecated, and is equivalent to:
@@ -2308,9 +2583,9 @@ ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dic
  *     ZSTD_DCtx_refDDict(zds, ddict);
  *
  * note : ddict is referenced, it must outlive decompression session
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  */
-ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset + ZSTD_DCtx_refDDict, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);
 
 /*!
  * This function is deprecated, and is equivalent to:
@@ -2318,9 +2593,9 @@ ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDi
  *     ZSTD_DCtx_reset(zds, ZSTD_reset_session_only);
  *
  * re-use decompression parameters from previous init; saves dictionary loading
- * Note : this prototype will be marked as deprecated and generate compilation warnings on reaching v1.5.x
  */
-ZSTDLIB_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
+ZSTD_DEPRECATED("use ZSTD_DCtx_reset, see zstd.h for detailed instructions")
+ZSTDLIB_STATIC_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
 
 
 /*********************************************************************
@@ -2340,7 +2615,6 @@ ZSTDLIB_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
 
   Start by initializing a context.
   Use ZSTD_compressBegin(), or ZSTD_compressBegin_usingDict() for dictionary compression.
-  It's also possible to duplicate a reference context which has already been initialized, using ZSTD_copyCCtx()
 
   Then, consume your input using ZSTD_compressContinue().
   There are some important considerations to keep in mind when using this advanced function :
@@ -2362,18 +2636,23 @@ ZSTDLIB_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);
 */
 
 /*=====   Buffer-less streaming compression functions  =====*/
-ZSTDLIB_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
-ZSTDLIB_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+ZSTDLIB_STATIC_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
 
-ZSTDLIB_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-ZSTDLIB_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTD_DEPRECATED("This function will likely be removed in a future release. It is misleading and has very limited utility.")
+ZSTDLIB_STATIC_API
+size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+
+ZSTDLIB_STATIC_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
 /* The ZSTD_compressBegin_advanced() and ZSTD_compressBegin_usingCDict_advanced() are now DEPRECATED and will generate a compiler warning */
 ZSTD_DEPRECATED("use advanced API to access custom parameters")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
 ZSTD_DEPRECATED("use advanced API to access custom parameters")
+ZSTDLIB_STATIC_API
 size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   /* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */
 /**
   Buffer-less streaming decompression (synchronous mode)
@@ -2386,8 +2665,8 @@ size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_
   Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
   Data fragment must be large enough to ensure successful decoding.
  `ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
-  @result : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
-           >0 : `srcSize` is too small, please provide at least @result bytes on next attempt.
+  result  : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
+           >0 : `srcSize` is too small, please provide at least result bytes on next attempt.
            errorCode, which can be tested using ZSTD_isError().
 
   It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
@@ -2406,7 +2685,7 @@ size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_
 
   The most memory efficient way is to use a round buffer of sufficient size.
   Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
-  which can @return an error code if required value is too large for current system (in 32-bits mode).
+  which can return an error code if required value is too large for current system (in 32-bits mode).
   In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
   up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
   which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
@@ -2426,7 +2705,7 @@ size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_
   ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
   ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
 
- @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
   It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
   It can also be an error code, which can be tested with ZSTD_isError().
 
@@ -2458,6 +2737,8 @@ typedef struct {
     unsigned headerSize;
     unsigned dictID;
     unsigned checksumFlag;
+    unsigned _reserved1;
+    unsigned _reserved2;
 } ZSTD_frameHeader;
 
 /*! ZSTD_getFrameHeader() :
@@ -2465,24 +2746,25 @@ typedef struct {
  * @return : 0, `zfhPtr` is correctly filled,
  *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
  *           or an error code, which can be tested using ZSTD_isError() */
-ZSTDLIB_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
 /*! ZSTD_getFrameHeader_advanced() :
  *  same as ZSTD_getFrameHeader(),
  *  with added capability to select a format (like ZSTD_f_zstd1_magicless) */
-ZSTDLIB_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
-ZSTDLIB_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_STATIC_API size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format);
+ZSTDLIB_STATIC_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
 
-ZSTDLIB_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
-ZSTDLIB_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
-ZSTDLIB_API size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
 
-ZSTDLIB_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
-ZSTDLIB_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
 /* misc */
-ZSTDLIB_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
+ZSTD_DEPRECATED("This function will likely be removed in the next minor release. It is misleading and has very limited utility.")
+ZSTDLIB_STATIC_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
 typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
-ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
+ZSTDLIB_STATIC_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 
 
 
@@ -2502,7 +2784,6 @@ ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
     - It is necessary to init context before starting
       + compression : any ZSTD_compressBegin*() variant, including with dictionary
       + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
-      + copyCCtx() and copyDCtx() can be used too
     - Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
       + If input is larger than a block size, it's necessary to split input data into multiple blocks
       + For inputs larger than a single block, consider using regular ZSTD_compress() instead.
@@ -2519,11 +2800,172 @@ ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 */
 
 /*=====   Raw zstd block functions  =====*/
-ZSTDLIB_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
-ZSTDLIB_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-ZSTDLIB_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-ZSTDLIB_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
-
+ZSTDLIB_STATIC_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
+ZSTDLIB_STATIC_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_STATIC_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
+
+
+/* ********************* BLOCK-LEVEL SEQUENCE PRODUCER API *********************
+ *
+ * *** OVERVIEW ***
+ * The Block-Level Sequence Producer API allows users to provide their own custom
+ * sequence producer which libzstd invokes to process each block. The produced list
+ * of sequences (literals and matches) is then post-processed by libzstd to produce
+ * valid compressed blocks.
+ *
+ * This block-level offload API is a more granular complement of the existing
+ * frame-level offload API compressSequences() (introduced in v1.5.1). It offers
+ * an easier migration story for applications already integrated with libzstd: the
+ * user application continues to invoke the same compression functions
+ * ZSTD_compress2() or ZSTD_compressStream2() as usual, and transparently benefits
+ * from the specific advantages of the external sequence producer. For example,
+ * the sequence producer could be tuned to take advantage of known characteristics
+ * of the input, to offer better speed / ratio, or could leverage hardware
+ * acceleration not available within libzstd itself.
+ *
+ * See contrib/externalSequenceProducer for an example program employing the
+ * Block-Level Sequence Producer API.
+ *
+ * *** USAGE ***
+ * The user is responsible for implementing a function of type
+ * ZSTD_sequenceProducer_F. For each block, zstd will pass the following
+ * arguments to the user-provided function:
+ *
+ *   - sequenceProducerState: a pointer to a user-managed state for the sequence
+ *     producer.
+ *
+ *   - outSeqs, outSeqsCapacity: an output buffer for the sequence producer.
+ *     outSeqsCapacity is guaranteed >= ZSTD_sequenceBound(srcSize). The memory
+ *     backing outSeqs is managed by the CCtx.
+ *
+ *   - src, srcSize: an input buffer for the sequence producer to parse.
+ *     srcSize is guaranteed to be <= ZSTD_BLOCKSIZE_MAX.
+ *
+ *   - dict, dictSize: a history buffer, which may be empty, which the sequence
+ *     producer may reference as it parses the src buffer. Currently, zstd will
+ *     always pass dictSize == 0 into external sequence producers, but this will
+ *     change in the future.
+ *
+ *   - compressionLevel: a signed integer representing the zstd compression level
+ *     set by the user for the current operation. The sequence producer may choose
+ *     to use this information to change its compression strategy and speed/ratio
+ *     tradeoff. Note: the compression level does not reflect zstd parameters set
+ *     through the advanced API.
+ *
+ *   - windowSize: a size_t representing the maximum allowed offset for external
+ *     sequences. Note that sequence offsets are sometimes allowed to exceed the
+ *     windowSize if a dictionary is present, see doc/zstd_compression_format.md
+ *     for details.
+ *
+ * The user-provided function shall return a size_t representing the number of
+ * sequences written to outSeqs. This return value will be treated as an error
+ * code if it is greater than outSeqsCapacity. The return value must be non-zero
+ * if srcSize is non-zero. The ZSTD_SEQUENCE_PRODUCER_ERROR macro is provided
+ * for convenience, but any value greater than outSeqsCapacity will be treated as
+ * an error code.
+ *
+ * If the user-provided function does not return an error code, the sequences
+ * written to outSeqs must be a valid parse of the src buffer. Data corruption may
+ * occur if the parse is not valid. A parse is defined to be valid if the
+ * following conditions hold:
+ *   - The sum of matchLengths and literalLengths must equal srcSize.
+ *   - All sequences in the parse, except for the final sequence, must have
+ *     matchLength >= ZSTD_MINMATCH_MIN. The final sequence must have
+ *     matchLength >= ZSTD_MINMATCH_MIN or matchLength == 0.
+ *   - All offsets must respect the windowSize parameter as specified in
+ *     doc/zstd_compression_format.md.
+ *   - If the final sequence has matchLength == 0, it must also have offset == 0.
+ *
+ * zstd will only validate these conditions (and fail compression if they do not
+ * hold) if the ZSTD_c_validateSequences cParam is enabled. Note that sequence
+ * validation has a performance cost.
+ *
+ * If the user-provided function returns an error, zstd will either fall back
+ * to an internal sequence producer or fail the compression operation. The user can
+ * choose between the two behaviors by setting the ZSTD_c_enableSeqProducerFallback
+ * cParam. Fallback compression will follow any other cParam settings, such as
+ * compression level, the same as in a normal compression operation.
+ *
+ * The user shall instruct zstd to use a particular ZSTD_sequenceProducer_F
+ * function by calling
+ *         ZSTD_registerSequenceProducer(cctx,
+ *                                       sequenceProducerState,
+ *                                       sequenceProducer)
+ * This setting will persist until the next parameter reset of the CCtx.
+ *
+ * The sequenceProducerState must be initialized by the user before calling
+ * ZSTD_registerSequenceProducer(). The user is responsible for destroying the
+ * sequenceProducerState.
+ *
+ * *** LIMITATIONS ***
+ * This API is compatible with all zstd compression APIs which respect advanced parameters.
+ * However, there are three limitations:
+ *
+ * First, the ZSTD_c_enableLongDistanceMatching cParam is not currently supported.
+ * COMPRESSION WILL FAIL if it is enabled and the user tries to compress with a block-level
+ * external sequence producer.
+ *   - Note that ZSTD_c_enableLongDistanceMatching is auto-enabled by default in some
+ *     cases (see its documentation for details). Users must explicitly set
+ *     ZSTD_c_enableLongDistanceMatching to ZSTD_ps_disable in such cases if an external
+ *     sequence producer is registered.
+ *   - As of this writing, ZSTD_c_enableLongDistanceMatching is disabled by default
+ *     whenever ZSTD_c_windowLog < 128MB, but that's subject to change. Users should
+ *     check the docs on ZSTD_c_enableLongDistanceMatching whenever the Block-Level Sequence
+ *     Producer API is used in conjunction with advanced settings (like ZSTD_c_windowLog).
+ *
+ * Second, history buffers are not currently supported. Concretely, zstd will always pass
+ * dictSize == 0 to the external sequence producer (for now). This has two implications:
+ *   - Dictionaries are not currently supported. Compression will *not* fail if the user
+ *     references a dictionary, but the dictionary won't have any effect.
+ *   - Stream history is not currently supported. All advanced compression APIs, including
+ *     streaming APIs, work with external sequence producers, but each block is treated as
+ *     an independent chunk without history from previous blocks.
+ *
+ * Third, multi-threading within a single compression is not currently supported. In other words,
+ * COMPRESSION WILL FAIL if ZSTD_c_nbWorkers > 0 and an external sequence producer is registered.
+ * Multi-threading across compressions is fine: simply create one CCtx per thread.
+ *
+ * Long-term, we plan to overcome all three limitations. There is no technical blocker to
+ * overcoming them. It is purely a question of engineering effort.
+ */
+
+#define ZSTD_SEQUENCE_PRODUCER_ERROR ((size_t)(-1))
+
+typedef size_t ZSTD_sequenceProducer_F (
+  void* sequenceProducerState,
+  ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
+  const void* src, size_t srcSize,
+  const void* dict, size_t dictSize,
+  int compressionLevel,
+  size_t windowSize
+);
+
+/*! ZSTD_registerSequenceProducer() :
+ * Instruct zstd to use a block-level external sequence producer function.
+ *
+ * The sequenceProducerState must be initialized by the caller, and the caller is
+ * responsible for managing its lifetime. This parameter is sticky across
+ * compressions. It will remain set until the user explicitly resets compression
+ * parameters.
+ *
+ * Sequence producer registration is considered to be an "advanced parameter",
+ * part of the "advanced API". This means it will only have an effect on compression
+ * APIs which respect advanced parameters, such as compress2() and compressStream2().
+ * Older compression APIs such as compressCCtx(), which predate the introduction of
+ * "advanced parameters", will ignore any external sequence producer setting.
+ *
+ * The sequence producer can be "cleared" by registering a NULL function pointer. This
+ * removes all limitations described above in the "LIMITATIONS" section of the API docs.
+ *
+ * The user is strongly encouraged to read the full API documentation (above) before
+ * calling this function. */
+ZSTDLIB_STATIC_API void
+ZSTD_registerSequenceProducer(
+  ZSTD_CCtx* cctx,
+  void* sequenceProducerState,
+  ZSTD_sequenceProducer_F* sequenceProducer
+);
 
 #endif   /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
 
diff --git a/priv/zstd/lib/zstd_errors.h b/priv/zstd/lib/zstd_errors.h
index fa3686b..dc75eee 100644
--- a/priv/zstd/lib/zstd_errors.h
+++ b/priv/zstd/lib/zstd_errors.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -20,19 +20,31 @@ extern "C" {
 
 
 /* =====   ZSTDERRORLIB_API : control library symbols visibility   ===== */
-#ifndef ZSTDERRORLIB_VISIBILITY
-#  if defined(__GNUC__) && (__GNUC__ >= 4)
-#    define ZSTDERRORLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#ifndef ZSTDERRORLIB_VISIBLE
+   /* Backwards compatibility with old macro name */
+#  ifdef ZSTDERRORLIB_VISIBILITY
+#    define ZSTDERRORLIB_VISIBLE ZSTDERRORLIB_VISIBILITY
+#  elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_VISIBLE __attribute__ ((visibility ("default")))
 #  else
-#    define ZSTDERRORLIB_VISIBILITY
+#    define ZSTDERRORLIB_VISIBLE
 #  endif
 #endif
+
+#ifndef ZSTDERRORLIB_HIDDEN
+#  if defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__MINGW32__)
+#    define ZSTDERRORLIB_HIDDEN __attribute__ ((visibility ("hidden")))
+#  else
+#    define ZSTDERRORLIB_HIDDEN
+#  endif
+#endif
+
 #if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBILITY
+#  define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBLE
 #elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
-#  define ZSTDERRORLIB_API __declspec(dllimport) ZSTDERRORLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#  define ZSTDERRORLIB_API __declspec(dllimport) ZSTDERRORLIB_VISIBLE /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
 #else
-#  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBILITY
+#  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBLE
 #endif
 
 /*-*********************************************
@@ -58,14 +70,17 @@ typedef enum {
   ZSTD_error_frameParameter_windowTooLarge = 16,
   ZSTD_error_corruption_detected = 20,
   ZSTD_error_checksum_wrong      = 22,
+  ZSTD_error_literals_headerWrong = 24,
   ZSTD_error_dictionary_corrupted      = 30,
   ZSTD_error_dictionary_wrong          = 32,
   ZSTD_error_dictionaryCreation_failed = 34,
   ZSTD_error_parameter_unsupported   = 40,
+  ZSTD_error_parameter_combination_unsupported = 41,
   ZSTD_error_parameter_outOfBound    = 42,
   ZSTD_error_tableLog_tooLarge       = 44,
   ZSTD_error_maxSymbolValue_tooLarge = 46,
   ZSTD_error_maxSymbolValue_tooSmall = 48,
+  ZSTD_error_stabilityCondition_notRespected = 50,
   ZSTD_error_stage_wrong       = 60,
   ZSTD_error_init_missing      = 62,
   ZSTD_error_memory_allocation = 64,
@@ -73,11 +88,15 @@ typedef enum {
   ZSTD_error_dstSize_tooSmall = 70,
   ZSTD_error_srcSize_wrong    = 72,
   ZSTD_error_dstBuffer_null   = 74,
+  ZSTD_error_noForwardProgress_destFull = 80,
+  ZSTD_error_noForwardProgress_inputEmpty = 82,
   /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
   ZSTD_error_frameIndex_tooLarge = 100,
   ZSTD_error_seekableIO          = 102,
   ZSTD_error_dstBuffer_wrong     = 104,
   ZSTD_error_srcBuffer_wrong     = 105,
+  ZSTD_error_sequenceProducer_failed = 106,
+  ZSTD_error_externalSequences_invalid = 107,
   ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
 } ZSTD_ErrorCode;
 
diff --git a/priv/zstd/programs/Makefile b/priv/zstd/programs/Makefile
index 599fb02..fcff41d 100644
--- a/priv/zstd/programs/Makefile
+++ b/priv/zstd/programs/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -9,7 +9,7 @@
 # ##########################################################################
 # zstd : Command Line Utility, supporting gzip-like arguments
 # zstd32 : Same as zstd, but forced to compile in 32-bits mode
-# zstd_nolegacy : zstd without support of decompression of legacy versions
+# zstd-nolegacy : zstd without support of decompression of legacy versions
 # zstd-small : minimal zstd without dictionary builder and benchmark
 # zstd-compress : compressor-only version of zstd
 # zstd-decompress : decompressor-only version of zstd
@@ -18,31 +18,9 @@
 .PHONY: default
 default: zstd-release
 
-# silent mode by default; verbose can be triggered by V=1 or VERBOSE=1
-$(V)$(VERBOSE).SILENT:
+LIBZSTD := ../lib
 
-
-ZSTDDIR := ../lib
-
-# Version numbers
-LIBVER_SRC := $(ZSTDDIR)/zstd.h
-LIBVER_MAJOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MAJOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
-LIBVER_MINOR_SCRIPT:=`sed -n '/define ZSTD_VERSION_MINOR/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
-LIBVER_PATCH_SCRIPT:=`sed -n '/define ZSTD_VERSION_RELEASE/s/.*[[:blank:]]\([0-9][0-9]*\).*/\1/p' < $(LIBVER_SRC)`
-LIBVER_SCRIPT:= $(LIBVER_MAJOR_SCRIPT).$(LIBVER_MINOR_SCRIPT).$(LIBVER_PATCH_SCRIPT)
-LIBVER_MAJOR := $(shell echo $(LIBVER_MAJOR_SCRIPT))
-LIBVER_MINOR := $(shell echo $(LIBVER_MINOR_SCRIPT))
-LIBVER_PATCH := $(shell echo $(LIBVER_PATCH_SCRIPT))
-LIBVER  := $(shell echo $(LIBVER_SCRIPT))
-
-ZSTD_VERSION = $(LIBVER)
-
-HAVE_COLORNEVER = $(shell echo a | grep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
-GREP_OPTIONS ?=
-ifeq ($HAVE_COLORNEVER, 1)
-  GREP_OPTIONS += --color=never
-endif
-GREP = grep $(GREP_OPTIONS)
+include $(LIBZSTD)/libzstd.mk
 
 ifeq ($(shell $(CC) -v 2>&1 | $(GREP) -c "gcc version "), 1)
   ALIGN_LOOP = -falign-loops=32
@@ -50,78 +28,25 @@ else
   ALIGN_LOOP =
 endif
 
-DEBUGLEVEL ?= 0
-CPPFLAGS += -DXXH_NAMESPACE=ZSTD_ -DDEBUGLEVEL=$(DEBUGLEVEL)
-ifeq ($(OS),Windows_NT)   # MinGW assumed
-  CPPFLAGS += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
-endif
-CFLAGS   ?= -O3
-DEBUGFLAGS+=-Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
-            -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
-            -Wstrict-prototypes -Wundef -Wpointer-arith \
-            -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
-            -Wredundant-decls -Wmissing-prototypes -Wc++-compat
-CFLAGS   += $(DEBUGFLAGS)
-CPPFLAGS += $(MOREFLAGS)
-LDFLAGS  += $(MOREFLAGS)
-FLAGS     = $(CFLAGS) $(CPPFLAGS) $(LDFLAGS)
-
-ZSTDLIB_COMMON := $(ZSTDDIR)/common
-ZSTDLIB_COMPRESS := $(ZSTDDIR)/compress
-ZSTDLIB_DECOMPRESS := $(ZSTDDIR)/decompress
-ZDICT_DIR := $(ZSTDDIR)/dictBuilder
-ZSTDLEGACY_DIR := $(ZSTDDIR)/legacy
-
-vpath %.c $(ZSTDLIB_COMMON) $(ZSTDLIB_COMPRESS) $(ZSTDLIB_DECOMPRESS) $(ZDICT_DIR) $(ZSTDLEGACY_DIR)
-
-ZSTDLIB_COMMON_C := $(wildcard $(ZSTDLIB_COMMON)/*.c)
-ZSTDLIB_COMPRESS_C := $(wildcard $(ZSTDLIB_COMPRESS)/*.c)
-ZSTDLIB_DECOMPRESS_C := $(wildcard $(ZSTDLIB_DECOMPRESS)/*.c)
-ZSTDLIB_CORE_SRC := $(ZSTDLIB_DECOMPRESS_C) $(ZSTDLIB_COMMON_C) $(ZSTDLIB_COMPRESS_C)
-ZDICT_SRC := $(wildcard $(ZDICT_DIR)/*.c)
-
-ZSTD_LEGACY_SUPPORT ?= 5
-ZSTDLEGACY_SRC :=
-ifneq ($(ZSTD_LEGACY_SUPPORT), 0)
-ifeq ($(shell test $(ZSTD_LEGACY_SUPPORT) -lt 8; echo $$?), 0)
-  ZSTDLEGACY_SRC += $(shell ls $(ZSTDLEGACY_DIR)/*.c | $(GREP) 'v0[$(ZSTD_LEGACY_SUPPORT)-7]')
-endif
-endif
+ZSTDLIB_COMMON_SRC := $(sort $(ZSTD_COMMON_FILES))
+ZSTDLIB_COMPRESS_SRC := $(sort $(ZSTD_COMPRESS_FILES))
+ZSTDLIB_DECOMPRESS_SRC := $(sort $(ZSTD_DECOMPRESS_FILES))
+ZSTDLIB_CORE_SRC := $(sort $(ZSTD_DECOMPRESS_FILES) $(ZSTD_COMMON_FILES) $(ZSTD_COMPRESS_FILES))
+ZDICT_SRC := $(sort $(ZSTD_DICTBUILDER_FILES))
+ZSTDLEGACY_SRC := $(sort $(ZSTD_LEGACY_FILES))
 
 # Sort files in alphabetical order for reproducible builds
 ZSTDLIB_FULL_SRC = $(sort $(ZSTDLIB_CORE_SRC) $(ZSTDLEGACY_SRC) $(ZDICT_SRC))
 ZSTDLIB_LOCAL_SRC = $(notdir $(ZSTDLIB_FULL_SRC))
-ZSTDLIB_LOCAL_OBJ := $(ZSTDLIB_LOCAL_SRC:.c=.o)
+ZSTDLIB_LOCAL_OBJ0 := $(ZSTDLIB_LOCAL_SRC:.c=.o)
+ZSTDLIB_LOCAL_OBJ := $(ZSTDLIB_LOCAL_OBJ0:.S=.o)
 
-ZSTD_CLI_SRC := $(wildcard *.c)
+ZSTD_CLI_SRC := $(sort $(wildcard *.c))
 ZSTD_CLI_OBJ := $(ZSTD_CLI_SRC:.c=.o)
 
 ZSTD_ALL_SRC = $(ZSTDLIB_LOCAL_SRC) $(ZSTD_CLI_SRC)
-ZSTD_ALL_OBJ := $(ZSTD_ALL_SRC:.c=.o)
-
-UNAME := $(shell uname)
-
-ifndef BUILD_DIR
-ifeq ($(UNAME), Darwin)
-  ifeq ($(shell md5 < /dev/null > /dev/null; echo $$?), 0)
-    HASH ?= md5
-  endif
-else ifeq ($(UNAME), FreeBSD)
-  HASH ?= gmd5sum
-else ifeq ($(UNAME), NetBSD)
-  HASH ?= md5 -n
-else ifeq ($(UNAME), OpenBSD)
-  HASH ?= md5
-endif
-HASH ?= md5sum
-HAVE_HASH :=$(shell echo 1 | $(HASH) > /dev/null && echo 1 || echo 0)
-
-HASH_DIR = conf_$(shell echo $(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $(LDLIBS) $(ZSTD_FILES) | $(HASH) | cut -f 1 -d " ")
-ifeq ($(HAVE_HASH),0)
-  $(info warning : could not find HASH ($(HASH)), needed to differentiate builds using different flags)
-  BUILD_DIR := obj/generic_noconf
-endif
-endif # BUILD_DIR
+ZSTD_ALL_OBJ0 := $(ZSTD_ALL_SRC:.c=.o)
+ZSTD_ALL_OBJ := $(ZSTD_ALL_OBJ0:.S=.o)
 
 # Define *.exe as extension for Windows systems
 ifneq (,$(filter Windows%,$(OS)))
@@ -137,11 +62,6 @@ else
   EXT =
 endif
 
-VOID = /dev/null
-
-# Make 4.3 doesn't support '\#' anymore (https://lwn.net/Articles/810071/)
-NUM_SYMBOL := \#
-
 # thread detection
 NO_THREAD_MSG := ==> no threads, building without multithreading support
 HAVE_PTHREAD := $(shell printf '$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_pthread.c && $(CC) $(FLAGS) -o have_pthread$(EXT) have_pthread.c -pthread 2> $(VOID) && rm have_pthread$(EXT) && echo 1 || echo 0; rm have_pthread.c)
@@ -156,7 +76,7 @@ endif
 
 # zlib detection
 NO_ZLIB_MSG := ==> no zlib, building zstd without .gz support
-HAVE_ZLIB := $(shell printf '$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_zlib.c && $(CC) $(FLAGS) -o have_zlib$(EXT) have_zlib.c -lz 2> $(VOID) && rm have_zlib$(EXT) && echo 1 || echo 0; rm have_zlib.c)
+HAVE_ZLIB ?= $(shell printf '$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_zlib.c && $(CC) $(FLAGS) -o have_zlib$(EXT) have_zlib.c -lz 2> $(VOID) && rm have_zlib$(EXT) && echo 1 || echo 0; rm have_zlib.c)
 ifeq ($(HAVE_ZLIB), 1)
   ZLIB_MSG := ==> building zstd with .gz compression support
   ZLIBCPP = -DZSTD_GZCOMPRESS -DZSTD_GZDECOMPRESS
@@ -167,7 +87,7 @@ endif
 
 # lzma detection
 NO_LZMA_MSG := ==> no liblzma, building zstd without .xz/.lzma support
-HAVE_LZMA := $(shell printf '$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_lzma.c && $(CC) $(FLAGS) -o have_lzma$(EXT) have_lzma.c -llzma 2> $(VOID) && rm have_lzma$(EXT) && echo 1 || echo 0; rm have_lzma.c)
+HAVE_LZMA ?= $(shell printf '$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_lzma.c && $(CC) $(FLAGS) -o have_lzma$(EXT) have_lzma.c -llzma 2> $(VOID) && rm have_lzma$(EXT) && echo 1 || echo 0; rm have_lzma.c)
 ifeq ($(HAVE_LZMA), 1)
   LZMA_MSG := ==> building zstd with .xz/.lzma compression support
   LZMACPP = -DZSTD_LZMACOMPRESS -DZSTD_LZMADECOMPRESS
@@ -178,7 +98,7 @@ endif
 
 # lz4 detection
 NO_LZ4_MSG := ==> no liblz4, building zstd without .lz4 support
-HAVE_LZ4 := $(shell printf '$(NUM_SYMBOL)include \n$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_lz4.c && $(CC) $(FLAGS) -o have_lz4$(EXT) have_lz4.c -llz4 2> $(VOID) && rm have_lz4$(EXT) && echo 1 || echo 0; rm have_lz4.c)
+HAVE_LZ4 ?= $(shell printf '$(NUM_SYMBOL)include \n$(NUM_SYMBOL)include \nint main(void) { return 0; }' > have_lz4.c && $(CC) $(FLAGS) -o have_lz4$(EXT) have_lz4.c -llz4 2> $(VOID) && rm have_lz4$(EXT) && echo 1 || echo 0; rm have_lz4.c)
 ifeq ($(HAVE_LZ4), 1)
   LZ4_MSG := ==> building zstd with .lz4 compression support
   LZ4CPP = -DZSTD_LZ4COMPRESS -DZSTD_LZ4DECOMPRESS
@@ -209,10 +129,10 @@ SET_CACHE_DIRECTORY = \
 
 
 .PHONY: all
-all: zstd
+all: zstd zstd-compress zstd-decompress zstd-small
 
 .PHONY: allVariants
-allVariants: zstd zstd-compress zstd-decompress zstd-small zstd-nolegacy zstd-dictBuilder
+allVariants: all zstd-frugal zstd-nolegacy zstd-dictBuilder
 
 .PHONY: zstd  # must always be run
 zstd : CPPFLAGS += $(THREAD_CPP) $(ZLIBCPP) $(LZMACPP) $(LZ4CPP)
@@ -242,8 +162,8 @@ $(BUILD_DIR)/zstd : $(ZSTD_OBJ)
 	$(CC) $(FLAGS) $^ $(LDLIBS) -o $@$(EXT)
 
 ifeq ($(HAVE_HASH),1)
-SRCBIN_HASH = $(shell cat $(BUILD_DIR)/zstd 2> $(VOID) | $(HASH) | cut -f 1 -d " ")
-DSTBIN_HASH = $(shell cat zstd 2> $(VOID) | $(HASH) | cut -f 1 -d " ")
+SRCBIN_HASH = $(shell cat $(BUILD_DIR)/zstd$(EXT) 2> $(VOID) | $(HASH) | cut -f 1 -d " ")
+DSTBIN_HASH = $(shell cat zstd$(EXT) 2> $(VOID) | $(HASH) | cut -f 1 -d " ")
 BIN_ISDIFFERENT = $(if $(filter $(SRCBIN_HASH),$(DSTBIN_HASH)),0,1)
 else
 BIN_ISDIFFERENT = 1
@@ -251,7 +171,7 @@ endif
 
 zstd : $(BUILD_DIR)/zstd
 	if [ $(BIN_ISDIFFERENT) -eq 1 ]; then \
-		cp -f $< $@; \
+		cp -f $<$(EXT) $@$(EXT); \
 		echo zstd build completed; \
 	else \
 		echo zstd already built; \
@@ -260,11 +180,13 @@ zstd : $(BUILD_DIR)/zstd
 endif  # BUILD_DIR
 
 
+CLEAN += zstd
 .PHONY: zstd-release
 zstd-release: DEBUGFLAGS := -DBACKTRACE_ENABLE=0
 zstd-release: DEBUGFLAGS_LD :=
 zstd-release: zstd
 
+CLEAN += zstd32
 zstd32 : CPPFLAGS += $(THREAD_CPP)
 zstd32 : LDFLAGS  += $(THREAD_LD)
 zstd32 : CPPFLAGS += -DZSTD_LEGACY_SUPPORT=$(ZSTD_LEGACY_SUPPORT)
@@ -274,8 +196,10 @@ endif
 zstd32 : $(ZSTDLIB_FULL_SRC) $(ZSTD_CLI_SRC)
 	$(CC) -m32 $(FLAGS) $^ -o $@$(EXT)
 
-## zstd-nolegacy: same scope as zstd, with just support of legacy formats removed
+## zstd-nolegacy: same scope as zstd, with removed support of legacy formats
+CLEAN += zstd-nolegacy
 zstd-nolegacy : LDFLAGS += $(THREAD_LD) $(ZLIBLD) $(LZMALD) $(LZ4LD) $(DEBUGFLAGS_LD)
+zstd-nolegacy : CPPFLAGS += -UZSTD_LEGACY_SUPPORT -DZSTD_LEGACY_SUPPORT=0
 zstd-nolegacy : $(ZSTDLIB_CORE_SRC) $(ZDICT_SRC) $(ZSTD_CLI_OBJ)
 	$(CC) $(FLAGS) $^ -o $@$(EXT) $(LDFLAGS)
 
@@ -299,7 +223,7 @@ zstd-noxz : zstd
 
 ## zstd-dll: zstd executable linked to dynamic library libzstd (must have same version)
 .PHONY: zstd-dll
-zstd-dll : LDFLAGS+= -L$(ZSTDDIR)
+zstd-dll : LDFLAGS+= -L$(LIBZSTD)
 zstd-dll : LDLIBS += -lzstd
 zstd-dll : ZSTDLIB_LOCAL_SRC = xxhash.c
 zstd-dll : zstd
@@ -307,35 +231,46 @@ zstd-dll : zstd
 
 ## zstd-pgo: zstd executable optimized with PGO.
 .PHONY: zstd-pgo
+zstd-pgo : LLVM_PROFDATA?=llvm-profdata
+zstd-pgo : PROF_GENERATE_FLAGS=-fprofile-generate $(if $(findstring gcc,$(CC)),-fprofile-dir=.)
+zstd-pgo : PROF_USE_FLAGS=-fprofile-use $(if $(findstring gcc,$(CC)),-fprofile-dir=. -Werror=missing-profile -Wno-error=coverage-mismatch)
 zstd-pgo :
-	$(MAKE) clean
-	$(MAKE) zstd MOREFLAGS=-fprofile-generate
+	$(MAKE) clean HASH_DIR=$(HASH_DIR)
+	$(MAKE) zstd HASH_DIR=$(HASH_DIR) MOREFLAGS="$(PROF_GENERATE_FLAGS)"
 	./zstd -b19i1 $(PROFILE_WITH)
 	./zstd -b16i1 $(PROFILE_WITH)
 	./zstd -b9i2 $(PROFILE_WITH)
 	./zstd -b $(PROFILE_WITH)
 	./zstd -b7i2 $(PROFILE_WITH)
 	./zstd -b5 $(PROFILE_WITH)
-	$(RM) zstd *.o
-	case $(CC) in *clang*) if ! [ -e default.profdata ]; then llvm-profdata merge -output=default.profdata default*.profraw; fi ;; esac
-	$(MAKE) zstd MOREFLAGS=-fprofile-use
+ifndef BUILD_DIR
+	$(RM) zstd obj/$(HASH_DIR)/zstd obj/$(HASH_DIR)/*.o
+else
+	$(RM) zstd $(BUILD_DIR)/zstd $(BUILD_DIR)/*.o
+endif
+	case $(CC) in *clang*) if ! [ -e default.profdata ]; then $(LLVM_PROFDATA) merge -output=default.profdata default*.profraw; fi ;; esac
+	$(MAKE) zstd HASH_DIR=$(HASH_DIR) MOREFLAGS="$(PROF_USE_FLAGS)"
 
 ## zstd-small: minimal target, supporting only zstd compression and decompression. no bench. no legacy. no other format.
-zstd-small: CFLAGS = -Os -s
-zstd-frugal zstd-small: $(ZSTDLIB_CORE_SRC) zstdcli.c util.c timefn.c fileio.c
-	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NOTRACE $^ -o $@$(EXT)
+CLEAN += zstd-small zstd-frugal
+zstd-small: CFLAGS = -Os -Wl,-s
+zstd-frugal zstd-small: $(ZSTDLIB_CORE_SRC) zstdcli.c util.c timefn.c fileio.c fileio_asyncio.c
+	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NOTRACE -UZSTD_LEGACY_SUPPORT -DZSTD_LEGACY_SUPPORT=0 $^ -o $@$(EXT)
 
-zstd-decompress: $(ZSTDLIB_COMMON_C) $(ZSTDLIB_DECOMPRESS_C) zstdcli.c util.c timefn.c fileio.c
-	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NOCOMPRESS -DZSTD_NOTRACE $^ -o $@$(EXT)
+CLEAN += zstd-decompress
+zstd-decompress: $(ZSTDLIB_COMMON_SRC) $(ZSTDLIB_DECOMPRESS_SRC) zstdcli.c util.c timefn.c fileio.c fileio_asyncio.c
+	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NOCOMPRESS -DZSTD_NOTRACE -UZSTD_LEGACY_SUPPORT -DZSTD_LEGACY_SUPPORT=0 $^ -o $@$(EXT)
 
-zstd-compress: $(ZSTDLIB_COMMON_C) $(ZSTDLIB_COMPRESS_C) zstdcli.c util.c timefn.c fileio.c
-	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NODECOMPRESS -DZSTD_NOTRACE $^ -o $@$(EXT)
+CLEAN += zstd-compress
+zstd-compress: $(ZSTDLIB_COMMON_SRC) $(ZSTDLIB_COMPRESS_SRC) zstdcli.c util.c timefn.c fileio.c fileio_asyncio.c
+	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODICT -DZSTD_NODECOMPRESS -DZSTD_NOTRACE -UZSTD_LEGACY_SUPPORT -DZSTD_LEGACY_SUPPORT=0 $^ -o $@$(EXT)
 
 ## zstd-dictBuilder: executable supporting dictionary creation and compression (only)
-zstd-dictBuilder: CPPFLAGS += -DZSTD_NOBENCH -DZSTD_NODECOMPRESS -DZSTD_NOTRACE
-zstd-dictBuilder: $(ZSTDLIB_COMMON_C) $(ZSTDLIB_COMPRESS_C) $(ZDICT_SRC) zstdcli.c util.c timefn.c fileio.c dibio.c
-	$(CC) $(FLAGS) $^ -o $@$(EXT)
+CLEAN += zstd-dictBuilder
+zstd-dictBuilder: $(ZSTDLIB_COMMON_SRC) $(ZSTDLIB_COMPRESS_SRC) $(ZDICT_SRC) zstdcli.c util.c timefn.c fileio.c fileio_asyncio.c dibio.c
+	$(CC) $(FLAGS) -DZSTD_NOBENCH -DZSTD_NODECOMPRESS -DZSTD_NOTRACE $^ -o $@$(EXT)
 
+CLEAN += zstdmt
 zstdmt: zstd
 	ln -sf zstd zstdmt
 
@@ -344,7 +279,7 @@ generate_res: $(RES64_FILE) $(RES32_FILE)
 
 ifneq (,$(filter Windows%,$(OS)))
 RC ?= windres
-# http://stackoverflow.com/questions/708238/how-do-i-add-an-icon-to-a-mingw-gcc-compiled-executable
+# https://stackoverflow.com/questions/708238/how-do-i-add-an-icon-to-a-mingw-gcc-compiled-executable
 $(RES64_FILE): windres/zstd.rc
 	$(RC) -o $@ -I ../lib -I windres -i $< -O coff -F pe-x86-64
 $(RES32_FILE): windres/zstd.rc
@@ -353,12 +288,8 @@ endif
 
 .PHONY: clean
 clean:
-	$(RM) core *.o tmp* result* *.gcda dictionary *.zst \
-        zstd$(EXT) zstd32$(EXT) zstd-dll$(EXT) \
-        zstd-compress$(EXT) zstd-decompress$(EXT) \
-        zstd-small$(EXT) zstd-frugal$(EXT) zstd-nolegacy$(EXT) zstd4$(EXT) \
-        zstd-dictBuilder$(EXT) \
-        *.gcda default*.profraw default.profdata have_zlib$(EXT)
+	$(RM) $(CLEAN) core *.o tmp* result* dictionary *.zst \
+        *.gcda default*.profraw default.profdata have_zlib
 	$(RM) -r obj/*
 	@echo Cleaning completed
 
@@ -398,6 +329,10 @@ $(BUILD_DIR)/%.o : %.c $(BUILD_DIR)/%.d | $(BUILD_DIR)
 	@echo CC $@
 	$(COMPILE.c) $(DEPFLAGS) $(BUILD_DIR)/$*.d $(OUTPUT_OPTION) $<
 
+$(BUILD_DIR)/%.o : %.S | $(BUILD_DIR)
+	@echo AS $@
+	$(COMPILE.S) $(OUTPUT_OPTION) $<
+
 MKDIR ?= mkdir
 $(BUILD_DIR): ; $(MKDIR) -p $@
 
@@ -411,11 +346,11 @@ include $(wildcard $(DEPFILES))
 #-----------------------------------------------------------------------------
 # make install is validated only for Linux, macOS, BSD, Hurd and Solaris targets
 #-----------------------------------------------------------------------------
-ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku))
+ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS Haiku AIX))
 
 HAVE_COLORNEVER = $(shell echo a | egrep --color=never a > /dev/null 2> /dev/null && echo 1 || echo 0)
 EGREP_OPTIONS ?=
-ifeq ($HAVE_COLORNEVER, 1)
+ifeq ($(HAVE_COLORNEVER), 1)
   EGREP_OPTIONS += --color=never
 endif
 EGREP = egrep $(EGREP_OPTIONS)
diff --git a/priv/zstd/programs/README.md b/priv/zstd/programs/README.md
index 7fd7104..1b9f47c 100644
--- a/priv/zstd/programs/README.md
+++ b/priv/zstd/programs/README.md
@@ -156,7 +156,7 @@ Arguments :
 
 Advanced arguments :
  -V     : display Version number and exit
- -c     : force write to standard output, even if it is the console
+ -c     : write to standard output (even if it is the console)
  -v     : verbose mode; specify multiple times to increase verbosity
  -q     : suppress warnings; specify twice to suppress errors too
 --no-progress : do not display the progress counter
@@ -164,6 +164,7 @@ Advanced arguments :
 --filelist FILE : read list of files to operate upon from FILE
 --output-dir-flat DIR : processed files are stored into DIR
 --output-dir-mirror DIR : processed files are stored into DIR respecting original directory structure
+--[no-]asyncio : use asynchronous IO (default: enabled)
 --[no-]check : during compression, add XXH64 integrity checksum to frame (default: enabled). If specified with -d, decompressor will ignore/validate checksums in compressed frame (default: validate).
 --      : All arguments after "--" are treated as files
 
@@ -172,6 +173,7 @@ Advanced compression arguments :
 --long[=#]: enable long distance matching with given window log (default: 27)
 --fast[=#]: switch to very fast compression levels (default: 1)
 --adapt : dynamically adapt compression level to I/O conditions
+--patch-from=FILE : specify the file to be used as a reference point for zstd's diff engine
  -T#    : spawns # compression threads (default: 1, 0==# cores)
  -B#    : select size of each job (default: 0==automatic)
 --single-thread : use a single thread for both I/O and compression (result slightly different than -T1)
@@ -207,7 +209,7 @@ Benchmark arguments :
  -b#    : benchmark file(s), using # compression level (default: 3)
  -e#    : test all compression levels successively from -b# to -e# (default: 1)
  -i#    : minimum evaluation time in seconds (default: 3s)
- -B#    : cut file into independent blocks of size # (default: no block)
+ -B#    : cut file into independent chunks of size # (default: no chunking)
  -S     : output one benchmark result per input file (default: consolidated result)
 --priority=rt : set process priority to real-time
 ```
@@ -274,7 +276,7 @@ compression speed (for lower levels) with minimal change in compression ratio.
 
 The below table illustrates this on the [Silesia compression corpus].
 
-[Silesia compression corpus]: http://sun.aei.polsl.pl/~sdeor/index.php?page=silesia
+[Silesia compression corpus]: https://sun.aei.polsl.pl//~sdeor/index.php?page=silesia
 
 | Method | Compression ratio | Compression speed | Decompression speed  |
 |:-------|------------------:|------------------:|---------------------:|
diff --git a/priv/zstd/programs/benchfn.c b/priv/zstd/programs/benchfn.c
index 1aadbdd..8e6726f 100644
--- a/priv/zstd/programs/benchfn.c
+++ b/priv/zstd/programs/benchfn.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -229,9 +229,9 @@ BMK_runOutcome_t BMK_benchTimedFn(BMK_timedFnState_t* cont,
             cont->timeSpent_ns += (unsigned long long)loopDuration_ns;
 
             /* estimate nbLoops for next run to last approximately 1 second */
-            if (loopDuration_ns > (runBudget_ns / 50)) {
+            if (loopDuration_ns > ((double)runBudget_ns / 50)) {
                 double const fastestRun_ns = MIN(bestRunTime.nanoSecPerRun, newRunTime.nanoSecPerRun);
-                cont->nbLoops = (unsigned)(runBudget_ns / fastestRun_ns) + 1;
+                cont->nbLoops = (unsigned)((double)runBudget_ns / fastestRun_ns) + 1;
             } else {
                 /* previous run was too short : blindly increase workload by x multiplier */
                 const unsigned multiplier = 10;
@@ -239,7 +239,7 @@ BMK_runOutcome_t BMK_benchTimedFn(BMK_timedFnState_t* cont,
                 cont->nbLoops *= multiplier;
             }
 
-            if(loopDuration_ns < runTimeMin_ns) {
+            if(loopDuration_ns < (double)runTimeMin_ns) {
                 /* don't report results for which benchmark run time was too small : increased risks of rounding errors */
                 assert(completed == 0);
                 continue;
diff --git a/priv/zstd/programs/benchfn.h b/priv/zstd/programs/benchfn.h
index 590f292..1bd93d1 100644
--- a/priv/zstd/programs/benchfn.h
+++ b/priv/zstd/programs/benchfn.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -123,7 +123,7 @@ BMK_runTime_t BMK_extract_runTime(BMK_runOutcome_t outcome);
 /* when benchmark failed, it means one invocation of `benchFn` failed.
  * The failure was detected by `errorFn`, operating on return values of `benchFn`.
  * Returns the faulty return value.
- * note : this function will abort() program execution if benchmark did not failed.
+ * note : this function will abort() program execution if benchmark did not fail.
  *        always check if benchmark failed first !
  */
 size_t BMK_extract_errorResult(BMK_runOutcome_t outcome);
diff --git a/priv/zstd/programs/benchzstd.c b/priv/zstd/programs/benchzstd.c
index 49c0349..a76db5f 100644
--- a/priv/zstd/programs/benchzstd.c
+++ b/priv/zstd/programs/benchzstd.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -13,7 +13,7 @@
 *  Tuning parameters
 ****************************************/
 #ifndef BMK_TIMETEST_DEFAULT_S   /* default minimum time per test */
-#define BMK_TIMETEST_DEFAULT_S 3
+# define BMK_TIMETEST_DEFAULT_S 3
 #endif
 
 
@@ -31,9 +31,14 @@
 #include "timefn.h"      /* UTIL_time_t */
 #include "benchfn.h"
 #include "../lib/common/mem.h"
+#ifndef ZSTD_STATIC_LINKING_ONLY
 #define ZSTD_STATIC_LINKING_ONLY
+#endif
 #include "../lib/zstd.h"
 #include "datagen.h"     /* RDG_genBuffer */
+#ifndef XXH_INLINE_ALL
+#define XXH_INLINE_ALL
+#endif
 #include "../lib/common/xxhash.h"
 #include "benchzstd.h"
 #include "../lib/zstd_errors.h"
@@ -70,6 +75,8 @@ static const size_t maxMemory = (sizeof(size_t)==4)  ?
 #define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush(NULL); }
 #define DISPLAYLEVEL(l, ...) if (displayLevel>=l) { DISPLAY(__VA_ARGS__); }
 /* 0 : no display;   1: errors;   2 : + result + interaction + warnings;   3 : + progression;   4 : + information */
+#define OUTPUT(...)          { fprintf(stdout, __VA_ARGS__); fflush(NULL); }
+#define OUTPUTLEVEL(l, ...)  if (displayLevel>=l) { OUTPUT(__VA_ARGS__); }
 
 
 /* *************************************
@@ -129,7 +136,7 @@ BMK_advancedParams_t BMK_initAdvancedParams(void) {
         0, /* ldmHashLog */
         0, /* ldmBuckSizeLog */
         0,  /* ldmHashRateLog */
-        ZSTD_lcm_auto, /* literalCompressionMode */
+        ZSTD_ps_auto, /* literalCompressionMode */
         0 /* useRowMatchFinder */
     };
     return res;
@@ -181,7 +188,7 @@ BMK_initCCtx(ZSTD_CCtx* ctx,
     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_minMatch, (int)comprParams->minMatch));
     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_targetLength, (int)comprParams->targetLength));
     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_literalCompressionMode, (int)adv->literalCompressionMode));
-    CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_strategy, comprParams->strategy));
+    CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_strategy, (int)comprParams->strategy));
     CHECK_Z(ZSTD_CCtx_loadDictionary(ctx, dictBuffer, dictBufferSize));
 }
 
@@ -320,26 +327,31 @@ BMK_benchMemAdvancedNoAlloc(
     /* init */
     memset(&benchResult, 0, sizeof(benchResult));
     if (strlen(displayName)>17) displayName += strlen(displayName) - 17;   /* display last 17 characters */
-    if (adv->mode == BMK_decodeOnly) {  /* benchmark only decompression : source must be already compressed */
+    if (adv->mode == BMK_decodeOnly) {
+        /* benchmark only decompression : source must be already compressed */
         const char* srcPtr = (const char*)srcBuffer;
         U64 totalDSize64 = 0;
         U32 fileNb;
         for (fileNb=0; fileNb decodedSize) {  /* size_t overflow */
+                RETURN_ERROR(32, BMK_benchOutcome_t, "decompressed size is too large for local system");
+            }
             *resultBufferPtr = malloc(decodedSize);
             if (!(*resultBufferPtr)) {
-                RETURN_ERROR(33, BMK_benchOutcome_t, "not enough memory");
-            }
-            if (totalDSize64 > decodedSize) {  /* size_t overflow */
-                free(*resultBufferPtr);
-                RETURN_ERROR(32, BMK_benchOutcome_t, "original size is too large");
+                RETURN_ERROR(33, BMK_benchOutcome_t, "allocation error: not enough memory");
             }
             cSize = srcSize;
             srcSize = decodedSize;
@@ -371,10 +383,7 @@ BMK_benchMemAdvancedNoAlloc(
                 if (adv->mode == BMK_decodeOnly) {
                     cSizes[nbBlocks] = thisBlockSize;
                     benchResult.cSize = thisBlockSize;
-                }
-            }
-        }
-    }
+    }   }   }   }
 
     /* warming up `compressedBuffer` */
     if (adv->mode == BMK_decodeOnly) {
@@ -383,6 +392,10 @@ BMK_benchMemAdvancedNoAlloc(
         RDG_genBuffer(compressedBuffer, maxCompressedSize, 0.10, 0.50, 1);
     }
 
+    if (!UTIL_support_MT_measurements() && adv->nbWorkers > 1) {
+        OUTPUTLEVEL(2, "Warning : time measurements may be incorrect in multithreading mode... \n")
+    }
+
     /* Bench */
     {   U64 const crcOrig = (adv->mode == BMK_decodeOnly) ? 0 : XXH64(srcBuffer, srcSize, 0);
 #       define NB_MARKS 4
@@ -429,20 +442,21 @@ BMK_benchMemAdvancedNoAlloc(
         dctxprep.dictBuffer = dictBuffer;
         dctxprep.dictBufferSize = dictBufferSize;
 
-        DISPLAYLEVEL(2, "\r%70s\r", "");   /* blank line */
-        DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->\r", marks[markNb], displayName, (unsigned)srcSize);
+        OUTPUTLEVEL(2, "\r%70s\r", "");   /* blank line */
+        assert(srcSize < UINT_MAX);
+        OUTPUTLEVEL(2, "%2s-%-17.17s :%10u -> \r", marks[markNb], displayName, (unsigned)srcSize);
 
         while (!(compressionCompleted && decompressionCompleted)) {
             if (!compressionCompleted) {
                 BMK_runOutcome_t const cOutcome = BMK_benchTimedFn( timeStateCompress, cbp);
 
                 if (!BMK_isSuccessful_runOutcome(cOutcome)) {
-                    return BMK_benchOutcome_error();
+                    RETURN_ERROR(30, BMK_benchOutcome_t, "compression error");
                 }
 
                 {   BMK_runTime_t const cResult = BMK_extract_runTime(cOutcome);
                     cSize = cResult.sumOfReturn;
-                    ratio = (double)srcSize / cSize;
+                    ratio = (double)srcSize / (double)cSize;
                     {   BMK_benchResult_t newResult;
                         newResult.cSpeed = (U64)((double)srcSize * TIMELOOP_NANOSEC / cResult.nanoSecPerRun);
                         benchResult.cSize = cSize;
@@ -451,11 +465,12 @@ BMK_benchMemAdvancedNoAlloc(
                 }   }
 
                 {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                    DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s\r",
+                    assert(cSize < UINT_MAX);
+                    OUTPUTLEVEL(2, "%2s-%-17.17s :%10u ->%10u (x%5.*f), %6.*f MB/s \r",
                             marks[markNb], displayName,
                             (unsigned)srcSize, (unsigned)cSize,
                             ratioAccuracy, ratio,
-                            benchResult.cSpeed < (10 MB) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT);
+                            benchResult.cSpeed < (10 * MB_UNIT) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT);
                 }
                 compressionCompleted = BMK_isCompleted_TimedFn(timeStateCompress);
             }
@@ -464,7 +479,7 @@ BMK_benchMemAdvancedNoAlloc(
                 BMK_runOutcome_t const dOutcome = BMK_benchTimedFn(timeStateDecompress, dbp);
 
                 if(!BMK_isSuccessful_runOutcome(dOutcome)) {
-                    return BMK_benchOutcome_error();
+                    RETURN_ERROR(30, BMK_benchOutcome_t, "decompression error");
                 }
 
                 {   BMK_runTime_t const dResult = BMK_extract_runTime(dOutcome);
@@ -474,11 +489,11 @@ BMK_benchMemAdvancedNoAlloc(
                 }
 
                 {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
-                    DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s ,%6.1f MB/s \r",
+                    OUTPUTLEVEL(2, "%2s-%-17.17s :%10u ->%10u (x%5.*f), %6.*f MB/s, %6.1f MB/s\r",
                             marks[markNb], displayName,
                             (unsigned)srcSize, (unsigned)cSize,
                             ratioAccuracy, ratio,
-                            benchResult.cSpeed < (10 MB) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT,
+                            benchResult.cSpeed < (10 * MB_UNIT) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT,
                             (double)benchResult.dSpeed / MB_UNIT);
                 }
                 decompressionCompleted = BMK_isCompleted_TimedFn(timeStateDecompress);
@@ -515,7 +530,7 @@ BMK_benchMemAdvancedNoAlloc(
                                 DISPLAY("%02X ", ((const BYTE*)srcBuffer)[u+n]);
                             DISPLAY(" \n");
                             DISPLAY("decode: ");
-                            for (n=lowest; n>0; n++)
+                            for (n=lowest; n>0; n--)
                                 DISPLAY("%02X ", resultBuffer[u-n]);
                             DISPLAY(" :%02X:  ", resultBuffer[u]);
                             for (n=1; n<3; n++)
@@ -535,13 +550,13 @@ BMK_benchMemAdvancedNoAlloc(
             double const cSpeed = (double)benchResult.cSpeed / MB_UNIT;
             double const dSpeed = (double)benchResult.dSpeed / MB_UNIT;
             if (adv->additionalParam) {
-                DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, adv->additionalParam);
+                OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, adv->additionalParam);
             } else {
-                DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName);
+                OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName);
             }
         }
 
-        DISPLAYLEVEL(2, "%2i#\n", cLevel);
+        OUTPUTLEVEL(2, "%2i#\n", cLevel);
     }   /* Bench */
 
     benchResult.cMem = (1ULL << (comprParams->windowLog)) + ZSTD_sizeof_CCtx(cctx);
@@ -588,7 +603,7 @@ BMK_benchOutcome_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
 
     void* resultBuffer = srcSize ? malloc(srcSize) : NULL;
 
-    int allocationincomplete = !srcPtrs || !srcSizes || !cPtrs ||
+    int const allocationincomplete = !srcPtrs || !srcSizes || !cPtrs ||
         !cSizes || !cCapacities || !resPtrs || !resSizes ||
         !timeStateCompress || !timeStateDecompress ||
         !cctx || !dctx ||
@@ -670,7 +685,7 @@ static BMK_benchOutcome_t BMK_benchCLevel(const void* srcBuffer, size_t benchedS
     }
 
     if (displayLevel == 1 && !adv->additionalParam)   /* --quiet mode */
-        DISPLAY("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n",
+        OUTPUT("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n",
                 ZSTD_VERSION_STRING, ZSTD_GIT_COMMIT_STRING,
                 (unsigned)benchedSize, adv->nbSeconds, (unsigned)(adv->blockSize>>10));
 
@@ -760,7 +775,7 @@ static int BMK_loadFiles(void* buffer, size_t bufferSize,
         }
         {   FILE* const f = fopen(fileNamesTable[n], "rb");
             if (f==NULL) RETURN_ERROR_INT(10, "impossible to open file %s", fileNamesTable[n]);
-            DISPLAYLEVEL(2, "Loading %s...       \r", fileNamesTable[n]);
+            OUTPUTLEVEL(2, "Loading %s...       \r", fileNamesTable[n]);
             if (fileSize > bufferSize-pos) fileSize = bufferSize-pos, nbFiles=n;   /* buffer too small - stop after this file */
             {   size_t const readSize = fread(((char*)buffer)+pos, 1, (size_t)fileSize, f);
                 if (readSize != (size_t)fileSize) RETURN_ERROR_INT(11, "could not read %s", fileNamesTable[n]);
@@ -797,6 +812,10 @@ BMK_benchOutcome_t BMK_benchFilesAdvanced(
         RETURN_ERROR(15, BMK_benchOutcome_t, "Invalid Compression Level");
     }
 
+    if (totalSizeToLoad == UTIL_FILESIZE_UNKNOWN) {
+        RETURN_ERROR(9, BMK_benchOutcome_t, "Error loading files");
+    }
+
     fileSizes = (size_t*)calloc(nbFiles, sizeof(size_t));
     if (!fileSizes) RETURN_ERROR(12, BMK_benchOutcome_t, "not enough memory for fileSizes");
 
diff --git a/priv/zstd/programs/benchzstd.h b/priv/zstd/programs/benchzstd.h
index 9b40dcc..aa683df 100644
--- a/priv/zstd/programs/benchzstd.h
+++ b/priv/zstd/programs/benchzstd.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -116,7 +116,7 @@ typedef struct {
     int ldmHashLog;
     int ldmBucketSizeLog;
     int ldmHashRateLog;
-    ZSTD_literalCompressionMode_e literalCompressionMode;
+    ZSTD_paramSwitch_e literalCompressionMode;
     int useRowMatchFinder;  /* use row-based matchfinder if possible */
 } BMK_advancedParams_t;
 
diff --git a/priv/zstd/programs/datagen.c b/priv/zstd/programs/datagen.c
index 3b4f9e5..ddc690b 100644
--- a/priv/zstd/programs/datagen.c
+++ b/priv/zstd/programs/datagen.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/programs/datagen.h b/priv/zstd/programs/datagen.h
index b76ae2a..ca72700 100644
--- a/priv/zstd/programs/datagen.h
+++ b/priv/zstd/programs/datagen.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/programs/dibio.c b/priv/zstd/programs/dibio.c
index d6c9f6d..26ebe5c 100644
--- a/priv/zstd/programs/dibio.c
+++ b/priv/zstd/programs/dibio.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -27,11 +27,11 @@
 #include          /* memset */
 #include           /* fprintf, fopen, ftello64 */
 #include           /* errno */
-#include 
 
 #include "timefn.h"         /* UTIL_time_t, UTIL_clockSpanMicro, UTIL_getTime */
+#include "../lib/common/debug.h" /* assert */
 #include "../lib/common/mem.h"  /* read */
-#include "../lib/common/error_private.h"
+#include "../lib/zstd_errors.h"
 #include "dibio.h"
 
 
@@ -49,6 +49,7 @@
 static const size_t g_maxMemory = (sizeof(size_t) == 4) ? (2 GB - 64 MB) : ((size_t)(512 MB) << sizeof(size_t));
 
 #define NOISELENGTH 32
+#define MAX_SAMPLES_SIZE (2 GB) /* training dataset limited to 2GB */
 
 
 /*-*************************************
@@ -88,6 +89,15 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
 #undef MIN
 #define MIN(a,b)    ((a) < (b) ? (a) : (b))
 
+/**
+  Returns the size of a file.
+  If error returns -1.
+*/
+static S64 DiB_getFileSize (const char * fileName)
+{
+    U64 const fileSize = UTIL_getFileSize(fileName);
+    return (fileSize == UTIL_FILESIZE_UNKNOWN) ? -1 : (S64)fileSize;
+}
 
 /* ********************************************************
 *  File related operations
@@ -101,47 +111,70 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
  * *bufferSizePtr is modified, it provides the amount data loaded within buffer.
  *  sampleSizes is filled with the size of each sample.
  */
-static unsigned DiB_loadFiles(void* buffer, size_t* bufferSizePtr,
-                              size_t* sampleSizes, unsigned sstSize,
-                              const char** fileNamesTable, unsigned nbFiles, size_t targetChunkSize,
-                              unsigned displayLevel)
+static int DiB_loadFiles(
+    void* buffer, size_t* bufferSizePtr,
+    size_t* sampleSizes, int sstSize,
+    const char** fileNamesTable, int nbFiles,
+    size_t targetChunkSize, int displayLevel )
 {
     char* const buff = (char*)buffer;
-    size_t pos = 0;
-    unsigned nbLoadedChunks = 0, fileIndex;
-
-    for (fileIndex=0; fileIndex *bufferSizePtr-pos) break;
-            {   size_t const readSize = fread(buff+pos, 1, toLoad, f);
-                if (readSize != toLoad) EXM_THROW(11, "Pb reading %s", fileName);
-                pos += readSize;
-                sampleSizes[nbLoadedChunks++] = toLoad;
-                remainingToLoad -= targetChunkSize;
-                if (nbLoadedChunks == sstSize) { /* no more space left in sampleSizes table */
-                    fileIndex = nbFiles;  /* stop there */
+    size_t totalDataLoaded = 0;
+    int nbSamplesLoaded = 0;
+    int fileIndex = 0;
+    FILE * f = NULL;
+
+    assert(targetChunkSize <= SAMPLESIZE_MAX);
+
+    while ( nbSamplesLoaded < sstSize && fileIndex < nbFiles ) {
+        size_t fileDataLoaded;
+        S64 const fileSize = DiB_getFileSize(fileNamesTable[fileIndex]);
+        if (fileSize <= 0) {
+            /* skip if zero-size or file error */
+            ++fileIndex;
+            continue;
+        }
+
+        f = fopen( fileNamesTable[fileIndex], "rb");
+        if (f == NULL)
+            EXM_THROW(10, "zstd: dictBuilder: %s %s ", fileNamesTable[fileIndex], strerror(errno));
+        DISPLAYUPDATE(2, "Loading %s...       \r", fileNamesTable[fileIndex]);
+
+        /* Load the first chunk of data from the file */
+        fileDataLoaded = targetChunkSize > 0 ?
+                            (size_t)MIN(fileSize, (S64)targetChunkSize) :
+                            (size_t)MIN(fileSize, SAMPLESIZE_MAX );
+        if (totalDataLoaded + fileDataLoaded > *bufferSizePtr)
+            break;
+        if (fread( buff+totalDataLoaded, 1, fileDataLoaded, f ) != fileDataLoaded)
+            EXM_THROW(11, "Pb reading %s", fileNamesTable[fileIndex]);
+        sampleSizes[nbSamplesLoaded++] = fileDataLoaded;
+        totalDataLoaded += fileDataLoaded;
+
+        /* If file-chunking is enabled, load the rest of the file as more samples */
+        if (targetChunkSize > 0) {
+            while( (S64)fileDataLoaded < fileSize && nbSamplesLoaded < sstSize ) {
+                size_t const chunkSize = MIN((size_t)(fileSize-fileDataLoaded), targetChunkSize);
+                if (totalDataLoaded + chunkSize > *bufferSizePtr) /* buffer is full */
                     break;
-                }
-                if (toLoad < targetChunkSize) {
-                    fseek(f, (long)(targetChunkSize - toLoad), SEEK_CUR);
-        }   }   }
-        fclose(f);
+
+                if (fread( buff+totalDataLoaded, 1, chunkSize, f ) != chunkSize)
+                    EXM_THROW(11, "Pb reading %s", fileNamesTable[fileIndex]);
+                sampleSizes[nbSamplesLoaded++] = chunkSize;
+                totalDataLoaded += chunkSize;
+                fileDataLoaded += chunkSize;
+            }
+        }
+        fileIndex += 1;
+        fclose(f); f = NULL;
     }
+    if (f != NULL)
+        fclose(f);
+
     DISPLAYLEVEL(2, "\r%79s\r", "");
-    *bufferSizePtr = pos;
-    DISPLAYLEVEL(4, "loaded : %u KB \n", (unsigned)(pos >> 10))
-    return nbLoadedChunks;
+    DISPLAYLEVEL(4, "Loaded %d KB total training data, %d nb samples \n",
+        (int)(totalDataLoaded / (1 KB)), nbSamplesLoaded );
+    *bufferSizePtr = totalDataLoaded;
+    return nbSamplesLoaded;
 }
 
 #define DiB_rotl32(x,r) ((x << r) | (x >> (32 - r)))
@@ -164,7 +197,8 @@ static U32 DiB_rand(U32* src)
 static void DiB_shuffle(const char** fileNamesTable, unsigned nbFiles) {
     U32 seed = 0xFD2FB528;
     unsigned i;
-    assert(nbFiles >= 1);
+    if (nbFiles == 0)
+        return;
     for (i = nbFiles - 1; i > 0; --i) {
         unsigned const j = DiB_rand(&seed) % (i + 1);
         const char* const tmp = fileNamesTable[j];
@@ -223,11 +257,10 @@ static void DiB_saveDict(const char* dictFileName,
       if (n!=0) EXM_THROW(5, "%s : flush error", dictFileName) }
 }
 
-
 typedef struct {
-    U64 totalSizeToLoad;
-    unsigned oneSampleTooLarge;
-    unsigned nbSamples;
+    S64 totalSizeToLoad;
+    int nbSamples;
+    int oneSampleTooLarge;
 } fileStats;
 
 /*! DiB_fileStats() :
@@ -235,48 +268,93 @@ typedef struct {
  *  provides the amount of data to be loaded and the resulting nb of samples.
  *  This is useful primarily for allocation purpose => sample buffer, and sample sizes table.
  */
-static fileStats DiB_fileStats(const char** fileNamesTable, unsigned nbFiles, size_t chunkSize, unsigned displayLevel)
+static fileStats DiB_fileStats(const char** fileNamesTable, int nbFiles, size_t chunkSize, int displayLevel)
 {
     fileStats fs;
-    unsigned n;
+    int n;
     memset(&fs, 0, sizeof(fs));
+
+    /* We assume that if chunking is requested, the chunk size is < SAMPLESIZE_MAX */
+    assert( chunkSize <= SAMPLESIZE_MAX );
+
     for (n=0; n 2*SAMPLESIZE_MAX);
-        fs.nbSamples += nbSamples;
+      S64 const fileSize = DiB_getFileSize(fileNamesTable[n]);
+      /* TODO: is there a minimum sample size? What if the file is 1-byte? */
+      if (fileSize == 0) {
+        DISPLAYLEVEL(3, "Sample file '%s' has zero size, skipping...\n", fileNamesTable[n]);
+        continue;
+      }
+
+      /* the case where we are breaking up files in sample chunks */
+      if (chunkSize > 0) {
+        /* TODO: is there a minimum sample size? Can we have a 1-byte sample? */
+        fs.nbSamples += (int)((fileSize + chunkSize-1) / chunkSize);
+        fs.totalSizeToLoad += fileSize;
+      }
+      else {
+      /* the case where one file is one sample */
+        if (fileSize > SAMPLESIZE_MAX) {
+          /* flag excessively large sample files */
+          fs.oneSampleTooLarge |= (fileSize > 2*SAMPLESIZE_MAX);
+
+          /* Limit to the first SAMPLESIZE_MAX (128kB) of the file */
+          DISPLAYLEVEL(3, "Sample file '%s' is too large, limiting to %d KB",
+              fileNamesTable[n], SAMPLESIZE_MAX / (1 KB));
+        }
+        fs.nbSamples += 1;
+        fs.totalSizeToLoad += MIN(fileSize, SAMPLESIZE_MAX);
+      }
     }
-    DISPLAYLEVEL(4, "Preparing to load : %u KB \n", (unsigned)(fs.totalSizeToLoad >> 10));
+    DISPLAYLEVEL(4, "Found training data %d files, %d KB, %d samples\n", nbFiles, (int)(fs.totalSizeToLoad / (1 KB)), fs.nbSamples);
     return fs;
 }
 
-
-int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
-                       const char** fileNamesTable, unsigned nbFiles, size_t chunkSize,
+int DiB_trainFromFiles(const char* dictFileName, size_t maxDictSize,
+                       const char** fileNamesTable, int nbFiles, size_t chunkSize,
                        ZDICT_legacy_params_t* params, ZDICT_cover_params_t* coverParams,
-                       ZDICT_fastCover_params_t* fastCoverParams, int optimize)
+                       ZDICT_fastCover_params_t* fastCoverParams, int optimize, unsigned memLimit)
 {
-    unsigned const displayLevel = params ? params->zParams.notificationLevel :
-                        coverParams ? coverParams->zParams.notificationLevel :
-                        fastCoverParams ? fastCoverParams->zParams.notificationLevel :
-                        0;   /* should never happen */
+    fileStats fs;
+    size_t* sampleSizes; /* vector of sample sizes. Each sample can be up to SAMPLESIZE_MAX */
+    int nbSamplesLoaded; /* nb of samples effectively loaded in srcBuffer */
+    size_t loadedSize; /* total data loaded in srcBuffer for all samples */
+    void* srcBuffer /* contiguous buffer with training data/samples */;
     void* const dictBuffer = malloc(maxDictSize);
-    fileStats const fs = DiB_fileStats(fileNamesTable, nbFiles, chunkSize, displayLevel);
-    size_t* const sampleSizes = (size_t*)malloc(fs.nbSamples * sizeof(size_t));
-    size_t const memMult = params ? MEMMULT :
-                           coverParams ? COVER_MEMMULT:
-                           FASTCOVER_MEMMULT;
-    size_t const maxMem =  DiB_findMaxMem(fs.totalSizeToLoad * memMult) / memMult;
-    size_t loadedSize = (size_t) MIN ((unsigned long long)maxMem, fs.totalSizeToLoad);
-    void* const srcBuffer = malloc(loadedSize+NOISELENGTH);
     int result = 0;
 
+    int const displayLevel = params ? params->zParams.notificationLevel :
+        coverParams ? coverParams->zParams.notificationLevel :
+        fastCoverParams ? fastCoverParams->zParams.notificationLevel : 0;
+
+    /* Shuffle input files before we start assessing how much sample datA to load.
+       The purpose of the shuffle is to pick random samples when the sample
+       set is larger than what we can load in memory. */
+    DISPLAYLEVEL(3, "Shuffling input files\n");
+    DiB_shuffle(fileNamesTable, nbFiles);
+
+    /* Figure out how much sample data to load with how many samples */
+    fs = DiB_fileStats(fileNamesTable, nbFiles, chunkSize, displayLevel);
+
+    {
+        int const memMult = params ? MEMMULT :
+                            coverParams ? COVER_MEMMULT:
+                            FASTCOVER_MEMMULT;
+        size_t const maxMem =  DiB_findMaxMem(fs.totalSizeToLoad * memMult) / memMult;
+        /* Limit the size of the training data to the free memory */
+        /* Limit the size of the training data to 2GB */
+        /* TODO: there is opportunity to stop DiB_fileStats() early when the data limit is reached */
+        loadedSize = (size_t)MIN( MIN((S64)maxMem, fs.totalSizeToLoad), MAX_SAMPLES_SIZE );
+        if (memLimit != 0) {
+            DISPLAYLEVEL(2, "!  Warning : setting manual memory limit for dictionary training data at %u MB \n",
+                (unsigned)(memLimit / (1 MB)));
+            loadedSize = (size_t)MIN(loadedSize, memLimit);
+        }
+        srcBuffer = malloc(loadedSize+NOISELENGTH);
+        sampleSizes = (size_t*)malloc(fs.nbSamples * sizeof(size_t));
+    }
+
     /* Checks */
-    if ((!sampleSizes) || (!srcBuffer) || (!dictBuffer))
+    if ((fs.nbSamples && !sampleSizes) || (!srcBuffer) || (!dictBuffer))
         EXM_THROW(12, "not enough memory for DiB_trainFiles");   /* should not happen */
     if (fs.oneSampleTooLarge) {
         DISPLAYLEVEL(2, "!  Warning : some sample(s) are very large \n");
@@ -289,31 +367,32 @@ int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
         DISPLAYLEVEL(2, "!  Alternatively, split files into fixed-size blocks representative of samples, with -B# \n");
         EXM_THROW(14, "nb of samples too low");   /* we now clearly forbid this case */
     }
-    if (fs.totalSizeToLoad < (unsigned long long)maxDictSize * 8) {
+    if (fs.totalSizeToLoad < (S64)maxDictSize * 8) {
         DISPLAYLEVEL(2, "!  Warning : data size of samples too small for target dictionary size \n");
         DISPLAYLEVEL(2, "!  Samples should be about 100x larger than target dictionary size \n");
     }
 
     /* init */
-    if (loadedSize < fs.totalSizeToLoad)
-        DISPLAYLEVEL(1, "Not enough memory; training on %u MB only...\n", (unsigned)(loadedSize >> 20));
+    if ((S64)loadedSize < fs.totalSizeToLoad)
+        DISPLAYLEVEL(1, "Training samples set too large (%u MB); training on %u MB only...\n",
+            (unsigned)(fs.totalSizeToLoad / (1 MB)),
+            (unsigned)(loadedSize / (1 MB)));
 
     /* Load input buffer */
-    DISPLAYLEVEL(3, "Shuffling input files\n");
-    DiB_shuffle(fileNamesTable, nbFiles);
+    nbSamplesLoaded = DiB_loadFiles(
+        srcBuffer, &loadedSize, sampleSizes, fs.nbSamples, fileNamesTable,
+        nbFiles, chunkSize, displayLevel);
 
-    DiB_loadFiles(srcBuffer, &loadedSize, sampleSizes, fs.nbSamples, fileNamesTable, nbFiles, chunkSize, displayLevel);
-
-    {   size_t dictSize;
+    {   size_t dictSize = ZSTD_error_GENERIC;
         if (params) {
             DiB_fillNoise((char*)srcBuffer + loadedSize, NOISELENGTH);   /* guard band, for end of buffer condition */
             dictSize = ZDICT_trainFromBuffer_legacy(dictBuffer, maxDictSize,
-                                                    srcBuffer, sampleSizes, fs.nbSamples,
+                                                    srcBuffer, sampleSizes, nbSamplesLoaded,
                                                     *params);
         } else if (coverParams) {
             if (optimize) {
               dictSize = ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, maxDictSize,
-                                                             srcBuffer, sampleSizes, fs.nbSamples,
+                                                             srcBuffer, sampleSizes, nbSamplesLoaded,
                                                              coverParams);
               if (!ZDICT_isError(dictSize)) {
                   unsigned splitPercentage = (unsigned)(coverParams->splitPoint * 100);
@@ -322,13 +401,12 @@ int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
               }
             } else {
               dictSize = ZDICT_trainFromBuffer_cover(dictBuffer, maxDictSize, srcBuffer,
-                                                     sampleSizes, fs.nbSamples, *coverParams);
+                                                     sampleSizes, nbSamplesLoaded, *coverParams);
             }
-        } else {
-            assert(fastCoverParams != NULL);
+        } else if (fastCoverParams != NULL) {
             if (optimize) {
               dictSize = ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, maxDictSize,
-                                                              srcBuffer, sampleSizes, fs.nbSamples,
+                                                              srcBuffer, sampleSizes, nbSamplesLoaded,
                                                               fastCoverParams);
               if (!ZDICT_isError(dictSize)) {
                 unsigned splitPercentage = (unsigned)(fastCoverParams->splitPoint * 100);
@@ -338,8 +416,10 @@ int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
               }
             } else {
               dictSize = ZDICT_trainFromBuffer_fastCover(dictBuffer, maxDictSize, srcBuffer,
-                                                        sampleSizes, fs.nbSamples, *fastCoverParams);
+                                                        sampleSizes, nbSamplesLoaded, *fastCoverParams);
             }
+        } else {
+            assert(0 /* Impossible */);
         }
         if (ZDICT_isError(dictSize)) {
             DISPLAYLEVEL(1, "dictionary training failed : %s \n", ZDICT_getErrorName(dictSize));   /* should not happen */
diff --git a/priv/zstd/programs/dibio.h b/priv/zstd/programs/dibio.h
index f65ed9b..a96104c 100644
--- a/priv/zstd/programs/dibio.h
+++ b/priv/zstd/programs/dibio.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -31,9 +31,9 @@
     `parameters` is optional and can be provided with values set to 0, meaning "default".
     @return : 0 == ok. Any other : error.
 */
-int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
-                       const char** fileNamesTable, unsigned nbFiles, size_t chunkSize,
+int DiB_trainFromFiles(const char* dictFileName, size_t maxDictSize,
+                       const char** fileNamesTable, int nbFiles, size_t chunkSize,
                        ZDICT_legacy_params_t* params, ZDICT_cover_params_t* coverParams,
-                       ZDICT_fastCover_params_t* fastCoverParams, int optimize);
+                       ZDICT_fastCover_params_t* fastCoverParams, int optimize, unsigned memLimit);
 
 #endif
diff --git a/priv/zstd/programs/fileio.c b/priv/zstd/programs/fileio.c
index 5693ac3..9a8300c 100644
--- a/priv/zstd/programs/fileio.c
+++ b/priv/zstd/programs/fileio.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -28,6 +28,7 @@
 #include       /* fprintf, open, fdopen, fread, _fileno, stdin, stdout */
 #include      /* malloc, free */
 #include      /* strcmp, strlen */
+#include        /* clock_t, to measure process time */
 #include       /* O_WRONLY */
 #include 
 #include       /* errno */
@@ -40,8 +41,12 @@
 #  include 
 #endif
 
-#include "../lib/common/mem.h"     /* U32, U64 */
 #include "fileio.h"
+#include "fileio_asyncio.h"
+#include "fileio_common.h"
+
+FIO_display_prefs_t g_display_prefs = {2, FIO_ps_auto};
+UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
 
 #define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_magicNumber, ZSTD_frameHeaderSize_max */
 #include "../lib/zstd.h"
@@ -65,6 +70,40 @@
 #  include 
 #endif
 
+char const* FIO_zlibVersion(void)
+{
+#if defined(ZSTD_GZCOMPRESS) || defined(ZSTD_GZDECOMPRESS)
+    return zlibVersion();
+#else
+    return "Unsupported";
+#endif
+}
+
+char const* FIO_lz4Version(void)
+{
+#if defined(ZSTD_LZ4COMPRESS) || defined(ZSTD_LZ4DECOMPRESS)
+    /* LZ4_versionString() added in v1.7.3 */
+#   if LZ4_VERSION_NUMBER >= 10703
+        return LZ4_versionString();
+#   else
+#       define ZSTD_LZ4_VERSION LZ4_VERSION_MAJOR.LZ4_VERSION_MINOR.LZ4_VERSION_RELEASE
+#       define ZSTD_LZ4_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LZ4_VERSION)
+        return ZSTD_LZ4_VERSION_STRING;
+#   endif
+#else
+    return "Unsupported";
+#endif
+}
+
+char const* FIO_lzmaVersion(void)
+{
+#if defined(ZSTD_LZMACOMPRESS) || defined(ZSTD_LZMADECOMPRESS)
+    return lzma_version_string();
+#else
+    return "Unsupported";
+#endif
+}
+
 
 /*-*************************************
 *  Constants
@@ -75,69 +114,17 @@
 #define FNSPACE 30
 
 /* Default file permissions 0666 (modulated by umask) */
+/* Temporary restricted file permissions are used when we're going to
+ * chmod/chown at the end of the operation. */
 #if !defined(_WIN32)
 /* These macros aren't defined on windows. */
 #define DEFAULT_FILE_PERMISSIONS (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP|S_IROTH|S_IWOTH)
+#define TEMPORARY_FILE_PERMISSIONS (S_IRUSR|S_IWUSR)
 #else
 #define DEFAULT_FILE_PERMISSIONS (0666)
+#define TEMPORARY_FILE_PERMISSIONS (0600)
 #endif
 
-/*-*************************************
-*  Macros
-***************************************/
-#define KB *(1 <<10)
-#define MB *(1 <<20)
-#define GB *(1U<<30)
-#undef MAX
-#define MAX(a,b) ((a)>(b) ? (a) : (b))
-
-struct FIO_display_prefs_s {
-    int displayLevel;   /* 0 : no display;  1: errors;  2: + result + interaction + warnings;  3: + progression;  4: + information */
-    FIO_progressSetting_e progressSetting;
-};
-
-static FIO_display_prefs_t g_display_prefs = {2, FIO_ps_auto};
-
-#define DISPLAY(...)         fprintf(stderr, __VA_ARGS__)
-#define DISPLAYOUT(...)      fprintf(stdout, __VA_ARGS__)
-#define DISPLAYLEVEL(l, ...) { if (g_display_prefs.displayLevel>=l) { DISPLAY(__VA_ARGS__); } }
-
-static const U64 g_refreshRate = SEC_TO_MICRO / 6;
-static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
-
-#define READY_FOR_UPDATE() ((g_display_prefs.progressSetting != FIO_ps_never) && UTIL_clockSpanMicro(g_displayClock) > g_refreshRate)
-#define DELAY_NEXT_UPDATE() { g_displayClock = UTIL_getTime(); }
-#define DISPLAYUPDATE(l, ...) {                              \
-        if (g_display_prefs.displayLevel>=l && (g_display_prefs.progressSetting != FIO_ps_never)) { \
-            if (READY_FOR_UPDATE() || (g_display_prefs.displayLevel>=4)) { \
-                DELAY_NEXT_UPDATE();                         \
-                DISPLAY(__VA_ARGS__);                        \
-                if (g_display_prefs.displayLevel>=4) fflush(stderr);       \
-    }   }   }
-
-#undef MIN  /* in case it would be already defined */
-#define MIN(a,b)    ((a) < (b) ? (a) : (b))
-
-
-#define EXM_THROW(error, ...)                                             \
-{                                                                         \
-    DISPLAYLEVEL(1, "zstd: ");                                            \
-    DISPLAYLEVEL(5, "Error defined at %s, line %i : \n", __FILE__, __LINE__); \
-    DISPLAYLEVEL(1, "error %i : ", error);                                \
-    DISPLAYLEVEL(1, __VA_ARGS__);                                         \
-    DISPLAYLEVEL(1, " \n");                                               \
-    exit(error);                                                          \
-}
-
-#define CHECK_V(v, f)                                \
-    v = f;                                           \
-    if (ZSTD_isError(v)) {                           \
-        DISPLAYLEVEL(5, "%s \n", #f);                \
-        EXM_THROW(11, "%s", ZSTD_getErrorName(v));   \
-    }
-#define CHECK(f) { size_t err; CHECK_V(err, f); }
-
-
 /*-************************************
 *  Signal (Ctrl-C trapping)
 **************************************/
@@ -237,7 +224,7 @@ static void ABRThandler(int sig) {
 }
 #endif
 
-void FIO_addAbortHandler()
+void FIO_addAbortHandler(void)
 {
 #if BACKTRACE_ENABLE
     signal(SIGABRT, ABRThandler);
@@ -248,94 +235,6 @@ void FIO_addAbortHandler()
 #endif
 }
 
-
-/*-************************************************************
-* Avoid fseek()'s 2GiB barrier with MSVC, macOS, *BSD, MinGW
-***************************************************************/
-#if defined(_MSC_VER) && _MSC_VER >= 1400
-#   define LONG_SEEK _fseeki64
-#   define LONG_TELL _ftelli64
-#elif !defined(__64BIT__) && (PLATFORM_POSIX_VERSION >= 200112L) /* No point defining Large file for 64 bit */
-#  define LONG_SEEK fseeko
-#  define LONG_TELL ftello
-#elif defined(__MINGW32__) && !defined(__STRICT_ANSI__) && !defined(__NO_MINGW_LFS) && defined(__MSVCRT__)
-#   define LONG_SEEK fseeko64
-#   define LONG_TELL ftello64
-#elif defined(_WIN32) && !defined(__DJGPP__)
-#   include 
-    static int LONG_SEEK(FILE* file, __int64 offset, int origin) {
-        LARGE_INTEGER off;
-        DWORD method;
-        off.QuadPart = offset;
-        if (origin == SEEK_END)
-            method = FILE_END;
-        else if (origin == SEEK_CUR)
-            method = FILE_CURRENT;
-        else
-            method = FILE_BEGIN;
-
-        if (SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, NULL, method))
-            return 0;
-        else
-            return -1;
-    }
-    static __int64 LONG_TELL(FILE* file) {
-        LARGE_INTEGER off, newOff;
-        off.QuadPart = 0;
-        newOff.QuadPart = 0;
-        SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, &newOff, FILE_CURRENT);
-        return newOff.QuadPart;
-    }
-#else
-#   define LONG_SEEK fseek
-#   define LONG_TELL ftell
-#endif
-
-
-/*-*************************************
-*  Parameters: FIO_prefs_t
-***************************************/
-
-/* typedef'd to FIO_prefs_t within fileio.h */
-struct FIO_prefs_s {
-
-    /* Algorithm preferences */
-    FIO_compressionType_t compressionType;
-    U32 sparseFileSupport;   /* 0: no sparse allowed; 1: auto (file yes, stdout no); 2: force sparse */
-    int dictIDFlag;
-    int checksumFlag;
-    int blockSize;
-    int overlapLog;
-    U32 adaptiveMode;
-    U32 useRowMatchFinder;
-    int rsyncable;
-    int minAdaptLevel;
-    int maxAdaptLevel;
-    int ldmFlag;
-    int ldmHashLog;
-    int ldmMinMatch;
-    int ldmBucketSizeLog;
-    int ldmHashRateLog;
-    size_t streamSrcSize;
-    size_t targetCBlockSize;
-    int srcSizeHint;
-    int testMode;
-    ZSTD_literalCompressionMode_e literalCompressionMode;
-
-    /* IO preferences */
-    U32 removeSrcFile;
-    U32 overwrite;
-
-    /* Computation resources preferences */
-    unsigned memLimit;
-    int nbWorkers;
-
-    int excludeCompressedFiles;
-    int patchFromMode;
-    int contentSize;
-    int allowBlockDevices;
-};
-
 /*-*************************************
 *  Parameters: FIO_ctx_t
 ***************************************/
@@ -355,6 +254,18 @@ struct FIO_ctx_s {
     size_t totalBytesOutput;
 };
 
+static int FIO_shouldDisplayFileSummary(FIO_ctx_t const* fCtx)
+{
+    return fCtx->nbFilesTotal <= 1 || g_display_prefs.displayLevel >= 3;
+}
+
+static int FIO_shouldDisplayMultipleFileSummary(FIO_ctx_t const* fCtx)
+{
+    int const shouldDisplay = (fCtx->nbFilesProcessed >= 1 && fCtx->nbFilesTotal > 1);
+    assert(shouldDisplay || FIO_shouldDisplayFileSummary(fCtx) || fCtx->nbFilesProcessed == 0);
+    return shouldDisplay;
+}
+
 
 /*-*************************************
 *  Parameters: Initialization
@@ -392,9 +303,11 @@ FIO_prefs_t* FIO_createPreferences(void)
     ret->targetCBlockSize = 0;
     ret->srcSizeHint = 0;
     ret->testMode = 0;
-    ret->literalCompressionMode = ZSTD_lcm_auto;
+    ret->literalCompressionMode = ZSTD_ps_auto;
     ret->excludeCompressedFiles = 0;
     ret->allowBlockDevices = 0;
+    ret->asyncIO = AIO_supported();
+    ret->passThrough = -1;
     return ret;
 }
 
@@ -443,13 +356,13 @@ void FIO_setCompressionType(FIO_prefs_t* const prefs, FIO_compressionType_t comp
 
 void FIO_overwriteMode(FIO_prefs_t* const prefs) { prefs->overwrite = 1; }
 
-void FIO_setSparseWrite(FIO_prefs_t* const prefs, unsigned sparse) { prefs->sparseFileSupport = sparse; }
+void FIO_setSparseWrite(FIO_prefs_t* const prefs, int sparse) { prefs->sparseFileSupport = sparse; }
 
 void FIO_setDictIDFlag(FIO_prefs_t* const prefs, int dictIDFlag) { prefs->dictIDFlag = dictIDFlag; }
 
 void FIO_setChecksumFlag(FIO_prefs_t* const prefs, int checksumFlag) { prefs->checksumFlag = checksumFlag; }
 
-void FIO_setRemoveSrcFile(FIO_prefs_t* const prefs, unsigned flag) { prefs->removeSrcFile = (flag>0); }
+void FIO_setRemoveSrcFile(FIO_prefs_t* const prefs, int flag) { prefs->removeSrcFile = (flag!=0); }
 
 void FIO_setMemLimit(FIO_prefs_t* const prefs, unsigned memLimit) { prefs->memLimit = memLimit; }
 
@@ -476,7 +389,7 @@ void FIO_setOverlapLog(FIO_prefs_t* const prefs, int overlapLog){
     prefs->overlapLog = overlapLog;
 }
 
-void FIO_setAdaptiveMode(FIO_prefs_t* const prefs, unsigned adapt) {
+void FIO_setAdaptiveMode(FIO_prefs_t* const prefs, int adapt) {
     if ((adapt>0) && (prefs->nbWorkers==0))
         EXM_THROW(1, "Adaptive mode is not compatible with single thread mode \n");
     prefs->adaptiveMode = adapt;
@@ -510,7 +423,7 @@ void FIO_setTestMode(FIO_prefs_t* const prefs, int testMode) {
 
 void FIO_setLiteralCompressionMode(
         FIO_prefs_t* const prefs,
-        ZSTD_literalCompressionMode_e mode) {
+        ZSTD_paramSwitch_e mode) {
     prefs->literalCompressionMode = mode;
 }
 
@@ -558,6 +471,20 @@ void FIO_setContentSize(FIO_prefs_t* const prefs, int value)
     prefs->contentSize = value != 0;
 }
 
+void FIO_setAsyncIOFlag(FIO_prefs_t* const prefs, int value) {
+#ifdef ZSTD_MULTITHREAD
+    prefs->asyncIO = value;
+#else
+    (void) prefs;
+    (void) value;
+    DISPLAYLEVEL(2, "Note : asyncio is disabled (lack of multithreading support) \n");
+#endif
+}
+
+void FIO_setPassThroughFlag(FIO_prefs_t* const prefs, int value) {
+    prefs->passThrough = (value != 0);
+}
+
 /* FIO_ctx_t functions */
 
 void FIO_setHasStdoutOutput(FIO_ctx_t* const fCtx, int value) {
@@ -608,26 +535,26 @@ static int FIO_removeFile(const char* path)
 /** FIO_openSrcFile() :
  *  condition : `srcFileName` must be non-NULL. `prefs` may be NULL.
  * @result : FILE* to `srcFileName`, or NULL if it fails */
-static FILE* FIO_openSrcFile(const FIO_prefs_t* const prefs, const char* srcFileName)
+static FILE* FIO_openSrcFile(const FIO_prefs_t* const prefs, const char* srcFileName, stat_t* statbuf)
 {
-    stat_t statbuf;
     int allowBlockDevices = prefs != NULL ? prefs->allowBlockDevices : 0;
     assert(srcFileName != NULL);
+    assert(statbuf != NULL);
     if (!strcmp (srcFileName, stdinmark)) {
         DISPLAYLEVEL(4,"Using stdin for input \n");
         SET_BINARY_MODE(stdin);
         return stdin;
     }
 
-    if (!UTIL_stat(srcFileName, &statbuf)) {
+    if (!UTIL_stat(srcFileName, statbuf)) {
         DISPLAYLEVEL(1, "zstd: can't stat %s : %s -- ignored \n",
                         srcFileName, strerror(errno));
         return NULL;
     }
 
-    if (!UTIL_isRegularFileStat(&statbuf)
-     && !UTIL_isFIFOStat(&statbuf)
-     && !(allowBlockDevices && UTIL_isBlockDevStat(&statbuf))
+    if (!UTIL_isRegularFileStat(statbuf)
+     && !UTIL_isFIFOStat(statbuf)
+     && !(allowBlockDevices && UTIL_isBlockDevStat(statbuf))
     ) {
         DISPLAYLEVEL(1, "zstd: %s is not a regular file -- ignored \n",
                         srcFileName);
@@ -685,7 +612,7 @@ FIO_openDstFile(FIO_ctx_t* fCtx, FIO_prefs_t* const prefs,
         if (!prefs->overwrite) {
             if (g_display_prefs.displayLevel <= 1) {
                 /* No interaction possible */
-                DISPLAY("zstd: %s already exists; not overwritten  \n",
+                DISPLAYLEVEL(1, "zstd: %s already exists; not overwritten  \n",
                         dstFileName);
                 return NULL;
             }
@@ -718,6 +645,17 @@ FIO_openDstFile(FIO_ctx_t* fCtx, FIO_prefs_t* const prefs,
         if (f == NULL) {
             DISPLAYLEVEL(1, "zstd: %s: %s\n", dstFileName, strerror(errno));
         }
+        /* An increased buffer size can provide a significant performance boost on some platforms.
+         * Note that providing a NULL buf with a size that's not 0 is not defined in ANSI C, but is defined
+         * in an extension. There are three possibilities here -
+         * 1. Libc supports the extended version and everything is good.
+         * 2. Libc ignores the size when buf is NULL, in which case everything will continue as if we didn't
+         *    call `setvbuf`.
+         * 3. We fail the call and execution continues but a warning message might be shown.
+         * In all cases due execution continues. For now, I believe that this is a more cost-effective
+         * solution than managing the buffers allocations ourselves (will require an API change). */
+        if(setvbuf(f, NULL, _IOFBF, 1 MB))
+            DISPLAYLEVEL(2, "Warning: setvbuf failed for %s\n", dstFileName);
         return f;
     }
 }
@@ -728,33 +666,47 @@ FIO_openDstFile(FIO_ctx_t* fCtx, FIO_prefs_t* const prefs,
  * @return : loaded size
  *  if fileName==NULL, returns 0 and a NULL pointer
  */
-static size_t FIO_createDictBuffer(void** bufferPtr, const char* fileName, FIO_prefs_t* const prefs)
+static size_t FIO_createDictBuffer(void** bufferPtr, const char* fileName, FIO_prefs_t* const prefs, stat_t* dictFileStat)
 {
     FILE* fileHandle;
     U64 fileSize;
 
     assert(bufferPtr != NULL);
+    assert(dictFileStat != NULL);
     *bufferPtr = NULL;
     if (fileName == NULL) return 0;
 
     DISPLAYLEVEL(4,"Loading %s as dictionary \n", fileName);
+
+    if (!UTIL_stat(fileName, dictFileStat)) {
+        EXM_THROW(31, "Stat failed on dictionary file %s: %s", fileName, strerror(errno));
+    }
+
+    if (!UTIL_isRegularFileStat(dictFileStat)) {
+        EXM_THROW(32, "Dictionary %s must be a regular file.", fileName);
+    }
+
     fileHandle = fopen(fileName, "rb");
-    if (fileHandle==NULL) EXM_THROW(31, "%s: %s", fileName, strerror(errno));
 
-    fileSize = UTIL_getFileSize(fileName);
+    if (fileHandle == NULL) {
+        EXM_THROW(33, "Couldn't open dictionary %s: %s", fileName, strerror(errno));
+    }
+
+    fileSize = UTIL_getFileSizeStat(dictFileStat);
     {
         size_t const dictSizeMax = prefs->patchFromMode ? prefs->memLimit : DICTSIZE_MAX;
         if (fileSize >  dictSizeMax) {
-            EXM_THROW(32, "Dictionary file %s is too large (> %u bytes)",
+            EXM_THROW(34, "Dictionary file %s is too large (> %u bytes)",
                             fileName,  (unsigned)dictSizeMax);   /* avoid extreme cases */
         }
     }
     *bufferPtr = malloc((size_t)fileSize);
     if (*bufferPtr==NULL) EXM_THROW(34, "%s", strerror(errno));
     {   size_t const readSize = fread(*bufferPtr, 1, (size_t)fileSize, fileHandle);
-        if (readSize != fileSize)
+        if (readSize != fileSize) {
             EXM_THROW(35, "Error reading dictionary file %s : %s",
                     fileName, strerror(errno));
+        }
     }
     fclose(fileHandle);
     return (size_t)fileSize;
@@ -771,7 +723,7 @@ int FIO_checkFilenameCollisions(const char** filenameTable, unsigned nbFiles) {
 
     filenameTableSorted = (const char**) malloc(sizeof(char*) * nbFiles);
     if (!filenameTableSorted) {
-        DISPLAY("Unable to malloc new str array, not checking for name collisions\n");
+        DISPLAYLEVEL(1, "Allocation error during filename collision checking \n");
         return 1;
     }
 
@@ -788,7 +740,7 @@ int FIO_checkFilenameCollisions(const char** filenameTable, unsigned nbFiles) {
     prevElem = filenameTableSorted[0];
     for (u = 1; u < nbFiles; ++u) {
         if (strcmp(prevElem, filenameTableSorted[u]) == 0) {
-            DISPLAY("WARNING: Two files have same filename: %s\n", prevElem);
+            DISPLAYLEVEL(2, "WARNING: Two files have same filename: %s\n", prevElem);
         }
         prevElem = filenameTableSorted[u];
     }
@@ -871,46 +823,89 @@ static void FIO_adjustMemLimitForPatchFromMode(FIO_prefs_t* const prefs,
     FIO_setMemLimit(prefs, (unsigned)maxSize);
 }
 
-/* FIO_removeMultiFilesWarning() :
+/* FIO_multiFilesConcatWarning() :
+ * This function handles logic when processing multiple files with -o or -c, displaying the appropriate warnings/prompts.
  * Returns 1 if the console should abort, 0 if console should proceed.
- * This function handles logic when processing multiple files with -o, displaying the appropriate warnings/prompts.
  *
- * If -f is specified, or there is just 1 file, zstd will always proceed as usual.
- * If --rm is specified, there will be a prompt asking for user confirmation.
- *         If -f is specified with --rm, zstd will proceed as usual
- *         If -q is specified with --rm, zstd will abort pre-emptively
- *         If neither flag is specified, zstd will prompt the user for confirmation to proceed.
- * If --rm is not specified, then zstd will print a warning to the user (which can be silenced with -q).
- * However, if the output is stdout, we will always abort rather than displaying the warning prompt.
+ * If output is stdout or test mode is active, check that `--rm` disabled.
+ *
+ * If there is just 1 file to process, zstd will proceed as usual.
+ * If each file get processed into its own separate destination file, proceed as usual.
+ *
+ * When multiple files are processed into a single output,
+ * display a warning message, then disable --rm if it's set.
+ *
+ * If -f is specified or if output is stdout, just proceed.
+ * If output is set with -o, prompt for confirmation.
  */
-static int FIO_removeMultiFilesWarning(FIO_ctx_t* const fCtx, const FIO_prefs_t* const prefs, const char* outFileName, int displayLevelCutoff)
+static int FIO_multiFilesConcatWarning(const FIO_ctx_t* fCtx, FIO_prefs_t* prefs, const char* outFileName, int displayLevelCutoff)
 {
-    int error = 0;
-    if (fCtx->nbFilesTotal > 1 && !prefs->overwrite) {
-        if (g_display_prefs.displayLevel <= displayLevelCutoff) {
-            if (prefs->removeSrcFile) {
-                DISPLAYLEVEL(1, "zstd: Aborting... not deleting files and processing into dst: %s", outFileName);
-                error =  1;
-            }
-        } else {
-            if (!strcmp(outFileName, stdoutmark)) {
-                DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into stdout. ");
-            } else {
-                DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into a single output file: %s ", outFileName);
-            }
-            DISPLAYLEVEL(2, "\nThe concatenated output CANNOT regenerate the original directory tree. ")
-            if (prefs->removeSrcFile) {
-                if (fCtx->hasStdoutOutput) {
-                    DISPLAYLEVEL(1, "\nAborting. Use -f if you really want to delete the files and output to stdout");
-                    error = 1;
-                } else {
-                    error = g_display_prefs.displayLevel > displayLevelCutoff && UTIL_requireUserConfirmation("This is a destructive operation. Proceed? (y/n): ", "Aborting...", "yY", fCtx->hasStdinInput);
-                }
-            }
-        }
-        DISPLAY("\n");
+    if (fCtx->hasStdoutOutput) {
+        if (prefs->removeSrcFile)
+            /* this should not happen ; hard fail, to protect user's data
+             * note: this should rather be an assert(), but we want to be certain that user's data will not be wiped out in case it nonetheless happen */
+            EXM_THROW(43, "It's not allowed to remove input files when processed output is piped to stdout. "
+                "This scenario is not supposed to be possible. "
+                "This is a programming error. File an issue for it to be fixed.");
+    }
+    if (prefs->testMode) {
+        if (prefs->removeSrcFile)
+            /* this should not happen ; hard fail, to protect user's data
+             * note: this should rather be an assert(), but we want to be certain that user's data will not be wiped out in case it nonetheless happen */
+            EXM_THROW(43, "Test mode shall not remove input files! "
+                 "This scenario is not supposed to be possible. "
+                 "This is a programming error. File an issue for it to be fixed.");
+        return 0;
     }
-    return error;
+
+    if (fCtx->nbFilesTotal == 1) return 0;
+    assert(fCtx->nbFilesTotal > 1);
+
+    if (!outFileName) return 0;
+
+    if (fCtx->hasStdoutOutput) {
+        DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into stdout. \n");
+    } else {
+        DISPLAYLEVEL(2, "zstd: WARNING: all input files will be processed and concatenated into a single output file: %s \n", outFileName);
+    }
+    DISPLAYLEVEL(2, "The concatenated output CANNOT regenerate original file names nor directory structure. \n")
+
+    /* multi-input into single output : --rm is not allowed */
+    if (prefs->removeSrcFile) {
+        DISPLAYLEVEL(2, "Since it's a destructive operation, input files will not be removed. \n");
+        prefs->removeSrcFile = 0;
+    }
+
+    if (fCtx->hasStdoutOutput) return 0;
+    if (prefs->overwrite) return 0;
+
+    /* multiple files concatenated into single destination file using -o without -f */
+    if (g_display_prefs.displayLevel <= displayLevelCutoff) {
+        /* quiet mode => no prompt => fail automatically */
+        DISPLAYLEVEL(1, "Concatenating multiple processed inputs into a single output loses file metadata. \n");
+        DISPLAYLEVEL(1, "Aborting. \n");
+        return 1;
+    }
+    /* normal mode => prompt */
+    return UTIL_requireUserConfirmation("Proceed? (y/n): ", "Aborting...", "yY", fCtx->hasStdinInput);
+}
+
+static ZSTD_inBuffer setInBuffer(const void* buf, size_t s, size_t pos)
+{
+    ZSTD_inBuffer i;
+    i.src = buf;
+    i.size = s;
+    i.pos = pos;
+    return i;
+}
+
+static ZSTD_outBuffer setOutBuffer(void* buf, size_t s, size_t pos)
+{
+    ZSTD_outBuffer o;
+    o.dst = buf;
+    o.size = s;
+    o.pos = pos;
+    return o;
 }
 
 #ifndef ZSTD_NOCOMPRESS
@@ -919,16 +914,13 @@ static int FIO_removeMultiFilesWarning(FIO_ctx_t* const fCtx, const FIO_prefs_t*
  *  Compression
  ************************************************************************/
 typedef struct {
-    FILE* srcFile;
-    FILE* dstFile;
-    void*  srcBuffer;
-    size_t srcBufferSize;
-    void*  dstBuffer;
-    size_t dstBufferSize;
     void* dictBuffer;
     size_t dictBufferSize;
     const char* dictFileName;
+    stat_t dictFileStat;
     ZSTD_CStream* cctx;
+    WritePoolCtx_t *writeCtx;
+    ReadPoolCtx_t *readCtx;
 } cRess_t;
 
 /** ZSTD_cycleLog() :
@@ -960,9 +952,9 @@ static void FIO_adjustParamsForPatchFromMode(FIO_prefs_t* const prefs,
     if (cParams.strategy >= ZSTD_btopt) {
         DISPLAYLEVEL(1, "[Optimal parser notes] Consider the following to improve patch size at the cost of speed:\n");
         DISPLAYLEVEL(1, "- Use --single-thread mode in the zstd cli\n");
-        DISPLAYLEVEL(1, "- Set a larger targetLength (eg. --zstd=targetLength=4096)\n");
-        DISPLAYLEVEL(1, "- Set a larger chainLog (eg. --zstd=chainLog=%u)\n", ZSTD_CHAINLOG_MAX);
-        DISPLAYLEVEL(1, "Also consdier playing around with searchLog and hashLog\n");
+        DISPLAYLEVEL(1, "- Set a larger targetLength (e.g. --zstd=targetLength=4096)\n");
+        DISPLAYLEVEL(1, "- Set a larger chainLog (e.g. --zstd=chainLog=%u)\n", ZSTD_CHAINLOG_MAX);
+        DISPLAYLEVEL(1, "Also consider playing around with searchLog and hashLog\n");
     }
 }
 
@@ -977,9 +969,6 @@ static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
     if (ress.cctx == NULL)
         EXM_THROW(30, "allocation error (%s): can't create ZSTD_CCtx",
                     strerror(errno));
-    ress.srcBufferSize = ZSTD_CStreamInSize();
-    ress.srcBuffer = malloc(ress.srcBufferSize);
-    ress.dstBufferSize = ZSTD_CStreamOutSize();
 
     /* need to update memLimit before calling createDictBuffer
      * because of memLimit check inside it */
@@ -987,10 +976,10 @@ static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
         unsigned long long const ssSize = (unsigned long long)prefs->streamSrcSize;
         FIO_adjustParamsForPatchFromMode(prefs, &comprParams, UTIL_getFileSize(dictFileName), ssSize > 0 ? ssSize : maxSrcFileSize, cLevel);
     }
-    ress.dstBuffer = malloc(ress.dstBufferSize);
-    ress.dictBufferSize = FIO_createDictBuffer(&ress.dictBuffer, dictFileName, prefs);   /* works with dictFileName==NULL */
-    if (!ress.srcBuffer || !ress.dstBuffer)
-        EXM_THROW(31, "allocation error : not enough memory");
+    ress.dictBufferSize = FIO_createDictBuffer(&ress.dictBuffer, dictFileName, prefs, &ress.dictFileStat);   /* works with dictFileName==NULL */
+
+    ress.writeCtx = AIO_WritePool_create(prefs, ZSTD_CStreamOutSize());
+    ress.readCtx = AIO_ReadPool_create(prefs, ZSTD_CStreamInSize());
 
     /* Advanced parameters, including dictionary */
     if (dictFileName && (ress.dictBuffer==NULL))
@@ -1053,9 +1042,9 @@ static cRess_t FIO_createCResources(FIO_prefs_t* const prefs,
 
 static void FIO_freeCResources(const cRess_t* const ress)
 {
-    free(ress->srcBuffer);
-    free(ress->dstBuffer);
     free(ress->dictBuffer);
+    AIO_WritePool_free(ress->writeCtx);
+    AIO_ReadPool_free(ress->readCtx);
     ZSTD_freeCStream(ress->cctx);   /* never fails */
 }
 
@@ -1068,6 +1057,7 @@ FIO_compressGzFrame(const cRess_t* ress,  /* buffers & handlers are used, but no
 {
     unsigned long long inFileSize = 0, outFileSize = 0;
     z_stream strm;
+    IOJob_t *writeJob = NULL;
 
     if (compressionLevel > Z_BEST_COMPRESSION)
         compressionLevel = Z_BEST_COMPRESSION;
@@ -1078,56 +1068,65 @@ FIO_compressGzFrame(const cRess_t* ress,  /* buffers & handlers are used, but no
 
     {   int const ret = deflateInit2(&strm, compressionLevel, Z_DEFLATED,
                         15 /* maxWindowLogSize */ + 16 /* gzip only */,
-                        8, Z_DEFAULT_STRATEGY); /* see http://www.zlib.net/manual.html */
+                        8, Z_DEFAULT_STRATEGY); /* see https://www.zlib.net/manual.html */
         if (ret != Z_OK) {
             EXM_THROW(71, "zstd: %s: deflateInit2 error %d \n", srcFileName, ret);
     }   }
 
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
     strm.next_in = 0;
     strm.avail_in = 0;
-    strm.next_out = (Bytef*)ress->dstBuffer;
-    strm.avail_out = (uInt)ress->dstBufferSize;
+    strm.next_out = (Bytef*)writeJob->buffer;
+    strm.avail_out = (uInt)writeJob->bufferSize;
 
     while (1) {
         int ret;
         if (strm.avail_in == 0) {
-            size_t const inSize = fread(ress->srcBuffer, 1, ress->srcBufferSize, ress->srcFile);
-            if (inSize == 0) break;
-            inFileSize += inSize;
-            strm.next_in = (z_const unsigned char*)ress->srcBuffer;
-            strm.avail_in = (uInt)inSize;
+            AIO_ReadPool_fillBuffer(ress->readCtx, ZSTD_CStreamInSize());
+            if (ress->readCtx->srcBufferLoaded == 0) break;
+            inFileSize += ress->readCtx->srcBufferLoaded;
+            strm.next_in = (z_const unsigned char*)ress->readCtx->srcBuffer;
+            strm.avail_in = (uInt)ress->readCtx->srcBufferLoaded;
         }
-        ret = deflate(&strm, Z_NO_FLUSH);
+
+        {
+            size_t const availBefore = strm.avail_in;
+            ret = deflate(&strm, Z_NO_FLUSH);
+            AIO_ReadPool_consumeBytes(ress->readCtx, availBefore - strm.avail_in);
+        }
+
         if (ret != Z_OK)
             EXM_THROW(72, "zstd: %s: deflate error %d \n", srcFileName, ret);
-        {   size_t const cSize = ress->dstBufferSize - strm.avail_out;
+        {   size_t const cSize = writeJob->bufferSize - strm.avail_out;
             if (cSize) {
-                if (fwrite(ress->dstBuffer, 1, cSize, ress->dstFile) != cSize)
-                    EXM_THROW(73, "Write error : cannot write to output file : %s ", strerror(errno));
+                writeJob->usedBufferSize = cSize;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 outFileSize += cSize;
-                strm.next_out = (Bytef*)ress->dstBuffer;
-                strm.avail_out = (uInt)ress->dstBufferSize;
-        }   }
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = (uInt)writeJob->bufferSize;
+            }   }
         if (srcFileSize == UTIL_FILESIZE_UNKNOWN) {
-            DISPLAYUPDATE(2, "\rRead : %u MB ==> %.2f%% ",
-                            (unsigned)(inFileSize>>20),
-                            (double)outFileSize/inFileSize*100)
+            DISPLAYUPDATE_PROGRESS(
+                    "\rRead : %u MB ==> %.2f%% ",
+                    (unsigned)(inFileSize>>20),
+                    (double)outFileSize/(double)inFileSize*100)
         } else {
-            DISPLAYUPDATE(2, "\rRead : %u / %u MB ==> %.2f%% ",
-                            (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
-                            (double)outFileSize/inFileSize*100);
+            DISPLAYUPDATE_PROGRESS(
+                    "\rRead : %u / %u MB ==> %.2f%% ",
+                    (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
+                    (double)outFileSize/(double)inFileSize*100);
     }   }
 
     while (1) {
         int const ret = deflate(&strm, Z_FINISH);
-        {   size_t const cSize = ress->dstBufferSize - strm.avail_out;
+        {   size_t const cSize = writeJob->bufferSize - strm.avail_out;
             if (cSize) {
-                if (fwrite(ress->dstBuffer, 1, cSize, ress->dstFile) != cSize)
-                    EXM_THROW(75, "Write error : %s ", strerror(errno));
+                writeJob->usedBufferSize = cSize;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 outFileSize += cSize;
-                strm.next_out = (Bytef*)ress->dstBuffer;
-                strm.avail_out = (uInt)ress->dstBufferSize;
-        }   }
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = (uInt)writeJob->bufferSize;
+            }   }
         if (ret == Z_STREAM_END) break;
         if (ret != Z_BUF_ERROR)
             EXM_THROW(77, "zstd: %s: deflate error %d \n", srcFileName, ret);
@@ -1138,6 +1137,8 @@ FIO_compressGzFrame(const cRess_t* ress,  /* buffers & handlers are used, but no
             EXM_THROW(79, "zstd: %s: deflateEnd error %d \n", srcFileName, ret);
     }   }
     *readsize = inFileSize;
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
     return outFileSize;
 }
 #endif
@@ -1153,6 +1154,7 @@ FIO_compressLzmaFrame(cRess_t* ress,
     lzma_stream strm = LZMA_STREAM_INIT;
     lzma_action action = LZMA_RUN;
     lzma_ret ret;
+    IOJob_t *writeJob = NULL;
 
     if (compressionLevel < 0) compressionLevel = 0;
     if (compressionLevel > 9) compressionLevel = 9;
@@ -1170,46 +1172,55 @@ FIO_compressLzmaFrame(cRess_t* ress,
             EXM_THROW(83, "zstd: %s: lzma_easy_encoder error %d", srcFileName, ret);
     }
 
+    writeJob =AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_out = (Bytef*)writeJob->buffer;
+    strm.avail_out = (uInt)writeJob->bufferSize;
     strm.next_in = 0;
     strm.avail_in = 0;
-    strm.next_out = (BYTE*)ress->dstBuffer;
-    strm.avail_out = ress->dstBufferSize;
 
     while (1) {
         if (strm.avail_in == 0) {
-            size_t const inSize = fread(ress->srcBuffer, 1, ress->srcBufferSize, ress->srcFile);
-            if (inSize == 0) action = LZMA_FINISH;
+            size_t const inSize = AIO_ReadPool_fillBuffer(ress->readCtx, ZSTD_CStreamInSize());
+            if (ress->readCtx->srcBufferLoaded == 0) action = LZMA_FINISH;
             inFileSize += inSize;
-            strm.next_in = (BYTE const*)ress->srcBuffer;
-            strm.avail_in = inSize;
+            strm.next_in = (BYTE const*)ress->readCtx->srcBuffer;
+            strm.avail_in = ress->readCtx->srcBufferLoaded;
+        }
+
+        {
+            size_t const availBefore = strm.avail_in;
+            ret = lzma_code(&strm, action);
+            AIO_ReadPool_consumeBytes(ress->readCtx, availBefore - strm.avail_in);
         }
 
-        ret = lzma_code(&strm, action);
 
         if (ret != LZMA_OK && ret != LZMA_STREAM_END)
             EXM_THROW(84, "zstd: %s: lzma_code encoding error %d", srcFileName, ret);
-        {   size_t const compBytes = ress->dstBufferSize - strm.avail_out;
+        {   size_t const compBytes = writeJob->bufferSize - strm.avail_out;
             if (compBytes) {
-                if (fwrite(ress->dstBuffer, 1, compBytes, ress->dstFile) != compBytes)
-                    EXM_THROW(85, "Write error : %s", strerror(errno));
+                writeJob->usedBufferSize = compBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 outFileSize += compBytes;
-                strm.next_out = (BYTE*)ress->dstBuffer;
-                strm.avail_out = ress->dstBufferSize;
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = writeJob->bufferSize;
         }   }
         if (srcFileSize == UTIL_FILESIZE_UNKNOWN)
-            DISPLAYUPDATE(2, "\rRead : %u MB ==> %.2f%%",
+            DISPLAYUPDATE_PROGRESS("\rRead : %u MB ==> %.2f%%",
                             (unsigned)(inFileSize>>20),
-                            (double)outFileSize/inFileSize*100)
+                            (double)outFileSize/(double)inFileSize*100)
         else
-            DISPLAYUPDATE(2, "\rRead : %u / %u MB ==> %.2f%%",
+            DISPLAYUPDATE_PROGRESS("\rRead : %u / %u MB ==> %.2f%%",
                             (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
-                            (double)outFileSize/inFileSize*100);
+                            (double)outFileSize/(double)inFileSize*100);
         if (ret == LZMA_STREAM_END) break;
     }
 
     lzma_end(&strm);
     *readsize = inFileSize;
 
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
+
     return outFileSize;
 }
 #endif
@@ -1235,15 +1246,18 @@ FIO_compressLz4Frame(cRess_t* ress,
     LZ4F_preferences_t prefs;
     LZ4F_compressionContext_t ctx;
 
+    IOJob_t* writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+
     LZ4F_errorCode_t const errorCode = LZ4F_createCompressionContext(&ctx, LZ4F_VERSION);
     if (LZ4F_isError(errorCode))
         EXM_THROW(31, "zstd: failed to create lz4 compression context");
 
     memset(&prefs, 0, sizeof(prefs));
 
-    assert(blockSize <= ress->srcBufferSize);
+    assert(blockSize <= ress->readCtx->base.jobBufferSize);
 
-    prefs.autoFlush = 1;
+    /* autoflush off to mitigate a bug in lz4<=1.9.3 for compression level 12 */
+    prefs.autoFlush = 0;
     prefs.compressionLevel = compressionLevel;
     prefs.frameInfo.blockMode = LZ4F_blockLinked;
     prefs.frameInfo.blockSizeID = LZ4F_max64KB;
@@ -1251,75 +1265,68 @@ FIO_compressLz4Frame(cRess_t* ress,
 #if LZ4_VERSION_NUMBER >= 10600
     prefs.frameInfo.contentSize = (srcFileSize==UTIL_FILESIZE_UNKNOWN) ? 0 : srcFileSize;
 #endif
-    assert(LZ4F_compressBound(blockSize, &prefs) <= ress->dstBufferSize);
+    assert(LZ4F_compressBound(blockSize, &prefs) <= writeJob->bufferSize);
 
     {
-        size_t readSize;
-        size_t headerSize = LZ4F_compressBegin(ctx, ress->dstBuffer, ress->dstBufferSize, &prefs);
+        size_t headerSize = LZ4F_compressBegin(ctx, writeJob->buffer, writeJob->bufferSize, &prefs);
         if (LZ4F_isError(headerSize))
             EXM_THROW(33, "File header generation failed : %s",
                             LZ4F_getErrorName(headerSize));
-        if (fwrite(ress->dstBuffer, 1, headerSize, ress->dstFile) != headerSize)
-            EXM_THROW(34, "Write error : %s (cannot write header)", strerror(errno));
+        writeJob->usedBufferSize = headerSize;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
         outFileSize += headerSize;
 
         /* Read first block */
-        readSize  = fread(ress->srcBuffer, (size_t)1, (size_t)blockSize, ress->srcFile);
-        inFileSize += readSize;
+        inFileSize += AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
 
         /* Main Loop */
-        while (readSize>0) {
-            size_t const outSize = LZ4F_compressUpdate(ctx,
-                                        ress->dstBuffer, ress->dstBufferSize,
-                                        ress->srcBuffer, readSize, NULL);
+        while (ress->readCtx->srcBufferLoaded) {
+            size_t inSize = MIN(blockSize, ress->readCtx->srcBufferLoaded);
+            size_t const outSize = LZ4F_compressUpdate(ctx, writeJob->buffer, writeJob->bufferSize,
+                                                       ress->readCtx->srcBuffer, inSize, NULL);
             if (LZ4F_isError(outSize))
                 EXM_THROW(35, "zstd: %s: lz4 compression failed : %s",
                             srcFileName, LZ4F_getErrorName(outSize));
             outFileSize += outSize;
             if (srcFileSize == UTIL_FILESIZE_UNKNOWN) {
-                DISPLAYUPDATE(2, "\rRead : %u MB ==> %.2f%%",
+                DISPLAYUPDATE_PROGRESS("\rRead : %u MB ==> %.2f%%",
                                 (unsigned)(inFileSize>>20),
-                                (double)outFileSize/inFileSize*100)
+                                (double)outFileSize/(double)inFileSize*100)
             } else {
-                DISPLAYUPDATE(2, "\rRead : %u / %u MB ==> %.2f%%",
+                DISPLAYUPDATE_PROGRESS("\rRead : %u / %u MB ==> %.2f%%",
                                 (unsigned)(inFileSize>>20), (unsigned)(srcFileSize>>20),
-                                (double)outFileSize/inFileSize*100);
+                                (double)outFileSize/(double)inFileSize*100);
             }
 
             /* Write Block */
-            {   size_t const sizeCheck = fwrite(ress->dstBuffer, 1, outSize, ress->dstFile);
-                if (sizeCheck != outSize)
-                    EXM_THROW(36, "Write error : %s", strerror(errno));
-            }
+            writeJob->usedBufferSize = outSize;
+            AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
 
             /* Read next block */
-            readSize  = fread(ress->srcBuffer, (size_t)1, (size_t)blockSize, ress->srcFile);
-            inFileSize += readSize;
+            AIO_ReadPool_consumeBytes(ress->readCtx, inSize);
+            inFileSize += AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
         }
-        if (ferror(ress->srcFile)) EXM_THROW(37, "Error reading %s ", srcFileName);
 
         /* End of Stream mark */
-        headerSize = LZ4F_compressEnd(ctx, ress->dstBuffer, ress->dstBufferSize, NULL);
+        headerSize = LZ4F_compressEnd(ctx, writeJob->buffer, writeJob->bufferSize, NULL);
         if (LZ4F_isError(headerSize))
             EXM_THROW(38, "zstd: %s: lz4 end of file generation failed : %s",
                         srcFileName, LZ4F_getErrorName(headerSize));
 
-        {   size_t const sizeCheck = fwrite(ress->dstBuffer, 1, headerSize, ress->dstFile);
-            if (sizeCheck != headerSize)
-                EXM_THROW(39, "Write error : %s (cannot write end of stream)",
-                            strerror(errno));
-        }
+        writeJob->usedBufferSize = headerSize;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
         outFileSize += headerSize;
     }
 
     *readsize = inFileSize;
     LZ4F_freeCompressionContext(ctx);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
 
     return outFileSize;
 }
 #endif
 
-
 static unsigned long long
 FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
                       FIO_prefs_t* const prefs,
@@ -1328,10 +1335,11 @@ FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
                       int compressionLevel, U64* readsize)
 {
     cRess_t const ress = *ressPtr;
-    FILE* const srcFile = ress.srcFile;
-    FILE* const dstFile = ress.dstFile;
+    IOJob_t *writeJob = AIO_WritePool_acquireJob(ressPtr->writeCtx);
+
     U64 compressedfilesize = 0;
     ZSTD_EndDirective directive = ZSTD_e_continue;
+    U64 pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
 
     /* stats */
     ZSTD_frameProgression previous_zfp_update = { 0, 0, 0, 0, 0, 0 };
@@ -1342,28 +1350,51 @@ FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
     unsigned inputPresented = 0;
     unsigned inputBlocked = 0;
     unsigned lastJobID = 0;
+    UTIL_time_t lastAdaptTime = UTIL_getTime();
+    U64 const adaptEveryMicro = REFRESH_RATE;
+
+    UTIL_HumanReadableSize_t const file_hrs = UTIL_makeHumanReadableSize(fileSize);
 
     DISPLAYLEVEL(6, "compression using zstd format \n");
 
     /* init */
     if (fileSize != UTIL_FILESIZE_UNKNOWN) {
+        pledgedSrcSize = fileSize;
         CHECK(ZSTD_CCtx_setPledgedSrcSize(ress.cctx, fileSize));
     } else if (prefs->streamSrcSize > 0) {
       /* unknown source size; use the declared stream size */
+      pledgedSrcSize = prefs->streamSrcSize;
       CHECK( ZSTD_CCtx_setPledgedSrcSize(ress.cctx, prefs->streamSrcSize) );
     }
+
+    {
+        int windowLog;
+        UTIL_HumanReadableSize_t windowSize;
+        CHECK(ZSTD_CCtx_getParameter(ress.cctx, ZSTD_c_windowLog, &windowLog));
+        if (windowLog == 0) {
+            if (prefs->ldmFlag) {
+                /* If long mode is set without a window size libzstd will set this size internally */
+                windowLog = ZSTD_WINDOWLOG_LIMIT_DEFAULT;
+            } else {
+                const ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, fileSize, 0);
+                windowLog = (int)cParams.windowLog;
+            }
+        }
+        windowSize = UTIL_makeHumanReadableSize(MAX(1ULL, MIN(1ULL << windowLog, pledgedSrcSize)));
+        DISPLAYLEVEL(4, "Decompression will require %.*f%s of memory\n", windowSize.precision, windowSize.value, windowSize.suffix);
+    }
     (void)srcFileName;
 
     /* Main compression loop */
     do {
         size_t stillToFlush;
         /* Fill input Buffer */
-        size_t const inSize = fread(ress.srcBuffer, (size_t)1, ress.srcBufferSize, srcFile);
-        ZSTD_inBuffer inBuff = { ress.srcBuffer, inSize, 0 };
+        size_t const inSize = AIO_ReadPool_fillBuffer(ress.readCtx, ZSTD_CStreamInSize());
+        ZSTD_inBuffer inBuff = setInBuffer( ress.readCtx->srcBuffer, ress.readCtx->srcBufferLoaded, 0 );
         DISPLAYLEVEL(6, "fread %u bytes from source \n", (unsigned)inSize);
         *readsize += inSize;
 
-        if ((inSize == 0) || (*readsize == fileSize))
+        if ((ress.readCtx->srcBufferLoaded == 0) || (*readsize == fileSize))
             directive = ZSTD_e_end;
 
         stillToFlush = 1;
@@ -1371,9 +1402,10 @@ FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
             || (directive == ZSTD_e_end && stillToFlush != 0) ) {
 
             size_t const oldIPos = inBuff.pos;
-            ZSTD_outBuffer outBuff = { ress.dstBuffer, ress.dstBufferSize, 0 };
+            ZSTD_outBuffer outBuff = setOutBuffer( writeJob->buffer, writeJob->bufferSize, 0 );
             size_t const toFlushNow = ZSTD_toFlushNow(ress.cctx);
             CHECK_V(stillToFlush, ZSTD_compressStream2(ress.cctx, &outBuff, &inBuff, directive));
+            AIO_ReadPool_consumeBytes(ress.readCtx, inBuff.pos - oldIPos);
 
             /* count stats */
             inputPresented++;
@@ -1382,148 +1414,155 @@ FIO_compressZstdFrame(FIO_ctx_t* const fCtx,
 
             /* Write compressed stream */
             DISPLAYLEVEL(6, "ZSTD_compress_generic(end:%u) => input pos(%u)<=(%u)size ; output generated %u bytes \n",
-                            (unsigned)directive, (unsigned)inBuff.pos, (unsigned)inBuff.size, (unsigned)outBuff.pos);
+                         (unsigned)directive, (unsigned)inBuff.pos, (unsigned)inBuff.size, (unsigned)outBuff.pos);
             if (outBuff.pos) {
-                size_t const sizeCheck = fwrite(ress.dstBuffer, 1, outBuff.pos, dstFile);
-                if (sizeCheck != outBuff.pos)
-                    EXM_THROW(25, "Write error : %s (cannot write compressed block)",
-                                    strerror(errno));
+                writeJob->usedBufferSize = outBuff.pos;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 compressedfilesize += outBuff.pos;
             }
 
-            /* display notification; and adapt compression level */
-            if (READY_FOR_UPDATE()) {
+            /* adaptive mode : statistics measurement and speed correction */
+            if (prefs->adaptiveMode && UTIL_clockSpanMicro(lastAdaptTime) > adaptEveryMicro) {
+                ZSTD_frameProgression const zfp = ZSTD_getFrameProgression(ress.cctx);
+
+                lastAdaptTime = UTIL_getTime();
+
+                /* check output speed */
+                if (zfp.currentJobID > 1) {  /* only possible if nbWorkers >= 1 */
+
+                    unsigned long long newlyProduced = zfp.produced - previous_zfp_update.produced;
+                    unsigned long long newlyFlushed = zfp.flushed - previous_zfp_update.flushed;
+                    assert(zfp.produced >= previous_zfp_update.produced);
+                    assert(prefs->nbWorkers >= 1);
+
+                    /* test if compression is blocked
+                        * either because output is slow and all buffers are full
+                        * or because input is slow and no job can start while waiting for at least one buffer to be filled.
+                        * note : exclude starting part, since currentJobID > 1 */
+                    if ( (zfp.consumed == previous_zfp_update.consumed)   /* no data compressed : no data available, or no more buffer to compress to, OR compression is really slow (compression of a single block is slower than update rate)*/
+                        && (zfp.nbActiveWorkers == 0)                       /* confirmed : no compression ongoing */
+                        ) {
+                        DISPLAYLEVEL(6, "all buffers full : compression stopped => slow down \n")
+                        speedChange = slower;
+                    }
+
+                    previous_zfp_update = zfp;
+
+                    if ( (newlyProduced > (newlyFlushed * 9 / 8))   /* compression produces more data than output can flush (though production can be spiky, due to work unit : (N==4)*block sizes) */
+                        && (flushWaiting == 0)                        /* flush speed was never slowed by lack of production, so it's operating at max capacity */
+                        ) {
+                        DISPLAYLEVEL(6, "compression faster than flush (%llu > %llu), and flushed was never slowed down by lack of production => slow down \n", newlyProduced, newlyFlushed);
+                        speedChange = slower;
+                    }
+                    flushWaiting = 0;
+                }
+
+                /* course correct only if there is at least one new job completed */
+                if (zfp.currentJobID > lastJobID) {
+                    DISPLAYLEVEL(6, "compression level adaptation check \n")
+
+                    /* check input speed */
+                    if (zfp.currentJobID > (unsigned)(prefs->nbWorkers+1)) {   /* warm up period, to fill all workers */
+                        if (inputBlocked <= 0) {
+                            DISPLAYLEVEL(6, "input is never blocked => input is slower than ingestion \n");
+                            speedChange = slower;
+                        } else if (speedChange == noChange) {
+                            unsigned long long newlyIngested = zfp.ingested - previous_zfp_correction.ingested;
+                            unsigned long long newlyConsumed = zfp.consumed - previous_zfp_correction.consumed;
+                            unsigned long long newlyProduced = zfp.produced - previous_zfp_correction.produced;
+                            unsigned long long newlyFlushed  = zfp.flushed  - previous_zfp_correction.flushed;
+                            previous_zfp_correction = zfp;
+                            assert(inputPresented > 0);
+                            DISPLAYLEVEL(6, "input blocked %u/%u(%.2f) - ingested:%u vs %u:consumed - flushed:%u vs %u:produced \n",
+                                            inputBlocked, inputPresented, (double)inputBlocked/inputPresented*100,
+                                            (unsigned)newlyIngested, (unsigned)newlyConsumed,
+                                            (unsigned)newlyFlushed, (unsigned)newlyProduced);
+                            if ( (inputBlocked > inputPresented / 8)     /* input is waiting often, because input buffers is full : compression or output too slow */
+                                && (newlyFlushed * 33 / 32 > newlyProduced)  /* flush everything that is produced */
+                                && (newlyIngested * 33 / 32 > newlyConsumed) /* input speed as fast or faster than compression speed */
+                            ) {
+                                DISPLAYLEVEL(6, "recommend faster as in(%llu) >= (%llu)comp(%llu) <= out(%llu) \n",
+                                                newlyIngested, newlyConsumed, newlyProduced, newlyFlushed);
+                                speedChange = faster;
+                            }
+                        }
+                        inputBlocked = 0;
+                        inputPresented = 0;
+                    }
+
+                    if (speedChange == slower) {
+                        DISPLAYLEVEL(6, "slower speed , higher compression \n")
+                        compressionLevel ++;
+                        if (compressionLevel > ZSTD_maxCLevel()) compressionLevel = ZSTD_maxCLevel();
+                        if (compressionLevel > prefs->maxAdaptLevel) compressionLevel = prefs->maxAdaptLevel;
+                        compressionLevel += (compressionLevel == 0);   /* skip 0 */
+                        ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, compressionLevel);
+                    }
+                    if (speedChange == faster) {
+                        DISPLAYLEVEL(6, "faster speed , lighter compression \n")
+                        compressionLevel --;
+                        if (compressionLevel < prefs->minAdaptLevel) compressionLevel = prefs->minAdaptLevel;
+                        compressionLevel -= (compressionLevel == 0);   /* skip 0 */
+                        ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, compressionLevel);
+                    }
+                    speedChange = noChange;
+
+                    lastJobID = zfp.currentJobID;
+                }  /* if (zfp.currentJobID > lastJobID) */
+            } /* if (prefs->adaptiveMode && UTIL_clockSpanMicro(lastAdaptTime) > adaptEveryMicro) */
+
+            /* display notification */
+            if (SHOULD_DISPLAY_PROGRESS() && READY_FOR_UPDATE()) {
                 ZSTD_frameProgression const zfp = ZSTD_getFrameProgression(ress.cctx);
                 double const cShare = (double)zfp.produced / (double)(zfp.consumed + !zfp.consumed/*avoid div0*/) * 100;
+                UTIL_HumanReadableSize_t const buffered_hrs = UTIL_makeHumanReadableSize(zfp.ingested - zfp.consumed);
+                UTIL_HumanReadableSize_t const consumed_hrs = UTIL_makeHumanReadableSize(zfp.consumed);
+                UTIL_HumanReadableSize_t const produced_hrs = UTIL_makeHumanReadableSize(zfp.produced);
+
+                DELAY_NEXT_UPDATE();
 
                 /* display progress notifications */
+                DISPLAY_PROGRESS("\r%79s\r", "");    /* Clear out the current displayed line */
                 if (g_display_prefs.displayLevel >= 3) {
-                    DISPLAYUPDATE(3, "\r(L%i) Buffered :%4u MB - Consumed :%4u MB - Compressed :%4u MB => %.2f%% ",
-                                compressionLevel,
-                                (unsigned)((zfp.ingested - zfp.consumed) >> 20),
-                                (unsigned)(zfp.consumed >> 20),
-                                (unsigned)(zfp.produced >> 20),
-                                cShare );
-                } else if (g_display_prefs.displayLevel >= 2 || g_display_prefs.progressSetting == FIO_ps_always) {
+                    /* Verbose progress update */
+                    DISPLAY_PROGRESS(
+                            "(L%i) Buffered:%5.*f%s - Consumed:%5.*f%s - Compressed:%5.*f%s => %.2f%% ",
+                            compressionLevel,
+                            buffered_hrs.precision, buffered_hrs.value, buffered_hrs.suffix,
+                            consumed_hrs.precision, consumed_hrs.value, consumed_hrs.suffix,
+                            produced_hrs.precision, produced_hrs.value, produced_hrs.suffix,
+                            cShare );
+                } else {
                     /* Require level 2 or forcibly displayed progress counter for summarized updates */
-                    DISPLAYLEVEL(1, "\r%79s\r", "");    /* Clear out the current displayed line */
                     if (fCtx->nbFilesTotal > 1) {
                         size_t srcFileNameSize = strlen(srcFileName);
                         /* Ensure that the string we print is roughly the same size each time */
                         if (srcFileNameSize > 18) {
                             const char* truncatedSrcFileName = srcFileName + srcFileNameSize - 15;
-                            DISPLAYLEVEL(1, "Compress: %u/%u files. Current: ...%s ",
+                            DISPLAY_PROGRESS("Compress: %u/%u files. Current: ...%s ",
                                         fCtx->currFileIdx+1, fCtx->nbFilesTotal, truncatedSrcFileName);
                         } else {
-                            DISPLAYLEVEL(1, "Compress: %u/%u files. Current: %*s ",
+                            DISPLAY_PROGRESS("Compress: %u/%u files. Current: %*s ",
                                         fCtx->currFileIdx+1, fCtx->nbFilesTotal, (int)(18-srcFileNameSize), srcFileName);
                         }
                     }
-                    DISPLAYLEVEL(1, "Read : %2u ", (unsigned)(zfp.consumed >> 20));
+                    DISPLAY_PROGRESS("Read:%6.*f%4s ", consumed_hrs.precision, consumed_hrs.value, consumed_hrs.suffix);
                     if (fileSize != UTIL_FILESIZE_UNKNOWN)
-                        DISPLAYLEVEL(2, "/ %2u ", (unsigned)(fileSize >> 20));
-                    DISPLAYLEVEL(1, "MB ==> %2.f%%", cShare);
-                    DELAY_NEXT_UPDATE();
+                        DISPLAY_PROGRESS("/%6.*f%4s", file_hrs.precision, file_hrs.value, file_hrs.suffix);
+                    DISPLAY_PROGRESS(" ==> %2.f%%", cShare);
                 }
-
-                /* adaptive mode : statistics measurement and speed correction */
-                if (prefs->adaptiveMode) {
-
-                    /* check output speed */
-                    if (zfp.currentJobID > 1) {  /* only possible if nbWorkers >= 1 */
-
-                        unsigned long long newlyProduced = zfp.produced - previous_zfp_update.produced;
-                        unsigned long long newlyFlushed = zfp.flushed - previous_zfp_update.flushed;
-                        assert(zfp.produced >= previous_zfp_update.produced);
-                        assert(prefs->nbWorkers >= 1);
-
-                        /* test if compression is blocked
-                         * either because output is slow and all buffers are full
-                         * or because input is slow and no job can start while waiting for at least one buffer to be filled.
-                         * note : exclude starting part, since currentJobID > 1 */
-                        if ( (zfp.consumed == previous_zfp_update.consumed)   /* no data compressed : no data available, or no more buffer to compress to, OR compression is really slow (compression of a single block is slower than update rate)*/
-                          && (zfp.nbActiveWorkers == 0)                       /* confirmed : no compression ongoing */
-                          ) {
-                            DISPLAYLEVEL(6, "all buffers full : compression stopped => slow down \n")
-                            speedChange = slower;
-                        }
-
-                        previous_zfp_update = zfp;
-
-                        if ( (newlyProduced > (newlyFlushed * 9 / 8))   /* compression produces more data than output can flush (though production can be spiky, due to work unit : (N==4)*block sizes) */
-                          && (flushWaiting == 0)                        /* flush speed was never slowed by lack of production, so it's operating at max capacity */
-                          ) {
-                            DISPLAYLEVEL(6, "compression faster than flush (%llu > %llu), and flushed was never slowed down by lack of production => slow down \n", newlyProduced, newlyFlushed);
-                            speedChange = slower;
-                        }
-                        flushWaiting = 0;
-                    }
-
-                    /* course correct only if there is at least one new job completed */
-                    if (zfp.currentJobID > lastJobID) {
-                        DISPLAYLEVEL(6, "compression level adaptation check \n")
-
-                        /* check input speed */
-                        if (zfp.currentJobID > (unsigned)(prefs->nbWorkers+1)) {   /* warm up period, to fill all workers */
-                            if (inputBlocked <= 0) {
-                                DISPLAYLEVEL(6, "input is never blocked => input is slower than ingestion \n");
-                                speedChange = slower;
-                            } else if (speedChange == noChange) {
-                                unsigned long long newlyIngested = zfp.ingested - previous_zfp_correction.ingested;
-                                unsigned long long newlyConsumed = zfp.consumed - previous_zfp_correction.consumed;
-                                unsigned long long newlyProduced = zfp.produced - previous_zfp_correction.produced;
-                                unsigned long long newlyFlushed  = zfp.flushed  - previous_zfp_correction.flushed;
-                                previous_zfp_correction = zfp;
-                                assert(inputPresented > 0);
-                                DISPLAYLEVEL(6, "input blocked %u/%u(%.2f) - ingested:%u vs %u:consumed - flushed:%u vs %u:produced \n",
-                                                inputBlocked, inputPresented, (double)inputBlocked/inputPresented*100,
-                                                (unsigned)newlyIngested, (unsigned)newlyConsumed,
-                                                (unsigned)newlyFlushed, (unsigned)newlyProduced);
-                                if ( (inputBlocked > inputPresented / 8)     /* input is waiting often, because input buffers is full : compression or output too slow */
-                                  && (newlyFlushed * 33 / 32 > newlyProduced)  /* flush everything that is produced */
-                                  && (newlyIngested * 33 / 32 > newlyConsumed) /* input speed as fast or faster than compression speed */
-                                ) {
-                                    DISPLAYLEVEL(6, "recommend faster as in(%llu) >= (%llu)comp(%llu) <= out(%llu) \n",
-                                                    newlyIngested, newlyConsumed, newlyProduced, newlyFlushed);
-                                    speedChange = faster;
-                                }
-                            }
-                            inputBlocked = 0;
-                            inputPresented = 0;
-                        }
-
-                        if (speedChange == slower) {
-                            DISPLAYLEVEL(6, "slower speed , higher compression \n")
-                            compressionLevel ++;
-                            if (compressionLevel > ZSTD_maxCLevel()) compressionLevel = ZSTD_maxCLevel();
-                            if (compressionLevel > prefs->maxAdaptLevel) compressionLevel = prefs->maxAdaptLevel;
-                            compressionLevel += (compressionLevel == 0);   /* skip 0 */
-                            ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, compressionLevel);
-                        }
-                        if (speedChange == faster) {
-                            DISPLAYLEVEL(6, "faster speed , lighter compression \n")
-                            compressionLevel --;
-                            if (compressionLevel < prefs->minAdaptLevel) compressionLevel = prefs->minAdaptLevel;
-                            compressionLevel -= (compressionLevel == 0);   /* skip 0 */
-                            ZSTD_CCtx_setParameter(ress.cctx, ZSTD_c_compressionLevel, compressionLevel);
-                        }
-                        speedChange = noChange;
-
-                        lastJobID = zfp.currentJobID;
-                    }  /* if (zfp.currentJobID > lastJobID) */
-                }  /* if (g_adaptiveMode) */
-            }  /* if (READY_FOR_UPDATE()) */
+            }  /* if (SHOULD_DISPLAY_PROGRESS() && READY_FOR_UPDATE()) */
         }  /* while ((inBuff.pos != inBuff.size) */
     } while (directive != ZSTD_e_end);
 
-    if (ferror(srcFile)) {
-        EXM_THROW(26, "Read error : I/O error");
-    }
     if (fileSize != UTIL_FILESIZE_UNKNOWN && *readsize != fileSize) {
         EXM_THROW(27, "Read error : Incomplete read : %llu / %llu B",
                 (unsigned long long)*readsize, (unsigned long long)fileSize);
     }
 
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ressPtr->writeCtx);
+
     return compressedfilesize;
 }
 
@@ -1588,20 +1627,22 @@ FIO_compressFilename_internal(FIO_ctx_t* const fCtx,
     /* Status */
     fCtx->totalBytesInput += (size_t)readsize;
     fCtx->totalBytesOutput += (size_t)compressedfilesize;
-    DISPLAYLEVEL(2, "\r%79s\r", "");
-    if (g_display_prefs.displayLevel >= 2 &&
-        !fCtx->hasStdoutOutput &&
-        (g_display_prefs.displayLevel >= 3 || fCtx->nbFilesTotal <= 1)) {
+    DISPLAY_PROGRESS("\r%79s\r", "");
+    if (FIO_shouldDisplayFileSummary(fCtx)) {
+        UTIL_HumanReadableSize_t hr_isize = UTIL_makeHumanReadableSize((U64) readsize);
+        UTIL_HumanReadableSize_t hr_osize = UTIL_makeHumanReadableSize((U64) compressedfilesize);
         if (readsize == 0) {
-            DISPLAYLEVEL(2,"%-20s :  (%6llu => %6llu bytes, %s) \n",
+            DISPLAY_SUMMARY("%-20s :  (%6.*f%s => %6.*f%s, %s) \n",
                 srcFileName,
-                (unsigned long long)readsize, (unsigned long long) compressedfilesize,
+                hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                hr_osize.precision, hr_osize.value, hr_osize.suffix,
                 dstFileName);
         } else {
-            DISPLAYLEVEL(2,"%-20s :%6.2f%%   (%6llu => %6llu bytes, %s) \n",
+            DISPLAY_SUMMARY("%-20s :%6.2f%%   (%6.*f%s => %6.*f%s, %s) \n",
                 srcFileName,
                 (double)compressedfilesize / (double)readsize * 100,
-                (unsigned long long)readsize, (unsigned long long) compressedfilesize,
+                hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                hr_osize.precision, hr_osize.value, hr_osize.suffix,
                 dstFileName);
         }
     }
@@ -1620,7 +1661,7 @@ FIO_compressFilename_internal(FIO_ctx_t* const fCtx,
 
 
 /*! FIO_compressFilename_dstFile() :
- *  open dstFileName, or pass-through if ress.dstFile != NULL,
+ *  open dstFileName, or pass-through if ress.file != NULL,
  *  then start compression with FIO_compressFilename_internal().
  *  Manages source removal (--rm) and file permissions transfer.
  *  note : ress.srcFile must be != NULL,
@@ -1633,24 +1674,29 @@ static int FIO_compressFilename_dstFile(FIO_ctx_t* const fCtx,
                                         cRess_t ress,
                                         const char* dstFileName,
                                         const char* srcFileName,
+                                        const stat_t* srcFileStat,
                                         int compressionLevel)
 {
     int closeDstFile = 0;
     int result;
-    stat_t statbuf;
-    assert(ress.srcFile != NULL);
-    if (ress.dstFile == NULL) {
-        int dstFilePermissions = DEFAULT_FILE_PERMISSIONS;
+    int transferStat = 0;
+    FILE *dstFile;
+
+    assert(AIO_ReadPool_getFile(ress.readCtx) != NULL);
+    if (AIO_WritePool_getFile(ress.writeCtx) == NULL) {
+        int dstFileInitialPermissions = DEFAULT_FILE_PERMISSIONS;
         if ( strcmp (srcFileName, stdinmark)
-          && UTIL_stat(srcFileName, &statbuf)
-          && UTIL_isRegularFileStat(&statbuf) ) {
-            dstFilePermissions = statbuf.st_mode;
+          && strcmp (dstFileName, stdoutmark)
+          && UTIL_isRegularFileStat(srcFileStat) ) {
+            transferStat = 1;
+            dstFileInitialPermissions = TEMPORARY_FILE_PERMISSIONS;
         }
 
         closeDstFile = 1;
         DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: opening dst: %s \n", dstFileName);
-        ress.dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName, dstFilePermissions);
-        if (ress.dstFile==NULL) return 1;  /* could not open dstFileName */
+        dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName, dstFileInitialPermissions);
+        if (dstFile==NULL) return 1;  /* could not open dstFileName */
+        AIO_WritePool_setFile(ress.writeCtx, dstFile);
         /* Must only be added after FIO_openDstFile() succeeds.
          * Otherwise we may delete the destination file if it already exists,
          * and the user presses Ctrl-C when asked if they wish to overwrite.
@@ -1661,16 +1707,16 @@ static int FIO_compressFilename_dstFile(FIO_ctx_t* const fCtx,
     result = FIO_compressFilename_internal(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
 
     if (closeDstFile) {
-        FILE* const dstFile = ress.dstFile;
-        ress.dstFile = NULL;
-
         clearHandler();
 
         DISPLAYLEVEL(6, "FIO_compressFilename_dstFile: closing dst: %s \n", dstFileName);
-        if (fclose(dstFile)) { /* error closing dstFile */
+        if (AIO_WritePool_closeFile(ress.writeCtx)) { /* error closing file */
             DISPLAYLEVEL(1, "zstd: %s: %s \n", dstFileName, strerror(errno));
             result=1;
         }
+        if (transferStat) {
+            UTIL_setFileStat(dstFileName, srcFileStat);
+        }
         if ( (result != 0)  /* operation failure */
           && strcmp(dstFileName, stdoutmark)  /* special case : don't remove() stdout */
           ) {
@@ -1710,18 +1756,27 @@ FIO_compressFilename_srcFile(FIO_ctx_t* const fCtx,
                              int compressionLevel)
 {
     int result;
+    FILE* srcFile;
+    stat_t srcFileStat;
+    U64 fileSize = UTIL_FILESIZE_UNKNOWN;
     DISPLAYLEVEL(6, "FIO_compressFilename_srcFile: %s \n", srcFileName);
 
-    /* ensure src is not a directory */
-    if (UTIL_isDirectory(srcFileName)) {
-        DISPLAYLEVEL(1, "zstd: %s is a directory -- ignored \n", srcFileName);
-        return 1;
-    }
+    if (strcmp(srcFileName, stdinmark)) {
+        if (UTIL_stat(srcFileName, &srcFileStat)) {
+            /* failure to stat at all is handled during opening */
 
-    /* ensure src is not the same as dict (if present) */
-    if (ress.dictFileName != NULL && UTIL_isSameFile(srcFileName, ress.dictFileName)) {
-        DISPLAYLEVEL(1, "zstd: cannot use %s as an input file and dictionary \n", srcFileName);
-        return 1;
+            /* ensure src is not a directory */
+            if (UTIL_isDirectoryStat(&srcFileStat)) {
+                DISPLAYLEVEL(1, "zstd: %s is a directory -- ignored \n", srcFileName);
+                return 1;
+            }
+
+            /* ensure src is not the same as dict (if present) */
+            if (ress.dictFileName != NULL && UTIL_isSameFileStat(srcFileName, ress.dictFileName, &srcFileStat, &ress.dictFileStat)) {
+                DISPLAYLEVEL(1, "zstd: cannot use %s as an input file and dictionary \n", srcFileName);
+                return 1;
+            }
+        }
     }
 
     /* Check if "srcFile" is compressed. Only done if --exclude-compressed flag is used
@@ -1733,16 +1788,30 @@ FIO_compressFilename_srcFile(FIO_ctx_t* const fCtx,
         return 0;
     }
 
-    ress.srcFile = FIO_openSrcFile(prefs, srcFileName);
-    if (ress.srcFile == NULL) return 1;   /* srcFile could not be opened */
+    srcFile = FIO_openSrcFile(prefs, srcFileName, &srcFileStat);
+    if (srcFile == NULL) return 1;   /* srcFile could not be opened */
 
-    result = FIO_compressFilename_dstFile(fCtx, prefs, ress, dstFileName, srcFileName, compressionLevel);
+    /* Don't use AsyncIO for small files */
+    if (strcmp(srcFileName, stdinmark)) /* Stdin doesn't have stats */
+        fileSize = UTIL_getFileSizeStat(&srcFileStat);
+    if(fileSize != UTIL_FILESIZE_UNKNOWN && fileSize < ZSTD_BLOCKSIZE_MAX * 3) {
+        AIO_ReadPool_setAsync(ress.readCtx, 0);
+        AIO_WritePool_setAsync(ress.writeCtx, 0);
+    } else {
+        AIO_ReadPool_setAsync(ress.readCtx, 1);
+        AIO_WritePool_setAsync(ress.writeCtx, 1);
+    }
 
-    fclose(ress.srcFile);
-    ress.srcFile = NULL;
-    if ( prefs->removeSrcFile   /* --rm */
-      && result == 0       /* success */
-      && strcmp(srcFileName, stdinmark)   /* exception : don't erase stdin */
+    AIO_ReadPool_setFile(ress.readCtx, srcFile);
+    result = FIO_compressFilename_dstFile(
+            fCtx, prefs, ress,
+            dstFileName, srcFileName,
+            &srcFileStat, compressionLevel);
+    AIO_ReadPool_closeFile(ress.readCtx);
+
+    if ( prefs->removeSrcFile  /* --rm */
+      && result == 0           /* success */
+      && strcmp(srcFileName, stdinmark)  /* exception : don't erase stdin */
       ) {
         /* We must clear the handler, since after this point calling it would
          * delete both the source and destination files.
@@ -1754,6 +1823,52 @@ FIO_compressFilename_srcFile(FIO_ctx_t* const fCtx,
     return result;
 }
 
+static const char*
+checked_index(const char* options[], size_t length, size_t index) {
+    assert(index < length);
+    /* Necessary to avoid warnings since -O3 will omit the above `assert` */
+    (void) length;
+    return options[index];
+}
+
+#define INDEX(options, index) checked_index((options), sizeof(options)  / sizeof(char*), (size_t)(index))
+
+void FIO_displayCompressionParameters(const FIO_prefs_t* prefs)
+{
+    static const char* formatOptions[5] = {ZSTD_EXTENSION, GZ_EXTENSION, XZ_EXTENSION,
+                                           LZMA_EXTENSION, LZ4_EXTENSION};
+    static const char* sparseOptions[3] = {" --no-sparse", "", " --sparse"};
+    static const char* checkSumOptions[3] = {" --no-check", "", " --check"};
+    static const char* rowMatchFinderOptions[3] = {"", " --no-row-match-finder", " --row-match-finder"};
+    static const char* compressLiteralsOptions[3] = {"", " --compress-literals", " --no-compress-literals"};
+
+    assert(g_display_prefs.displayLevel >= 4);
+
+    DISPLAY("--format=%s", formatOptions[prefs->compressionType]);
+    DISPLAY("%s", INDEX(sparseOptions, prefs->sparseFileSupport));
+    DISPLAY("%s", prefs->dictIDFlag ? "" : " --no-dictID");
+    DISPLAY("%s", INDEX(checkSumOptions, prefs->checksumFlag));
+    DISPLAY(" --block-size=%d", prefs->blockSize);
+    if (prefs->adaptiveMode)
+        DISPLAY(" --adapt=min=%d,max=%d", prefs->minAdaptLevel, prefs->maxAdaptLevel);
+    DISPLAY("%s", INDEX(rowMatchFinderOptions, prefs->useRowMatchFinder));
+    DISPLAY("%s", prefs->rsyncable ? " --rsyncable" : "");
+    if (prefs->streamSrcSize)
+        DISPLAY(" --stream-size=%u", (unsigned) prefs->streamSrcSize);
+    if (prefs->srcSizeHint)
+        DISPLAY(" --size-hint=%d", prefs->srcSizeHint);
+    if (prefs->targetCBlockSize)
+        DISPLAY(" --target-compressed-block-size=%u", (unsigned) prefs->targetCBlockSize);
+    DISPLAY("%s", INDEX(compressLiteralsOptions, prefs->literalCompressionMode));
+    DISPLAY(" --memory=%u", prefs->memLimit ? prefs->memLimit : 128 MB);
+    DISPLAY(" --threads=%d", prefs->nbWorkers);
+    DISPLAY("%s", prefs->excludeCompressedFiles ? " --exclude-compressed" : "");
+    DISPLAY(" --%scontent-size", prefs->contentSize ? "" : "no-");
+    DISPLAY("\n");
+}
+
+#undef INDEX
+
 int FIO_compressFilename(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs, const char* dstFileName,
                          const char* srcFileName, const char* dictFileName,
                          int compressionLevel, ZSTD_compressionParameters comprParams)
@@ -1780,6 +1895,11 @@ FIO_determineCompressedName(const char* srcFileName, const char* outDirName, con
     char* outDirFilename = NULL;
     size_t sfnSize = strlen(srcFileName);
     size_t const srcSuffixLen = strlen(suffix);
+
+    if(!strcmp(srcFileName, stdinmark)) {
+        return stdoutmark;
+    }
+
     if (outDirName) {
         outDirFilename = FIO_createFilename_fromOutDir(srcFileName, outDirName, srcSuffixLen);
         sfnSize = strlen(outDirFilename);
@@ -1842,23 +1962,24 @@ int FIO_compressMultipleFilenames(FIO_ctx_t* const fCtx,
     /* init */
     assert(outFileName != NULL || suffix != NULL);
     if (outFileName != NULL) {   /* output into a single destination (stdout typically) */
-        if (FIO_removeMultiFilesWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
+        FILE *dstFile;
+        if (FIO_multiFilesConcatWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
             FIO_freeCResources(&ress);
             return 1;
         }
-        ress.dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName, DEFAULT_FILE_PERMISSIONS);
-        if (ress.dstFile == NULL) {  /* could not open outFileName */
+        dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName, DEFAULT_FILE_PERMISSIONS);
+        if (dstFile == NULL) {  /* could not open outFileName */
             error = 1;
         } else {
+            AIO_WritePool_setFile(ress.writeCtx, dstFile);
             for (; fCtx->currFileIdx < fCtx->nbFilesTotal; ++fCtx->currFileIdx) {
                 status = FIO_compressFilename_srcFile(fCtx, prefs, ress, outFileName, inFileNamesTable[fCtx->currFileIdx], compressionLevel);
                 if (!status) fCtx->nbFilesProcessed++;
                 error |= status;
             }
-            if (fclose(ress.dstFile))
+            if (AIO_WritePool_closeFile(ress.writeCtx))
                 EXM_THROW(29, "Write error (%s) : cannot properly close %s",
                             strerror(errno), outFileName);
-            ress.dstFile = NULL;
         }
     } else {
         if (outMirroredRootDirName)
@@ -1889,11 +2010,23 @@ int FIO_compressMultipleFilenames(FIO_ctx_t* const fCtx,
             FIO_checkFilenameCollisions(inFileNamesTable , (unsigned)fCtx->nbFilesTotal);
     }
 
-    if (fCtx->nbFilesProcessed >= 1 && fCtx->nbFilesTotal > 1 && fCtx->totalBytesInput != 0) {
-        DISPLAYLEVEL(2, "\r%79s\r", "");
-        DISPLAYLEVEL(2, "%d files compressed : %.2f%%  (%6zu => %6zu bytes)\n", fCtx->nbFilesProcessed,
-                        (double)fCtx->totalBytesOutput/((double)fCtx->totalBytesInput)*100,
-                        fCtx->totalBytesInput, fCtx->totalBytesOutput);
+    if (FIO_shouldDisplayMultipleFileSummary(fCtx)) {
+        UTIL_HumanReadableSize_t hr_isize = UTIL_makeHumanReadableSize((U64) fCtx->totalBytesInput);
+        UTIL_HumanReadableSize_t hr_osize = UTIL_makeHumanReadableSize((U64) fCtx->totalBytesOutput);
+
+        DISPLAY_PROGRESS("\r%79s\r", "");
+        if (fCtx->totalBytesInput == 0) {
+            DISPLAY_SUMMARY("%3d files compressed : (%6.*f%4s => %6.*f%4s)\n",
+                            fCtx->nbFilesProcessed,
+                            hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                            hr_osize.precision, hr_osize.value, hr_osize.suffix);
+        } else {
+            DISPLAY_SUMMARY("%3d files compressed : %.2f%% (%6.*f%4s => %6.*f%4s)\n",
+                            fCtx->nbFilesProcessed,
+                            (double)fCtx->totalBytesOutput/((double)fCtx->totalBytesInput)*100,
+                            hr_isize.precision, hr_isize.value, hr_isize.suffix,
+                            hr_osize.precision, hr_osize.value, hr_osize.suffix);
+        }
     }
 
     FIO_freeCResources(&ress);
@@ -1910,13 +2043,9 @@ int FIO_compressMultipleFilenames(FIO_ctx_t* const fCtx,
  *  Decompression
  ***************************************************************************/
 typedef struct {
-    void*  srcBuffer;
-    size_t srcBufferSize;
-    size_t srcBufferLoaded;
-    void*  dstBuffer;
-    size_t dstBufferSize;
     ZSTD_DStream* dctx;
-    FILE*  dstFile;
+    WritePoolCtx_t *writeCtx;
+    ReadPoolCtx_t *readCtx;
 } dRess_t;
 
 static dRess_t FIO_createDResources(FIO_prefs_t* const prefs, const char* dictFileName)
@@ -1934,159 +2063,49 @@ static dRess_t FIO_createDResources(FIO_prefs_t* const prefs, const char* dictFi
     CHECK( ZSTD_DCtx_setMaxWindowSize(ress.dctx, prefs->memLimit) );
     CHECK( ZSTD_DCtx_setParameter(ress.dctx, ZSTD_d_forceIgnoreChecksum, !prefs->checksumFlag));
 
-    ress.srcBufferSize = ZSTD_DStreamInSize();
-    ress.srcBuffer = malloc(ress.srcBufferSize);
-    ress.dstBufferSize = ZSTD_DStreamOutSize();
-    ress.dstBuffer = malloc(ress.dstBufferSize);
-    if (!ress.srcBuffer || !ress.dstBuffer)
-        EXM_THROW(61, "Allocation error : not enough memory");
-
     /* dictionary */
     {   void* dictBuffer;
-        size_t const dictBufferSize = FIO_createDictBuffer(&dictBuffer, dictFileName, prefs);
-        CHECK( ZSTD_initDStream_usingDict(ress.dctx, dictBuffer, dictBufferSize) );
+        stat_t statbuf;
+        size_t const dictBufferSize = FIO_createDictBuffer(&dictBuffer, dictFileName, prefs, &statbuf);
+        CHECK( ZSTD_DCtx_reset(ress.dctx, ZSTD_reset_session_only) );
+        CHECK( ZSTD_DCtx_loadDictionary(ress.dctx, dictBuffer, dictBufferSize) );
         free(dictBuffer);
     }
 
+    ress.writeCtx = AIO_WritePool_create(prefs, ZSTD_DStreamOutSize());
+    ress.readCtx = AIO_ReadPool_create(prefs, ZSTD_DStreamInSize());
+
     return ress;
 }
 
 static void FIO_freeDResources(dRess_t ress)
 {
     CHECK( ZSTD_freeDStream(ress.dctx) );
-    free(ress.srcBuffer);
-    free(ress.dstBuffer);
+    AIO_WritePool_free(ress.writeCtx);
+    AIO_ReadPool_free(ress.readCtx);
 }
 
-
-/** FIO_fwriteSparse() :
-*  @return : storedSkips,
-*            argument for next call to FIO_fwriteSparse() or FIO_fwriteSparseEnd() */
-static unsigned
-FIO_fwriteSparse(FILE* file,
-                 const void* buffer, size_t bufferSize,
-                 const FIO_prefs_t* const prefs,
-                 unsigned storedSkips)
-{
-    const size_t* const bufferT = (const size_t*)buffer;   /* Buffer is supposed malloc'ed, hence aligned on size_t */
-    size_t bufferSizeT = bufferSize / sizeof(size_t);
-    const size_t* const bufferTEnd = bufferT + bufferSizeT;
-    const size_t* ptrT = bufferT;
-    static const size_t segmentSizeT = (32 KB) / sizeof(size_t);   /* check every 32 KB */
-
-    if (prefs->testMode) return 0;  /* do not output anything in test mode */
-
-    if (!prefs->sparseFileSupport) {  /* normal write */
-        size_t const sizeCheck = fwrite(buffer, 1, bufferSize, file);
-        if (sizeCheck != bufferSize)
-            EXM_THROW(70, "Write error : cannot write decoded block : %s",
-                            strerror(errno));
-        return 0;
-    }
-
-    /* avoid int overflow */
-    if (storedSkips > 1 GB) {
-        if (LONG_SEEK(file, 1 GB, SEEK_CUR) != 0)
-            EXM_THROW(91, "1 GB skip error (sparse file support)");
-        storedSkips -= 1 GB;
-    }
-
-    while (ptrT < bufferTEnd) {
-        size_t nb0T;
-
-        /* adjust last segment if < 32 KB */
-        size_t seg0SizeT = segmentSizeT;
-        if (seg0SizeT > bufferSizeT) seg0SizeT = bufferSizeT;
-        bufferSizeT -= seg0SizeT;
-
-        /* count leading zeroes */
-        for (nb0T=0; (nb0T < seg0SizeT) && (ptrT[nb0T] == 0); nb0T++) ;
-        storedSkips += (unsigned)(nb0T * sizeof(size_t));
-
-        if (nb0T != seg0SizeT) {   /* not all 0s */
-            size_t const nbNon0ST = seg0SizeT - nb0T;
-            /* skip leading zeros */
-            if (LONG_SEEK(file, storedSkips, SEEK_CUR) != 0)
-                EXM_THROW(92, "Sparse skip error ; try --no-sparse");
-            storedSkips = 0;
-            /* write the rest */
-            if (fwrite(ptrT + nb0T, sizeof(size_t), nbNon0ST, file) != nbNon0ST)
-                EXM_THROW(93, "Write error : cannot write decoded block : %s",
-                            strerror(errno));
-        }
-        ptrT += seg0SizeT;
-    }
-
-    {   static size_t const maskT = sizeof(size_t)-1;
-        if (bufferSize & maskT) {
-            /* size not multiple of sizeof(size_t) : implies end of block */
-            const char* const restStart = (const char*)bufferTEnd;
-            const char* restPtr = restStart;
-            const char* const restEnd = (const char*)buffer + bufferSize;
-            assert(restEnd > restStart && restEnd < restStart + sizeof(size_t));
-            for ( ; (restPtr < restEnd) && (*restPtr == 0); restPtr++) ;
-            storedSkips += (unsigned) (restPtr - restStart);
-            if (restPtr != restEnd) {
-                /* not all remaining bytes are 0 */
-                size_t const restSize = (size_t)(restEnd - restPtr);
-                if (LONG_SEEK(file, storedSkips, SEEK_CUR) != 0)
-                    EXM_THROW(92, "Sparse skip error ; try --no-sparse");
-                if (fwrite(restPtr, 1, restSize, file) != restSize)
-                    EXM_THROW(95, "Write error : cannot write end of decoded block : %s",
-                        strerror(errno));
-                storedSkips = 0;
-    }   }   }
-
-    return storedSkips;
-}
-
-static void
-FIO_fwriteSparseEnd(const FIO_prefs_t* const prefs, FILE* file, unsigned storedSkips)
-{
-    if (prefs->testMode) assert(storedSkips == 0);
-    if (storedSkips>0) {
-        assert(prefs->sparseFileSupport > 0);  /* storedSkips>0 implies sparse support is enabled */
-        (void)prefs;   /* assert can be disabled, in which case prefs becomes unused */
-        if (LONG_SEEK(file, storedSkips-1, SEEK_CUR) != 0)
-            EXM_THROW(69, "Final skip error (sparse file support)");
-        /* last zero must be explicitly written,
-         * so that skipped ones get implicitly translated as zero by FS */
-        {   const char lastZeroByte[1] = { 0 };
-            if (fwrite(lastZeroByte, 1, 1, file) != 1)
-                EXM_THROW(69, "Write error : cannot write last zero : %s", strerror(errno));
-    }   }
-}
-
-
 /** FIO_passThrough() : just copy input into output, for compatibility with gzip -df mode
     @return : 0 (no error) */
-static int FIO_passThrough(const FIO_prefs_t* const prefs,
-                           FILE* foutput, FILE* finput,
-                           void* buffer, size_t bufferSize,
-                           size_t alreadyLoaded)
-{
-    size_t const blockSize = MIN(64 KB, bufferSize);
-    size_t readFromInput;
-    unsigned storedSkips = 0;
-
-    /* assumption : ress->srcBufferLoaded bytes already loaded and stored within buffer */
-    {   size_t const sizeCheck = fwrite(buffer, 1, alreadyLoaded, foutput);
-        if (sizeCheck != alreadyLoaded) {
-            DISPLAYLEVEL(1, "Pass-through write error : %s\n", strerror(errno));
-            return 1;
-    }   }
-
-    do {
-        readFromInput = fread(buffer, 1, blockSize, finput);
-        storedSkips = FIO_fwriteSparse(foutput, buffer, readFromInput, prefs, storedSkips);
-    } while (readFromInput == blockSize);
-    if (ferror(finput)) {
-        DISPLAYLEVEL(1, "Pass-through read error : %s\n", strerror(errno));
-        return 1;
-    }
-    assert(feof(finput));
-
-    FIO_fwriteSparseEnd(prefs, foutput, storedSkips);
+static int FIO_passThrough(dRess_t *ress)
+{
+    size_t const blockSize = MIN(MIN(64 KB, ZSTD_DStreamInSize()), ZSTD_DStreamOutSize());
+    IOJob_t *writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
+
+    while(ress->readCtx->srcBufferLoaded) {
+        size_t writeSize;
+        writeSize = MIN(blockSize, ress->readCtx->srcBufferLoaded);
+        assert(writeSize <= writeJob->bufferSize);
+        memcpy(writeJob->buffer, ress->readCtx->srcBuffer, writeSize);
+        writeJob->usedBufferSize = writeSize;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
+        AIO_ReadPool_consumeBytes(ress->readCtx, writeSize);
+        AIO_ReadPool_fillBuffer(ress->readCtx, blockSize);
+    }
+    assert(ress->readCtx->reachedEof);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
     return 0;
 }
 
@@ -2104,7 +2123,7 @@ FIO_zstdErrorHelp(const FIO_prefs_t* const prefs,
         return;
 
     /* Try to decode the frame header */
-    err = ZSTD_getFrameHeader(&header, ress->srcBuffer, ress->srcBufferLoaded);
+    err = ZSTD_getFrameHeader(&header, ress->readCtx->srcBuffer, ress->readCtx->srcBufferLoaded);
     if (err == 0) {
         unsigned long long const windowSize = header.windowSize;
         unsigned const windowLog = FIO_highbit64(windowSize) + ((windowSize & (windowSize - 1)) != 0);
@@ -2127,13 +2146,13 @@ FIO_zstdErrorHelp(const FIO_prefs_t* const prefs,
  */
 #define FIO_ERROR_FRAME_DECODING   ((unsigned long long)(-2))
 static unsigned long long
-FIO_decompressZstdFrame(FIO_ctx_t* const fCtx, dRess_t* ress, FILE* finput,
+FIO_decompressZstdFrame(FIO_ctx_t* const fCtx, dRess_t* ress,
                         const FIO_prefs_t* const prefs,
                         const char* srcFileName,
                         U64 alreadyDecoded)  /* for multi-frames streams */
 {
     U64 frameSize = 0;
-    U32 storedSkips = 0;
+    IOJob_t *writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
 
     /* display last 20 characters only */
     {   size_t const srcFileLength = strlen(srcFileName);
@@ -2143,67 +2162,60 @@ FIO_decompressZstdFrame(FIO_ctx_t* const fCtx, dRess_t* ress, FILE* finput,
     ZSTD_DCtx_reset(ress->dctx, ZSTD_reset_session_only);
 
     /* Header loading : ensures ZSTD_getFrameHeader() will succeed */
-    {   size_t const toDecode = ZSTD_FRAMEHEADERSIZE_MAX;
-        if (ress->srcBufferLoaded < toDecode) {
-            size_t const toRead = toDecode - ress->srcBufferLoaded;
-            void* const startPosition = (char*)ress->srcBuffer + ress->srcBufferLoaded;
-            ress->srcBufferLoaded += fread(startPosition, 1, toRead, finput);
-    }   }
+    AIO_ReadPool_fillBuffer(ress->readCtx, ZSTD_FRAMEHEADERSIZE_MAX);
 
     /* Main decompression Loop */
     while (1) {
-        ZSTD_inBuffer  inBuff = { ress->srcBuffer, ress->srcBufferLoaded, 0 };
-        ZSTD_outBuffer outBuff= { ress->dstBuffer, ress->dstBufferSize, 0 };
+        ZSTD_inBuffer  inBuff = setInBuffer( ress->readCtx->srcBuffer, ress->readCtx->srcBufferLoaded, 0 );
+        ZSTD_outBuffer outBuff= setOutBuffer( writeJob->buffer, writeJob->bufferSize, 0 );
         size_t const readSizeHint = ZSTD_decompressStream(ress->dctx, &outBuff, &inBuff);
+        UTIL_HumanReadableSize_t const hrs = UTIL_makeHumanReadableSize(alreadyDecoded+frameSize);
         if (ZSTD_isError(readSizeHint)) {
             DISPLAYLEVEL(1, "%s : Decoding error (36) : %s \n",
                             srcFileName, ZSTD_getErrorName(readSizeHint));
             FIO_zstdErrorHelp(prefs, ress, readSizeHint, srcFileName);
+            AIO_WritePool_releaseIoJob(writeJob);
             return FIO_ERROR_FRAME_DECODING;
         }
 
         /* Write block */
-        storedSkips = FIO_fwriteSparse(ress->dstFile, ress->dstBuffer, outBuff.pos, prefs, storedSkips);
+        writeJob->usedBufferSize = outBuff.pos;
+        AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
         frameSize += outBuff.pos;
-        if (!fCtx->hasStdoutOutput || g_display_prefs.progressSetting == FIO_ps_always) {
-            if (fCtx->nbFilesTotal > 1) {
-                size_t srcFileNameSize = strlen(srcFileName);
-                if (srcFileNameSize > 18) {
-                    const char* truncatedSrcFileName = srcFileName + srcFileNameSize - 15;
-                    DISPLAYUPDATE(2, "\rDecompress: %2u/%2u files. Current: ...%s : %u MB...    ",
-                                    fCtx->currFileIdx+1, fCtx->nbFilesTotal, truncatedSrcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
-                } else {
-                    DISPLAYUPDATE(2, "\rDecompress: %2u/%2u files. Current: %s : %u MB...    ",
-                                fCtx->currFileIdx+1, fCtx->nbFilesTotal, srcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
-                }
+        if (fCtx->nbFilesTotal > 1) {
+            size_t srcFileNameSize = strlen(srcFileName);
+            if (srcFileNameSize > 18) {
+                const char* truncatedSrcFileName = srcFileName + srcFileNameSize - 15;
+                DISPLAYUPDATE_PROGRESS(
+                        "\rDecompress: %2u/%2u files. Current: ...%s : %.*f%s...    ",
+                        fCtx->currFileIdx+1, fCtx->nbFilesTotal, truncatedSrcFileName, hrs.precision, hrs.value, hrs.suffix);
             } else {
-                DISPLAYUPDATE(2, "\r%-20.20s : %u MB...     ",
-                                srcFileName, (unsigned)((alreadyDecoded+frameSize)>>20) );
+                DISPLAYUPDATE_PROGRESS("\rDecompress: %2u/%2u files. Current: %s : %.*f%s...    ",
+                            fCtx->currFileIdx+1, fCtx->nbFilesTotal, srcFileName, hrs.precision, hrs.value, hrs.suffix);
             }
+        } else {
+            DISPLAYUPDATE_PROGRESS("\r%-20.20s : %.*f%s...     ",
+                            srcFileName, hrs.precision, hrs.value, hrs.suffix);
         }
 
-        if (inBuff.pos > 0) {
-            memmove(ress->srcBuffer, (char*)ress->srcBuffer + inBuff.pos, inBuff.size - inBuff.pos);
-            ress->srcBufferLoaded -= inBuff.pos;
-        }
+        AIO_ReadPool_consumeBytes(ress->readCtx, inBuff.pos);
 
         if (readSizeHint == 0) break;   /* end of frame */
 
         /* Fill input buffer */
-        {   size_t const toDecode = MIN(readSizeHint, ress->srcBufferSize);  /* support large skippable frames */
-            if (ress->srcBufferLoaded < toDecode) {
-                size_t const toRead = toDecode - ress->srcBufferLoaded;   /* > 0 */
-                void* const startPosition = (char*)ress->srcBuffer + ress->srcBufferLoaded;
-                size_t const readSize = fread(startPosition, 1, toRead, finput);
+        {   size_t const toDecode = MIN(readSizeHint, ZSTD_DStreamInSize());  /* support large skippable frames */
+            if (ress->readCtx->srcBufferLoaded < toDecode) {
+                size_t const readSize = AIO_ReadPool_fillBuffer(ress->readCtx, toDecode);
                 if (readSize==0) {
                     DISPLAYLEVEL(1, "%s : Read error (39) : premature end \n",
-                                    srcFileName);
+                                 srcFileName);
+                    AIO_WritePool_releaseIoJob(writeJob);
                     return FIO_ERROR_FRAME_DECODING;
                 }
-                ress->srcBufferLoaded += readSize;
-    }   }   }
+            }   }   }
 
-    FIO_fwriteSparseEnd(prefs, ress->dstFile, storedSkips);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
 
     return frameSize;
 }
@@ -2211,37 +2223,36 @@ FIO_decompressZstdFrame(FIO_ctx_t* const fCtx, dRess_t* ress, FILE* finput,
 
 #ifdef ZSTD_GZDECOMPRESS
 static unsigned long long
-FIO_decompressGzFrame(dRess_t* ress, FILE* srcFile,
-                      const FIO_prefs_t* const prefs,
-                      const char* srcFileName)
+FIO_decompressGzFrame(dRess_t* ress, const char* srcFileName)
 {
     unsigned long long outFileSize = 0;
     z_stream strm;
     int flush = Z_NO_FLUSH;
     int decodingError = 0;
-    unsigned storedSkips = 0;
+    IOJob_t *writeJob = NULL;
 
     strm.zalloc = Z_NULL;
     strm.zfree = Z_NULL;
     strm.opaque = Z_NULL;
     strm.next_in = 0;
     strm.avail_in = 0;
-    /* see http://www.zlib.net/manual.html */
+    /* see https://www.zlib.net/manual.html */
     if (inflateInit2(&strm, 15 /* maxWindowLogSize */ + 16 /* gzip only */) != Z_OK)
         return FIO_ERROR_FRAME_DECODING;
 
-    strm.next_out = (Bytef*)ress->dstBuffer;
-    strm.avail_out = (uInt)ress->dstBufferSize;
-    strm.avail_in = (uInt)ress->srcBufferLoaded;
-    strm.next_in = (z_const unsigned char*)ress->srcBuffer;
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_out = (Bytef*)writeJob->buffer;
+    strm.avail_out = (uInt)writeJob->bufferSize;
+    strm.avail_in = (uInt)ress->readCtx->srcBufferLoaded;
+    strm.next_in = (z_const unsigned char*)ress->readCtx->srcBuffer;
 
     for ( ; ; ) {
         int ret;
         if (strm.avail_in == 0) {
-            ress->srcBufferLoaded = fread(ress->srcBuffer, 1, ress->srcBufferSize, srcFile);
-            if (ress->srcBufferLoaded == 0) flush = Z_FINISH;
-            strm.next_in = (z_const unsigned char*)ress->srcBuffer;
-            strm.avail_in = (uInt)ress->srcBufferLoaded;
+            AIO_ReadPool_consumeAndRefill(ress->readCtx);
+            if (ress->readCtx->srcBufferLoaded == 0) flush = Z_FINISH;
+            strm.next_in = (z_const unsigned char*)ress->readCtx->srcBuffer;
+            strm.avail_in = (uInt)ress->readCtx->srcBufferLoaded;
         }
         ret = inflate(&strm, flush);
         if (ret == Z_BUF_ERROR) {
@@ -2252,35 +2263,34 @@ FIO_decompressGzFrame(dRess_t* ress, FILE* srcFile,
             DISPLAYLEVEL(1, "zstd: %s: inflate error %d \n", srcFileName, ret);
             decodingError = 1; break;
         }
-        {   size_t const decompBytes = ress->dstBufferSize - strm.avail_out;
+        {   size_t const decompBytes = writeJob->bufferSize - strm.avail_out;
             if (decompBytes) {
-                storedSkips = FIO_fwriteSparse(ress->dstFile, ress->dstBuffer, decompBytes, prefs, storedSkips);
+                writeJob->usedBufferSize = decompBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 outFileSize += decompBytes;
-                strm.next_out = (Bytef*)ress->dstBuffer;
-                strm.avail_out = (uInt)ress->dstBufferSize;
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = (uInt)writeJob->bufferSize;
             }
         }
         if (ret == Z_STREAM_END) break;
     }
 
-    if (strm.avail_in > 0)
-        memmove(ress->srcBuffer, strm.next_in, strm.avail_in);
-    ress->srcBufferLoaded = strm.avail_in;
+    AIO_ReadPool_consumeBytes(ress->readCtx, ress->readCtx->srcBufferLoaded - strm.avail_in);
+
     if ( (inflateEnd(&strm) != Z_OK)  /* release resources ; error detected */
       && (decodingError==0) ) {
         DISPLAYLEVEL(1, "zstd: %s: inflateEnd error \n", srcFileName);
         decodingError = 1;
     }
-    FIO_fwriteSparseEnd(prefs, ress->dstFile, storedSkips);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
     return decodingError ? FIO_ERROR_FRAME_DECODING : outFileSize;
 }
 #endif
 
-
 #ifdef ZSTD_LZMADECOMPRESS
 static unsigned long long
-FIO_decompressLzmaFrame(dRess_t* ress, FILE* srcFile,
-                        const FIO_prefs_t* const prefs,
+FIO_decompressLzmaFrame(dRess_t* ress,
                         const char* srcFileName, int plain_lzma)
 {
     unsigned long long outFileSize = 0;
@@ -2288,7 +2298,7 @@ FIO_decompressLzmaFrame(dRess_t* ress, FILE* srcFile,
     lzma_action action = LZMA_RUN;
     lzma_ret initRet;
     int decodingError = 0;
-    unsigned storedSkips = 0;
+    IOJob_t *writeJob = NULL;
 
     strm.next_in = 0;
     strm.avail_in = 0;
@@ -2305,18 +2315,19 @@ FIO_decompressLzmaFrame(dRess_t* ress, FILE* srcFile,
         return FIO_ERROR_FRAME_DECODING;
     }
 
-    strm.next_out = (BYTE*)ress->dstBuffer;
-    strm.avail_out = ress->dstBufferSize;
-    strm.next_in = (BYTE const*)ress->srcBuffer;
-    strm.avail_in = ress->srcBufferLoaded;
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
+    strm.next_out = (Bytef*)writeJob->buffer;
+    strm.avail_out = (uInt)writeJob->bufferSize;
+    strm.next_in = (BYTE const*)ress->readCtx->srcBuffer;
+    strm.avail_in = ress->readCtx->srcBufferLoaded;
 
     for ( ; ; ) {
         lzma_ret ret;
         if (strm.avail_in == 0) {
-            ress->srcBufferLoaded = fread(ress->srcBuffer, 1, ress->srcBufferSize, srcFile);
-            if (ress->srcBufferLoaded == 0) action = LZMA_FINISH;
-            strm.next_in = (BYTE const*)ress->srcBuffer;
-            strm.avail_in = ress->srcBufferLoaded;
+            AIO_ReadPool_consumeAndRefill(ress->readCtx);
+            if (ress->readCtx->srcBufferLoaded == 0) action = LZMA_FINISH;
+            strm.next_in = (BYTE const*)ress->readCtx->srcBuffer;
+            strm.avail_in = ress->readCtx->srcBufferLoaded;
         }
         ret = lzma_code(&strm, action);
 
@@ -2329,102 +2340,90 @@ FIO_decompressLzmaFrame(dRess_t* ress, FILE* srcFile,
                             srcFileName, ret);
             decodingError = 1; break;
         }
-        {   size_t const decompBytes = ress->dstBufferSize - strm.avail_out;
+        {   size_t const decompBytes = writeJob->bufferSize - strm.avail_out;
             if (decompBytes) {
-                storedSkips = FIO_fwriteSparse(ress->dstFile, ress->dstBuffer, decompBytes, prefs, storedSkips);
+                writeJob->usedBufferSize = decompBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 outFileSize += decompBytes;
-                strm.next_out = (BYTE*)ress->dstBuffer;
-                strm.avail_out = ress->dstBufferSize;
+                strm.next_out = (Bytef*)writeJob->buffer;
+                strm.avail_out = writeJob->bufferSize;
         }   }
         if (ret == LZMA_STREAM_END) break;
     }
 
-    if (strm.avail_in > 0)
-        memmove(ress->srcBuffer, strm.next_in, strm.avail_in);
-    ress->srcBufferLoaded = strm.avail_in;
+    AIO_ReadPool_consumeBytes(ress->readCtx, ress->readCtx->srcBufferLoaded - strm.avail_in);
     lzma_end(&strm);
-    FIO_fwriteSparseEnd(prefs, ress->dstFile, storedSkips);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
     return decodingError ? FIO_ERROR_FRAME_DECODING : outFileSize;
 }
 #endif
 
 #ifdef ZSTD_LZ4DECOMPRESS
 static unsigned long long
-FIO_decompressLz4Frame(dRess_t* ress, FILE* srcFile,
-                       const FIO_prefs_t* const prefs,
-                       const char* srcFileName)
+FIO_decompressLz4Frame(dRess_t* ress, const char* srcFileName)
 {
     unsigned long long filesize = 0;
-    LZ4F_errorCode_t nextToLoad;
+    LZ4F_errorCode_t nextToLoad = 4;
     LZ4F_decompressionContext_t dCtx;
     LZ4F_errorCode_t const errorCode = LZ4F_createDecompressionContext(&dCtx, LZ4F_VERSION);
     int decodingError = 0;
-    unsigned storedSkips = 0;
+    IOJob_t *writeJob = NULL;
 
     if (LZ4F_isError(errorCode)) {
         DISPLAYLEVEL(1, "zstd: failed to create lz4 decompression context \n");
         return FIO_ERROR_FRAME_DECODING;
     }
 
-    /* Init feed with magic number (already consumed from FILE* sFile) */
-    {   size_t inSize = 4;
-        size_t outSize= 0;
-        MEM_writeLE32(ress->srcBuffer, LZ4_MAGICNUMBER);
-        nextToLoad = LZ4F_decompress(dCtx, ress->dstBuffer, &outSize, ress->srcBuffer, &inSize, NULL);
-        if (LZ4F_isError(nextToLoad)) {
-            DISPLAYLEVEL(1, "zstd: %s: lz4 header error : %s \n",
-                            srcFileName, LZ4F_getErrorName(nextToLoad));
-            LZ4F_freeDecompressionContext(dCtx);
-            return FIO_ERROR_FRAME_DECODING;
-    }   }
+    writeJob = AIO_WritePool_acquireJob(ress->writeCtx);
 
     /* Main Loop */
     for (;nextToLoad;) {
-        size_t readSize;
         size_t pos = 0;
-        size_t decodedBytes = ress->dstBufferSize;
+        size_t decodedBytes = writeJob->bufferSize;
+        int fullBufferDecoded = 0;
 
         /* Read input */
-        if (nextToLoad > ress->srcBufferSize) nextToLoad = ress->srcBufferSize;
-        readSize = fread(ress->srcBuffer, 1, nextToLoad, srcFile);
-        if (!readSize) break;   /* reached end of file or stream */
+        AIO_ReadPool_fillBuffer(ress->readCtx, nextToLoad);
+        if(!ress->readCtx->srcBufferLoaded) break; /* reached end of file */
 
-        while ((pos < readSize) || (decodedBytes == ress->dstBufferSize)) {  /* still to read, or still to flush */
+        while ((pos < ress->readCtx->srcBufferLoaded) || fullBufferDecoded) {  /* still to read, or still to flush */
             /* Decode Input (at least partially) */
-            size_t remaining = readSize - pos;
-            decodedBytes = ress->dstBufferSize;
-            nextToLoad = LZ4F_decompress(dCtx, ress->dstBuffer, &decodedBytes, (char*)(ress->srcBuffer)+pos, &remaining, NULL);
+            size_t remaining = ress->readCtx->srcBufferLoaded - pos;
+            decodedBytes = writeJob->bufferSize;
+            nextToLoad = LZ4F_decompress(dCtx, writeJob->buffer, &decodedBytes, (char*)(ress->readCtx->srcBuffer)+pos,
+                                         &remaining, NULL);
             if (LZ4F_isError(nextToLoad)) {
                 DISPLAYLEVEL(1, "zstd: %s: lz4 decompression error : %s \n",
                                 srcFileName, LZ4F_getErrorName(nextToLoad));
                 decodingError = 1; nextToLoad = 0; break;
             }
             pos += remaining;
+            assert(pos <= ress->readCtx->srcBufferLoaded);
+            fullBufferDecoded = decodedBytes == writeJob->bufferSize;
 
             /* Write Block */
             if (decodedBytes) {
-                storedSkips = FIO_fwriteSparse(ress->dstFile, ress->dstBuffer, decodedBytes, prefs, storedSkips);
+                UTIL_HumanReadableSize_t hrs;
+                writeJob->usedBufferSize = decodedBytes;
+                AIO_WritePool_enqueueAndReacquireWriteJob(&writeJob);
                 filesize += decodedBytes;
-                DISPLAYUPDATE(2, "\rDecompressed : %u MB  ", (unsigned)(filesize>>20));
+                hrs = UTIL_makeHumanReadableSize(filesize);
+                DISPLAYUPDATE_PROGRESS("\rDecompressed : %.*f%s  ", hrs.precision, hrs.value, hrs.suffix);
             }
 
             if (!nextToLoad) break;
         }
+        AIO_ReadPool_consumeBytes(ress->readCtx, pos);
     }
-    /* can be out because readSize == 0, which could be an fread() error */
-    if (ferror(srcFile)) {
-        DISPLAYLEVEL(1, "zstd: %s: read error \n", srcFileName);
-        decodingError=1;
-    }
-
     if (nextToLoad!=0) {
         DISPLAYLEVEL(1, "zstd: %s: unfinished lz4 stream \n", srcFileName);
         decodingError=1;
     }
 
     LZ4F_freeDecompressionContext(dCtx);
-    ress->srcBufferLoaded = 0; /* LZ4F will reach exact frame boundary */
-    FIO_fwriteSparseEnd(prefs, ress->dstFile, storedSkips);
+    AIO_WritePool_releaseIoJob(writeJob);
+    AIO_WritePool_sparseWriteEnd(ress->writeCtx);
 
     return decodingError ? FIO_ERROR_FRAME_DECODING : filesize;
 }
@@ -2439,23 +2438,30 @@ FIO_decompressLz4Frame(dRess_t* ress, FILE* srcFile,
  *           1 : error
  */
 static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
-                          dRess_t ress, FILE* srcFile,
-                          const FIO_prefs_t* const prefs,
-                          const char* dstFileName, const char* srcFileName)
+                                dRess_t ress, const FIO_prefs_t* const prefs,
+                                const char* dstFileName, const char* srcFileName)
 {
     unsigned readSomething = 0;
     unsigned long long filesize = 0;
-    assert(srcFile != NULL);
+    int passThrough = prefs->passThrough;
+
+    if (passThrough == -1) {
+        /* If pass-through mode is not explicitly enabled or disabled,
+         * default to the legacy behavior of enabling it if we are writing
+         * to stdout with the overwrite flag enabled.
+         */
+        passThrough = prefs->overwrite && !strcmp(dstFileName, stdoutmark);
+    }
+    assert(passThrough == 0 || passThrough == 1);
 
     /* for each frame */
     for ( ; ; ) {
         /* check magic number -> version */
         size_t const toRead = 4;
-        const BYTE* const buf = (const BYTE*)ress.srcBuffer;
-        if (ress.srcBufferLoaded < toRead)  /* load up to 4 bytes for header */
-            ress.srcBufferLoaded += fread((char*)ress.srcBuffer + ress.srcBufferLoaded,
-                                          (size_t)1, toRead - ress.srcBufferLoaded, srcFile);
-        if (ress.srcBufferLoaded==0) {
+        const BYTE* buf;
+        AIO_ReadPool_fillBuffer(ress.readCtx, toRead);
+        buf = (const BYTE*)ress.readCtx->srcBuffer;
+        if (ress.readCtx->srcBufferLoaded==0) {
             if (readSomething==0) {  /* srcFile is empty (which is invalid) */
                 DISPLAYLEVEL(1, "zstd: %s: unexpected end of file \n", srcFileName);
                 return 1;
@@ -2463,17 +2469,20 @@ static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
             break;   /* no more input */
         }
         readSomething = 1;   /* there is at least 1 byte in srcFile */
-        if (ress.srcBufferLoaded < toRead) {
+        if (ress.readCtx->srcBufferLoaded < toRead) { /* not enough input to check magic number */
+            if (passThrough) {
+                return FIO_passThrough(&ress);
+            }
             DISPLAYLEVEL(1, "zstd: %s: unknown header \n", srcFileName);
             return 1;
         }
-        if (ZSTD_isFrame(buf, ress.srcBufferLoaded)) {
-            unsigned long long const frameSize = FIO_decompressZstdFrame(fCtx, &ress, srcFile, prefs, srcFileName, filesize);
+        if (ZSTD_isFrame(buf, ress.readCtx->srcBufferLoaded)) {
+            unsigned long long const frameSize = FIO_decompressZstdFrame(fCtx, &ress, prefs, srcFileName, filesize);
             if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
             filesize += frameSize;
         } else if (buf[0] == 31 && buf[1] == 139) { /* gz magic number */
 #ifdef ZSTD_GZDECOMPRESS
-            unsigned long long const frameSize = FIO_decompressGzFrame(&ress, srcFile, prefs, srcFileName);
+            unsigned long long const frameSize = FIO_decompressGzFrame(&ress, srcFileName);
             if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
             filesize += frameSize;
 #else
@@ -2483,7 +2492,7 @@ static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
         } else if ((buf[0] == 0xFD && buf[1] == 0x37)  /* xz magic number */
                 || (buf[0] == 0x5D && buf[1] == 0x00)) { /* lzma header (no magic number) */
 #ifdef ZSTD_LZMADECOMPRESS
-            unsigned long long const frameSize = FIO_decompressLzmaFrame(&ress, srcFile, prefs, srcFileName, buf[0] != 0xFD);
+            unsigned long long const frameSize = FIO_decompressLzmaFrame(&ress, srcFileName, buf[0] != 0xFD);
             if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
             filesize += frameSize;
 #else
@@ -2492,18 +2501,15 @@ static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
 #endif
         } else if (MEM_readLE32(buf) == LZ4_MAGICNUMBER) {
 #ifdef ZSTD_LZ4DECOMPRESS
-            unsigned long long const frameSize = FIO_decompressLz4Frame(&ress, srcFile, prefs, srcFileName);
+            unsigned long long const frameSize = FIO_decompressLz4Frame(&ress, srcFileName);
             if (frameSize == FIO_ERROR_FRAME_DECODING) return 1;
             filesize += frameSize;
 #else
             DISPLAYLEVEL(1, "zstd: %s: lz4 file cannot be uncompressed (zstd compiled without HAVE_LZ4) -- ignored \n", srcFileName);
             return 1;
 #endif
-        } else if ((prefs->overwrite) && !strcmp (dstFileName, stdoutmark)) {  /* pass-through mode */
-            return FIO_passThrough(prefs,
-                                   ress.dstFile, srcFile,
-                                   ress.srcBuffer, ress.srcBufferSize,
-                                   ress.srcBufferLoaded);
+        } else if (passThrough) {
+            return FIO_passThrough(&ress);
         } else {
             DISPLAYLEVEL(1, "zstd: %s: unsupported format \n", srcFileName);
             return 1;
@@ -2511,45 +2517,45 @@ static int FIO_decompressFrames(FIO_ctx_t* const fCtx,
 
     /* Final Status */
     fCtx->totalBytesOutput += (size_t)filesize;
-    DISPLAYLEVEL(2, "\r%79s\r", "");
-    /* No status message in pipe mode (stdin - stdout) or multi-files mode */
-    if (g_display_prefs.displayLevel >= 2) {
-        if (fCtx->nbFilesTotal <= 1 || g_display_prefs.displayLevel >= 3) {
-            DISPLAYLEVEL(2, "%-20s: %llu bytes \n", srcFileName, filesize);
-        }
-    }
+    DISPLAY_PROGRESS("\r%79s\r", "");
+    if (FIO_shouldDisplayFileSummary(fCtx))
+        DISPLAY_SUMMARY("%-20s: %llu bytes \n", srcFileName, filesize);
 
     return 0;
 }
 
 /** FIO_decompressDstFile() :
-    open `dstFileName`,
-    or path-through if ress.dstFile is already != 0,
+    open `dstFileName`, or pass-through if writeCtx's file is already != 0,
     then start decompression process (FIO_decompressFrames()).
     @return : 0 : OK
               1 : operation aborted
 */
 static int FIO_decompressDstFile(FIO_ctx_t* const fCtx,
                                  FIO_prefs_t* const prefs,
-                                 dRess_t ress, FILE* srcFile,
-                                 const char* dstFileName, const char* srcFileName)
+                                 dRess_t ress,
+                                 const char* dstFileName,
+                                 const char* srcFileName,
+                                 const stat_t* srcFileStat)
 {
     int result;
-    stat_t statbuf;
     int releaseDstFile = 0;
+    int transferStat = 0;
 
-    if ((ress.dstFile == NULL) && (prefs->testMode==0)) {
+    if ((AIO_WritePool_getFile(ress.writeCtx) == NULL) && (prefs->testMode == 0)) {
+        FILE *dstFile;
         int dstFilePermissions = DEFAULT_FILE_PERMISSIONS;
         if ( strcmp(srcFileName, stdinmark)   /* special case : don't transfer permissions from stdin */
-          && UTIL_stat(srcFileName, &statbuf)
-          && UTIL_isRegularFileStat(&statbuf) ) {
-            dstFilePermissions = statbuf.st_mode;
+          && strcmp(dstFileName, stdoutmark)
+          && UTIL_isRegularFileStat(srcFileStat) ) {
+            transferStat = 1;
+            dstFilePermissions = TEMPORARY_FILE_PERMISSIONS;
         }
 
         releaseDstFile = 1;
 
-        ress.dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName, dstFilePermissions);
-        if (ress.dstFile==NULL) return 1;
+        dstFile = FIO_openDstFile(fCtx, prefs, srcFileName, dstFileName, dstFilePermissions);
+        if (dstFile==NULL) return 1;
+        AIO_WritePool_setFile(ress.writeCtx, dstFile);
 
         /* Must only be added after FIO_openDstFile() succeeds.
          * Otherwise we may delete the destination file if it already exists,
@@ -2558,17 +2564,19 @@ static int FIO_decompressDstFile(FIO_ctx_t* const fCtx,
         addHandler(dstFileName);
     }
 
-    result = FIO_decompressFrames(fCtx, ress, srcFile, prefs, dstFileName, srcFileName);
+    result = FIO_decompressFrames(fCtx, ress, prefs, dstFileName, srcFileName);
 
     if (releaseDstFile) {
-        FILE* const dstFile = ress.dstFile;
         clearHandler();
-        ress.dstFile = NULL;
-        if (fclose(dstFile)) {
+        if (AIO_WritePool_closeFile(ress.writeCtx)) {
             DISPLAYLEVEL(1, "zstd: %s: %s \n", dstFileName, strerror(errno));
             result = 1;
         }
 
+        if (transferStat) {
+            UTIL_setFileStat(dstFileName, srcFileStat);
+        }
+
         if ( (result != 0)  /* operation failure */
           && strcmp(dstFileName, stdoutmark)  /* special case : don't remove() stdout */
           ) {
@@ -2588,18 +2596,34 @@ static int FIO_decompressDstFile(FIO_ctx_t* const fCtx,
 static int FIO_decompressSrcFile(FIO_ctx_t* const fCtx, FIO_prefs_t* const prefs, dRess_t ress, const char* dstFileName, const char* srcFileName)
 {
     FILE* srcFile;
+    stat_t srcFileStat;
     int result;
+    U64 fileSize = UTIL_FILESIZE_UNKNOWN;
 
     if (UTIL_isDirectory(srcFileName)) {
         DISPLAYLEVEL(1, "zstd: %s is a directory -- ignored \n", srcFileName);
         return 1;
     }
 
-    srcFile = FIO_openSrcFile(prefs, srcFileName);
+    srcFile = FIO_openSrcFile(prefs, srcFileName, &srcFileStat);
     if (srcFile==NULL) return 1;
-    ress.srcBufferLoaded = 0;
 
-    result = FIO_decompressDstFile(fCtx, prefs, ress, srcFile, dstFileName, srcFileName);
+    /* Don't use AsyncIO for small files */
+    if (strcmp(srcFileName, stdinmark)) /* Stdin doesn't have stats */
+        fileSize = UTIL_getFileSizeStat(&srcFileStat);
+    if(fileSize != UTIL_FILESIZE_UNKNOWN && fileSize < ZSTD_BLOCKSIZE_MAX * 3) {
+        AIO_ReadPool_setAsync(ress.readCtx, 0);
+        AIO_WritePool_setAsync(ress.writeCtx, 0);
+    } else {
+        AIO_ReadPool_setAsync(ress.readCtx, 1);
+        AIO_WritePool_setAsync(ress.writeCtx, 1);
+    }
+
+    AIO_ReadPool_setFile(ress.readCtx, srcFile);
+
+    result = FIO_decompressDstFile(fCtx, prefs, ress, dstFileName, srcFileName, &srcFileStat);
+
+    AIO_ReadPool_setFile(ress.readCtx, NULL);
 
     /* Close file */
     if (fclose(srcFile)) {
@@ -2688,6 +2712,11 @@ FIO_determineDstName(const char* srcFileName, const char* outDirName)
 
     size_t srcSuffixLen;
     const char* const srcSuffix = strrchr(srcFileName, '.');
+
+    if(!strcmp(srcFileName, stdinmark)) {
+        return stdoutmark;
+    }
+
     if (srcSuffix == NULL) {
         DISPLAYLEVEL(1,
             "zstd: %s: unknown suffix (%s expected). "
@@ -2769,20 +2798,21 @@ FIO_decompressMultipleFilenames(FIO_ctx_t* const fCtx,
     dRess_t ress = FIO_createDResources(prefs, dictFileName);
 
     if (outFileName) {
-        if (FIO_removeMultiFilesWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
+        if (FIO_multiFilesConcatWarning(fCtx, prefs, outFileName, 1 /* displayLevelCutoff */)) {
             FIO_freeDResources(ress);
             return 1;
         }
         if (!prefs->testMode) {
-            ress.dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName, DEFAULT_FILE_PERMISSIONS);
-            if (ress.dstFile == 0) EXM_THROW(19, "cannot open %s", outFileName);
+            FILE* dstFile = FIO_openDstFile(fCtx, prefs, NULL, outFileName, DEFAULT_FILE_PERMISSIONS);
+            if (dstFile == 0) EXM_THROW(19, "cannot open %s", outFileName);
+            AIO_WritePool_setFile(ress.writeCtx, dstFile);
         }
         for (; fCtx->currFileIdx < fCtx->nbFilesTotal; fCtx->currFileIdx++) {
             status = FIO_decompressSrcFile(fCtx, prefs, ress, outFileName, srcNamesTable[fCtx->currFileIdx]);
             if (!status) fCtx->nbFilesProcessed++;
             error |= status;
         }
-        if ((!prefs->testMode) && (fclose(ress.dstFile)))
+        if ((!prefs->testMode) && (AIO_WritePool_closeFile(ress.writeCtx)))
             EXM_THROW(72, "Write error : %s : cannot properly close output file",
                         strerror(errno));
     } else {
@@ -2812,8 +2842,11 @@ FIO_decompressMultipleFilenames(FIO_ctx_t* const fCtx,
             FIO_checkFilenameCollisions(srcNamesTable , (unsigned)fCtx->nbFilesTotal);
     }
 
-    if (fCtx->nbFilesProcessed >= 1  && fCtx->nbFilesTotal > 1 && fCtx->totalBytesOutput != 0)
-        DISPLAYLEVEL(2, "%d files decompressed : %6zu bytes total \n", fCtx->nbFilesProcessed, fCtx->totalBytesOutput);
+    if (FIO_shouldDisplayMultipleFileSummary(fCtx)) {
+        DISPLAY_PROGRESS("\r%79s\r", "");
+        DISPLAY_SUMMARY("%d files decompressed : %6llu bytes total \n",
+            fCtx->nbFilesProcessed, (unsigned long long)fCtx->totalBytesOutput);
+    }
 
     FIO_freeDResources(ress);
     return error;
@@ -2831,7 +2864,9 @@ typedef struct {
     int numSkippableFrames;
     int decompUnavailable;
     int usesCheck;
+    BYTE checksum[4];
     U32 nbFiles;
+    unsigned dictID;
 } fileInfo_t;
 
 typedef enum {
@@ -2839,7 +2874,7 @@ typedef enum {
   info_frame_error=1,
   info_not_zstd=2,
   info_file_error=3,
-  info_truncated_input=4,
+  info_truncated_input=4
 } InfoError;
 
 #define ERROR_IF(c,n,...) {             \
@@ -2886,6 +2921,12 @@ FIO_analyzeFrames(fileInfo_t* info, FILE* const srcFile)
                 }
                 ERROR_IF(ZSTD_getFrameHeader(&header, headerBuffer, numBytesRead) != 0,
                         info_frame_error, "Error: could not decode frame header");
+                if (info->dictID != 0 && info->dictID != header.dictID) {
+                    DISPLAY("WARNING: File contains multiple frames with different dictionary IDs. Showing dictID 0 instead");
+                    info->dictID = 0;
+                } else {
+                    info->dictID = header.dictID;
+                }
                 info->windowSize = header.windowSize;
                 /* move to the end of the frame header */
                 {   size_t const headerSize = ZSTD_frameHeaderSize(headerBuffer, numBytesRead);
@@ -2918,8 +2959,8 @@ FIO_analyzeFrames(fileInfo_t* info, FILE* const srcFile)
                     int const contentChecksumFlag = (frameHeaderDescriptor & (1 << 2)) >> 2;
                     if (contentChecksumFlag) {
                         info->usesCheck = 1;
-                        ERROR_IF(fseek(srcFile, 4, SEEK_CUR) != 0,
-                                info_frame_error, "Error: could not skip past checksum");
+                        ERROR_IF(fread(info->checksum, 1, 4, srcFile) != 4,
+                                info_frame_error, "Error: could not read checksum");
                 }   }
                 info->numActualFrames++;
             }
@@ -2945,10 +2986,11 @@ static InfoError
 getFileInfo_fileConfirmed(fileInfo_t* info, const char* inFileName)
 {
     InfoError status;
-    FILE* const srcFile = FIO_openSrcFile(NULL, inFileName);
+    stat_t srcFileStat;
+    FILE* const srcFile = FIO_openSrcFile(NULL, inFileName, &srcFileStat);
     ERROR_IF(srcFile == NULL, info_file_error, "Error: could not open source file %s", inFileName);
 
-    info->compressedSize = UTIL_getFileSize(inFileName);
+    info->compressedSize = UTIL_getFileSizeStat(&srcFileStat);
     status = FIO_analyzeFrames(info, srcFile);
 
     fclose(srcFile);
@@ -2973,25 +3015,24 @@ getFileInfo(fileInfo_t* info, const char* srcFileName)
 static void
 displayInfo(const char* inFileName, const fileInfo_t* info, int displayLevel)
 {
-    unsigned const unit = info->compressedSize < (1 MB) ? (1 KB) : (1 MB);
-    const char* const unitStr = info->compressedSize < (1 MB) ? "KB" : "MB";
-    double const windowSizeUnit = (double)info->windowSize / unit;
-    double const compressedSizeUnit = (double)info->compressedSize / unit;
-    double const decompressedSizeUnit = (double)info->decompressedSize / unit;
+    UTIL_HumanReadableSize_t const window_hrs = UTIL_makeHumanReadableSize(info->windowSize);
+    UTIL_HumanReadableSize_t const compressed_hrs = UTIL_makeHumanReadableSize(info->compressedSize);
+    UTIL_HumanReadableSize_t const decompressed_hrs = UTIL_makeHumanReadableSize(info->decompressedSize);
     double const ratio = (info->compressedSize == 0) ? 0 : ((double)info->decompressedSize)/(double)info->compressedSize;
     const char* const checkString = (info->usesCheck ? "XXH64" : "None");
     if (displayLevel <= 2) {
         if (!info->decompUnavailable) {
-            DISPLAYOUT("%6d  %5d  %7.2f %2s  %9.2f %2s  %5.3f  %5s  %s\n",
+            DISPLAYOUT("%6d  %5d  %6.*f%4s  %8.*f%4s  %5.3f  %5s  %s\n",
                     info->numSkippableFrames + info->numActualFrames,
                     info->numSkippableFrames,
-                    compressedSizeUnit, unitStr, decompressedSizeUnit, unitStr,
+                    compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                    decompressed_hrs.precision, decompressed_hrs.value, decompressed_hrs.suffix,
                     ratio, checkString, inFileName);
         } else {
-            DISPLAYOUT("%6d  %5d  %7.2f %2s                       %5s  %s\n",
+            DISPLAYOUT("%6d  %5d  %6.*f%4s                       %5s  %s\n",
                     info->numSkippableFrames + info->numActualFrames,
                     info->numSkippableFrames,
-                    compressedSizeUnit, unitStr,
+                    compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
                     checkString, inFileName);
         }
     } else {
@@ -2999,19 +3040,29 @@ displayInfo(const char* inFileName, const fileInfo_t* info, int displayLevel)
         DISPLAYOUT("# Zstandard Frames: %d\n", info->numActualFrames);
         if (info->numSkippableFrames)
             DISPLAYOUT("# Skippable Frames: %d\n", info->numSkippableFrames);
-        DISPLAYOUT("Window Size: %.2f %2s (%llu B)\n",
-                   windowSizeUnit, unitStr,
+        DISPLAYOUT("DictID: %u\n", info->dictID);
+        DISPLAYOUT("Window Size: %.*f%s (%llu B)\n",
+                   window_hrs.precision, window_hrs.value, window_hrs.suffix,
                    (unsigned long long)info->windowSize);
-        DISPLAYOUT("Compressed Size: %.2f %2s (%llu B)\n",
-                    compressedSizeUnit, unitStr,
+        DISPLAYOUT("Compressed Size: %.*f%s (%llu B)\n",
+                    compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
                     (unsigned long long)info->compressedSize);
         if (!info->decompUnavailable) {
-            DISPLAYOUT("Decompressed Size: %.2f %2s (%llu B)\n",
-                    decompressedSizeUnit, unitStr,
+            DISPLAYOUT("Decompressed Size: %.*f%s (%llu B)\n",
+                    decompressed_hrs.precision, decompressed_hrs.value, decompressed_hrs.suffix,
                     (unsigned long long)info->decompressedSize);
             DISPLAYOUT("Ratio: %.4f\n", ratio);
         }
-        DISPLAYOUT("Check: %s\n", checkString);
+
+        if (info->usesCheck && info->numActualFrames == 1) {
+            DISPLAYOUT("Check: %s %02x%02x%02x%02x\n", checkString,
+                info->checksum[3], info->checksum[2],
+                info->checksum[1], info->checksum[0]
+            );
+        } else {
+            DISPLAYOUT("Check: %s\n", checkString);
+        }
+
         DISPLAYOUT("\n");
     }
 }
@@ -3075,7 +3126,7 @@ int FIO_listMultipleFiles(unsigned numFiles, const char** filenameTable, int dis
     }   }
 
     if (numFiles == 0) {
-        if (!IS_CONSOLE(stdin)) {
+        if (!UTIL_isConsole(stdin)) {
             DISPLAYLEVEL(1, "zstd: --list does not support reading from standard input \n");
         }
         DISPLAYLEVEL(1, "No files given \n");
@@ -3095,24 +3146,23 @@ int FIO_listMultipleFiles(unsigned numFiles, const char** filenameTable, int dis
                 error |= FIO_listFile(&total, filenameTable[u], displayLevel);
         }   }
         if (numFiles > 1 && displayLevel <= 2) {   /* display total */
-            unsigned const unit = total.compressedSize < (1 MB) ? (1 KB) : (1 MB);
-            const char* const unitStr = total.compressedSize < (1 MB) ? "KB" : "MB";
-            double const compressedSizeUnit = (double)total.compressedSize / unit;
-            double const decompressedSizeUnit = (double)total.decompressedSize / unit;
+            UTIL_HumanReadableSize_t const compressed_hrs = UTIL_makeHumanReadableSize(total.compressedSize);
+            UTIL_HumanReadableSize_t const decompressed_hrs = UTIL_makeHumanReadableSize(total.decompressedSize);
             double const ratio = (total.compressedSize == 0) ? 0 : ((double)total.decompressedSize)/(double)total.compressedSize;
             const char* const checkString = (total.usesCheck ? "XXH64" : "");
             DISPLAYOUT("----------------------------------------------------------------- \n");
             if (total.decompUnavailable) {
-                DISPLAYOUT("%6d  %5d  %7.2f %2s                       %5s  %u files\n",
+                DISPLAYOUT("%6d  %5d  %6.*f%4s                       %5s  %u files\n",
                         total.numSkippableFrames + total.numActualFrames,
                         total.numSkippableFrames,
-                        compressedSizeUnit, unitStr,
+                        compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
                         checkString, (unsigned)total.nbFiles);
             } else {
-                DISPLAYOUT("%6d  %5d  %7.2f %2s  %9.2f %2s  %5.3f  %5s  %u files\n",
+                DISPLAYOUT("%6d  %5d  %6.*f%4s  %8.*f%4s  %5.3f  %5s  %u files\n",
                         total.numSkippableFrames + total.numActualFrames,
                         total.numSkippableFrames,
-                        compressedSizeUnit, unitStr, decompressedSizeUnit, unitStr,
+                        compressed_hrs.precision, compressed_hrs.value, compressed_hrs.suffix,
+                        decompressed_hrs.precision, decompressed_hrs.value, decompressed_hrs.suffix,
                         ratio, checkString, (unsigned)total.nbFiles);
         }   }
         return error;
diff --git a/priv/zstd/programs/fileio.h b/priv/zstd/programs/fileio.h
index 9d97ec8..291d4d4 100644
--- a/priv/zstd/programs/fileio.h
+++ b/priv/zstd/programs/fileio.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -12,6 +12,8 @@
 #ifndef FILEIO_H_23981798732
 #define FILEIO_H_23981798732
 
+#include "fileio_types.h"
+#include "util.h"                  /* FileNamesTable */
 #define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_compressionParameters */
 #include "../lib/zstd.h"           /* ZSTD_* */
 
@@ -53,10 +55,6 @@ extern "C" {
 /*-*************************************
 *  Types
 ***************************************/
-typedef enum { FIO_zstdCompression, FIO_gzipCompression, FIO_xzCompression, FIO_lzmaCompression, FIO_lz4Compression } FIO_compressionType_t;
-
-typedef struct FIO_prefs_s FIO_prefs_t;
-
 FIO_prefs_t* FIO_createPreferences(void);
 void FIO_freePreferences(FIO_prefs_t* const prefs);
 
@@ -66,9 +64,6 @@ typedef struct FIO_ctx_s FIO_ctx_t;
 FIO_ctx_t* FIO_createContext(void);
 void FIO_freeContext(FIO_ctx_t* const fCtx);
 
-typedef struct FIO_display_prefs_s FIO_display_prefs_t;
-
-typedef enum { FIO_ps_auto, FIO_ps_never, FIO_ps_always } FIO_progressSetting_e;
 
 /*-*************************************
 *  Parameters
@@ -76,7 +71,7 @@ typedef enum { FIO_ps_auto, FIO_ps_never, FIO_ps_always } FIO_progressSetting_e;
 /* FIO_prefs_t functions */
 void FIO_setCompressionType(FIO_prefs_t* const prefs, FIO_compressionType_t compressionType);
 void FIO_overwriteMode(FIO_prefs_t* const prefs);
-void FIO_setAdaptiveMode(FIO_prefs_t* const prefs, unsigned adapt);
+void FIO_setAdaptiveMode(FIO_prefs_t* const prefs, int adapt);
 void FIO_setAdaptMin(FIO_prefs_t* const prefs, int minCLevel);
 void FIO_setAdaptMax(FIO_prefs_t* const prefs, int maxCLevel);
 void FIO_setUseRowMatchFinder(FIO_prefs_t* const prefs, int useRowMatchFinder);
@@ -91,8 +86,8 @@ void FIO_setLdmMinMatch(FIO_prefs_t* const prefs, int ldmMinMatch);
 void FIO_setMemLimit(FIO_prefs_t* const prefs, unsigned memLimit);
 void FIO_setNbWorkers(FIO_prefs_t* const prefs, int nbWorkers);
 void FIO_setOverlapLog(FIO_prefs_t* const prefs, int overlapLog);
-void FIO_setRemoveSrcFile(FIO_prefs_t* const prefs, unsigned flag);
-void FIO_setSparseWrite(FIO_prefs_t* const prefs, unsigned sparse);  /**< 0: no sparse; 1: disable on stdout; 2: always enabled */
+void FIO_setRemoveSrcFile(FIO_prefs_t* const prefs, int flag);
+void FIO_setSparseWrite(FIO_prefs_t* const prefs, int sparse);  /**< 0: no sparse; 1: disable on stdout; 2: always enabled */
 void FIO_setRsyncable(FIO_prefs_t* const prefs, int rsyncable);
 void FIO_setStreamSrcSize(FIO_prefs_t* const prefs, size_t streamSrcSize);
 void FIO_setTargetCBlockSize(FIO_prefs_t* const prefs, size_t targetCBlockSize);
@@ -100,7 +95,7 @@ void FIO_setSrcSizeHint(FIO_prefs_t* const prefs, size_t srcSizeHint);
 void FIO_setTestMode(FIO_prefs_t* const prefs, int testMode);
 void FIO_setLiteralCompressionMode(
         FIO_prefs_t* const prefs,
-        ZSTD_literalCompressionMode_e mode);
+        ZSTD_paramSwitch_e mode);
 
 void FIO_setProgressSetting(FIO_progressSetting_e progressSetting);
 void FIO_setNotificationLevel(int level);
@@ -108,6 +103,9 @@ void FIO_setExcludeCompressedFile(FIO_prefs_t* const prefs, int excludeCompresse
 void FIO_setAllowBlockDevices(FIO_prefs_t* const prefs, int allowBlockDevices);
 void FIO_setPatchFromMode(FIO_prefs_t* const prefs, int value);
 void FIO_setContentSize(FIO_prefs_t* const prefs, int value);
+void FIO_displayCompressionParameters(const FIO_prefs_t* prefs);
+void FIO_setAsyncIOFlag(FIO_prefs_t* const prefs, int value);
+void FIO_setPassThroughFlag(FIO_prefs_t* const prefs, int value);
 
 /* FIO_ctx_t functions */
 void FIO_setNbFilesTotal(FIO_ctx_t* const fCtx, int value);
@@ -170,6 +168,9 @@ int FIO_checkFilenameCollisions(const char** filenameTable, unsigned nbFiles);
 /* custom crash signal handler */
 void FIO_addAbortHandler(void);
 
+char const* FIO_zlibVersion(void);
+char const* FIO_lz4Version(void);
+char const* FIO_lzmaVersion(void);
 
 
 #if defined (__cplusplus)
diff --git a/priv/zstd/programs/fileio_asyncio.c b/priv/zstd/programs/fileio_asyncio.c
new file mode 100644
index 0000000..fe9cca9
--- /dev/null
+++ b/priv/zstd/programs/fileio_asyncio.c
@@ -0,0 +1,663 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "platform.h"
+#include       /* fprintf, open, fdopen, fread, _fileno, stdin, stdout */
+#include      /* malloc, free */
+#include 
+#include       /* errno */
+
+#if defined (_MSC_VER)
+#  include 
+#  include 
+#endif
+
+#include "fileio_asyncio.h"
+#include "fileio_common.h"
+
+/* **********************************************************************
+ *  Sparse write
+ ************************************************************************/
+
+/** AIO_fwriteSparse() :
+*  @return : storedSkips,
+*            argument for next call to AIO_fwriteSparse() or AIO_fwriteSparseEnd() */
+static unsigned
+AIO_fwriteSparse(FILE* file,
+                 const void* buffer, size_t bufferSize,
+                 const FIO_prefs_t* const prefs,
+                 unsigned storedSkips)
+{
+    const size_t* const bufferT = (const size_t*)buffer;   /* Buffer is supposed malloc'ed, hence aligned on size_t */
+    size_t bufferSizeT = bufferSize / sizeof(size_t);
+    const size_t* const bufferTEnd = bufferT + bufferSizeT;
+    const size_t* ptrT = bufferT;
+    static const size_t segmentSizeT = (32 KB) / sizeof(size_t);   /* check every 32 KB */
+
+    if (prefs->testMode) return 0;  /* do not output anything in test mode */
+
+    if (!prefs->sparseFileSupport) {  /* normal write */
+        size_t const sizeCheck = fwrite(buffer, 1, bufferSize, file);
+        if (sizeCheck != bufferSize)
+            EXM_THROW(70, "Write error : cannot write block : %s",
+                      strerror(errno));
+        return 0;
+    }
+
+    /* avoid int overflow */
+    if (storedSkips > 1 GB) {
+        if (LONG_SEEK(file, 1 GB, SEEK_CUR) != 0)
+        EXM_THROW(91, "1 GB skip error (sparse file support)");
+        storedSkips -= 1 GB;
+    }
+
+    while (ptrT < bufferTEnd) {
+        size_t nb0T;
+
+        /* adjust last segment if < 32 KB */
+        size_t seg0SizeT = segmentSizeT;
+        if (seg0SizeT > bufferSizeT) seg0SizeT = bufferSizeT;
+        bufferSizeT -= seg0SizeT;
+
+        /* count leading zeroes */
+        for (nb0T=0; (nb0T < seg0SizeT) && (ptrT[nb0T] == 0); nb0T++) ;
+        storedSkips += (unsigned)(nb0T * sizeof(size_t));
+
+        if (nb0T != seg0SizeT) {   /* not all 0s */
+            size_t const nbNon0ST = seg0SizeT - nb0T;
+            /* skip leading zeros */
+            if (LONG_SEEK(file, storedSkips, SEEK_CUR) != 0)
+                EXM_THROW(92, "Sparse skip error ; try --no-sparse");
+            storedSkips = 0;
+            /* write the rest */
+            if (fwrite(ptrT + nb0T, sizeof(size_t), nbNon0ST, file) != nbNon0ST)
+                EXM_THROW(93, "Write error : cannot write block : %s",
+                          strerror(errno));
+        }
+        ptrT += seg0SizeT;
+    }
+
+    {   static size_t const maskT = sizeof(size_t)-1;
+        if (bufferSize & maskT) {
+            /* size not multiple of sizeof(size_t) : implies end of block */
+            const char* const restStart = (const char*)bufferTEnd;
+            const char* restPtr = restStart;
+            const char* const restEnd = (const char*)buffer + bufferSize;
+            assert(restEnd > restStart && restEnd < restStart + sizeof(size_t));
+            for ( ; (restPtr < restEnd) && (*restPtr == 0); restPtr++) ;
+            storedSkips += (unsigned) (restPtr - restStart);
+            if (restPtr != restEnd) {
+                /* not all remaining bytes are 0 */
+                size_t const restSize = (size_t)(restEnd - restPtr);
+                if (LONG_SEEK(file, storedSkips, SEEK_CUR) != 0)
+                    EXM_THROW(92, "Sparse skip error ; try --no-sparse");
+                if (fwrite(restPtr, 1, restSize, file) != restSize)
+                    EXM_THROW(95, "Write error : cannot write end of decoded block : %s",
+                              strerror(errno));
+                storedSkips = 0;
+            }   }   }
+
+    return storedSkips;
+}
+
+static void
+AIO_fwriteSparseEnd(const FIO_prefs_t* const prefs, FILE* file, unsigned storedSkips)
+{
+    if (prefs->testMode) assert(storedSkips == 0);
+    if (storedSkips>0) {
+        assert(prefs->sparseFileSupport > 0);  /* storedSkips>0 implies sparse support is enabled */
+        (void)prefs;   /* assert can be disabled, in which case prefs becomes unused */
+        if (LONG_SEEK(file, storedSkips-1, SEEK_CUR) != 0)
+            EXM_THROW(69, "Final skip error (sparse file support)");
+        /* last zero must be explicitly written,
+         * so that skipped ones get implicitly translated as zero by FS */
+        {   const char lastZeroByte[1] = { 0 };
+            if (fwrite(lastZeroByte, 1, 1, file) != 1)
+                EXM_THROW(69, "Write error : cannot write last zero : %s", strerror(errno));
+        }   }
+}
+
+
+/* **********************************************************************
+ *  AsyncIO functionality
+ ************************************************************************/
+
+/* AIO_supported:
+ * Returns 1 if AsyncIO is supported on the system, 0 otherwise. */
+int AIO_supported(void) {
+#ifdef ZSTD_MULTITHREAD
+    return 1;
+#else
+    return 0;
+#endif
+}
+
+/* ***********************************
+ *  Generic IoPool implementation
+ *************************************/
+
+static IOJob_t *AIO_IOPool_createIoJob(IOPoolCtx_t *ctx, size_t bufferSize) {
+    IOJob_t* const job  = (IOJob_t*) malloc(sizeof(IOJob_t));
+    void* const buffer = malloc(bufferSize);
+    if(!job || !buffer)
+        EXM_THROW(101, "Allocation error : not enough memory");
+    job->buffer = buffer;
+    job->bufferSize = bufferSize;
+    job->usedBufferSize = 0;
+    job->file = NULL;
+    job->ctx = ctx;
+    job->offset = 0;
+    return job;
+}
+
+
+/* AIO_IOPool_createThreadPool:
+ * Creates a thread pool and a mutex for threaded IO pool.
+ * Displays warning if asyncio is requested but MT isn't available. */
+static void AIO_IOPool_createThreadPool(IOPoolCtx_t* ctx, const FIO_prefs_t* prefs) {
+    ctx->threadPool = NULL;
+    ctx->threadPoolActive = 0;
+    if(prefs->asyncIO) {
+        if (ZSTD_pthread_mutex_init(&ctx->ioJobsMutex, NULL))
+            EXM_THROW(102,"Failed creating ioJobsMutex mutex");
+        /* We want MAX_IO_JOBS-2 queue items because we need to always have 1 free buffer to
+         * decompress into and 1 buffer that's actively written to disk and owned by the writing thread. */
+        assert(MAX_IO_JOBS >= 2);
+        ctx->threadPool = POOL_create(1, MAX_IO_JOBS - 2);
+        ctx->threadPoolActive = 1;
+        if (!ctx->threadPool)
+            EXM_THROW(104, "Failed creating I/O thread pool");
+    }
+}
+
+/* AIO_IOPool_init:
+ * Allocates and sets and a new I/O thread pool including its included availableJobs. */
+static void AIO_IOPool_init(IOPoolCtx_t* ctx, const FIO_prefs_t* prefs, POOL_function poolFunction, size_t bufferSize) {
+    int i;
+    AIO_IOPool_createThreadPool(ctx, prefs);
+    ctx->prefs = prefs;
+    ctx->poolFunction = poolFunction;
+    ctx->totalIoJobs = ctx->threadPool ? MAX_IO_JOBS : 2;
+    ctx->availableJobsCount = ctx->totalIoJobs;
+    for(i=0; i < ctx->availableJobsCount; i++) {
+        ctx->availableJobs[i] = AIO_IOPool_createIoJob(ctx, bufferSize);
+    }
+    ctx->jobBufferSize = bufferSize;
+    ctx->file = NULL;
+}
+
+
+/* AIO_IOPool_threadPoolActive:
+ * Check if current operation uses thread pool.
+ * Note that in some cases we have a thread pool initialized but choose not to use it. */
+static int AIO_IOPool_threadPoolActive(IOPoolCtx_t* ctx) {
+    return ctx->threadPool && ctx->threadPoolActive;
+}
+
+
+/* AIO_IOPool_lockJobsMutex:
+ * Locks the IO jobs mutex if threading is active */
+static void AIO_IOPool_lockJobsMutex(IOPoolCtx_t* ctx) {
+    if(AIO_IOPool_threadPoolActive(ctx))
+        ZSTD_pthread_mutex_lock(&ctx->ioJobsMutex);
+}
+
+/* AIO_IOPool_unlockJobsMutex:
+ * Unlocks the IO jobs mutex if threading is active */
+static void AIO_IOPool_unlockJobsMutex(IOPoolCtx_t* ctx) {
+    if(AIO_IOPool_threadPoolActive(ctx))
+        ZSTD_pthread_mutex_unlock(&ctx->ioJobsMutex);
+}
+
+/* AIO_IOPool_releaseIoJob:
+ * Releases an acquired job back to the pool. Doesn't execute the job. */
+static void AIO_IOPool_releaseIoJob(IOJob_t* job) {
+    IOPoolCtx_t* const ctx = (IOPoolCtx_t *) job->ctx;
+    AIO_IOPool_lockJobsMutex(ctx);
+    assert(ctx->availableJobsCount < ctx->totalIoJobs);
+    ctx->availableJobs[ctx->availableJobsCount++] = job;
+    AIO_IOPool_unlockJobsMutex(ctx);
+}
+
+/* AIO_IOPool_join:
+ * Waits for all tasks in the pool to finish executing. */
+static void AIO_IOPool_join(IOPoolCtx_t* ctx) {
+    if(AIO_IOPool_threadPoolActive(ctx))
+        POOL_joinJobs(ctx->threadPool);
+}
+
+/* AIO_IOPool_setThreaded:
+ * Allows (de)activating threaded mode, to be used when the expected overhead
+ * of threading costs more than the expected gains. */
+static void AIO_IOPool_setThreaded(IOPoolCtx_t* ctx, int threaded) {
+    assert(threaded == 0 || threaded == 1);
+    assert(ctx != NULL);
+    if(ctx->threadPoolActive != threaded) {
+        AIO_IOPool_join(ctx);
+        ctx->threadPoolActive = threaded;
+    }
+}
+
+/* AIO_IOPool_free:
+ * Release a previously allocated IO thread pool. Makes sure all tasks are done and released. */
+static void AIO_IOPool_destroy(IOPoolCtx_t* ctx) {
+    int i;
+    if(ctx->threadPool) {
+        /* Make sure we finish all tasks and then free the resources */
+        AIO_IOPool_join(ctx);
+        /* Make sure we are not leaking availableJobs */
+        assert(ctx->availableJobsCount == ctx->totalIoJobs);
+        POOL_free(ctx->threadPool);
+        ZSTD_pthread_mutex_destroy(&ctx->ioJobsMutex);
+    }
+    assert(ctx->file == NULL);
+    for(i=0; iavailableJobsCount; i++) {
+        IOJob_t* job = (IOJob_t*) ctx->availableJobs[i];
+        free(job->buffer);
+        free(job);
+    }
+}
+
+/* AIO_IOPool_acquireJob:
+ * Returns an available io job to be used for a future io. */
+static IOJob_t* AIO_IOPool_acquireJob(IOPoolCtx_t* ctx) {
+    IOJob_t *job;
+    assert(ctx->file != NULL || ctx->prefs->testMode);
+    AIO_IOPool_lockJobsMutex(ctx);
+    assert(ctx->availableJobsCount > 0);
+    job = (IOJob_t*) ctx->availableJobs[--ctx->availableJobsCount];
+    AIO_IOPool_unlockJobsMutex(ctx);
+    job->usedBufferSize = 0;
+    job->file = ctx->file;
+    job->offset = 0;
+    return job;
+}
+
+
+/* AIO_IOPool_setFile:
+ * Sets the destination file for future files in the pool.
+ * Requires completion of all queued jobs and release of all otherwise acquired jobs. */
+static void AIO_IOPool_setFile(IOPoolCtx_t* ctx, FILE* file) {
+    assert(ctx!=NULL);
+    AIO_IOPool_join(ctx);
+    assert(ctx->availableJobsCount == ctx->totalIoJobs);
+    ctx->file = file;
+}
+
+static FILE* AIO_IOPool_getFile(const IOPoolCtx_t* ctx) {
+    return ctx->file;
+}
+
+/* AIO_IOPool_enqueueJob:
+ * Enqueues an io job for execution.
+ * The queued job shouldn't be used directly after queueing it. */
+static void AIO_IOPool_enqueueJob(IOJob_t* job) {
+    IOPoolCtx_t* const ctx = (IOPoolCtx_t *)job->ctx;
+    if(AIO_IOPool_threadPoolActive(ctx))
+        POOL_add(ctx->threadPool, ctx->poolFunction, job);
+    else
+        ctx->poolFunction(job);
+}
+
+/* ***********************************
+ *  WritePool implementation
+ *************************************/
+
+/* AIO_WritePool_acquireJob:
+ * Returns an available write job to be used for a future write. */
+IOJob_t* AIO_WritePool_acquireJob(WritePoolCtx_t* ctx) {
+    return AIO_IOPool_acquireJob(&ctx->base);
+}
+
+/* AIO_WritePool_enqueueAndReacquireWriteJob:
+ * Queues a write job for execution and acquires a new one.
+ * After execution `job`'s pointed value would change to the newly acquired job.
+ * Make sure to set `usedBufferSize` to the wanted length before call.
+ * The queued job shouldn't be used directly after queueing it. */
+void AIO_WritePool_enqueueAndReacquireWriteJob(IOJob_t **job) {
+    AIO_IOPool_enqueueJob(*job);
+    *job = AIO_IOPool_acquireJob((IOPoolCtx_t *)(*job)->ctx);
+}
+
+/* AIO_WritePool_sparseWriteEnd:
+ * Ends sparse writes to the current file.
+ * Blocks on completion of all current write jobs before executing. */
+void AIO_WritePool_sparseWriteEnd(WritePoolCtx_t* ctx) {
+    assert(ctx != NULL);
+    AIO_IOPool_join(&ctx->base);
+    AIO_fwriteSparseEnd(ctx->base.prefs, ctx->base.file, ctx->storedSkips);
+    ctx->storedSkips = 0;
+}
+
+/* AIO_WritePool_setFile:
+ * Sets the destination file for future writes in the pool.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.
+ * Also requires ending of sparse write if a previous file was used in sparse mode. */
+void AIO_WritePool_setFile(WritePoolCtx_t* ctx, FILE* file) {
+    AIO_IOPool_setFile(&ctx->base, file);
+    assert(ctx->storedSkips == 0);
+}
+
+/* AIO_WritePool_getFile:
+ * Returns the file the writePool is currently set to write to. */
+FILE* AIO_WritePool_getFile(const WritePoolCtx_t* ctx) {
+    return AIO_IOPool_getFile(&ctx->base);
+}
+
+/* AIO_WritePool_releaseIoJob:
+ * Releases an acquired job back to the pool. Doesn't execute the job. */
+void AIO_WritePool_releaseIoJob(IOJob_t* job) {
+    AIO_IOPool_releaseIoJob(job);
+}
+
+/* AIO_WritePool_closeFile:
+ * Ends sparse write and closes the writePool's current file and sets the file to NULL.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.  */
+int AIO_WritePool_closeFile(WritePoolCtx_t* ctx) {
+    FILE* const dstFile = ctx->base.file;
+    assert(dstFile!=NULL || ctx->base.prefs->testMode!=0);
+    AIO_WritePool_sparseWriteEnd(ctx);
+    AIO_IOPool_setFile(&ctx->base, NULL);
+    return fclose(dstFile);
+}
+
+/* AIO_WritePool_executeWriteJob:
+ * Executes a write job synchronously. Can be used as a function for a thread pool. */
+static void AIO_WritePool_executeWriteJob(void* opaque){
+    IOJob_t* const job = (IOJob_t*) opaque;
+    WritePoolCtx_t* const ctx = (WritePoolCtx_t*) job->ctx;
+    ctx->storedSkips = AIO_fwriteSparse(job->file, job->buffer, job->usedBufferSize, ctx->base.prefs, ctx->storedSkips);
+    AIO_IOPool_releaseIoJob(job);
+}
+
+/* AIO_WritePool_create:
+ * Allocates and sets and a new write pool including its included jobs. */
+WritePoolCtx_t* AIO_WritePool_create(const FIO_prefs_t* prefs, size_t bufferSize) {
+    WritePoolCtx_t* const ctx = (WritePoolCtx_t*) malloc(sizeof(WritePoolCtx_t));
+    if(!ctx) EXM_THROW(100, "Allocation error : not enough memory");
+    AIO_IOPool_init(&ctx->base, prefs, AIO_WritePool_executeWriteJob, bufferSize);
+    ctx->storedSkips = 0;
+    return ctx;
+}
+
+/* AIO_WritePool_free:
+ * Frees and releases a writePool and its resources. Closes destination file if needs to. */
+void AIO_WritePool_free(WritePoolCtx_t* ctx) {
+    /* Make sure we finish all tasks and then free the resources */
+    if(AIO_WritePool_getFile(ctx))
+        AIO_WritePool_closeFile(ctx);
+    AIO_IOPool_destroy(&ctx->base);
+    assert(ctx->storedSkips==0);
+    free(ctx);
+}
+
+/* AIO_WritePool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_WritePool_setAsync(WritePoolCtx_t* ctx, int async) {
+    AIO_IOPool_setThreaded(&ctx->base, async);
+}
+
+
+/* ***********************************
+ *  ReadPool implementation
+ *************************************/
+static void AIO_ReadPool_releaseAllCompletedJobs(ReadPoolCtx_t* ctx) {
+    int i;
+    for(i=0; icompletedJobsCount; i++) {
+        IOJob_t* job = (IOJob_t*) ctx->completedJobs[i];
+        AIO_IOPool_releaseIoJob(job);
+    }
+    ctx->completedJobsCount = 0;
+}
+
+static void AIO_ReadPool_addJobToCompleted(IOJob_t* job) {
+    ReadPoolCtx_t* const ctx = (ReadPoolCtx_t *)job->ctx;
+    AIO_IOPool_lockJobsMutex(&ctx->base);
+    assert(ctx->completedJobsCount < MAX_IO_JOBS);
+    ctx->completedJobs[ctx->completedJobsCount++] = job;
+    if(AIO_IOPool_threadPoolActive(&ctx->base)) {
+        ZSTD_pthread_cond_signal(&ctx->jobCompletedCond);
+    }
+    AIO_IOPool_unlockJobsMutex(&ctx->base);
+}
+
+/* AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked:
+ * Looks through the completed jobs for a job matching the waitingOnOffset and returns it,
+ * if job wasn't found returns NULL.
+ * IMPORTANT: assumes ioJobsMutex is locked. */
+static IOJob_t* AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked(ReadPoolCtx_t* ctx) {
+    IOJob_t *job = NULL;
+    int i;
+    /* This implementation goes through all completed jobs and looks for the one matching the next offset.
+     * While not strictly needed for a single threaded reader implementation (as in such a case we could expect
+     * reads to be completed in order) this implementation was chosen as it better fits other asyncio
+     * interfaces (such as io_uring) that do not provide promises regarding order of completion. */
+    for (i=0; icompletedJobsCount; i++) {
+        job = (IOJob_t *) ctx->completedJobs[i];
+        if (job->offset == ctx->waitingOnOffset) {
+            ctx->completedJobs[i] = ctx->completedJobs[--ctx->completedJobsCount];
+            return job;
+        }
+    }
+    return NULL;
+}
+
+/* AIO_ReadPool_numReadsInFlight:
+ * Returns the number of IO read jobs currently in flight. */
+static size_t AIO_ReadPool_numReadsInFlight(ReadPoolCtx_t* ctx) {
+    const size_t jobsHeld = (ctx->currentJobHeld==NULL ? 0 : 1);
+    return ctx->base.totalIoJobs - (ctx->base.availableJobsCount + ctx->completedJobsCount + jobsHeld);
+}
+
+/* AIO_ReadPool_getNextCompletedJob:
+ * Returns a completed IOJob_t for the next read in line based on waitingOnOffset and advances waitingOnOffset.
+ * Would block. */
+static IOJob_t* AIO_ReadPool_getNextCompletedJob(ReadPoolCtx_t* ctx) {
+    IOJob_t *job = NULL;
+    AIO_IOPool_lockJobsMutex(&ctx->base);
+
+    job = AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked(ctx);
+
+    /* As long as we didn't find the job matching the next read, and we have some reads in flight continue waiting */
+    while (!job && (AIO_ReadPool_numReadsInFlight(ctx) > 0)) {
+        assert(ctx->base.threadPool != NULL); /* we shouldn't be here if we work in sync mode */
+        ZSTD_pthread_cond_wait(&ctx->jobCompletedCond, &ctx->base.ioJobsMutex);
+        job = AIO_ReadPool_findNextWaitingOffsetCompletedJob_locked(ctx);
+    }
+
+    if(job) {
+        assert(job->offset == ctx->waitingOnOffset);
+        ctx->waitingOnOffset += job->usedBufferSize;
+    }
+
+    AIO_IOPool_unlockJobsMutex(&ctx->base);
+    return job;
+}
+
+
+/* AIO_ReadPool_executeReadJob:
+ * Executes a read job synchronously. Can be used as a function for a thread pool. */
+static void AIO_ReadPool_executeReadJob(void* opaque){
+    IOJob_t* const job = (IOJob_t*) opaque;
+    ReadPoolCtx_t* const ctx = (ReadPoolCtx_t *)job->ctx;
+    if(ctx->reachedEof) {
+        job->usedBufferSize = 0;
+        AIO_ReadPool_addJobToCompleted(job);
+        return;
+    }
+    job->usedBufferSize = fread(job->buffer, 1, job->bufferSize, job->file);
+    if(job->usedBufferSize < job->bufferSize) {
+        if(ferror(job->file)) {
+            EXM_THROW(37, "Read error");
+        } else if(feof(job->file)) {
+            ctx->reachedEof = 1;
+        } else {
+            EXM_THROW(37, "Unexpected short read");
+        }
+    }
+    AIO_ReadPool_addJobToCompleted(job);
+}
+
+static void AIO_ReadPool_enqueueRead(ReadPoolCtx_t* ctx) {
+    IOJob_t* const job = AIO_IOPool_acquireJob(&ctx->base);
+    job->offset = ctx->nextReadOffset;
+    ctx->nextReadOffset += job->bufferSize;
+    AIO_IOPool_enqueueJob(job);
+}
+
+static void AIO_ReadPool_startReading(ReadPoolCtx_t* ctx) {
+    int i;
+    for (i = 0; i < ctx->base.availableJobsCount; i++) {
+        AIO_ReadPool_enqueueRead(ctx);
+    }
+}
+
+/* AIO_ReadPool_setFile:
+ * Sets the source file for future read in the pool. Initiates reading immediately if file is not NULL.
+ * Waits for all current enqueued tasks to complete if a previous file was set. */
+void AIO_ReadPool_setFile(ReadPoolCtx_t* ctx, FILE* file) {
+    assert(ctx!=NULL);
+    AIO_IOPool_join(&ctx->base);
+    AIO_ReadPool_releaseAllCompletedJobs(ctx);
+    if (ctx->currentJobHeld) {
+        AIO_IOPool_releaseIoJob((IOJob_t *)ctx->currentJobHeld);
+        ctx->currentJobHeld = NULL;
+    }
+    AIO_IOPool_setFile(&ctx->base, file);
+    ctx->nextReadOffset = 0;
+    ctx->waitingOnOffset = 0;
+    ctx->srcBuffer = ctx->coalesceBuffer;
+    ctx->srcBufferLoaded = 0;
+    ctx->reachedEof = 0;
+    if(file != NULL)
+        AIO_ReadPool_startReading(ctx);
+}
+
+/* AIO_ReadPool_create:
+ * Allocates and sets and a new readPool including its included jobs.
+ * bufferSize should be set to the maximal buffer we want to read at a time, will also be used
+ * as our basic read size. */
+ReadPoolCtx_t* AIO_ReadPool_create(const FIO_prefs_t* prefs, size_t bufferSize) {
+    ReadPoolCtx_t* const ctx = (ReadPoolCtx_t*) malloc(sizeof(ReadPoolCtx_t));
+    if(!ctx) EXM_THROW(100, "Allocation error : not enough memory");
+    AIO_IOPool_init(&ctx->base, prefs, AIO_ReadPool_executeReadJob, bufferSize);
+
+    ctx->coalesceBuffer = (U8*) malloc(bufferSize * 2);
+    ctx->srcBuffer = ctx->coalesceBuffer;
+    ctx->srcBufferLoaded = 0;
+    ctx->completedJobsCount = 0;
+    ctx->currentJobHeld = NULL;
+
+    if(ctx->base.threadPool)
+        if (ZSTD_pthread_cond_init(&ctx->jobCompletedCond, NULL))
+            EXM_THROW(103,"Failed creating jobCompletedCond cond");
+
+    return ctx;
+}
+
+/* AIO_ReadPool_free:
+ * Frees and releases a readPool and its resources. Closes source file. */
+void AIO_ReadPool_free(ReadPoolCtx_t* ctx) {
+    if(AIO_ReadPool_getFile(ctx))
+        AIO_ReadPool_closeFile(ctx);
+    if(ctx->base.threadPool)
+        ZSTD_pthread_cond_destroy(&ctx->jobCompletedCond);
+    AIO_IOPool_destroy(&ctx->base);
+    free(ctx->coalesceBuffer);
+    free(ctx);
+}
+
+/* AIO_ReadPool_consumeBytes:
+ * Consumes byes from srcBuffer's beginning and updates srcBufferLoaded accordingly. */
+void AIO_ReadPool_consumeBytes(ReadPoolCtx_t* ctx, size_t n) {
+    assert(n <= ctx->srcBufferLoaded);
+    ctx->srcBufferLoaded -= n;
+    ctx->srcBuffer += n;
+}
+
+/* AIO_ReadPool_releaseCurrentlyHeldAndGetNext:
+ * Release the current held job and get the next one, returns NULL if no next job available. */
+static IOJob_t* AIO_ReadPool_releaseCurrentHeldAndGetNext(ReadPoolCtx_t* ctx) {
+    if (ctx->currentJobHeld) {
+        AIO_IOPool_releaseIoJob((IOJob_t *)ctx->currentJobHeld);
+        ctx->currentJobHeld = NULL;
+        AIO_ReadPool_enqueueRead(ctx);
+    }
+    ctx->currentJobHeld = AIO_ReadPool_getNextCompletedJob(ctx);
+    return (IOJob_t*) ctx->currentJobHeld;
+}
+
+/* AIO_ReadPool_fillBuffer:
+ * Tries to fill the buffer with at least n or jobBufferSize bytes (whichever is smaller).
+ * Returns if srcBuffer has at least the expected number of bytes loaded or if we've reached the end of the file.
+ * Return value is the number of bytes added to the buffer.
+ * Note that srcBuffer might have up to 2 times jobBufferSize bytes. */
+size_t AIO_ReadPool_fillBuffer(ReadPoolCtx_t* ctx, size_t n) {
+    IOJob_t *job;
+    int useCoalesce = 0;
+    if(n > ctx->base.jobBufferSize)
+        n = ctx->base.jobBufferSize;
+
+    /* We are good, don't read anything */
+    if (ctx->srcBufferLoaded >= n)
+        return 0;
+
+    /* We still have bytes loaded, but not enough to satisfy caller. We need to get the next job
+     * and coalesce the remaining bytes with the next job's buffer */
+    if (ctx->srcBufferLoaded > 0) {
+        useCoalesce = 1;
+        memcpy(ctx->coalesceBuffer, ctx->srcBuffer, ctx->srcBufferLoaded);
+        ctx->srcBuffer = ctx->coalesceBuffer;
+    }
+
+    /* Read the next chunk */
+    job = AIO_ReadPool_releaseCurrentHeldAndGetNext(ctx);
+    if(!job)
+        return 0;
+    if(useCoalesce) {
+        assert(ctx->srcBufferLoaded + job->usedBufferSize <= 2*ctx->base.jobBufferSize);
+        memcpy(ctx->coalesceBuffer + ctx->srcBufferLoaded, job->buffer, job->usedBufferSize);
+        ctx->srcBufferLoaded += job->usedBufferSize;
+    }
+    else {
+        ctx->srcBuffer = (U8 *) job->buffer;
+        ctx->srcBufferLoaded = job->usedBufferSize;
+    }
+    return job->usedBufferSize;
+}
+
+/* AIO_ReadPool_consumeAndRefill:
+ * Consumes the current buffer and refills it with bufferSize bytes. */
+size_t AIO_ReadPool_consumeAndRefill(ReadPoolCtx_t* ctx) {
+    AIO_ReadPool_consumeBytes(ctx, ctx->srcBufferLoaded);
+    return AIO_ReadPool_fillBuffer(ctx, ctx->base.jobBufferSize);
+}
+
+/* AIO_ReadPool_getFile:
+ * Returns the current file set for the read pool. */
+FILE* AIO_ReadPool_getFile(const ReadPoolCtx_t* ctx) {
+    return AIO_IOPool_getFile(&ctx->base);
+}
+
+/* AIO_ReadPool_closeFile:
+ * Closes the current set file. Waits for all current enqueued tasks to complete and resets state. */
+int AIO_ReadPool_closeFile(ReadPoolCtx_t* ctx) {
+    FILE* const file = AIO_ReadPool_getFile(ctx);
+    AIO_ReadPool_setFile(ctx, NULL);
+    return fclose(file);
+}
+
+/* AIO_ReadPool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_ReadPool_setAsync(ReadPoolCtx_t* ctx, int async) {
+    AIO_IOPool_setThreaded(&ctx->base, async);
+}
diff --git a/priv/zstd/programs/fileio_asyncio.h b/priv/zstd/programs/fileio_asyncio.h
new file mode 100644
index 0000000..feb25a3
--- /dev/null
+++ b/priv/zstd/programs/fileio_asyncio.h
@@ -0,0 +1,203 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*
+  * FileIO AsyncIO exposes read/write IO pools that allow doing IO asynchronously.
+  * Current implementation relies on having one thread that reads and one that
+  * writes.
+  * Each IO pool supports up to `MAX_IO_JOBS` that can be enqueued for work, but
+  * are performed serially by the appropriate worker thread.
+  * Most systems exposes better primitives to perform asynchronous IO, such as
+  * io_uring on newer linux systems. The API is built in such a way that in the
+  * future we could replace the threads with better solutions when available.
+  */
+
+#ifndef ZSTD_FILEIO_ASYNCIO_H
+#define ZSTD_FILEIO_ASYNCIO_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "../lib/common/mem.h"     /* U32, U64 */
+#include "fileio_types.h"
+#include "platform.h"
+#include "util.h"
+#include "../lib/common/pool.h"
+#include "../lib/common/threading.h"
+
+#define MAX_IO_JOBS          (10)
+
+typedef struct {
+    /* These struct fields should be set only on creation and not changed afterwards */
+    POOL_ctx* threadPool;
+    int threadPoolActive;
+    int totalIoJobs;
+    const FIO_prefs_t* prefs;
+    POOL_function poolFunction;
+
+    /* Controls the file we currently write to, make changes only by using provided utility functions */
+    FILE* file;
+
+    /* The jobs and availableJobsCount fields are accessed by both the main and worker threads and should
+     * only be mutated after locking the mutex */
+    ZSTD_pthread_mutex_t ioJobsMutex;
+    void* availableJobs[MAX_IO_JOBS];
+    int availableJobsCount;
+    size_t jobBufferSize;
+} IOPoolCtx_t;
+
+typedef struct {
+    IOPoolCtx_t base;
+
+    /* State regarding the currently read file */
+    int reachedEof;
+    U64 nextReadOffset;
+    U64 waitingOnOffset;
+
+    /* We may hold an IOJob object as needed if we actively expose its buffer. */
+    void *currentJobHeld;
+
+    /* Coalesce buffer is used to join two buffers in case where we need to read more bytes than left in
+     * the first of them. Shouldn't be accessed from outside ot utility functions. */
+    U8 *coalesceBuffer;
+
+    /* Read buffer can be used by consumer code, take care when copying this pointer aside as it might
+     * change when consuming / refilling buffer. */
+    U8 *srcBuffer;
+    size_t srcBufferLoaded;
+
+    /* We need to know what tasks completed so we can use their buffers when their time comes.
+     * Should only be accessed after locking base.ioJobsMutex . */
+    void* completedJobs[MAX_IO_JOBS];
+    int completedJobsCount;
+    ZSTD_pthread_cond_t jobCompletedCond;
+} ReadPoolCtx_t;
+
+typedef struct {
+    IOPoolCtx_t base;
+    unsigned storedSkips;
+} WritePoolCtx_t;
+
+typedef struct {
+    /* These fields are automatically set and shouldn't be changed by non WritePool code. */
+    void *ctx;
+    FILE* file;
+    void *buffer;
+    size_t bufferSize;
+
+    /* This field should be changed before a job is queued for execution and should contain the number
+     * of bytes to write from the buffer. */
+    size_t usedBufferSize;
+    U64 offset;
+} IOJob_t;
+
+/* AIO_supported:
+ * Returns 1 if AsyncIO is supported on the system, 0 otherwise. */
+int AIO_supported(void);
+
+
+/* AIO_WritePool_releaseIoJob:
+ * Releases an acquired job back to the pool. Doesn't execute the job. */
+void AIO_WritePool_releaseIoJob(IOJob_t *job);
+
+/* AIO_WritePool_acquireJob:
+ * Returns an available write job to be used for a future write. */
+IOJob_t* AIO_WritePool_acquireJob(WritePoolCtx_t *ctx);
+
+/* AIO_WritePool_enqueueAndReacquireWriteJob:
+ * Enqueues a write job for execution and acquires a new one.
+ * After execution `job`'s pointed value would change to the newly acquired job.
+ * Make sure to set `usedBufferSize` to the wanted length before call.
+ * The queued job shouldn't be used directly after queueing it. */
+void AIO_WritePool_enqueueAndReacquireWriteJob(IOJob_t **job);
+
+/* AIO_WritePool_sparseWriteEnd:
+ * Ends sparse writes to the current file.
+ * Blocks on completion of all current write jobs before executing. */
+void AIO_WritePool_sparseWriteEnd(WritePoolCtx_t *ctx);
+
+/* AIO_WritePool_setFile:
+ * Sets the destination file for future writes in the pool.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.
+ * Also requires ending of sparse write if a previous file was used in sparse mode. */
+void AIO_WritePool_setFile(WritePoolCtx_t *ctx, FILE* file);
+
+/* AIO_WritePool_getFile:
+ * Returns the file the writePool is currently set to write to. */
+FILE* AIO_WritePool_getFile(const WritePoolCtx_t* ctx);
+
+/* AIO_WritePool_closeFile:
+ * Ends sparse write and closes the writePool's current file and sets the file to NULL.
+ * Requires completion of all queues write jobs and release of all otherwise acquired jobs.  */
+int AIO_WritePool_closeFile(WritePoolCtx_t *ctx);
+
+/* AIO_WritePool_create:
+ * Allocates and sets and a new write pool including its included jobs.
+ * bufferSize should be set to the maximal buffer we want to write to at a time. */
+WritePoolCtx_t* AIO_WritePool_create(const FIO_prefs_t* prefs, size_t bufferSize);
+
+/* AIO_WritePool_free:
+ * Frees and releases a writePool and its resources. Closes destination file. */
+void AIO_WritePool_free(WritePoolCtx_t* ctx);
+
+/* AIO_WritePool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_WritePool_setAsync(WritePoolCtx_t* ctx, int async);
+
+/* AIO_ReadPool_create:
+ * Allocates and sets and a new readPool including its included jobs.
+ * bufferSize should be set to the maximal buffer we want to read at a time, will also be used
+ * as our basic read size. */
+ReadPoolCtx_t* AIO_ReadPool_create(const FIO_prefs_t* prefs, size_t bufferSize);
+
+/* AIO_ReadPool_free:
+ * Frees and releases a readPool and its resources. Closes source file. */
+void AIO_ReadPool_free(ReadPoolCtx_t* ctx);
+
+/* AIO_ReadPool_setAsync:
+ * Allows (de)activating async mode, to be used when the expected overhead
+ * of asyncio costs more than the expected gains. */
+void AIO_ReadPool_setAsync(ReadPoolCtx_t* ctx, int async);
+
+/* AIO_ReadPool_consumeBytes:
+ * Consumes byes from srcBuffer's beginning and updates srcBufferLoaded accordingly. */
+void AIO_ReadPool_consumeBytes(ReadPoolCtx_t *ctx, size_t n);
+
+/* AIO_ReadPool_fillBuffer:
+ * Makes sure buffer has at least n bytes loaded (as long as n is not bigger than the initialized bufferSize).
+ * Returns if srcBuffer has at least n bytes loaded or if we've reached the end of the file.
+ * Return value is the number of bytes added to the buffer.
+ * Note that srcBuffer might have up to 2 times bufferSize bytes. */
+size_t AIO_ReadPool_fillBuffer(ReadPoolCtx_t *ctx, size_t n);
+
+/* AIO_ReadPool_consumeAndRefill:
+ * Consumes the current buffer and refills it with bufferSize bytes. */
+size_t AIO_ReadPool_consumeAndRefill(ReadPoolCtx_t *ctx);
+
+/* AIO_ReadPool_setFile:
+ * Sets the source file for future read in the pool. Initiates reading immediately if file is not NULL.
+ * Waits for all current enqueued tasks to complete if a previous file was set. */
+void AIO_ReadPool_setFile(ReadPoolCtx_t *ctx, FILE* file);
+
+/* AIO_ReadPool_getFile:
+ * Returns the current file set for the read pool. */
+FILE* AIO_ReadPool_getFile(const ReadPoolCtx_t *ctx);
+
+/* AIO_ReadPool_closeFile:
+ * Closes the current set file. Waits for all current enqueued tasks to complete and resets state. */
+int AIO_ReadPool_closeFile(ReadPoolCtx_t *ctx);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FILEIO_ASYNCIO_H */
diff --git a/priv/zstd/programs/fileio_common.h b/priv/zstd/programs/fileio_common.h
new file mode 100644
index 0000000..55491b8
--- /dev/null
+++ b/priv/zstd/programs/fileio_common.h
@@ -0,0 +1,125 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_FILEIO_COMMON_H
+#define ZSTD_FILEIO_COMMON_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "../lib/common/mem.h"     /* U32, U64 */
+#include "fileio_types.h"
+#include "platform.h"
+#include "timefn.h"     /* UTIL_getTime, UTIL_clockSpanMicro */
+
+/*-*************************************
+*  Macros
+***************************************/
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+#undef MAX
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+extern FIO_display_prefs_t g_display_prefs;
+
+#define DISPLAY(...)         fprintf(stderr, __VA_ARGS__)
+#define DISPLAYOUT(...)      fprintf(stdout, __VA_ARGS__)
+#define DISPLAYLEVEL(l, ...) { if (g_display_prefs.displayLevel>=l) { DISPLAY(__VA_ARGS__); } }
+
+extern UTIL_time_t g_displayClock;
+
+#define REFRESH_RATE  ((U64)(SEC_TO_MICRO / 6))
+#define READY_FOR_UPDATE() (UTIL_clockSpanMicro(g_displayClock) > REFRESH_RATE || g_display_prefs.displayLevel >= 4)
+#define DELAY_NEXT_UPDATE() { g_displayClock = UTIL_getTime(); }
+#define DISPLAYUPDATE(l, ...) {                              \
+        if (g_display_prefs.displayLevel>=l && (g_display_prefs.progressSetting != FIO_ps_never)) { \
+            if (READY_FOR_UPDATE()) { \
+                DELAY_NEXT_UPDATE();                         \
+                DISPLAY(__VA_ARGS__);                        \
+                if (g_display_prefs.displayLevel>=4) fflush(stderr);       \
+    }   }   }
+
+#define SHOULD_DISPLAY_SUMMARY() \
+    (g_display_prefs.displayLevel >= 2 || g_display_prefs.progressSetting == FIO_ps_always)
+#define SHOULD_DISPLAY_PROGRESS()                       \
+    (g_display_prefs.progressSetting != FIO_ps_never && SHOULD_DISPLAY_SUMMARY())
+#define DISPLAY_PROGRESS(...) { if (SHOULD_DISPLAY_PROGRESS()) { DISPLAYLEVEL(1, __VA_ARGS__); }}
+#define DISPLAYUPDATE_PROGRESS(...) { if (SHOULD_DISPLAY_PROGRESS()) { DISPLAYUPDATE(1, __VA_ARGS__); }}
+#define DISPLAY_SUMMARY(...) { if (SHOULD_DISPLAY_SUMMARY()) { DISPLAYLEVEL(1, __VA_ARGS__); } }
+
+#undef MIN  /* in case it would be already defined */
+#define MIN(a,b)    ((a) < (b) ? (a) : (b))
+
+
+#define EXM_THROW(error, ...)                                             \
+{                                                                         \
+    DISPLAYLEVEL(1, "zstd: ");                                            \
+    DISPLAYLEVEL(5, "Error defined at %s, line %i : \n", __FILE__, __LINE__); \
+    DISPLAYLEVEL(1, "error %i : ", error);                                \
+    DISPLAYLEVEL(1, __VA_ARGS__);                                         \
+    DISPLAYLEVEL(1, " \n");                                               \
+    exit(error);                                                          \
+}
+
+#define CHECK_V(v, f)                                \
+    v = f;                                           \
+    if (ZSTD_isError(v)) {                           \
+        DISPLAYLEVEL(5, "%s \n", #f);                \
+        EXM_THROW(11, "%s", ZSTD_getErrorName(v));   \
+    }
+#define CHECK(f) { size_t err; CHECK_V(err, f); }
+
+
+/* Avoid fseek()'s 2GiB barrier with MSVC, macOS, *BSD, MinGW */
+#if defined(_MSC_VER) && _MSC_VER >= 1400
+#   define LONG_SEEK _fseeki64
+#   define LONG_TELL _ftelli64
+#elif !defined(__64BIT__) && (PLATFORM_POSIX_VERSION >= 200112L) /* No point defining Large file for 64 bit */
+#  define LONG_SEEK fseeko
+#  define LONG_TELL ftello
+#elif defined(__MINGW32__) && !defined(__STRICT_ANSI__) && !defined(__NO_MINGW_LFS) && defined(__MSVCRT__)
+#   define LONG_SEEK fseeko64
+#   define LONG_TELL ftello64
+#elif defined(_WIN32) && !defined(__DJGPP__)
+#   include 
+    static int LONG_SEEK(FILE* file, __int64 offset, int origin) {
+        LARGE_INTEGER off;
+        DWORD method;
+        off.QuadPart = offset;
+        if (origin == SEEK_END)
+            method = FILE_END;
+        else if (origin == SEEK_CUR)
+            method = FILE_CURRENT;
+        else
+            method = FILE_BEGIN;
+
+        if (SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, NULL, method))
+            return 0;
+        else
+            return -1;
+    }
+    static __int64 LONG_TELL(FILE* file) {
+        LARGE_INTEGER off, newOff;
+        off.QuadPart = 0;
+        newOff.QuadPart = 0;
+        SetFilePointerEx((HANDLE) _get_osfhandle(_fileno(file)), off, &newOff, FILE_CURRENT);
+        return newOff.QuadPart;
+    }
+#else
+#   define LONG_SEEK fseek
+#   define LONG_TELL ftell
+#endif
+
+#if defined (__cplusplus)
+}
+#endif
+#endif /* ZSTD_FILEIO_COMMON_H */
diff --git a/priv/zstd/programs/fileio_types.h b/priv/zstd/programs/fileio_types.h
new file mode 100644
index 0000000..c1f42f1
--- /dev/null
+++ b/priv/zstd/programs/fileio_types.h
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef FILEIO_TYPES_HEADER
+#define FILEIO_TYPES_HEADER
+
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_compressionParameters */
+#include "../lib/zstd.h"           /* ZSTD_* */
+
+/*-*************************************
+*  Parameters: FIO_prefs_t
+***************************************/
+
+typedef struct FIO_display_prefs_s FIO_display_prefs_t;
+
+typedef enum { FIO_ps_auto, FIO_ps_never, FIO_ps_always } FIO_progressSetting_e;
+
+struct FIO_display_prefs_s {
+    int displayLevel;   /* 0 : no display;  1: errors;  2: + result + interaction + warnings;  3: + progression;  4: + information */
+    FIO_progressSetting_e progressSetting;
+};
+
+
+typedef enum { FIO_zstdCompression, FIO_gzipCompression, FIO_xzCompression, FIO_lzmaCompression, FIO_lz4Compression } FIO_compressionType_t;
+
+typedef struct FIO_prefs_s {
+
+    /* Algorithm preferences */
+    FIO_compressionType_t compressionType;
+    int sparseFileSupport;   /* 0: no sparse allowed; 1: auto (file yes, stdout no); 2: force sparse */
+    int dictIDFlag;
+    int checksumFlag;
+    int blockSize;
+    int overlapLog;
+    int adaptiveMode;
+    int useRowMatchFinder;
+    int rsyncable;
+    int minAdaptLevel;
+    int maxAdaptLevel;
+    int ldmFlag;
+    int ldmHashLog;
+    int ldmMinMatch;
+    int ldmBucketSizeLog;
+    int ldmHashRateLog;
+    size_t streamSrcSize;
+    size_t targetCBlockSize;
+    int srcSizeHint;
+    int testMode;
+    ZSTD_paramSwitch_e literalCompressionMode;
+
+    /* IO preferences */
+    int removeSrcFile;
+    int overwrite;
+    int asyncIO;
+
+    /* Computation resources preferences */
+    unsigned memLimit;
+    int nbWorkers;
+
+    int excludeCompressedFiles;
+    int patchFromMode;
+    int contentSize;
+    int allowBlockDevices;
+    int passThrough;
+} FIO_prefs_t;
+
+#endif /* FILEIO_TYPES_HEADER */
diff --git a/priv/zstd/programs/platform.h b/priv/zstd/programs/platform.h
index b858e3b..18a3587 100644
--- a/priv/zstd/programs/platform.h
+++ b/priv/zstd/programs/platform.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -33,7 +33,7 @@ extern "C" {
 
 /* **************************************
 *  Detect 64-bit OS
-*  http://nadeausoftware.com/articles/2012/02/c_c_tip_how_detect_processor_type_using_compiler_predefined_macros
+*  https://nadeausoftware.com/articles/2012/02/c_c_tip_how_detect_processor_type_using_compiler_predefined_macros
 ****************************************/
 #if defined __ia64 || defined _M_IA64                                                                               /* Intel Itanium */ \
   || defined __powerpc64__ || defined __ppc64__ || defined __PPC64__                                                /* POWER 64-bit */  \
@@ -80,7 +80,7 @@ extern "C" {
       * note: it's better to use unistd.h's _POSIX_VERSION whenever possible */
 #    define PLATFORM_POSIX_VERSION 200112L
 
-/* try to determine posix version through official unistd.h's _POSIX_VERSION (http://pubs.opengroup.org/onlinepubs/7908799/xsh/unistd.h.html).
+/* try to determine posix version through official unistd.h's _POSIX_VERSION (https://pubs.opengroup.org/onlinepubs/7908799/xsh/unistd.h.html).
  * note : there is no simple way to know in advance if  is present or not on target system,
  * Posix specification mandates its presence and its content, but target system must respect this spec.
  * It's necessary to _not_ #include  whenever target OS is not unix-like
@@ -127,6 +127,10 @@ extern "C" {
 
 /*-*********************************************
 *  Detect if isatty() and fileno() are available
+*
+*  Note: Use UTIL_isConsole() for the zstd CLI
+*  instead, as it allows faking is console for
+*  testing.
 ************************************************/
 #if (defined(__linux__) && (PLATFORM_POSIX_VERSION > 1)) \
  || (PLATFORM_POSIX_VERSION >= 200112L) \
@@ -192,13 +196,13 @@ static __inline int IS_CONSOLE(FILE* stdStream) {
 
 
 #ifndef ZSTD_SETPRIORITY_SUPPORT
-   /* mandates presence of  and support for setpriority() : http://man7.org/linux/man-pages/man2/setpriority.2.html */
+   /* mandates presence of  and support for setpriority() : https://man7.org/linux/man-pages/man2/setpriority.2.html */
 #  define ZSTD_SETPRIORITY_SUPPORT (PLATFORM_POSIX_VERSION >= 200112L)
 #endif
 
 
 #ifndef ZSTD_NANOSLEEP_SUPPORT
-   /* mandates support of nanosleep() within  : http://man7.org/linux/man-pages/man2/nanosleep.2.html */
+   /* mandates support of nanosleep() within  : https://man7.org/linux/man-pages/man2/nanosleep.2.html */
 #  if (defined(__linux__) && (PLATFORM_POSIX_VERSION >= 199309L)) \
    || (PLATFORM_POSIX_VERSION >= 200112L)
 #     define ZSTD_NANOSLEEP_SUPPORT 1
diff --git a/priv/zstd/programs/timefn.c b/priv/zstd/programs/timefn.c
index 64577b0..f941e57 100644
--- a/priv/zstd/programs/timefn.c
+++ b/priv/zstd/programs/timefn.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -12,7 +12,8 @@
 /* ===  Dependencies  === */
 
 #include "timefn.h"
-
+#include "platform.h" /* set _POSIX_C_SOURCE */
+#include      /* CLOCK_MONOTONIC, TIME_UTC */
 
 /*-****************************************
 *  Time functions
@@ -20,12 +21,11 @@
 
 #if defined(_WIN32)   /* Windows */
 
+#include   /* LARGE_INTEGER */
 #include    /* abort */
 #include     /* perror */
 
-UTIL_time_t UTIL_getTime(void) { UTIL_time_t x; QueryPerformanceCounter(&x); return x; }
-
-PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd)
+UTIL_time_t UTIL_getTime(void)
 {
     static LARGE_INTEGER ticksPerSecond;
     static int init = 0;
@@ -36,30 +36,20 @@ PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd)
         }
         init = 1;
     }
-    return 1000000ULL*(clockEnd.QuadPart - clockStart.QuadPart)/ticksPerSecond.QuadPart;
-}
-
-PTime UTIL_getSpanTimeNano(UTIL_time_t clockStart, UTIL_time_t clockEnd)
-{
-    static LARGE_INTEGER ticksPerSecond;
-    static int init = 0;
-    if (!init) {
-        if (!QueryPerformanceFrequency(&ticksPerSecond)) {
-            perror("timefn::QueryPerformanceFrequency");
-            abort();
-        }
-        init = 1;
+    {   UTIL_time_t r;
+        LARGE_INTEGER x;
+        QueryPerformanceCounter(&x);
+        r.t = (PTime)(x.QuadPart * 1000000000ULL / ticksPerSecond.QuadPart);
+        return r;
     }
-    return 1000000000ULL*(clockEnd.QuadPart - clockStart.QuadPart)/ticksPerSecond.QuadPart;
 }
 
 
-
 #elif defined(__APPLE__) && defined(__MACH__)
 
-UTIL_time_t UTIL_getTime(void) { return mach_absolute_time(); }
+#include  /* mach_timebase_info_data_t, mach_timebase_info, mach_absolute_time */
 
-PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd)
+UTIL_time_t UTIL_getTime(void)
 {
     static mach_timebase_info_data_t rate;
     static int init = 0;
@@ -67,23 +57,39 @@ PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd)
         mach_timebase_info(&rate);
         init = 1;
     }
-    return (((clockEnd - clockStart) * (PTime)rate.numer) / ((PTime)rate.denom))/1000ULL;
+    {   UTIL_time_t r;
+        r.t = mach_absolute_time() * (PTime)rate.numer / (PTime)rate.denom;
+        return r;
+    }
 }
 
-PTime UTIL_getSpanTimeNano(UTIL_time_t clockStart, UTIL_time_t clockEnd)
+/* POSIX.1-2001 (optional) */
+#elif defined(CLOCK_MONOTONIC)
+
+#include    /* abort */
+#include     /* perror */
+
+UTIL_time_t UTIL_getTime(void)
 {
-    static mach_timebase_info_data_t rate;
-    static int init = 0;
-    if (!init) {
-        mach_timebase_info(&rate);
-        init = 1;
+    /* time must be initialized, othersize it may fail msan test.
+     * No good reason, likely a limitation of timespec_get() for some target */
+    struct timespec time = { 0, 0 };
+    if (clock_gettime(CLOCK_MONOTONIC, &time) != 0) {
+        perror("timefn::clock_gettime(CLOCK_MONOTONIC)");
+        abort();
+    }
+    {   UTIL_time_t r;
+        r.t = (PTime)time.tv_sec * 1000000000ULL + (PTime)time.tv_nsec;
+        return r;
     }
-    return ((clockEnd - clockStart) * (PTime)rate.numer) / ((PTime)rate.denom);
 }
 
 
-/* C11 requires timespec_get, but FreeBSD 11 lacks it, while still claiming C11 compliance.
-   Android also lacks it but does define TIME_UTC. */
+/* C11 requires support of timespec_get().
+ * However, FreeBSD 11 claims C11 compliance while lacking timespec_get().
+ * Double confirm timespec_get() support by checking the definition of TIME_UTC.
+ * However, some versions of Android manage to simultanously define TIME_UTC
+ * and lack timespec_get() support... */
 #elif (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */) \
     && defined(TIME_UTC) && !defined(__ANDROID__)
 
@@ -94,65 +100,49 @@ UTIL_time_t UTIL_getTime(void)
 {
     /* time must be initialized, othersize it may fail msan test.
      * No good reason, likely a limitation of timespec_get() for some target */
-    UTIL_time_t time = UTIL_TIME_INITIALIZER;
+    struct timespec time = { 0, 0 };
     if (timespec_get(&time, TIME_UTC) != TIME_UTC) {
-        perror("timefn::timespec_get");
+        perror("timefn::timespec_get(TIME_UTC)");
         abort();
     }
-    return time;
-}
-
-static UTIL_time_t UTIL_getSpanTime(UTIL_time_t begin, UTIL_time_t end)
-{
-    UTIL_time_t diff;
-    if (end.tv_nsec < begin.tv_nsec) {
-        diff.tv_sec = (end.tv_sec - 1) - begin.tv_sec;
-        diff.tv_nsec = (end.tv_nsec + 1000000000ULL) - begin.tv_nsec;
-    } else {
-        diff.tv_sec = end.tv_sec - begin.tv_sec;
-        diff.tv_nsec = end.tv_nsec - begin.tv_nsec;
+    {   UTIL_time_t r;
+        r.t = (PTime)time.tv_sec * 1000000000ULL + (PTime)time.tv_nsec;
+        return r;
     }
-    return diff;
 }
 
-PTime UTIL_getSpanTimeMicro(UTIL_time_t begin, UTIL_time_t end)
-{
-    UTIL_time_t const diff = UTIL_getSpanTime(begin, end);
-    PTime micro = 0;
-    micro += 1000000ULL * diff.tv_sec;
-    micro += diff.tv_nsec / 1000ULL;
-    return micro;
-}
 
-PTime UTIL_getSpanTimeNano(UTIL_time_t begin, UTIL_time_t end)
+#else   /* relies on standard C90 (note : clock_t produces wrong measurements for multi-threaded workloads) */
+
+UTIL_time_t UTIL_getTime(void)
 {
-    UTIL_time_t const diff = UTIL_getSpanTime(begin, end);
-    PTime nano = 0;
-    nano += 1000000000ULL * diff.tv_sec;
-    nano += diff.tv_nsec;
-    return nano;
+    UTIL_time_t r;
+    r.t = (PTime)clock() * 1000000000ULL / CLOCKS_PER_SEC;
+    return r;
 }
 
-
-
-#else   /* relies on standard C90 (note : clock_t measurements can be wrong when using multi-threading) */
-
-UTIL_time_t UTIL_getTime(void) { return clock(); }
-PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd) { return 1000000ULL * (clockEnd - clockStart) / CLOCKS_PER_SEC; }
-PTime UTIL_getSpanTimeNano(UTIL_time_t clockStart, UTIL_time_t clockEnd) { return 1000000000ULL * (clockEnd - clockStart) / CLOCKS_PER_SEC; }
+#define TIME_MT_MEASUREMENTS_NOT_SUPPORTED
 
 #endif
 
+/* ==== Common functions, valid for all time API ==== */
 
+PTime UTIL_getSpanTimeNano(UTIL_time_t clockStart, UTIL_time_t clockEnd)
+{
+    return clockEnd.t - clockStart.t;
+}
+
+PTime UTIL_getSpanTimeMicro(UTIL_time_t begin, UTIL_time_t end)
+{
+    return UTIL_getSpanTimeNano(begin, end) / 1000ULL;
+}
 
-/* returns time span in microseconds */
 PTime UTIL_clockSpanMicro(UTIL_time_t clockStart )
 {
     UTIL_time_t const clockEnd = UTIL_getTime();
     return UTIL_getSpanTimeMicro(clockStart, clockEnd);
 }
 
-/* returns time span in microseconds */
 PTime UTIL_clockSpanNano(UTIL_time_t clockStart )
 {
     UTIL_time_t const clockEnd = UTIL_getTime();
@@ -167,3 +157,12 @@ void UTIL_waitForNextTick(void)
         clockEnd = UTIL_getTime();
     } while (UTIL_getSpanTimeNano(clockStart, clockEnd) == 0);
 }
+
+int UTIL_support_MT_measurements(void)
+{
+# if defined(TIME_MT_MEASUREMENTS_NOT_SUPPORTED)
+    return 0;
+# else
+    return 1;
+# endif
+}
diff --git a/priv/zstd/programs/timefn.h b/priv/zstd/programs/timefn.h
index 3fcd78a..b814ff8 100644
--- a/priv/zstd/programs/timefn.h
+++ b/priv/zstd/programs/timefn.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -16,70 +16,51 @@ extern "C" {
 #endif
 
 
-/*-****************************************
-*  Dependencies
-******************************************/
-#include          /* clock_t, clock, CLOCKS_PER_SEC */
-
-
 
 /*-****************************************
-*  Local Types
+*  Types
 ******************************************/
 
 #if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
 # if defined(_AIX)
 #  include 
 # else
-#  include  /* intptr_t */
+#  include  /* uint64_t */
 # endif
   typedef uint64_t           PTime;  /* Precise Time */
 #else
   typedef unsigned long long PTime;  /* does not support compilers without long long support */
 #endif
 
+/* UTIL_time_t contains a nanosecond time counter.
+ * The absolute value is not meaningful.
+ * It's only valid to compute the difference between 2 measurements. */
+typedef struct { PTime t; } UTIL_time_t;
+#define UTIL_TIME_INITIALIZER { 0 }
 
 
 /*-****************************************
 *  Time functions
 ******************************************/
-#if defined(_WIN32)   /* Windows */
-
-    #include    /* LARGE_INTEGER */
-    typedef LARGE_INTEGER UTIL_time_t;
-    #define UTIL_TIME_INITIALIZER { { 0, 0 } }
-
-#elif defined(__APPLE__) && defined(__MACH__)
-
-    #include 
-    typedef PTime UTIL_time_t;
-    #define UTIL_TIME_INITIALIZER 0
-
-/* C11 requires timespec_get, but FreeBSD 11 lacks it, while still claiming C11 compliance.
-   Android also lacks it but does define TIME_UTC. */
-#elif (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */) \
-    && defined(TIME_UTC) && !defined(__ANDROID__)
-
-    typedef struct timespec UTIL_time_t;
-    #define UTIL_TIME_INITIALIZER { 0, 0 }
-
-#else   /* relies on standard C90 (note : clock_t measurements can be wrong when using multi-threading) */
 
-    typedef clock_t UTIL_time_t;
-    #define UTIL_TIME_INITIALIZER 0
-
-#endif
+UTIL_time_t UTIL_getTime(void);
 
+/* Timer resolution can be low on some platforms.
+ * To improve accuracy, it's recommended to wait for a new tick
+ * before starting benchmark measurements */
+void UTIL_waitForNextTick(void);
+/* tells if timefn will return correct time measurements
+ * in presence of multi-threaded workload.
+ * note : this is not the case if only C90 clock_t measurements are available */
+int UTIL_support_MT_measurements(void);
 
-UTIL_time_t UTIL_getTime(void);
-PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd);
 PTime UTIL_getSpanTimeNano(UTIL_time_t clockStart, UTIL_time_t clockEnd);
+PTime UTIL_clockSpanNano(UTIL_time_t clockStart);
 
-#define SEC_TO_MICRO ((PTime)1000000)
+PTime UTIL_getSpanTimeMicro(UTIL_time_t clockStart, UTIL_time_t clockEnd);
 PTime UTIL_clockSpanMicro(UTIL_time_t clockStart);
-PTime UTIL_clockSpanNano(UTIL_time_t clockStart);
 
-void UTIL_waitForNextTick(void);
+#define SEC_TO_MICRO ((PTime)1000000)  /* nb of microseconds in a second */
 
 
 #if defined (__cplusplus)
diff --git a/priv/zstd/programs/util.c b/priv/zstd/programs/util.c
index 8d190c6..e017772 100644
--- a/priv/zstd/programs/util.c
+++ b/priv/zstd/programs/util.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -66,6 +66,27 @@ extern "C" {
 #define UTIL_DISPLAY(...)         fprintf(stderr, __VA_ARGS__)
 #define UTIL_DISPLAYLEVEL(l, ...) { if (g_utilDisplayLevel>=l) { UTIL_DISPLAY(__VA_ARGS__); } }
 
+static int g_traceDepth = 0;
+int g_traceFileStat = 0;
+
+#define UTIL_TRACE_CALL(...)                                         \
+    {                                                                \
+        if (g_traceFileStat) {                                       \
+            UTIL_DISPLAY("Trace:FileStat: %*s> ", g_traceDepth, ""); \
+            UTIL_DISPLAY(__VA_ARGS__);                               \
+            UTIL_DISPLAY("\n");                                      \
+            ++g_traceDepth;                                          \
+        }                                                            \
+    }
+
+#define UTIL_TRACE_RET(ret)                                                     \
+    {                                                                           \
+        if (g_traceFileStat) {                                                  \
+            --g_traceDepth;                                                     \
+            UTIL_DISPLAY("Trace:FileStat: %*s< %d\n", g_traceDepth, "", (ret)); \
+        }                                                                      \
+    }
+
 /* A modified version of realloc().
  * If UTIL_realloc() fails the original block is freed.
  */
@@ -100,7 +121,7 @@ int UTIL_requireUserConfirmation(const char* prompt, const char* abortMsg,
     ch = getchar();
     result = 0;
     if (strchr(acceptableLetters, ch) == NULL) {
-        UTIL_DISPLAY("%s", abortMsg);
+        UTIL_DISPLAY("%s \n", abortMsg);
         result = 1;
     }
     /* flush the rest */
@@ -121,21 +142,34 @@ int UTIL_requireUserConfirmation(const char* prompt, const char* abortMsg,
 *  Functions
 ***************************************/
 
+void UTIL_traceFileStat(void)
+{
+    g_traceFileStat = 1;
+}
+
 int UTIL_stat(const char* filename, stat_t* statbuf)
 {
+    int ret;
+    UTIL_TRACE_CALL("UTIL_stat(%s)", filename);
 #if defined(_MSC_VER)
-    return !_stat64(filename, statbuf);
+    ret = !_stat64(filename, statbuf);
 #elif defined(__MINGW32__) && defined (__MSVCRT__)
-    return !_stati64(filename, statbuf);
+    ret = !_stati64(filename, statbuf);
 #else
-    return !stat(filename, statbuf);
+    ret = !stat(filename, statbuf);
 #endif
+    UTIL_TRACE_RET(ret);
+    return ret;
 }
 
 int UTIL_isRegularFile(const char* infilename)
 {
     stat_t statbuf;
-    return UTIL_stat(infilename, &statbuf) && UTIL_isRegularFileStat(&statbuf);
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isRegularFile(%s)", infilename);
+    ret = UTIL_stat(infilename, &statbuf) && UTIL_isRegularFileStat(&statbuf);
+    UTIL_TRACE_RET(ret);
+    return ret;
 }
 
 int UTIL_isRegularFileStat(const stat_t* statbuf)
@@ -151,23 +185,32 @@ int UTIL_isRegularFileStat(const stat_t* statbuf)
 int UTIL_chmod(char const* filename, const stat_t* statbuf, mode_t permissions)
 {
     stat_t localStatBuf;
+    UTIL_TRACE_CALL("UTIL_chmod(%s, %#4o)", filename, (unsigned)permissions);
     if (statbuf == NULL) {
-        if (!UTIL_stat(filename, &localStatBuf)) return 0;
+        if (!UTIL_stat(filename, &localStatBuf)) {
+            UTIL_TRACE_RET(0);
+            return 0;
+        }
         statbuf = &localStatBuf;
     }
-    if (!UTIL_isRegularFileStat(statbuf)) return 0; /* pretend success, but don't change anything */
-    return chmod(filename, permissions);
+    if (!UTIL_isRegularFileStat(statbuf)) {
+        UTIL_TRACE_RET(0);
+        return 0; /* pretend success, but don't change anything */
+    }
+    UTIL_TRACE_CALL("chmod");
+    {
+        int const ret = chmod(filename, permissions);
+        UTIL_TRACE_RET(ret);
+        UTIL_TRACE_RET(ret);
+        return ret;
+    }
 }
 
-int UTIL_setFileStat(const char *filename, const stat_t *statbuf)
+/* set access and modification times */
+int UTIL_utime(const char* filename, const stat_t *statbuf)
 {
-    int res = 0;
-
-    stat_t curStatBuf;
-    if (!UTIL_stat(filename, &curStatBuf) || !UTIL_isRegularFileStat(&curStatBuf))
-        return -1;
-
-    /* set access and modification times */
+    int ret;
+    UTIL_TRACE_CALL("UTIL_utime(%s)", filename);
     /* We check that st_mtime is a macro here in order to give us confidence
      * that struct stat has a struct timespec st_mtim member. We need this
      * check because there are some platforms that claim to be POSIX 2008
@@ -177,40 +220,79 @@ int UTIL_setFileStat(const char *filename, const stat_t *statbuf)
         /* (atime, mtime) */
         struct timespec timebuf[2] = { {0, UTIME_NOW} };
         timebuf[1] = statbuf->st_mtim;
-        res += utimensat(AT_FDCWD, filename, timebuf, 0);
+        ret = utimensat(AT_FDCWD, filename, timebuf, 0);
     }
 #else
     {
         struct utimbuf timebuf;
         timebuf.actime = time(NULL);
         timebuf.modtime = statbuf->st_mtime;
-        res += utime(filename, &timebuf);
+        ret = utime(filename, &timebuf);
     }
 #endif
+    errno = 0;
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_setFileStat(const char *filename, const stat_t *statbuf)
+{
+    int res = 0;
+    stat_t curStatBuf;
+    UTIL_TRACE_CALL("UTIL_setFileStat(%s)", filename);
+
+    if (!UTIL_stat(filename, &curStatBuf) || !UTIL_isRegularFileStat(&curStatBuf)) {
+        UTIL_TRACE_RET(-1);
+        return -1;
+    }
+
+    /* set access and modification times */
+    res += UTIL_utime(filename, statbuf);
+
+    /* Mimic gzip's behavior:
+     *
+     * "Change the group first, then the permissions, then the owner.
+     * That way, the permissions will be correct on systems that allow
+     * users to give away files, without introducing a security hole.
+     * Security depends on permissions not containing the setuid or
+     * setgid bits." */
 
 #if !defined(_WIN32)
-    res += chown(filename, statbuf->st_uid, statbuf->st_gid);  /* Copy ownership */
+    res += chown(filename, -1, statbuf->st_gid);  /* Apply group ownership */
 #endif
 
-    res += UTIL_chmod(filename, &curStatBuf, statbuf->st_mode & 07777);  /* Copy file permissions */
+    res += UTIL_chmod(filename, &curStatBuf, statbuf->st_mode & 0777);  /* Copy file permissions */
+
+#if !defined(_WIN32)
+    res += chown(filename, statbuf->st_uid, -1);  /* Apply user ownership */
+#endif
 
     errno = 0;
+    UTIL_TRACE_RET(-res);
     return -res; /* number of errors is returned */
 }
 
 int UTIL_isDirectory(const char* infilename)
 {
     stat_t statbuf;
-    return UTIL_stat(infilename, &statbuf) && UTIL_isDirectoryStat(&statbuf);
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isDirectory(%s)", infilename);
+    ret = UTIL_stat(infilename, &statbuf) && UTIL_isDirectoryStat(&statbuf);
+    UTIL_TRACE_RET(ret);
+    return ret;
 }
 
 int UTIL_isDirectoryStat(const stat_t* statbuf)
 {
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isDirectoryStat()");
 #if defined(_MSC_VER)
-    return (statbuf->st_mode & _S_IFDIR) != 0;
+    ret = (statbuf->st_mode & _S_IFDIR) != 0;
 #else
-    return S_ISDIR(statbuf->st_mode) != 0;
+    ret = S_ISDIR(statbuf->st_mode) != 0;
 #endif
+    UTIL_TRACE_RET(ret);
+    return ret;
 }
 
 int UTIL_compareStr(const void *p1, const void *p2) {
@@ -219,33 +301,68 @@ int UTIL_compareStr(const void *p1, const void *p2) {
 
 int UTIL_isSameFile(const char* fName1, const char* fName2)
 {
+    int ret;
     assert(fName1 != NULL); assert(fName2 != NULL);
+    UTIL_TRACE_CALL("UTIL_isSameFile(%s, %s)", fName1, fName2);
 #if defined(_MSC_VER) || defined(_WIN32)
     /* note : Visual does not support file identification by inode.
      *        inode does not work on Windows, even with a posix layer, like msys2.
      *        The following work-around is limited to detecting exact name repetition only,
      *        aka `filename` is considered different from `subdir/../filename` */
-    return !strcmp(fName1, fName2);
+    ret = !strcmp(fName1, fName2);
 #else
     {   stat_t file1Stat;
         stat_t file2Stat;
-        return UTIL_stat(fName1, &file1Stat)
+        ret =  UTIL_stat(fName1, &file1Stat)
             && UTIL_stat(fName2, &file2Stat)
-            && (file1Stat.st_dev == file2Stat.st_dev)
-            && (file1Stat.st_ino == file2Stat.st_ino);
+            && UTIL_isSameFileStat(fName1, fName2, &file1Stat, &file2Stat);
+    }
+#endif
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+int UTIL_isSameFileStat(
+        const char* fName1, const char* fName2,
+        const stat_t* file1Stat, const stat_t* file2Stat)
+{
+    int ret;
+    assert(fName1 != NULL); assert(fName2 != NULL);
+    UTIL_TRACE_CALL("UTIL_isSameFileStat(%s, %s)", fName1, fName2);
+#if defined(_MSC_VER) || defined(_WIN32)
+    /* note : Visual does not support file identification by inode.
+     *        inode does not work on Windows, even with a posix layer, like msys2.
+     *        The following work-around is limited to detecting exact name repetition only,
+     *        aka `filename` is considered different from `subdir/../filename` */
+    (void)file1Stat;
+    (void)file2Stat;
+    ret = !strcmp(fName1, fName2);
+#else
+    {
+        ret =  (file1Stat->st_dev == file2Stat->st_dev)
+            && (file1Stat->st_ino == file2Stat->st_ino);
     }
 #endif
+    UTIL_TRACE_RET(ret);
+    return ret;
 }
 
 /* UTIL_isFIFO : distinguish named pipes */
 int UTIL_isFIFO(const char* infilename)
 {
+    UTIL_TRACE_CALL("UTIL_isFIFO(%s)", infilename);
 /* macro guards, as defined in : https://linux.die.net/man/2/lstat */
 #if PLATFORM_POSIX_VERSION >= 200112L
-    stat_t statbuf;
-    if (UTIL_stat(infilename, &statbuf) && UTIL_isFIFOStat(&statbuf)) return 1;
+    {
+        stat_t statbuf;
+        if (UTIL_stat(infilename, &statbuf) && UTIL_isFIFOStat(&statbuf)) {
+            UTIL_TRACE_RET(1);
+            return 1;
+        }
+    }
 #endif
     (void)infilename;
+    UTIL_TRACE_RET(0);
     return 0;
 }
 
@@ -273,21 +390,69 @@ int UTIL_isBlockDevStat(const stat_t* statbuf)
 
 int UTIL_isLink(const char* infilename)
 {
+    UTIL_TRACE_CALL("UTIL_isLink(%s)", infilename);
 /* macro guards, as defined in : https://linux.die.net/man/2/lstat */
 #if PLATFORM_POSIX_VERSION >= 200112L
-    stat_t statbuf;
-    int const r = lstat(infilename, &statbuf);
-    if (!r && S_ISLNK(statbuf.st_mode)) return 1;
+    {
+        stat_t statbuf;
+        int const r = lstat(infilename, &statbuf);
+        if (!r && S_ISLNK(statbuf.st_mode)) {
+            UTIL_TRACE_RET(1);
+            return 1;
+        }
+    }
 #endif
     (void)infilename;
+    UTIL_TRACE_RET(0);
     return 0;
 }
 
+static int g_fakeStdinIsConsole = 0;
+static int g_fakeStderrIsConsole = 0;
+static int g_fakeStdoutIsConsole = 0;
+
+int UTIL_isConsole(FILE* file)
+{
+    int ret;
+    UTIL_TRACE_CALL("UTIL_isConsole(%d)", fileno(file));
+    if (file == stdin && g_fakeStdinIsConsole)
+        ret = 1;
+    else if (file == stderr && g_fakeStderrIsConsole)
+        ret = 1;
+    else if (file == stdout && g_fakeStdoutIsConsole)
+        ret = 1;
+    else
+        ret = IS_CONSOLE(file);
+    UTIL_TRACE_RET(ret);
+    return ret;
+}
+
+void UTIL_fakeStdinIsConsole(void)
+{
+    g_fakeStdinIsConsole = 1;
+}
+void UTIL_fakeStdoutIsConsole(void)
+{
+    g_fakeStdoutIsConsole = 1;
+}
+void UTIL_fakeStderrIsConsole(void)
+{
+    g_fakeStderrIsConsole = 1;
+}
+
 U64 UTIL_getFileSize(const char* infilename)
 {
     stat_t statbuf;
-    if (!UTIL_stat(infilename, &statbuf)) return UTIL_FILESIZE_UNKNOWN;
-    return UTIL_getFileSizeStat(&statbuf);
+    UTIL_TRACE_CALL("UTIL_getFileSize(%s)", infilename);
+    if (!UTIL_stat(infilename, &statbuf)) {
+        UTIL_TRACE_RET(-1);
+        return UTIL_FILESIZE_UNKNOWN;
+    }
+    {
+        U64 const size = UTIL_getFileSizeStat(&statbuf);
+        UTIL_TRACE_RET((int)size);
+        return size;
+    }
 }
 
 U64 UTIL_getFileSizeStat(const stat_t* statbuf)
@@ -303,16 +468,77 @@ U64 UTIL_getFileSizeStat(const stat_t* statbuf)
     return (U64)statbuf->st_size;
 }
 
+UTIL_HumanReadableSize_t UTIL_makeHumanReadableSize(U64 size)
+{
+    UTIL_HumanReadableSize_t hrs;
+
+    if (g_utilDisplayLevel > 3) {
+        /* In verbose mode, do not scale sizes down, except in the case of
+         * values that exceed the integral precision of a double. */
+        if (size >= (1ull << 53)) {
+            hrs.value = (double)size / (1ull << 20);
+            hrs.suffix = " MiB";
+            /* At worst, a double representation of a maximal size will be
+             * accurate to better than tens of kilobytes. */
+            hrs.precision = 2;
+        } else {
+            hrs.value = (double)size;
+            hrs.suffix = " B";
+            hrs.precision = 0;
+        }
+    } else {
+        /* In regular mode, scale sizes down and use suffixes. */
+        if (size >= (1ull << 60)) {
+            hrs.value = (double)size / (1ull << 60);
+            hrs.suffix = " EiB";
+        } else if (size >= (1ull << 50)) {
+            hrs.value = (double)size / (1ull << 50);
+            hrs.suffix = " PiB";
+        } else if (size >= (1ull << 40)) {
+            hrs.value = (double)size / (1ull << 40);
+            hrs.suffix = " TiB";
+        } else if (size >= (1ull << 30)) {
+            hrs.value = (double)size / (1ull << 30);
+            hrs.suffix = " GiB";
+        } else if (size >= (1ull << 20)) {
+            hrs.value = (double)size / (1ull << 20);
+            hrs.suffix = " MiB";
+        } else if (size >= (1ull << 10)) {
+            hrs.value = (double)size / (1ull << 10);
+            hrs.suffix = " KiB";
+        } else {
+            hrs.value = (double)size;
+            hrs.suffix = " B";
+        }
+
+        if (hrs.value >= 100 || (U64)hrs.value == size) {
+            hrs.precision = 0;
+        } else if (hrs.value >= 10) {
+            hrs.precision = 1;
+        } else if (hrs.value > 1) {
+            hrs.precision = 2;
+        } else {
+            hrs.precision = 3;
+        }
+    }
+
+    return hrs;
+}
 
 U64 UTIL_getTotalFileSize(const char* const * fileNamesTable, unsigned nbFiles)
 {
     U64 total = 0;
     unsigned n;
+    UTIL_TRACE_CALL("UTIL_getTotalFileSize(%u)", nbFiles);
     for (n=0; ntableSize; i++) {
+        if(!strcmp(table->fileNames[i], name)) {
+            return (int)i;
+        }
+    }
+    return -1;
+}
+
 void UTIL_refFilename(FileNamesTable* fnt, const char* filename)
 {
     assert(fnt->tableSize < fnt->tableCapacity);
@@ -498,7 +734,7 @@ UTIL_mergeFileNamesTable(FileNamesTable* table1, FileNamesTable* table2)
         for( idx2=0 ; (idx2 < table2->tableSize) && table2->fileNames[idx2] && (pos < newTotalTableSize) ; ++idx2, ++newTableIdx) {
             size_t const curLen = strlen(table2->fileNames[idx2]);
             memcpy(buf+pos, table2->fileNames[idx2], curLen);
-            assert(newTableIdx <= newTable->tableSize);
+            assert(newTableIdx < newTable->tableSize);
             newTable->fileNames[newTableIdx] = buf+pos;
             pos += curLen+1;
     }   }
@@ -622,8 +858,11 @@ static int UTIL_prepareFileList(const char *dirName,
                 ptrdiff_t newListSize = (*bufEnd - *bufStart) + LIST_SIZE_INCREASE;
                 assert(newListSize >= 0);
                 *bufStart = (char*)UTIL_realloc(*bufStart, (size_t)newListSize);
-                *bufEnd = *bufStart + newListSize;
-                if (*bufStart == NULL) { free(path); closedir(dir); return 0; }
+                if (*bufStart != NULL) {
+                    *bufEnd = *bufStart + newListSize;
+                } else {
+                    free(path); closedir(dir); return 0;
+                }
             }
             if (*bufStart + *pos + pathLength < *bufEnd) {
                 memcpy(*bufStart + *pos, path, pathLength + 1);  /* with final \0 */
@@ -809,30 +1048,30 @@ static const char * trimPath(const char *pathname)
 
 static char* mallocAndJoin2Dir(const char *dir1, const char *dir2)
 {
-    const size_t dir1Size = strlen(dir1);
-    const size_t dir2Size = strlen(dir2);
-    char *outDirBuffer, *buffer, trailingChar;
-
     assert(dir1 != NULL && dir2 != NULL);
-    outDirBuffer = (char *) malloc(dir1Size + dir2Size + 2);
-    CONTROL(outDirBuffer != NULL);
+    {   const size_t dir1Size = strlen(dir1);
+        const size_t dir2Size = strlen(dir2);
+        char *outDirBuffer, *buffer;
 
-    memcpy(outDirBuffer, dir1, dir1Size);
-    outDirBuffer[dir1Size] = '\0';
+        outDirBuffer = (char *) malloc(dir1Size + dir2Size + 2);
+        CONTROL(outDirBuffer != NULL);
 
-    if (dir2[0] == '.')
-        return outDirBuffer;
+        memcpy(outDirBuffer, dir1, dir1Size);
+        outDirBuffer[dir1Size] = '\0';
 
-    buffer = outDirBuffer + dir1Size;
-    trailingChar = *(buffer - 1);
-    if (trailingChar != PATH_SEP) {
-        *buffer = PATH_SEP;
-        buffer++;
-    }
-    memcpy(buffer, dir2, dir2Size);
-    buffer[dir2Size] = '\0';
+        if (dir2[0] == '.')
+            return outDirBuffer;
 
-    return outDirBuffer;
+        buffer = outDirBuffer + dir1Size;
+        if (dir1Size > 0 && *(buffer - 1) != PATH_SEP) {
+            *buffer = PATH_SEP;
+            buffer++;
+        }
+        memcpy(buffer, dir2, dir2Size);
+        buffer[dir2Size] = '\0';
+
+        return outDirBuffer;
+    }
 }
 
 /* this function will return NULL if input srcFileName is not valid name for mirrored output path */
@@ -931,15 +1170,15 @@ makeUniqueMirroredDestDirs(char** srcDirNames, unsigned nbFile, const char* outD
         char* prevDirName = srcDirNames[i - 1];
         char* currDirName = srcDirNames[i];
 
-        /* note: we alwasy compare trimmed path, i.e.:
+        /* note: we always compare trimmed path, i.e.:
          * src dir of "./foo" and "/foo" will be both saved into:
          * "outDirName/foo/" */
         if (!firstIsParentOrSameDirOfSecond(trimPath(prevDirName),
                                             trimPath(currDirName)))
             uniqueDirNr++;
 
-        /* we need maintain original src dir name instead of trimmed
-         * dir, so we can retrive the original src dir's mode_t */
+        /* we need to maintain original src dir name instead of trimmed
+         * dir, so we can retrieve the original src dir's mode_t */
         uniqueDirNames[uniqueDirNr - 1] = currDirName;
     }
 
@@ -1048,7 +1287,7 @@ FileNamesTable* UTIL_createFNT_fromROTable(const char** filenames, size_t nbFile
 
 
 /*-****************************************
-*  count the number of physical cores
+*  count the number of cores
 ******************************************/
 
 #if defined(_WIN32) || defined(WIN32)
@@ -1057,10 +1296,26 @@ FileNamesTable* UTIL_createFNT_fromROTable(const char** filenames, size_t nbFile
 
 typedef BOOL(WINAPI* LPFN_GLPI)(PSYSTEM_LOGICAL_PROCESSOR_INFORMATION, PDWORD);
 
-int UTIL_countPhysicalCores(void)
+DWORD CountSetBits(ULONG_PTR bitMask)
 {
-    static int numPhysicalCores = 0;
-    if (numPhysicalCores != 0) return numPhysicalCores;
+    DWORD LSHIFT = sizeof(ULONG_PTR)*8 - 1;
+    DWORD bitSetCount = 0;
+    ULONG_PTR bitTest = (ULONG_PTR)1 << LSHIFT;
+    DWORD i;
+
+    for (i = 0; i <= LSHIFT; ++i)
+    {
+        bitSetCount += ((bitMask & bitTest)?1:0);
+        bitTest/=2;
+    }
+
+    return bitSetCount;
+}
+
+int UTIL_countCores(int logical)
+{
+    static int numCores = 0;
+    if (numCores != 0) return numCores;
 
     {   LPFN_GLPI glpi;
         BOOL done = FALSE;
@@ -1106,7 +1361,10 @@ int UTIL_countPhysicalCores(void)
         while (byteOffset + sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION) <= returnLength) {
 
             if (ptr->Relationship == RelationProcessorCore) {
-                numPhysicalCores++;
+                if (logical)
+                    numCores += CountSetBits(ptr->ProcessorMask);
+                else
+                    numCores++;
             }
 
             ptr++;
@@ -1115,17 +1373,17 @@ int UTIL_countPhysicalCores(void)
 
         free(buffer);
 
-        return numPhysicalCores;
+        return numCores;
     }
 
 failed:
     /* try to fall back on GetSystemInfo */
     {   SYSTEM_INFO sysinfo;
         GetSystemInfo(&sysinfo);
-        numPhysicalCores = sysinfo.dwNumberOfProcessors;
-        if (numPhysicalCores == 0) numPhysicalCores = 1; /* just in case */
+        numCores = sysinfo.dwNumberOfProcessors;
+        if (numCores == 0) numCores = 1; /* just in case */
     }
-    return numPhysicalCores;
+    return numCores;
 }
 
 #elif defined(__APPLE__)
@@ -1134,24 +1392,24 @@ int UTIL_countPhysicalCores(void)
 
 /* Use apple-provided syscall
  * see: man 3 sysctl */
-int UTIL_countPhysicalCores(void)
+int UTIL_countCores(int logical)
 {
-    static S32 numPhysicalCores = 0; /* apple specifies int32_t */
-    if (numPhysicalCores != 0) return numPhysicalCores;
+    static S32 numCores = 0; /* apple specifies int32_t */
+    if (numCores != 0) return numCores;
 
     {   size_t size = sizeof(S32);
-        int const ret = sysctlbyname("hw.physicalcpu", &numPhysicalCores, &size, NULL, 0);
+        int const ret = sysctlbyname(logical ? "hw.logicalcpu" : "hw.physicalcpu", &numCores, &size, NULL, 0);
         if (ret != 0) {
             if (errno == ENOENT) {
                 /* entry not present, fall back on 1 */
-                numPhysicalCores = 1;
+                numCores = 1;
             } else {
-                perror("zstd: can't get number of physical cpus");
+                perror("zstd: can't get number of cpus");
                 exit(1);
             }
         }
 
-        return numPhysicalCores;
+        return numCores;
     }
 }
 
@@ -1160,16 +1418,16 @@ int UTIL_countPhysicalCores(void)
 /* parse /proc/cpuinfo
  * siblings / cpu cores should give hyperthreading ratio
  * otherwise fall back on sysconf */
-int UTIL_countPhysicalCores(void)
+int UTIL_countCores(int logical)
 {
-    static int numPhysicalCores = 0;
+    static int numCores = 0;
 
-    if (numPhysicalCores != 0) return numPhysicalCores;
+    if (numCores != 0) return numCores;
 
-    numPhysicalCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
-    if (numPhysicalCores == -1) {
+    numCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
+    if (numCores == -1) {
         /* value not queryable, fall back on 1 */
-        return numPhysicalCores = 1;
+        return numCores = 1;
     }
 
     /* try to determine if there's hyperthreading */
@@ -1183,7 +1441,7 @@ int UTIL_countPhysicalCores(void)
 
         if (cpuinfo == NULL) {
             /* fall back on the sysconf value */
-            return numPhysicalCores;
+            return numCores;
         }
 
         /* assume the cpu cores/siblings values will be constant across all
@@ -1216,13 +1474,13 @@ int UTIL_countPhysicalCores(void)
             ratio = siblings / cpu_cores;
         }
 
-        if (ratio && numPhysicalCores > ratio) {
-            numPhysicalCores = numPhysicalCores / ratio;
+        if (ratio && numCores > ratio && !logical) {
+            numCores = numCores / ratio;
         }
 
 failed:
         fclose(cpuinfo);
-        return numPhysicalCores;
+        return numCores;
     }
 }
 
@@ -1233,59 +1491,90 @@ int UTIL_countPhysicalCores(void)
 
 /* Use physical core sysctl when available
  * see: man 4 smp, man 3 sysctl */
-int UTIL_countPhysicalCores(void)
+int UTIL_countCores(int logical)
 {
-    static int numPhysicalCores = 0; /* freebsd sysctl is native int sized */
-    if (numPhysicalCores != 0) return numPhysicalCores;
+    static int numCores = 0; /* freebsd sysctl is native int sized */
+#if __FreeBSD_version >= 1300008
+    static int perCore = 1;
+#endif
+    if (numCores != 0) return numCores;
 
 #if __FreeBSD_version >= 1300008
-    {   size_t size = sizeof(numPhysicalCores);
-        int ret = sysctlbyname("kern.smp.cores", &numPhysicalCores, &size, NULL, 0);
-        if (ret == 0) return numPhysicalCores;
+    {   size_t size = sizeof(numCores);
+        int ret = sysctlbyname("kern.smp.cores", &numCores, &size, NULL, 0);
+        if (ret == 0) {
+            if (logical) {
+                ret = sysctlbyname("kern.smp.threads_per_core", &perCore, &size, NULL, 0);
+                /* default to physical cores if logical cannot be read */
+                if (ret == 0)
+                    numCores *= perCore;
+            }
+
+            return numCores;
+        }
         if (errno != ENOENT) {
-            perror("zstd: can't get number of physical cpus");
+            perror("zstd: can't get number of cpus");
             exit(1);
         }
         /* sysctl not present, fall through to older sysconf method */
     }
+#else
+    /* suppress unused parameter warning */
+    (void) logical;
 #endif
 
-    numPhysicalCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
-    if (numPhysicalCores == -1) {
+    numCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
+    if (numCores == -1) {
         /* value not queryable, fall back on 1 */
-        numPhysicalCores = 1;
+        numCores = 1;
     }
-    return numPhysicalCores;
+    return numCores;
 }
 
 #elif defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) || defined(__CYGWIN__)
 
 /* Use POSIX sysconf
  * see: man 3 sysconf */
-int UTIL_countPhysicalCores(void)
+int UTIL_countCores(int logical)
 {
-    static int numPhysicalCores = 0;
+    static int numCores = 0;
 
-    if (numPhysicalCores != 0) return numPhysicalCores;
+    /* suppress unused parameter warning */
+    (void)logical;
 
-    numPhysicalCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
-    if (numPhysicalCores == -1) {
+    if (numCores != 0) return numCores;
+
+    numCores = (int)sysconf(_SC_NPROCESSORS_ONLN);
+    if (numCores == -1) {
         /* value not queryable, fall back on 1 */
-        return numPhysicalCores = 1;
+        return numCores = 1;
     }
-    return numPhysicalCores;
+    return numCores;
 }
 
 #else
 
-int UTIL_countPhysicalCores(void)
+int UTIL_countCores(int logical)
 {
+    /* suppress unused parameter warning */
+    (void)logical;
+
     /* assume 1 */
     return 1;
 }
 
 #endif
 
+int UTIL_countPhysicalCores(void)
+{
+    return UTIL_countCores(0);
+}
+
+int UTIL_countLogicalCores(void)
+{
+    return UTIL_countCores(1);
+}
+
 #if defined (__cplusplus)
 }
 #endif
diff --git a/priv/zstd/programs/util.h b/priv/zstd/programs/util.h
index 24cce44..4ec5413 100644
--- a/priv/zstd/programs/util.h
+++ b/priv/zstd/programs/util.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -64,7 +64,7 @@ extern "C" {
 #    define SET_REALTIME_PRIORITY /* disabled */
 #  endif
 
-#else  /* unknown non-unix operating systen */
+#else  /* unknown non-unix operating system */
 #  define UTIL_sleep(s)          /* disabled */
 #  define UTIL_sleepMilli(milli) /* disabled */
 #  define SET_REALTIME_PRIORITY  /* disabled */
@@ -122,6 +122,7 @@ int UTIL_requireUserConfirmation(const char* prompt, const char* abortMsg, const
 #define STRDUP(s) strdup(s)
 #endif
 
+
 /**
  * Calls platform's equivalent of stat() on filename and writes info to statbuf.
  * Returns success (1) or failure (0).
@@ -135,6 +136,14 @@ int UTIL_stat(const char* filename, stat_t* statbuf);
  */
 int UTIL_setFileStat(const char* filename, const stat_t* statbuf);
 
+/**
+ * Set atime to now and mtime to the st_mtim in statbuf.
+ *
+ * Directly wraps utime() or utimensat(). Returns -1 on error.
+ * Does not validate filename is valid.
+ */
+int UTIL_utime(const char* filename, const stat_t *statbuf);
+
 /*
  * These helpers operate on a pre-populated stat_t, i.e., the result of
  * calling one of the above functions.
@@ -162,14 +171,51 @@ int UTIL_chmod(char const* filename, const stat_t* statbuf, mode_t permissions);
 int UTIL_isRegularFile(const char* infilename);
 int UTIL_isDirectory(const char* infilename);
 int UTIL_isSameFile(const char* file1, const char* file2);
+int UTIL_isSameFileStat(const char* file1, const char* file2, const stat_t* file1Stat, const stat_t* file2Stat);
 int UTIL_isCompressedFile(const char* infilename, const char *extensionList[]);
 int UTIL_isLink(const char* infilename);
 int UTIL_isFIFO(const char* infilename);
 
+/**
+ * Returns with the given file descriptor is a console.
+ * Allows faking whether stdin/stdout/stderr is a console
+ * using UTIL_fake*IsConsole().
+ */
+int UTIL_isConsole(FILE* file);
+
+/**
+ * Pretends that stdin/stdout/stderr is a console for testing.
+ */
+void UTIL_fakeStdinIsConsole(void);
+void UTIL_fakeStdoutIsConsole(void);
+void UTIL_fakeStderrIsConsole(void);
+
+/**
+ * Emit traces for functions that read, or modify file metadata.
+ */
+void UTIL_traceFileStat(void);
+
 #define UTIL_FILESIZE_UNKNOWN  ((U64)(-1))
 U64 UTIL_getFileSize(const char* infilename);
 U64 UTIL_getTotalFileSize(const char* const * fileNamesTable, unsigned nbFiles);
 
+/**
+ * Take @size in bytes,
+ * prepare the components to pretty-print it in a scaled way.
+ * The components in the returned struct should be passed in
+ * precision, value, suffix order to a "%.*f%s" format string.
+ * Output policy is sensible to @g_utilDisplayLevel,
+ * for verbose mode (@g_utilDisplayLevel >= 4),
+ * does not scale down.
+ */
+typedef struct {
+  double value;
+  int precision;
+  const char* suffix;
+} UTIL_HumanReadableSize_t;
+
+UTIL_HumanReadableSize_t UTIL_makeHumanReadableSize(U64 size);
+
 int UTIL_compareStr(const void *p1, const void *p2);
 const char* UTIL_getFileExtension(const char* infilename);
 void  UTIL_mirrorSourceFilesDirectories(const char** fileNamesTable, unsigned int nbFiles, const char *outDirName);
@@ -222,7 +268,6 @@ UTIL_mergeFileNamesTable(FileNamesTable* table1, FileNamesTable* table2);
 /*! UTIL_expandFNT() :
  *  read names from @fnt, and expand those corresponding to directories
  *  update @fnt, now containing only file names,
- * @return : 0 in case of success, 1 if error
  *  note : in case of error, @fnt[0] is NULL
  */
 void UTIL_expandFNT(FileNamesTable** fnt, int followLinks);
@@ -243,6 +288,11 @@ UTIL_createFNT_fromROTable(const char** filenames, size_t nbFilenames);
  */
 FileNamesTable* UTIL_allocateFileNamesTable(size_t tableSize);
 
+/*! UTIL_searchFileNamesTable() :
+ *  Searched through entries in FileNamesTable for a specific name.
+ * @return : index of entry if found or -1 if not found
+ */
+int UTIL_searchFileNamesTable(FileNamesTable* table, char const* name);
 
 /*! UTIL_refFilename() :
  *  Add a reference to read-only name into @fnt table.
@@ -275,13 +325,19 @@ void UTIL_refFilename(FileNamesTable* fnt, const char* filename);
 FileNamesTable*
 UTIL_createExpandedFNT(const char* const* filenames, size_t nbFilenames, int followLinks);
 
+#if defined(_WIN32) || defined(WIN32)
+DWORD CountSetBits(ULONG_PTR bitMask);
+#endif
 
 /*-****************************************
  *  System
  ******************************************/
 
+int UTIL_countCores(int logical);
+
 int UTIL_countPhysicalCores(void);
 
+int UTIL_countLogicalCores(void);
 
 #if defined (__cplusplus)
 }
diff --git a/priv/zstd/programs/windres/verrsrc.h b/priv/zstd/programs/windres/verrsrc.h
index c1b60e9..61b1f3d 100644
--- a/priv/zstd/programs/windres/verrsrc.h
+++ b/priv/zstd/programs/windres/verrsrc.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/programs/windres/zstd.rc b/priv/zstd/programs/windres/zstd.rc
index f5e4047..a2118c2 100644
--- a/priv/zstd/programs/windres/zstd.rc
+++ b/priv/zstd/programs/windres/zstd.rc
@@ -32,11 +32,11 @@ BEGIN
     BEGIN
         BLOCK "040904B0"
         BEGIN
-            VALUE "CompanyName", "Yann Collet, Facebook, Inc."
+            VALUE "CompanyName", "Meta Platforms, Inc."
             VALUE "FileDescription", "Zstandard - Fast and efficient compression algorithm"
             VALUE "FileVersion", ZSTD_VERSION_STRING
             VALUE "InternalName", "zstd.exe"
-            VALUE "LegalCopyright", "Copyright (c) 2013-present, Yann Collet, Facebook, Inc."
+            VALUE "LegalCopyright", "Copyright (c) Meta Platforms, Inc. and affiliates."
             VALUE "OriginalFilename", "zstd.exe"
             VALUE "ProductName", "Zstandard"
             VALUE "ProductVersion", ZSTD_VERSION_STRING
diff --git a/priv/zstd/programs/zstd.1 b/priv/zstd/programs/zstd.1
index 861f938..edc128f 100644
--- a/priv/zstd/programs/zstd.1
+++ b/priv/zstd/programs/zstd.1
@@ -1,491 +1,381 @@
-.
-.TH "ZSTD" "1" "May 2021" "zstd 1.5.0" "User Commands"
-.
+.TH "ZSTD" "1" "February 2023" "zstd 1.5.4" "User Commands"
 .SH "NAME"
 \fBzstd\fR \- zstd, zstdmt, unzstd, zstdcat \- Compress or decompress \.zst files
-.
 .SH "SYNOPSIS"
-\fBzstd\fR [\fIOPTIONS\fR] [\-|\fIINPUT\-FILE\fR] [\-o \fIOUTPUT\-FILE\fR]
-.
+.TS
+allbox;
+\fBzstd\fR [\fIOPTIONS\fR] [\-	\fIINPUT\-FILE\fR] [\-o \fIOUTPUT\-FILE\fR]
+.TE
 .P
 \fBzstdmt\fR is equivalent to \fBzstd \-T0\fR
-.
 .P
 \fBunzstd\fR is equivalent to \fBzstd \-d\fR
-.
 .P
 \fBzstdcat\fR is equivalent to \fBzstd \-dcf\fR
-.
 .SH "DESCRIPTION"
-\fBzstd\fR is a fast lossless compression algorithm and data compression tool, with command line syntax similar to \fBgzip (1)\fR and \fBxz (1)\fR\. It is based on the \fBLZ77\fR family, with further FSE & huff0 entropy stages\. \fBzstd\fR offers highly configurable compression speed, with fast modes at > 200 MB/s per core, and strong modes nearing lzma compression ratios\. It also features a very fast decoder, with speeds > 500 MB/s per core\.
-.
+\fBzstd\fR is a fast lossless compression algorithm and data compression tool, with command line syntax similar to \fBgzip\fR(1) and \fBxz\fR(1)\. It is based on the \fBLZ77\fR family, with further FSE & huff0 entropy stages\. \fBzstd\fR offers highly configurable compression speed, from fast modes at > 200 MB/s per core, to strong modes with excellent compression ratios\. It also features a very fast decoder, with speeds > 500 MB/s per core\.
 .P
-\fBzstd\fR command line syntax is generally similar to gzip, but features the following differences :
-.
-.IP "\(bu" 4
+\fBzstd\fR command line syntax is generally similar to gzip, but features the following differences:
+.IP "\[ci]" 4
 Source files are preserved by default\. It\'s possible to remove them automatically by using the \fB\-\-rm\fR command\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 When compressing a single file, \fBzstd\fR displays progress notifications and result summary by default\. Use \fB\-q\fR to turn them off\.
-.
-.IP "\(bu" 4
-\fBzstd\fR does not accept input from console, but it properly accepts \fBstdin\fR when it\'s not the console\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fBzstd\fR displays a short help page when command line is an error\. Use \fB\-q\fR to turn it off\.
-.
+.IP "\[ci]" 4
+\fBzstd\fR does not accept input from console, though it does accept \fBstdin\fR when it\'s not the console\.
+.IP "\[ci]" 4
+\fBzstd\fR does not store the input\'s filename or attributes, only its contents\.
 .IP "" 0
-.
 .P
-\fBzstd\fR compresses or decompresses each \fIfile\fR according to the selected operation mode\. If no \fIfiles\fR are given or \fIfile\fR is \fB\-\fR, \fBzstd\fR reads from standard input and writes the processed data to standard output\. \fBzstd\fR will refuse to write compressed data to standard output if it is a terminal : it will display an error message and skip the \fIfile\fR\. Similarly, \fBzstd\fR will refuse to read compressed data from standard input if it is a terminal\.
-.
+\fBzstd\fR processes each \fIfile\fR according to the selected operation mode\. If no \fIfiles\fR are given or \fIfile\fR is \fB\-\fR, \fBzstd\fR reads from standard input and writes the processed data to standard output\. \fBzstd\fR will refuse to write compressed data to standard output if it is a terminal: it will display an error message and skip the file\. Similarly, \fBzstd\fR will refuse to read compressed data from standard input if it is a terminal\.
 .P
 Unless \fB\-\-stdout\fR or \fB\-o\fR is specified, \fIfiles\fR are written to a new file whose name is derived from the source \fIfile\fR name:
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 When compressing, the suffix \fB\.zst\fR is appended to the source filename to get the target filename\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 When decompressing, the \fB\.zst\fR suffix is removed from the source filename to get the target filename
-.
 .IP "" 0
-.
-.SS "Concatenation with \.zst files"
-It is possible to concatenate \fB\.zst\fR files as is\. \fBzstd\fR will decompress such files as if they were a single \fB\.zst\fR file\.
-.
+.SS "Concatenation with \.zst Files"
+It is possible to concatenate multiple \fB\.zst\fR files\. \fBzstd\fR will decompress such agglomerated file as if it was a single \fB\.zst\fR file\.
 .SH "OPTIONS"
-.
-.SS "Integer suffixes and special values"
+.SS "Integer Suffixes and Special Values"
 In most places where an integer argument is expected, an optional suffix is supported to easily indicate large integers\. There must be no space between the integer and the suffix\.
-.
 .TP
 \fBKiB\fR
-Multiply the integer by 1,024 (2^10)\. \fBKi\fR, \fBK\fR, and \fBKB\fR are accepted as synonyms for \fBKiB\fR\.
-.
+Multiply the integer by 1,024 (2\e^10)\. \fBKi\fR, \fBK\fR, and \fBKB\fR are accepted as synonyms for \fBKiB\fR\.
 .TP
 \fBMiB\fR
-Multiply the integer by 1,048,576 (2^20)\. \fBMi\fR, \fBM\fR, and \fBMB\fR are accepted as synonyms for \fBMiB\fR\.
-.
-.SS "Operation mode"
+Multiply the integer by 1,048,576 (2\e^20)\. \fBMi\fR, \fBM\fR, and \fBMB\fR are accepted as synonyms for \fBMiB\fR\.
+.SS "Operation Mode"
 If multiple operation mode options are given, the last one takes effect\.
-.
 .TP
 \fB\-z\fR, \fB\-\-compress\fR
 Compress\. This is the default operation mode when no operation mode option is specified and no other operation mode is implied from the command name (for example, \fBunzstd\fR implies \fB\-\-decompress\fR)\.
-.
 .TP
 \fB\-d\fR, \fB\-\-decompress\fR, \fB\-\-uncompress\fR
 Decompress\.
-.
 .TP
 \fB\-t\fR, \fB\-\-test\fR
-Test the integrity of compressed \fIfiles\fR\. This option is equivalent to \fB\-\-decompress \-\-stdout\fR except that the decompressed data is discarded instead of being written to standard output\. No files are created or removed\.
-.
+Test the integrity of compressed \fIfiles\fR\. This option is equivalent to \fB\-\-decompress \-\-stdout > /dev/null\fR, decompressed data is discarded and checksummed for errors\. No files are created or removed\.
 .TP
 \fB\-b#\fR
-Benchmark file(s) using compression level #
-.
+Benchmark file(s) using compression level \fI#\fR\. See \fIBENCHMARK\fR below for a description of this operation\.
 .TP
-\fB\-\-train FILEs\fR
-Use FILEs as a training set to create a dictionary\. The training set should contain a lot of small files (> 100)\.
-.
+\fB\-\-train FILES\fR
+Use \fIFILES\fR as a training set to create a dictionary\. The training set should contain a lot of small files (> 100)\. See \fIDICTIONARY BUILDER\fR below for a description of this operation\.
 .TP
 \fB\-l\fR, \fB\-\-list\fR
-Display information related to a zstd compressed file, such as size, ratio, and checksum\. Some of these fields may not be available\. This command can be augmented with the \fB\-v\fR modifier\.
-.
-.SS "Operation modifiers"
-.
-.IP "\(bu" 4
-\fB\-#\fR: \fB#\fR compression level [1\-19] (default: 3)
-.
-.IP "\(bu" 4
+Display information related to a zstd compressed file, such as size, ratio, and checksum\. Some of these fields may not be available\. This command\'s output can be augmented with the \fB\-v\fR modifier\.
+.SS "Operation Modifiers"
+.IP "\[ci]" 4
+\fB\-#\fR: selects \fB#\fR compression level [1\-19] (default: 3)
+.IP "\[ci]" 4
 \fB\-\-ultra\fR: unlocks high compression levels 20+ (maximum 22), using a lot more memory\. Note that decompression will also require more memory when using these levels\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-\-fast[=#]\fR: switch to ultra\-fast compression levels\. If \fB=#\fR is not present, it defaults to \fB1\fR\. The higher the value, the faster the compression speed, at the cost of some compression ratio\. This setting overwrites compression level if one was set previously\. Similarly, if a compression level is set after \fB\-\-fast\fR, it overrides it\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-T#\fR, \fB\-\-threads=#\fR: Compress using \fB#\fR working threads (default: 1)\. If \fB#\fR is 0, attempt to detect and use the number of physical CPU cores\. In all cases, the nb of threads is capped to \fBZSTDMT_NBWORKERS_MAX\fR, which is either 64 in 32\-bit mode, or 256 for 64\-bit environments\. This modifier does nothing if \fBzstd\fR is compiled without multithread support\.
-.
-.IP "\(bu" 4
-\fB\-\-single\-thread\fR: Does not spawn a thread for compression, use a single thread for both I/O and compression\. In this mode, compression is serialized with I/O, which is slightly slower\. (This is different from \fB\-T1\fR, which spawns 1 compression thread in parallel of I/O)\. This mode is the only one available when multithread support is disabled\. Single\-thread mode features lower memory usage\. Final compressed result is slightly different from \fB\-T1\fR\.
-.
-.IP "\(bu" 4
-\fB\-\-adapt[=min=#,max=#]\fR : \fBzstd\fR will dynamically adapt compression level to perceived I/O conditions\. Compression level adaptation can be observed live by using command \fB\-v\fR\. Adaptation can be constrained between supplied \fBmin\fR and \fBmax\fR levels\. The feature works when combined with multi\-threading and \fB\-\-long\fR mode\. It does not work with \fB\-\-single\-thread\fR\. It sets window size to 8 MB by default (can be changed manually, see \fBwlog\fR)\. Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible\. \fInote\fR : at the time of this writing, \fB\-\-adapt\fR can remain stuck at low speed when combined with multiple worker threads (>=2)\.
-.
-.IP "\(bu" 4
-\fB\-\-long[=#]\fR: enables long distance matching with \fB#\fR \fBwindowLog\fR, if not \fB#\fR is not present it defaults to \fB27\fR\. This increases the window size (\fBwindowLog\fR) and memory usage for both the compressor and decompressor\. This setting is designed to improve the compression ratio for files with long matches at a large distance\.
-.
+.IP "\[ci]" 4
+\fB\-\-single\-thread\fR: Use a single thread for both I/O and compression\. As compression is serialized with I/O, this can be slightly slower\. Single\-thread mode features significantly lower memory usage, which can be useful for systems with limited amount of memory, such as 32\-bit systems\.
+.IP
+Note 1: this mode is the only available one when multithread support is disabled\.
+.IP
+Note 2: this mode is different from \fB\-T1\fR, which spawns 1 compression thread in parallel with I/O\. Final compressed result is also slightly different from \fB\-T1\fR\.
+.IP "\[ci]" 4
+\fB\-\-auto\-threads={physical,logical} (default: physical)\fR: When using a default amount of threads via \fB\-T0\fR, choose the default based on the number of detected physical or logical cores\.
+.IP "\[ci]" 4
+\fB\-\-adapt[=min=#,max=#]\fR: \fBzstd\fR will dynamically adapt compression level to perceived I/O conditions\. Compression level adaptation can be observed live by using command \fB\-v\fR\. Adaptation can be constrained between supplied \fBmin\fR and \fBmax\fR levels\. The feature works when combined with multi\-threading and \fB\-\-long\fR mode\. It does not work with \fB\-\-single\-thread\fR\. It sets window size to 8 MiB by default (can be changed manually, see \fBwlog\fR)\. Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible\.
+.IP
+\fINote\fR: at the time of this writing, \fB\-\-adapt\fR can remain stuck at low speed when combined with multiple worker threads (>=2)\.
+.IP "\[ci]" 4
+\fB\-\-long[=#]\fR: enables long distance matching with \fB#\fR \fBwindowLog\fR, if \fB#\fR is not present it defaults to \fB27\fR\. This increases the window size (\fBwindowLog\fR) and memory usage for both the compressor and decompressor\. This setting is designed to improve the compression ratio for files with long matches at a large distance\.
 .IP
 Note: If \fBwindowLog\fR is set to larger than 27, \fB\-\-long=windowLog\fR or \fB\-\-memory=windowSize\fR needs to be passed to the decompressor\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-D DICT\fR: use \fBDICT\fR as Dictionary to compress or decompress FILE(s)
-.
-.IP "\(bu" 4
-\fB\-\-patch\-from FILE\fR: Specify the file to be used as a reference point for zstd\'s diff engine\. This is effectively dictionary compression with some convenient parameter selection, namely that windowSize > srcSize\.
-.
-.IP
-Note: cannot use both this and \-D together Note: \fB\-\-long\fR mode will be automatically activated if chainLog < fileLog (fileLog being the windowLog required to cover the whole file)\. You can also manually force it\. Node: for all levels, you can use \-\-patch\-from in \-\-single\-thread mode to improve compression ratio at the cost of speed Note: for level 19, you can get increased compression ratio at the cost of speed by specifying \fB\-\-zstd=targetLength=\fR to be something large (i\.e 4096), and by setting a large \fB\-\-zstd=chainLog=\fR
-.
-.IP "\(bu" 4
-\fB\-\-rsyncable\fR : \fBzstd\fR will periodically synchronize the compression state to make the compressed file more rsync\-friendly\. There is a negligible impact to compression ratio, and the faster compression levels will see a small compression speed hit\. This feature does not work with \fB\-\-single\-thread\fR\. You probably don\'t want to use it with long range mode, since it will decrease the effectiveness of the synchronization points, but your milage may vary\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-\-patch\-from FILE\fR: Specify the file to be used as a reference point for zstd\'s diff engine\. This is effectively dictionary compression with some convenient parameter selection, namely that \fIwindowSize\fR > \fIsrcSize\fR\.
+.IP
+Note: cannot use both this and \fB\-D\fR together\.
+.IP
+Note: \fB\-\-long\fR mode will be automatically activated if \fIchainLog\fR < \fIfileLog\fR (\fIfileLog\fR being the \fIwindowLog\fR required to cover the whole file)\. You can also manually force it\.
+.IP
+Note: for all levels, you can use \fB\-\-patch\-from\fR in \fB\-\-single\-thread\fR mode to improve compression ratio at the cost of speed\.
+.IP
+Note: for level 19, you can get increased compression ratio at the cost of speed by specifying \fB\-\-zstd=targetLength=\fR to be something large (i\.e\. 4096), and by setting a large \fB\-\-zstd=chainLog=\fR\.
+.IP "\[ci]" 4
+\fB\-\-rsyncable\fR: \fBzstd\fR will periodically synchronize the compression state to make the compressed file more rsync\-friendly\. There is a negligible impact to compression ratio, and the faster compression levels will see a small compression speed hit\. This feature does not work with \fB\-\-single\-thread\fR\. You probably don\'t want to use it with long range mode, since it will decrease the effectiveness of the synchronization points, but your mileage may vary\.
+.IP "\[ci]" 4
 \fB\-C\fR, \fB\-\-[no\-]check\fR: add integrity check computed from uncompressed data (default: enabled)
-.
-.IP "\(bu" 4
-\fB\-\-[no\-]content\-size\fR: enable / disable whether or not the original size of the file is placed in the header of the compressed file\. The default option is \-\-content\-size (meaning that the original size will be placed in the header)\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-\-[no\-]content\-size\fR: enable / disable whether or not the original size of the file is placed in the header of the compressed file\. The default option is \fB\-\-content\-size\fR (meaning that the original size will be placed in the header)\.
+.IP "\[ci]" 4
 \fB\-\-no\-dictID\fR: do not store dictionary ID within frame header (dictionary compression)\. The decoder will have to rely on implicit knowledge about which dictionary to use, it won\'t be able to check if it\'s correct\.
-.
-.IP "\(bu" 4
-\fB\-M#\fR, \fB\-\-memory=#\fR: Set a memory usage limit\. By default, Zstandard uses 128 MB for decompression as the maximum amount of memory the decompressor is allowed to use, but you can override this manually if need be in either direction (ie\. you can increase or decrease it)\.
-.
-.IP
-This is also used during compression when using with \-\-patch\-from=\. In this case, this parameter overrides that maximum size allowed for a dictionary\. (128 MB)\.
-.
-.IP "\(bu" 4
-\fB\-\-stream\-size=#\fR : Sets the pledged source size of input coming from a stream\. This value must be exact, as it will be included in the produced frame header\. Incorrect stream sizes will cause an error\. This information will be used to better optimize compression parameters, resulting in better and potentially faster compression, especially for smaller source sizes\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-M#\fR, \fB\-\-memory=#\fR: Set a memory usage limit\. By default, \fBzstd\fR uses 128 MiB for decompression as the maximum amount of memory the decompressor is allowed to use, but you can override this manually if need be in either direction (i\.e\. you can increase or decrease it)\.
+.IP
+This is also used during compression when using with \fB\-\-patch\-from=\fR\. In this case, this parameter overrides that maximum size allowed for a dictionary\. (128 MiB)\.
+.IP
+Additionally, this can be used to limit memory for dictionary training\. This parameter overrides the default limit of 2 GiB\. zstd will load training samples up to the memory limit and ignore the rest\.
+.IP "\[ci]" 4
+\fB\-\-stream\-size=#\fR: Sets the pledged source size of input coming from a stream\. This value must be exact, as it will be included in the produced frame header\. Incorrect stream sizes will cause an error\. This information will be used to better optimize compression parameters, resulting in better and potentially faster compression, especially for smaller source sizes\.
+.IP "\[ci]" 4
 \fB\-\-size\-hint=#\fR: When handling input from a stream, \fBzstd\fR must guess how large the source size will be when optimizing compression parameters\. If the stream size is relatively small, this guess may be a poor one, resulting in a higher compression ratio than expected\. This feature allows for controlling the guess when needed\. Exact guesses result in better compression ratios\. Overestimates result in slightly degraded compression ratios, while underestimates may result in significant degradation\.
-.
-.IP "\(bu" 4
-\fB\-o FILE\fR: save result into \fBFILE\fR
-.
-.IP "\(bu" 4
-\fB\-f\fR, \fB\-\-force\fR: disable input and output checks\. Allows overwriting existing files, input from console, output to stdout, operating on links, block devices, etc\.
-.
-.IP "\(bu" 4
-\fB\-c\fR, \fB\-\-stdout\fR: force write to standard output, even if it is the console
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-o FILE\fR: save result into \fBFILE\fR\.
+.IP "\[ci]" 4
+\fB\-f\fR, \fB\-\-force\fR: disable input and output checks\. Allows overwriting existing files, input from console, output to stdout, operating on links, block devices, etc\. During decompression and when the output destination is stdout, pass\-through unrecognized formats as\-is\.
+.IP "\[ci]" 4
+\fB\-c\fR, \fB\-\-stdout\fR: write to standard output (even if it is the console); keep original files unchanged\.
+.IP "\[ci]" 4
 \fB\-\-[no\-]sparse\fR: enable / disable sparse FS support, to make files with many zeroes smaller on disk\. Creating sparse files may save disk space and speed up decompression by reducing the amount of disk I/O\. default: enabled when output is into a file, and disabled when output is stdout\. This setting overrides default and can force sparse mode over stdout\.
-.
-.IP "\(bu" 4
-\fB\-\-rm\fR: remove source file(s) after successful compression or decompression\. If used in combination with \-o, will trigger a confirmation prompt (which can be silenced with \-f), as this is a destructive operation\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-\-[no\-]pass\-through\fR enable / disable passing through uncompressed files as\-is\. During decompression when pass\-through is enabled, unrecognized formats will be copied as\-is from the input to the output\. By default, pass\-through will occur when the output destination is stdout and the force (\fB\-f\fR) option is set\.
+.IP "\[ci]" 4
+\fB\-\-rm\fR: remove source file(s) after successful compression or decompression\. This command is silently ignored if output is \fBstdout\fR\. If used in combination with \fB\-o\fR, triggers a confirmation prompt (which can be silenced with \fB\-f\fR), as this is a destructive operation\.
+.IP "\[ci]" 4
 \fB\-k\fR, \fB\-\-keep\fR: keep source file(s) after successful compression or decompression\. This is the default behavior\.
-.
-.IP "\(bu" 4
-\fB\-r\fR: operate recursively on directories
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-r\fR: operate recursively on directories\. It selects all files in the named directory and all its subdirectories\. This can be useful both to reduce command line typing, and to circumvent shell expansion limitations, when there are a lot of files and naming breaks the maximum size of a command line\.
+.IP "\[ci]" 4
 \fB\-\-filelist FILE\fR read a list of files to process as content from \fBFILE\fR\. Format is compatible with \fBls\fR output, with one file per line\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-\-output\-dir\-flat DIR\fR: resulting files are stored into target \fBDIR\fR directory, instead of same directory as origin file\. Be aware that this command can introduce name collision issues, if multiple files, from different directories, end up having the same name\. Collision resolution ensures first file with a given name will be present in \fBDIR\fR, while in combination with \fB\-f\fR, the last file will be present instead\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-\-output\-dir\-mirror DIR\fR: similar to \fB\-\-output\-dir\-flat\fR, the output files are stored underneath target \fBDIR\fR directory, but this option will replicate input directory hierarchy into output \fBDIR\fR\.
-.
 .IP
 If input directory contains "\.\.", the files in this directory will be ignored\. If input directory is an absolute directory (i\.e\. "/var/tmp/abc"), it will be stored into the "output\-dir/var/tmp/abc"\. If there are multiple input files or directories, name collision resolution will follow the same rules as \fB\-\-output\-dir\-flat\fR\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-\-format=FORMAT\fR: compress and decompress in other formats\. If compiled with support, zstd can compress to or decompress from other compression algorithm formats\. Possibly available options are \fBzstd\fR, \fBgzip\fR, \fBxz\fR, \fBlzma\fR, and \fBlz4\fR\. If no such format is provided, \fBzstd\fR is the default\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-h\fR/\fB\-H\fR, \fB\-\-help\fR: display help/long help and exit
-.
-.IP "\(bu" 4
-\fB\-V\fR, \fB\-\-version\fR: display version number and exit\. Advanced : \fB\-vV\fR also displays supported formats\. \fB\-vvV\fR also displays POSIX support\. \fB\-q\fR will only display the version number, suitable for machine reading\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-V\fR, \fB\-\-version\fR: display version number and exit\. Advanced: \fB\-vV\fR also displays supported formats\. \fB\-vvV\fR also displays POSIX support\. \fB\-q\fR will only display the version number, suitable for machine reading\.
+.IP "\[ci]" 4
 \fB\-v\fR, \fB\-\-verbose\fR: verbose mode, display more information
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-q\fR, \fB\-\-quiet\fR: suppress warnings, interactivity, and notifications\. specify twice to suppress errors too\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
 \fB\-\-no\-progress\fR: do not display the progress bar, but keep all other messages\.
-.
-.IP "\(bu" 4
-\fB\-\-show\-default\-cparams\fR: Shows the default compression parameters that will be used for a particular src file\. If the provided src file is not a regular file (eg\. named pipe), the cli will just output the default parameters\. That is, the parameters that are used when the src size is unknown\.
-.
-.IP "\(bu" 4
+.IP "\[ci]" 4
+\fB\-\-show\-default\-cparams\fR: shows the default compression parameters that will be used for a particular input file, based on the provided compression level and the input size\. If the provided file is not a regular file (e\.g\. a pipe), this flag will output the parameters used for inputs of unknown size\.
+.IP "\[ci]" 4
 \fB\-\-\fR: All arguments after \fB\-\-\fR are treated as files
-.
 .IP "" 0
-.
-.SS "Restricted usage of Environment Variables"
-Using environment variables to set parameters has security implications\. Therefore, this avenue is intentionally restricted\. Only \fBZSTD_CLEVEL\fR and \fBZSTD_NBTHREADS\fR are currently supported\. They set the compression level and number of threads to use during compression, respectively\.
-.
+.SS "gzip Operation Modifiers"
+When invoked via a \fBgzip\fR symlink, \fBzstd\fR will support further options that intend to mimic the \fBgzip\fR behavior:
+.TP
+\fB\-n\fR, \fB\-\-no\-name\fR
+do not store the original filename and timestamps when compressing a file\. This is the default behavior and hence a no\-op\.
+.TP
+\fB\-\-best\fR
+alias to the option \fB\-9\fR\.
+.SS "Environment Variables"
+Employing environment variables to set parameters has security implications\. Therefore, this avenue is intentionally limited\. Only \fBZSTD_CLEVEL\fR and \fBZSTD_NBTHREADS\fR are currently supported\. They set the compression level and number of threads to use during compression, respectively\.
 .P
 \fBZSTD_CLEVEL\fR can be used to set the level between 1 and 19 (the "normal" range)\. If the value of \fBZSTD_CLEVEL\fR is not a valid integer, it will be ignored with a warning message\. \fBZSTD_CLEVEL\fR just replaces the default compression level (\fB3\fR)\.
-.
 .P
 \fBZSTD_NBTHREADS\fR can be used to set the number of threads \fBzstd\fR will attempt to use during compression\. If the value of \fBZSTD_NBTHREADS\fR is not a valid unsigned integer, it will be ignored with a warning message\. \fBZSTD_NBTHREADS\fR has a default value of (\fB1\fR), and is capped at ZSTDMT_NBWORKERS_MAX==200\. \fBzstd\fR must be compiled with multithread support for this to have any effect\.
-.
 .P
 They can both be overridden by corresponding command line arguments: \fB\-#\fR for compression level and \fB\-T#\fR for number of compression threads\.
-.
 .SH "DICTIONARY BUILDER"
-\fBzstd\fR offers \fIdictionary\fR compression, which greatly improves efficiency on small files and messages\. It\'s possible to train \fBzstd\fR with a set of samples, the result of which is saved into a file called a \fBdictionary\fR\. Then during compression and decompression, reference the same dictionary, using command \fB\-D dictionaryFileName\fR\. Compression of small files similar to the sample set will be greatly improved\.
-.
+\fBzstd\fR offers \fIdictionary\fR compression, which greatly improves efficiency on small files and messages\. It\'s possible to train \fBzstd\fR with a set of samples, the result of which is saved into a file called a \fBdictionary\fR\. Then, during compression and decompression, reference the same dictionary, using command \fB\-D dictionaryFileName\fR\. Compression of small files similar to the sample set will be greatly improved\.
 .TP
 \fB\-\-train FILEs\fR
-Use FILEs as training set to create a dictionary\. The training set should contain a lot of small files (> 100), and weight typically 100x the target dictionary size (for example, 10 MB for a 100 KB dictionary)\.
-.
+Use FILEs as training set to create a dictionary\. The training set should ideally contain a lot of samples (> 100), and weight typically 100x the target dictionary size (for example, ~10 MB for a 100 KB dictionary)\. \fB\-\-train\fR can be combined with \fB\-r\fR to indicate a directory rather than listing all the files, which can be useful to circumvent shell expansion limits\.
 .IP
-Supports multithreading if \fBzstd\fR is compiled with threading support\. Additional parameters can be specified with \fB\-\-train\-fastcover\fR\. The legacy dictionary builder can be accessed with \fB\-\-train\-legacy\fR\. The cover dictionary builder can be accessed with \fB\-\-train\-cover\fR\. Equivalent to \fB\-\-train\-fastcover=d=8,steps=4\fR\.
-.
+Since dictionary compression is mostly effective for small files, the expectation is that the training set will only contain small files\. In the case where some samples happen to be large, only the first 128 KiB of these samples will be used for training\.
+.IP
+\fB\-\-train\fR supports multithreading if \fBzstd\fR is compiled with threading support (default)\. Additional advanced parameters can be specified with \fB\-\-train\-fastcover\fR\. The legacy dictionary builder can be accessed with \fB\-\-train\-legacy\fR\. The slower cover dictionary builder can be accessed with \fB\-\-train\-cover\fR\. Default \fB\-\-train\fR is equivalent to \fB\-\-train\-fastcover=d=8,steps=4\fR\.
 .TP
-\fB\-o file\fR
-Dictionary saved into \fBfile\fR (default name: dictionary)\.
-.
+\fB\-o FILE\fR
+Dictionary saved into \fBFILE\fR (default name: dictionary)\.
 .TP
 \fB\-\-maxdict=#\fR
-Limit dictionary to specified size (default: 112640)\.
-.
+Limit dictionary to specified size (default: 112640 bytes)\. As usual, quantities are expressed in bytes by default, and it\'s possible to employ suffixes (like \fBKB\fR or \fBMB\fR) to specify larger values\.
 .TP
 \fB\-#\fR
 Use \fB#\fR compression level during training (optional)\. Will generate statistics more tuned for selected compression level, resulting in a \fIsmall\fR compression ratio improvement for this level\.
-.
 .TP
 \fB\-B#\fR
-Split input files in blocks of size # (default: no split)
-.
+Split input files into blocks of size # (default: no split)
+.TP
+\fB\-M#\fR, \fB\-\-memory=#\fR
+Limit the amount of sample data loaded for training (default: 2 GB)\. Note that the default (2 GB) is also the maximum\. This parameter can be useful in situations where the training set size is not well controlled and could be potentially very large\. Since speed of the training process is directly correlated to the size of the training sample set, a smaller sample set leads to faster training\.
+.IP
+In situations where the training set is larger than maximum memory, the CLI will randomly select samples among the available ones, up to the maximum allowed memory budget\. This is meant to improve dictionary relevance by mitigating the potential impact of clustering, such as selecting only files from the beginning of a list sorted by modification date, or sorted by alphabetical order\. The randomization process is deterministic, so training of the same list of files with the same parameters will lead to the creation of the same dictionary\.
 .TP
 \fB\-\-dictID=#\fR
-A dictionary ID is a locally unique ID that a decoder can use to verify it is using the right dictionary\. By default, zstd will create a 4\-bytes random number ID\. It\'s possible to give a precise number instead\. Short numbers have an advantage : an ID < 256 will only need 1 byte in the compressed frame header, and an ID < 65536 will only need 2 bytes\. This compares favorably to 4 bytes default\. However, it\'s up to the dictionary manager to not assign twice the same ID to 2 different dictionaries\.
-.
+A dictionary ID is a locally unique ID\. The decoder will use this value to verify it is using the right dictionary\. By default, zstd will create a 4\-bytes random number ID\. It\'s possible to provide an explicit number ID instead\. It\'s up to the dictionary manager to not assign twice the same ID to 2 different dictionaries\. Note that short numbers have an advantage: an ID < 256 will only need 1 byte in the compressed frame header, and an ID < 65536 will only need 2 bytes\. This compares favorably to 4 bytes default\.
+.IP
+Note that RFC8878 reserves IDs less than 32768 and greater than or equal to 2\e^31, so they should not be used in public\.
 .TP
 \fB\-\-train\-cover[=k#,d=#,steps=#,split=#,shrink[=#]]\fR
 Select parameters for the default dictionary builder algorithm named cover\. If \fId\fR is not specified, then it tries \fId\fR = 6 and \fId\fR = 8\. If \fIk\fR is not specified, then it tries \fIsteps\fR values in the range [50, 2000]\. If \fIsteps\fR is not specified, then the default value of 40 is used\. If \fIsplit\fR is not specified or split <= 0, then the default value of 100 is used\. Requires that \fId\fR <= \fIk\fR\. If \fIshrink\fR flag is not used, then the default value for \fIshrinkDict\fR of 0 is used\. If \fIshrink\fR is not specified, then the default value for \fIshrinkDictMaxRegression\fR of 1 is used\.
-.
 .IP
 Selects segments of size \fIk\fR with highest score to put in the dictionary\. The score of a segment is computed by the sum of the frequencies of all the subsegments of size \fId\fR\. Generally \fId\fR should be in the range [6, 8], occasionally up to 16, but the algorithm will run faster with d <= \fI8\fR\. Good values for \fIk\fR vary widely based on the input data, but a safe range is [2 * \fId\fR, 2000]\. If \fIsplit\fR is 100, all input samples are used for both training and testing to find optimal \fId\fR and \fIk\fR to build dictionary\. Supports multithreading if \fBzstd\fR is compiled with threading support\. Having \fIshrink\fR enabled takes a truncated dictionary of minimum size and doubles in size until compression ratio of the truncated dictionary is at most \fIshrinkDictMaxRegression%\fR worse than the compression ratio of the largest dictionary\.
-.
 .IP
 Examples:
-.
 .IP
 \fBzstd \-\-train\-cover FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-cover=k=50,d=8 FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-cover=d=8,steps=500 FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-cover=k=50 FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-cover=k=50,split=60 FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-cover=shrink FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-cover=shrink=2 FILEs\fR
-.
 .TP
 \fB\-\-train\-fastcover[=k#,d=#,f=#,steps=#,split=#,accel=#]\fR
 Same as cover but with extra parameters \fIf\fR and \fIaccel\fR and different default value of split If \fIsplit\fR is not specified, then it tries \fIsplit\fR = 75\. If \fIf\fR is not specified, then it tries \fIf\fR = 20\. Requires that 0 < \fIf\fR < 32\. If \fIaccel\fR is not specified, then it tries \fIaccel\fR = 1\. Requires that 0 < \fIaccel\fR <= 10\. Requires that \fId\fR = 6 or \fId\fR = 8\.
-.
 .IP
 \fIf\fR is log of size of array that keeps track of frequency of subsegments of size \fId\fR\. The subsegment is hashed to an index in the range [0,2^\fIf\fR \- 1]\. It is possible that 2 different subsegments are hashed to the same index, and they are considered as the same subsegment when computing frequency\. Using a higher \fIf\fR reduces collision but takes longer\.
-.
 .IP
 Examples:
-.
 .IP
 \fBzstd \-\-train\-fastcover FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-fastcover=d=8,f=15,accel=2 FILEs\fR
-.
 .TP
 \fB\-\-train\-legacy[=selectivity=#]\fR
-Use legacy dictionary builder algorithm with the given dictionary \fIselectivity\fR (default: 9)\. The smaller the \fIselectivity\fR value, the denser the dictionary, improving its efficiency but reducing its possible maximum size\. \fB\-\-train\-legacy=s=#\fR is also accepted\.
-.
+Use legacy dictionary builder algorithm with the given dictionary \fIselectivity\fR (default: 9)\. The smaller the \fIselectivity\fR value, the denser the dictionary, improving its efficiency but reducing its achievable maximum size\. \fB\-\-train\-legacy=s=#\fR is also accepted\.
 .IP
 Examples:
-.
 .IP
 \fBzstd \-\-train\-legacy FILEs\fR
-.
 .IP
 \fBzstd \-\-train\-legacy=selectivity=8 FILEs\fR
-.
 .SH "BENCHMARK"
-.
 .TP
 \fB\-b#\fR
 benchmark file(s) using compression level #
-.
 .TP
 \fB\-e#\fR
 benchmark file(s) using multiple compression levels, from \fB\-b#\fR to \fB\-e#\fR (inclusive)
-.
 .TP
 \fB\-i#\fR
 minimum evaluation time, in seconds (default: 3s), benchmark mode only
-.
 .TP
 \fB\-B#\fR, \fB\-\-block\-size=#\fR
-cut file(s) into independent blocks of size # (default: no block)
-.
+cut file(s) into independent chunks of size # (default: no chunking)
 .TP
 \fB\-\-priority=rt\fR
 set process priority to real\-time
-.
 .P
-\fBOutput Format:\fR CompressionLevel#Filename : IntputSize \-> OutputSize (CompressionRatio), CompressionSpeed, DecompressionSpeed
-.
+\fBOutput Format:\fR CompressionLevel#Filename: InputSize \-> OutputSize (CompressionRatio), CompressionSpeed, DecompressionSpeed
 .P
 \fBMethodology:\fR For both compression and decompression speed, the entire input is compressed/decompressed in\-memory to measure speed\. A run lasts at least 1 sec, so when files are small, they are compressed/decompressed several times per run, in order to improve measurement accuracy\.
-.
 .SH "ADVANCED COMPRESSION OPTIONS"
-.
-.SS "\-B#:"
-Select the size of each compression job\. This parameter is only available when multi\-threading is enabled\. Each compression job is run in parallel, so this value indirectly impacts the nb of active threads\. Default job size varies depending on compression level (generally \fB4 * windowSize\fR)\. \fB\-B#\fR makes it possible to manually select a custom size\. Note that job size must respect a minimum value which is enforced transparently\. This minimum is either 512 KB, or \fBoverlapSize\fR, whichever is largest\. Different job sizes will lead to (slightly) different compressed frames\.
-.
+### \-B#: Specify the size of each compression job\. This parameter is only available when multi\-threading is enabled\. Each compression job is run in parallel, so this value indirectly impacts the nb of active threads\. Default job size varies depending on compression level (generally \fB4 * windowSize\fR)\. \fB\-B#\fR makes it possible to manually select a custom size\. Note that job size must respect a minimum value which is enforced transparently\. This minimum is either 512 KB, or \fBoverlapSize\fR, whichever is largest\. Different job sizes will lead to non\-identical compressed frames\.
 .SS "\-\-zstd[=options]:"
-\fBzstd\fR provides 22 predefined compression levels\. The selected or default predefined compression level can be changed with advanced compression options\. The \fIoptions\fR are provided as a comma\-separated list\. You may specify only the options you want to change and the rest will be taken from the selected or default compression level\. The list of available \fIoptions\fR:
-.
+\fBzstd\fR provides 22 predefined regular compression levels plus the fast levels\. This compression level is translated internally into a number of specific parameters that actually control the behavior of the compressor\. (You can see the result of this translation with \fB\-\-show\-default\-cparams\fR\.) These specific parameters can be overridden with advanced compression options\. The \fIoptions\fR are provided as a comma\-separated list\. You may specify only the options you want to change and the rest will be taken from the selected or default compression level\. The list of available \fIoptions\fR:
 .TP
 \fBstrategy\fR=\fIstrat\fR, \fBstrat\fR=\fIstrat\fR
 Specify a strategy used by a match finder\.
-.
 .IP
-There are 9 strategies numbered from 1 to 9, from faster to stronger: 1=ZSTD_fast, 2=ZSTD_dfast, 3=ZSTD_greedy, 4=ZSTD_lazy, 5=ZSTD_lazy2, 6=ZSTD_btlazy2, 7=ZSTD_btopt, 8=ZSTD_btultra, 9=ZSTD_btultra2\.
-.
+There are 9 strategies numbered from 1 to 9, from fastest to strongest: 1=\fBZSTD_fast\fR, 2=\fBZSTD_dfast\fR, 3=\fBZSTD_greedy\fR, 4=\fBZSTD_lazy\fR, 5=\fBZSTD_lazy2\fR, 6=\fBZSTD_btlazy2\fR, 7=\fBZSTD_btopt\fR, 8=\fBZSTD_btultra\fR, 9=\fBZSTD_btultra2\fR\.
 .TP
 \fBwindowLog\fR=\fIwlog\fR, \fBwlog\fR=\fIwlog\fR
 Specify the maximum number of bits for a match distance\.
-.
 .IP
 The higher number of increases the chance to find a match which usually improves compression ratio\. It also increases memory requirements for the compressor and decompressor\. The minimum \fIwlog\fR is 10 (1 KiB) and the maximum is 30 (1 GiB) on 32\-bit platforms and 31 (2 GiB) on 64\-bit platforms\.
-.
 .IP
 Note: If \fBwindowLog\fR is set to larger than 27, \fB\-\-long=windowLog\fR or \fB\-\-memory=windowSize\fR needs to be passed to the decompressor\.
-.
 .TP
 \fBhashLog\fR=\fIhlog\fR, \fBhlog\fR=\fIhlog\fR
 Specify the maximum number of bits for a hash table\.
-.
 .IP
-Bigger hash tables cause less collisions which usually makes compression faster, but requires more memory during compression\.
-.
+Bigger hash tables cause fewer collisions which usually makes compression faster, but requires more memory during compression\.
 .IP
-The minimum \fIhlog\fR is 6 (64 B) and the maximum is 30 (1 GiB)\.
-.
+The minimum \fIhlog\fR is 6 (64 entries / 256 B) and the maximum is 30 (1B entries / 4 GiB)\.
 .TP
 \fBchainLog\fR=\fIclog\fR, \fBclog\fR=\fIclog\fR
-Specify the maximum number of bits for a hash chain or a binary tree\.
-.
+Specify the maximum number of bits for the secondary search structure, whose form depends on the selected \fBstrategy\fR\.
 .IP
-Higher numbers of bits increases the chance to find a match which usually improves compression ratio\. It also slows down compression speed and increases memory requirements for compression\. This option is ignored for the ZSTD_fast strategy\.
-.
+Higher numbers of bits increases the chance to find a match which usually improves compression ratio\. It also slows down compression speed and increases memory requirements for compression\. This option is ignored for the \fBZSTD_fast\fR \fBstrategy\fR, which only has the primary hash table\.
 .IP
-The minimum \fIclog\fR is 6 (64 B) and the maximum is 29 (524 Mib) on 32\-bit platforms and 30 (1 Gib) on 64\-bit platforms\.
-.
+The minimum \fIclog\fR is 6 (64 entries / 256 B) and the maximum is 29 (512M entries / 2 GiB) on 32\-bit platforms and 30 (1B entries / 4 GiB) on 64\-bit platforms\.
 .TP
 \fBsearchLog\fR=\fIslog\fR, \fBslog\fR=\fIslog\fR
 Specify the maximum number of searches in a hash chain or a binary tree using logarithmic scale\.
-.
 .IP
 More searches increases the chance to find a match which usually increases compression ratio but decreases compression speed\.
-.
 .IP
 The minimum \fIslog\fR is 1 and the maximum is \'windowLog\' \- 1\.
-.
 .TP
 \fBminMatch\fR=\fImml\fR, \fBmml\fR=\fImml\fR
 Specify the minimum searched length of a match in a hash table\.
-.
 .IP
 Larger search lengths usually decrease compression ratio but improve decompression speed\.
-.
 .IP
 The minimum \fImml\fR is 3 and the maximum is 7\.
-.
 .TP
 \fBtargetLength\fR=\fItlen\fR, \fBtlen\fR=\fItlen\fR
 The impact of this field vary depending on selected strategy\.
-.
 .IP
-For ZSTD_btopt, ZSTD_btultra and ZSTD_btultra2, it specifies the minimum match length that causes match finder to stop searching\. A larger \fBtargetLength\fR usually improves compression ratio but decreases compression speed\. t For ZSTD_fast, it triggers ultra\-fast mode when > 0\. The value represents the amount of data skipped between match sampling\. Impact is reversed : a larger \fBtargetLength\fR increases compression speed but decreases compression ratio\.
-.
+For \fBZSTD_btopt\fR, \fBZSTD_btultra\fR and \fBZSTD_btultra2\fR, it specifies the minimum match length that causes match finder to stop searching\. A larger \fBtargetLength\fR usually improves compression ratio but decreases compression speed\.
+.IP
+For \fBZSTD_fast\fR, it triggers ultra\-fast mode when > 0\. The value represents the amount of data skipped between match sampling\. Impact is reversed: a larger \fBtargetLength\fR increases compression speed but decreases compression ratio\.
 .IP
 For all other strategies, this field has no impact\.
-.
 .IP
-The minimum \fItlen\fR is 0 and the maximum is 128 Kib\.
-.
+The minimum \fItlen\fR is 0 and the maximum is 128 KiB\.
 .TP
 \fBoverlapLog\fR=\fIovlog\fR, \fBovlog\fR=\fIovlog\fR
 Determine \fBoverlapSize\fR, amount of data reloaded from previous job\. This parameter is only available when multithreading is enabled\. Reloading more data improves compression ratio, but decreases speed\.
-.
 .IP
-The minimum \fIovlog\fR is 0, and the maximum is 9\. 1 means "no overlap", hence completely independent jobs\. 9 means "full overlap", meaning up to \fBwindowSize\fR is reloaded from previous job\. Reducing \fIovlog\fR by 1 reduces the reloaded amount by a factor 2\. For example, 8 means "windowSize/2", and 6 means "windowSize/8"\. Value 0 is special and means "default" : \fIovlog\fR is automatically determined by \fBzstd\fR\. In which case, \fIovlog\fR will range from 6 to 9, depending on selected \fIstrat\fR\.
-.
+The minimum \fIovlog\fR is 0, and the maximum is 9\. 1 means "no overlap", hence completely independent jobs\. 9 means "full overlap", meaning up to \fBwindowSize\fR is reloaded from previous job\. Reducing \fIovlog\fR by 1 reduces the reloaded amount by a factor 2\. For example, 8 means "windowSize/2", and 6 means "windowSize/8"\. Value 0 is special and means "default": \fIovlog\fR is automatically determined by \fBzstd\fR\. In which case, \fIovlog\fR will range from 6 to 9, depending on selected \fIstrat\fR\.
 .TP
 \fBldmHashLog\fR=\fIlhlog\fR, \fBlhlog\fR=\fIlhlog\fR
 Specify the maximum size for a hash table used for long distance matching\.
-.
 .IP
 This option is ignored unless long distance matching is enabled\.
-.
 .IP
 Bigger hash tables usually improve compression ratio at the expense of more memory during compression and a decrease in compression speed\.
-.
 .IP
 The minimum \fIlhlog\fR is 6 and the maximum is 30 (default: 20)\.
-.
 .TP
 \fBldmMinMatch\fR=\fIlmml\fR, \fBlmml\fR=\fIlmml\fR
 Specify the minimum searched length of a match for long distance matching\.
-.
 .IP
 This option is ignored unless long distance matching is enabled\.
-.
 .IP
 Larger/very small values usually decrease compression ratio\.
-.
 .IP
 The minimum \fIlmml\fR is 4 and the maximum is 4096 (default: 64)\.
-.
 .TP
 \fBldmBucketSizeLog\fR=\fIlblog\fR, \fBlblog\fR=\fIlblog\fR
 Specify the size of each bucket for the hash table used for long distance matching\.
-.
 .IP
 This option is ignored unless long distance matching is enabled\.
-.
 .IP
 Larger bucket sizes improve collision resolution but decrease compression speed\.
-.
 .IP
 The minimum \fIlblog\fR is 1 and the maximum is 8 (default: 3)\.
-.
 .TP
 \fBldmHashRateLog\fR=\fIlhrlog\fR, \fBlhrlog\fR=\fIlhrlog\fR
 Specify the frequency of inserting entries into the long distance matching hash table\.
-.
 .IP
 This option is ignored unless long distance matching is enabled\.
-.
 .IP
 Larger values will improve compression speed\. Deviating far from the default value will likely result in a decrease in compression ratio\.
-.
 .IP
 The default value is \fBwlog \- lhlog\fR\.
-.
 .SS "Example"
 The following parameters sets advanced compression options to something similar to predefined level 19 for files bigger than 256 KB:
-.
 .P
 \fB\-\-zstd\fR=wlog=23,clog=23,hlog=22,slog=6,mml=3,tlen=48,strat=6
-.
+.SH "SEE ALSO"
+\fBzstdgrep\fR(1), \fBzstdless\fR(1), \fBgzip\fR(1), \fBxz\fR(1)
+.P
+The \fIzstandard\fR format is specified in Y\. Collet, "Zstandard Compression and the \'application/zstd\' Media Type", https://www\.ietf\.org/rfc/rfc8878\.txt, Internet RFC 8878 (February 2021)\.
 .SH "BUGS"
 Report bugs at: https://github\.com/facebook/zstd/issues
-.
 .SH "AUTHOR"
 Yann Collet
diff --git a/priv/zstd/programs/zstd.1.md b/priv/zstd/programs/zstd.1.md
index ae50928..3b7f24f 100644
--- a/priv/zstd/programs/zstd.1.md
+++ b/priv/zstd/programs/zstd.1.md
@@ -4,7 +4,7 @@ zstd(1) -- zstd, zstdmt, unzstd, zstdcat - Compress or decompress .zst files
 SYNOPSIS
 --------
 
-`zstd` [*OPTIONS*] [-|_INPUT-FILE_] [-o _OUTPUT-FILE_]
+`zstd` [] [-|] [-o ]
 
 `zstdmt` is equivalent to `zstd -T0`
 
@@ -16,32 +16,32 @@ SYNOPSIS
 DESCRIPTION
 -----------
 `zstd` is a fast lossless compression algorithm and data compression tool,
-with command line syntax similar to `gzip (1)` and `xz (1)`.
+with command line syntax similar to `gzip`(1) and `xz`(1).
 It is based on the **LZ77** family, with further FSE & huff0 entropy stages.
 `zstd` offers highly configurable compression speed,
-with fast modes at > 200 MB/s per core,
-and strong modes nearing lzma compression ratios.
+from fast modes at > 200 MB/s per core,
+to strong modes with excellent compression ratios.
 It also features a very fast decoder, with speeds > 500 MB/s per core.
 
 `zstd` command line syntax is generally similar to gzip,
-but features the following differences :
+but features the following differences:
 
   - Source files are preserved by default.
     It's possible to remove them automatically by using the `--rm` command.
   - When compressing a single file, `zstd` displays progress notifications
     and result summary by default.
     Use `-q` to turn them off.
-  - `zstd` does not accept input from console,
-    but it properly accepts `stdin` when it's not the console.
   - `zstd` displays a short help page when command line is an error.
     Use `-q` to turn it off.
+  - `zstd` does not accept input from console,
+    though it does accept `stdin` when it's not the console.
+  - `zstd` does not store the input's filename or attributes, only its contents.
 
-`zstd` compresses or decompresses each _file_ according to the selected
-operation mode.
+`zstd` processes each _file_ according to the selected operation mode.
 If no _files_ are given or _file_ is `-`, `zstd` reads from standard input
 and writes the processed data to standard output.
 `zstd` will refuse to write compressed data to standard output
-if it is a terminal : it will display an error message and skip the _file_.
+if it is a terminal: it will display an error message and skip the file.
 Similarly, `zstd` will refuse to read compressed data from standard input
 if it is a terminal.
 
@@ -53,14 +53,15 @@ whose name is derived from the source _file_ name:
 * When decompressing, the `.zst` suffix is removed from the source filename to
   get the target filename
 
-### Concatenation with .zst files
-It is possible to concatenate `.zst` files as is.
-`zstd` will decompress such files as if they were a single `.zst` file.
+### Concatenation with .zst Files
+It is possible to concatenate multiple `.zst` files. `zstd` will decompress
+such agglomerated file as if it was a single `.zst` file.
 
 OPTIONS
 -------
 
-### Integer suffixes and special values
+### Integer Suffixes and Special Values
+
 In most places where an integer argument is expected,
 an optional suffix is supported to easily indicate large integers.
 There must be no space between the integer and the suffix.
@@ -72,7 +73,8 @@ There must be no space between the integer and the suffix.
     Multiply the integer by 1,048,576 (2\^20).
     `Mi`, `M`, and `MB` are accepted as synonyms for `MiB`.
 
-### Operation mode
+### Operation Mode
+
 If multiple operation mode options are given,
 the last one takes effect.
 
@@ -85,23 +87,25 @@ the last one takes effect.
     Decompress.
 * `-t`, `--test`:
     Test the integrity of compressed _files_.
-    This option is equivalent to `--decompress --stdout` except that the
-    decompressed data is discarded instead of being written to standard output.
+    This option is equivalent to `--decompress --stdout > /dev/null`,
+    decompressed data is discarded and checksummed for errors.
     No files are created or removed.
 * `-b#`:
-    Benchmark file(s) using compression level #
-* `--train FILEs`:
-    Use FILEs as a training set to create a dictionary.
+    Benchmark file(s) using compression level _#_.
+    See _BENCHMARK_ below for a description of this operation.
+* `--train FILES`:
+    Use _FILES_ as a training set to create a dictionary.
     The training set should contain a lot of small files (> 100).
+    See _DICTIONARY BUILDER_ below for a description of this operation.
 * `-l`, `--list`:
     Display information related to a zstd compressed file, such as size, ratio, and checksum.
     Some of these fields may not be available.
-    This command can be augmented with the `-v` modifier.
+    This command's output can be augmented with the `-v` modifier.
 
-### Operation modifiers
+### Operation Modifiers
 
 * `-#`:
-    `#` compression level \[1-19] (default: 3)
+    selects `#` compression level \[1-19\] (default: 3)
 * `--ultra`:
     unlocks high compression levels 20+ (maximum 22), using a lot more memory.
     Note that decompression will also require more memory when using these levels.
@@ -119,24 +123,31 @@ the last one takes effect.
     which is either 64 in 32-bit mode, or 256 for 64-bit environments.
     This modifier does nothing if `zstd` is compiled without multithread support.
 * `--single-thread`:
-    Does not spawn a thread for compression, use a single thread for both I/O and compression.
-    In this mode, compression is serialized with I/O, which is slightly slower.
-    (This is different from `-T1`, which spawns 1 compression thread in parallel of I/O).
-    This mode is the only one available when multithread support is disabled.
-    Single-thread mode features lower memory usage.
-    Final compressed result is slightly different from `-T1`.
-* `--adapt[=min=#,max=#]` :
+    Use a single thread for both I/O and compression.
+    As compression is serialized with I/O, this can be slightly slower.
+    Single-thread mode features significantly lower memory usage,
+    which can be useful for systems with limited amount of memory, such as 32-bit systems.
+
+    Note 1: this mode is the only available one when multithread support is disabled.
+
+    Note 2: this mode is different from `-T1`, which spawns 1 compression thread in parallel with I/O.
+    Final compressed result is also slightly different from `-T1`.
+* `--auto-threads={physical,logical} (default: physical)`:
+    When using a default amount of threads via `-T0`, choose the default based on the number
+    of detected physical or logical cores.
+* `--adapt[=min=#,max=#]`:
     `zstd` will dynamically adapt compression level to perceived I/O conditions.
     Compression level adaptation can be observed live by using command `-v`.
     Adaptation can be constrained between supplied `min` and `max` levels.
     The feature works when combined with multi-threading and `--long` mode.
     It does not work with `--single-thread`.
-    It sets window size to 8 MB by default (can be changed manually, see `wlog`).
+    It sets window size to 8 MiB by default (can be changed manually, see `wlog`).
     Due to the chaotic nature of dynamic adaptation, compressed result is not reproducible.
-    _note_ : at the time of this writing, `--adapt` can remain stuck at low speed
+
+    _Note_: at the time of this writing, `--adapt` can remain stuck at low speed
     when combined with multiple worker threads (>=2).
 * `--long[=#]`:
-    enables long distance matching with `#` `windowLog`, if not `#` is not
+    enables long distance matching with `#` `windowLog`, if `#` is not
     present it defaults to `27`.
     This increases the window size (`windowLog`) and memory usage for both the
     compressor and decompressor.
@@ -150,44 +161,51 @@ the last one takes effect.
 * `--patch-from FILE`:
     Specify the file to be used as a reference point for zstd's diff engine.
     This is effectively dictionary compression with some convenient parameter
-    selection, namely that windowSize > srcSize.
+    selection, namely that _windowSize_ > _srcSize_.
+
+    Note: cannot use both this and `-D` together.
 
-    Note: cannot use both this and -D together
-    Note: `--long` mode will be automatically activated if chainLog < fileLog
-        (fileLog being the windowLog required to cover the whole file). You
+    Note: `--long` mode will be automatically activated if _chainLog_ < _fileLog_
+        (_fileLog_ being the _windowLog_ required to cover the whole file). You
         can also manually force it.
-	Node: for all levels, you can use --patch-from in --single-thread mode
-		to improve compression ratio at the cost of speed
+
+    Note: for all levels, you can use `--patch-from` in `--single-thread` mode
+        to improve compression ratio at the cost of speed.
+
     Note: for level 19, you can get increased compression ratio at the cost
         of speed by specifying `--zstd=targetLength=` to be something large
-        (i.e 4096), and by setting a large `--zstd=chainLog=`
-* `--rsyncable` :
+        (i.e. 4096), and by setting a large `--zstd=chainLog=`.
+* `--rsyncable`:
     `zstd` will periodically synchronize the compression state to make the
     compressed file more rsync-friendly. There is a negligible impact to
     compression ratio, and the faster compression levels will see a small
     compression speed hit.
     This feature does not work with `--single-thread`. You probably don't want
     to use it with long range mode, since it will decrease the effectiveness of
-    the synchronization points, but your milage may vary.
+    the synchronization points, but your mileage may vary.
 * `-C`, `--[no-]check`:
     add integrity check computed from uncompressed data (default: enabled)
 * `--[no-]content-size`:
     enable / disable whether or not the original size of the file is placed in
     the header of the compressed file. The default option is
-    --content-size (meaning that the original size will be placed in the header).
+    `--content-size` (meaning that the original size will be placed in the header).
 * `--no-dictID`:
     do not store dictionary ID within frame header (dictionary compression).
     The decoder will have to rely on implicit knowledge about which dictionary to use,
     it won't be able to check if it's correct.
 * `-M#`, `--memory=#`:
-    Set a memory usage limit. By default, Zstandard uses 128 MB for decompression
+    Set a memory usage limit. By default, `zstd` uses 128 MiB for decompression
     as the maximum amount of memory the decompressor is allowed to use, but you can
-    override this manually if need be in either direction (ie. you can increase or
+    override this manually if need be in either direction (i.e. you can increase or
     decrease it).
 
-    This is also used during compression when using with --patch-from=. In this case,
-    this parameter overrides that maximum size allowed for a dictionary. (128 MB).
-* `--stream-size=#` :
+    This is also used during compression when using with `--patch-from=`. In this case,
+    this parameter overrides that maximum size allowed for a dictionary. (128 MiB).
+
+    Additionally, this can be used to limit memory for dictionary training. This parameter
+    overrides the default limit of 2 GiB. zstd will load training samples up to the memory limit
+    and ignore the rest.
+* `--stream-size=#`:
     Sets the pledged source size of input coming from a stream. This value must be exact, as it
     will be included in the produced frame header. Incorrect stream sizes will cause an error.
     This information will be used to better optimize compression parameters, resulting in
@@ -200,12 +218,14 @@ the last one takes effect.
     Exact guesses result in better compression ratios. Overestimates result in slightly
     degraded compression ratios, while underestimates may result in significant degradation.
 * `-o FILE`:
-    save result into `FILE`
+    save result into `FILE`.
 * `-f`, `--force`:
     disable input and output checks. Allows overwriting existing files, input
     from console, output to stdout, operating on links, block devices, etc.
+    During decompression and when the output destination is stdout, pass-through
+    unrecognized formats as-is.
 * `-c`, `--stdout`:
-    force write to standard output, even if it is the console
+    write to standard output (even if it is the console); keep original files unchanged.
 * `--[no-]sparse`:
     enable / disable sparse FS support,
     to make files with many zeroes smaller on disk.
@@ -214,14 +234,26 @@ the last one takes effect.
     default: enabled when output is into a file,
     and disabled when output is stdout.
     This setting overrides default and can force sparse mode over stdout.
+* `--[no-]pass-through`
+    enable / disable passing through uncompressed files as-is. During
+    decompression when pass-through is enabled, unrecognized formats will be
+    copied as-is from the input to the output. By default, pass-through will
+    occur when the output destination is stdout and the force (`-f`) option is
+    set.
 * `--rm`:
-    remove source file(s) after successful compression or decompression. If used in combination with
-    -o, will trigger a confirmation prompt (which can be silenced with -f), as this is a destructive operation.
+    remove source file(s) after successful compression or decompression.
+    This command is silently ignored if output is `stdout`.
+    If used in combination with `-o`,
+    triggers a confirmation prompt (which can be silenced with `-f`), as this is a destructive operation.
 * `-k`, `--keep`:
     keep source file(s) after successful compression or decompression.
     This is the default behavior.
 * `-r`:
-    operate recursively on directories
+    operate recursively on directories.
+    It selects all files in the named directory and all its subdirectories.
+    This can be useful both to reduce command line typing,
+    and to circumvent shell expansion limitations,
+    when there are a lot of files and naming breaks the maximum size of a command line.
 * `--filelist FILE`
     read a list of files to process as content from `FILE`.
     Format is compatible with `ls` output, with one file per line.
@@ -251,7 +283,7 @@ the last one takes effect.
     display help/long help and exit
 * `-V`, `--version`:
     display version number and exit.
-    Advanced : `-vV` also displays supported formats.
+    Advanced: `-vV` also displays supported formats.
     `-vvV` also displays POSIX support.
     `-q` will only display the version number, suitable for machine reading.
 * `-v`, `--verbose`:
@@ -262,17 +294,27 @@ the last one takes effect.
 * `--no-progress`:
     do not display the progress bar, but keep all other messages.
 * `--show-default-cparams`:
-    Shows the default compression parameters that will be used for a
-    particular src file. If the provided src file is not a regular file
-    (eg. named pipe), the cli will just output the default parameters.
-    That is, the parameters that are used when the src size is unknown.
+    shows the default compression parameters that will be used for a particular input file, based on the provided compression level and the input size.
+    If the provided file is not a regular file (e.g. a pipe), this flag will output the parameters used for inputs of unknown size.
 * `--`:
     All arguments after `--` are treated as files
 
-### Restricted usage of Environment Variables
 
-Using environment variables to set parameters has security implications.
-Therefore, this avenue is intentionally restricted.
+### gzip Operation Modifiers
+When invoked via a `gzip` symlink, `zstd` will support further
+options that intend to mimic the `gzip` behavior:
+
+* `-n`, `--no-name`:
+    do not store the original filename and timestamps when compressing
+    a file. This is the default behavior and hence a no-op.
+* `--best`:
+    alias to the option `-9`.
+
+
+### Environment Variables
+
+Employing environment variables to set parameters has security implications.
+Therefore, this avenue is intentionally limited.
 Only `ZSTD_CLEVEL` and `ZSTD_NBTHREADS` are currently supported.
 They set the compression level and number of threads to use during compression, respectively.
 
@@ -282,8 +324,8 @@ If the value of `ZSTD_CLEVEL` is not a valid integer, it will be ignored with a
 
 `ZSTD_NBTHREADS` can be used to set the number of threads `zstd` will attempt to use during compression.
 If the value of `ZSTD_NBTHREADS` is not a valid unsigned integer, it will be ignored with a warning message.
-`ZSTD_NBTHREADS` has a default value of (`1`), and is capped at ZSTDMT_NBWORKERS_MAX==200. `zstd` must be
-compiled with multithread support for this to have any effect.
+`ZSTD_NBTHREADS` has a default value of (`1`), and is capped at ZSTDMT_NBWORKERS_MAX==200.
+`zstd` must be compiled with multithread support for this to have any effect.
 
 They can both be overridden by corresponding command line arguments:
 `-#` for compression level and `-T#` for number of compression threads.
@@ -295,41 +337,76 @@ DICTIONARY BUILDER
 which greatly improves efficiency on small files and messages.
 It's possible to train `zstd` with a set of samples,
 the result of which is saved into a file called a `dictionary`.
-Then during compression and decompression, reference the same dictionary,
+Then, during compression and decompression, reference the same dictionary,
 using command `-D dictionaryFileName`.
 Compression of small files similar to the sample set will be greatly improved.
 
 * `--train FILEs`:
     Use FILEs as training set to create a dictionary.
-    The training set should contain a lot of small files (> 100),
+    The training set should ideally contain a lot of samples (> 100),
     and weight typically 100x the target dictionary size
-    (for example, 10 MB for a 100 KB dictionary).
+    (for example, ~10 MB for a 100 KB dictionary).
+    `--train` can be combined with `-r` to indicate a directory rather than listing all the files,
+    which can be useful to circumvent shell expansion limits.
 
-    Supports multithreading if `zstd` is compiled with threading support.
-    Additional parameters can be specified with `--train-fastcover`.
+    Since dictionary compression is mostly effective for small files,
+    the expectation is that the training set will only contain small files.
+    In the case where some samples happen to be large,
+    only the first 128 KiB of these samples will be used for training.
+
+    `--train` supports multithreading if `zstd` is compiled with threading support (default).
+    Additional advanced parameters can be specified with `--train-fastcover`.
     The legacy dictionary builder can be accessed with `--train-legacy`.
-    The cover dictionary builder can be accessed with `--train-cover`.
-    Equivalent to `--train-fastcover=d=8,steps=4`.
-* `-o file`:
-    Dictionary saved into `file` (default name: dictionary).
+    The slower cover dictionary builder can be accessed with `--train-cover`.
+    Default `--train` is equivalent to `--train-fastcover=d=8,steps=4`.
+
+* `-o FILE`:
+    Dictionary saved into `FILE` (default name: dictionary).
 * `--maxdict=#`:
-    Limit dictionary to specified size (default: 112640).
+    Limit dictionary to specified size (default: 112640 bytes).
+    As usual, quantities are expressed in bytes by default,
+    and it's possible to employ suffixes (like `KB` or `MB`)
+    to specify larger values.
 * `-#`:
     Use `#` compression level during training (optional).
     Will generate statistics more tuned for selected compression level,
     resulting in a _small_ compression ratio improvement for this level.
 * `-B#`:
-    Split input files in blocks of size # (default: no split)
+    Split input files into blocks of size # (default: no split)
+* `-M#`, `--memory=#`:
+    Limit the amount of sample data loaded for training (default: 2 GB).
+    Note that the default (2 GB) is also the maximum.
+    This parameter can be useful in situations where the training set size
+    is not well controlled and could be potentially very large.
+    Since speed of the training process is directly correlated to
+    the size of the training sample set,
+    a smaller sample set leads to faster training.
+
+    In situations where the training set is larger than maximum memory,
+    the CLI will randomly select samples among the available ones,
+    up to the maximum allowed memory budget.
+    This is meant to improve dictionary relevance
+    by mitigating the potential impact of clustering,
+    such as selecting only files from the beginning of a list
+    sorted by modification date, or sorted by alphabetical order.
+    The randomization process is deterministic, so
+    training of the same list of files with the same parameters
+    will lead to the creation of the same dictionary.
+
 * `--dictID=#`:
-    A dictionary ID is a locally unique ID that a decoder can use to verify it is
-    using the right dictionary.
+    A dictionary ID is a locally unique ID.
+    The decoder will use this value to verify it is using the right dictionary.
     By default, zstd will create a 4-bytes random number ID.
-    It's possible to give a precise number instead.
-    Short numbers have an advantage : an ID < 256 will only need 1 byte in the
-    compressed frame header, and an ID < 65536 will only need 2 bytes.
-    This compares favorably to 4 bytes default.
-    However, it's up to the dictionary manager to not assign twice the same ID to
+    It's possible to provide an explicit number ID instead.
+    It's up to the dictionary manager to not assign twice the same ID to
     2 different dictionaries.
+    Note that short numbers have an advantage:
+    an ID < 256 will only need 1 byte in the compressed frame header,
+    and an ID < 65536 will only need 2 bytes.
+    This compares favorably to 4 bytes default.
+
+    Note that RFC8878 reserves IDs less than 32768 and greater than or equal to 2\^31, so they should not be used in public.
+
 * `--train-cover[=k#,d=#,steps=#,split=#,shrink[=#]]`:
     Select parameters for the default dictionary builder algorithm named cover.
     If _d_ is not specified, then it tries _d_ = 6 and _d_ = 8.
@@ -394,7 +471,7 @@ Compression of small files similar to the sample set will be greatly improved.
     Use legacy dictionary builder algorithm with the given dictionary
     _selectivity_ (default: 9).
     The smaller the _selectivity_ value, the denser the dictionary,
-    improving its efficiency but reducing its possible maximum size.
+    improving its efficiency but reducing its achievable maximum size.
     `--train-legacy=s=#` is also accepted.
 
     Examples:
@@ -414,30 +491,31 @@ BENCHMARK
 * `-i#`:
     minimum evaluation time, in seconds (default: 3s), benchmark mode only
 * `-B#`, `--block-size=#`:
-    cut file(s) into independent blocks of size # (default: no block)
+    cut file(s) into independent chunks of size # (default: no chunking)
 * `--priority=rt`:
     set process priority to real-time
 
-**Output Format:** CompressionLevel#Filename : IntputSize -> OutputSize (CompressionRatio), CompressionSpeed, DecompressionSpeed
+**Output Format:** CompressionLevel#Filename: InputSize -> OutputSize (CompressionRatio), CompressionSpeed, DecompressionSpeed
 
 **Methodology:** For both compression and decompression speed, the entire input is compressed/decompressed in-memory to measure speed. A run lasts at least 1 sec, so when files are small, they are compressed/decompressed several times per run, in order to improve measurement accuracy.
 
 ADVANCED COMPRESSION OPTIONS
 ----------------------------
 ### -B#:
-Select the size of each compression job.
+Specify the size of each compression job.
 This parameter is only available when multi-threading is enabled.
 Each compression job is run in parallel, so this value indirectly impacts the nb of active threads.
 Default job size varies depending on compression level (generally  `4 * windowSize`).
 `-B#` makes it possible to manually select a custom size.
 Note that job size must respect a minimum value which is enforced transparently.
 This minimum is either 512 KB, or `overlapSize`, whichever is largest.
-Different job sizes will lead to (slightly) different compressed frames.
+Different job sizes will lead to non-identical compressed frames.
 
 ### --zstd[=options]:
-`zstd` provides 22 predefined compression levels.
-The selected or default predefined compression level can be changed with
-advanced compression options.
+`zstd` provides 22 predefined regular compression levels plus the fast levels.
+This compression level is translated internally into a number of specific parameters that actually control the behavior of the compressor.
+(You can see the result of this translation with `--show-default-cparams`.)
+These specific parameters can be overridden with advanced compression options.
 The _options_ are provided as a comma-separated list.
 You may specify only the options you want to change and the rest will be
 taken from the selected or default compression level.
@@ -446,10 +524,10 @@ The list of available _options_:
 - `strategy`=_strat_, `strat`=_strat_:
     Specify a strategy used by a match finder.
 
-    There are 9 strategies numbered from 1 to 9, from faster to stronger:
-    1=ZSTD\_fast, 2=ZSTD\_dfast, 3=ZSTD\_greedy,
-    4=ZSTD\_lazy, 5=ZSTD\_lazy2, 6=ZSTD\_btlazy2,
-    7=ZSTD\_btopt, 8=ZSTD\_btultra, 9=ZSTD\_btultra2.
+    There are 9 strategies numbered from 1 to 9, from fastest to strongest:
+    1=`ZSTD_fast`, 2=`ZSTD_dfast`, 3=`ZSTD_greedy`,
+    4=`ZSTD_lazy`, 5=`ZSTD_lazy2`, 6=`ZSTD_btlazy2`,
+    7=`ZSTD_btopt`, 8=`ZSTD_btultra`, 9=`ZSTD_btultra2`.
 
 - `windowLog`=_wlog_, `wlog`=_wlog_:
     Specify the maximum number of bits for a match distance.
@@ -466,22 +544,23 @@ The list of available _options_:
 - `hashLog`=_hlog_, `hlog`=_hlog_:
     Specify the maximum number of bits for a hash table.
 
-    Bigger hash tables cause less collisions which usually makes compression
+    Bigger hash tables cause fewer collisions which usually makes compression
     faster, but requires more memory during compression.
 
-    The minimum _hlog_ is 6 (64 B) and the maximum is 30 (1 GiB).
+    The minimum _hlog_ is 6 (64 entries / 256 B) and the maximum is 30 (1B entries / 4 GiB).
 
 - `chainLog`=_clog_, `clog`=_clog_:
-    Specify the maximum number of bits for a hash chain or a binary tree.
+    Specify the maximum number of bits for the secondary search structure,
+    whose form depends on the selected `strategy`.
 
     Higher numbers of bits increases the chance to find a match which usually
     improves compression ratio.
     It also slows down compression speed and increases memory requirements for
     compression.
-    This option is ignored for the ZSTD_fast strategy.
+    This option is ignored for the `ZSTD_fast` `strategy`, which only has the primary hash table.
 
-    The minimum _clog_ is 6 (64 B) and the maximum is 29 (524 Mib) on 32-bit platforms
-    and 30 (1 Gib) on 64-bit platforms.
+    The minimum _clog_ is 6 (64 entries / 256 B) and the maximum is 29 (512M entries / 2 GiB) on 32-bit platforms
+    and 30 (1B entries / 4 GiB) on 64-bit platforms.
 
 - `searchLog`=_slog_, `slog`=_slog_:
     Specify the maximum number of searches in a hash chain or a binary tree
@@ -503,19 +582,19 @@ The list of available _options_:
 - `targetLength`=_tlen_, `tlen`=_tlen_:
     The impact of this field vary depending on selected strategy.
 
-    For ZSTD\_btopt, ZSTD\_btultra and ZSTD\_btultra2, it specifies
+    For `ZSTD_btopt`, `ZSTD_btultra` and `ZSTD_btultra2`, it specifies
     the minimum match length that causes match finder to stop searching.
     A larger `targetLength` usually improves compression ratio
     but decreases compression speed.
-t
-    For ZSTD\_fast, it triggers ultra-fast mode when > 0.
+
+    For `ZSTD_fast`, it triggers ultra-fast mode when > 0.
     The value represents the amount of data skipped between match sampling.
-    Impact is reversed : a larger `targetLength` increases compression speed
+    Impact is reversed: a larger `targetLength` increases compression speed
     but decreases compression ratio.
 
     For all other strategies, this field has no impact.
 
-    The minimum _tlen_ is 0 and the maximum is 128 Kib.
+    The minimum _tlen_ is 0 and the maximum is 128 KiB.
 
 - `overlapLog`=_ovlog_,  `ovlog`=_ovlog_:
     Determine `overlapSize`, amount of data reloaded from previous job.
@@ -527,7 +606,7 @@ t
     9 means "full overlap", meaning up to `windowSize` is reloaded from previous job.
     Reducing _ovlog_ by 1 reduces the reloaded amount by a factor 2.
     For example, 8 means "windowSize/2", and 6 means "windowSize/8".
-    Value 0 is special and means "default" : _ovlog_ is automatically determined by `zstd`.
+    Value 0 is special and means "default": _ovlog_ is automatically determined by `zstd`.
     In which case, _ovlog_ will range from 6 to 9, depending on selected _strat_.
 
 - `ldmHashLog`=_lhlog_, `lhlog`=_lhlog_:
@@ -577,6 +656,11 @@ similar to predefined level 19 for files bigger than 256 KB:
 
 `--zstd`=wlog=23,clog=23,hlog=22,slog=6,mml=3,tlen=48,strat=6
 
+SEE ALSO
+--------
+`zstdgrep`(1), `zstdless`(1), `gzip`(1), `xz`(1)
+
+The  format is specified in Y. Collet, "Zstandard Compression and the 'application/zstd' Media Type", https://www.ietf.org/rfc/rfc8878.txt, Internet RFC 8878 (February 2021).
 
 BUGS
 ----
diff --git a/priv/zstd/programs/zstdcli.c b/priv/zstd/programs/zstdcli.c
index 239aaf4..93f75e2 100644
--- a/priv/zstd/programs/zstdcli.c
+++ b/priv/zstd/programs/zstdcli.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -27,8 +27,8 @@
 /*-************************************
 *  Dependencies
 **************************************/
-#include "platform.h" /* IS_CONSOLE, PLATFORM_POSIX_VERSION */
-#include "util.h"     /* UTIL_HAS_CREATEFILELIST, UTIL_createFileList */
+#include "platform.h" /* PLATFORM_POSIX_VERSION */
+#include "util.h"     /* UTIL_HAS_CREATEFILELIST, UTIL_createFileList, UTIL_isConsole */
 #include    /* getenv */
 #include    /* strcmp, strlen */
 #include     /* fprintf(), stdin, stdout, stderr */
@@ -46,17 +46,18 @@
 #  include "zstdcli_trace.h"
 #endif
 #include "../lib/zstd.h"  /* ZSTD_VERSION_STRING, ZSTD_minCLevel, ZSTD_maxCLevel */
+#include "fileio_asyncio.h"
 
 
 /*-************************************
 *  Constants
 **************************************/
-#define COMPRESSOR_NAME "zstd command line interface"
+#define COMPRESSOR_NAME "Zstandard CLI"
 #ifndef ZSTD_VERSION
 #  define ZSTD_VERSION "v" ZSTD_VERSION_STRING
 #endif
 #define AUTHOR "Yann Collet"
-#define WELCOME_MESSAGE "*** %s %i-bits %s, by %s ***\n", COMPRESSOR_NAME, (int)(sizeof(size_t)*8), ZSTD_VERSION, AUTHOR
+#define WELCOME_MESSAGE "*** %s (%i-bit) %s, by %s ***\n", COMPRESSOR_NAME, (int)(sizeof(size_t)*8), ZSTD_VERSION, AUTHOR
 
 #define ZSTD_ZSTDMT "zstdmt"
 #define ZSTD_UNZSTD "unzstd"
@@ -124,6 +125,15 @@ static void checkLibVersion(void)
 }
 
 
+/*! exeNameMatch() :
+    @return : a non-zero value if exeName matches test, excluding the extension
+   */
+static int exeNameMatch(const char* exeName, const char* test)
+{
+    return !strncmp(exeName, test, strlen(test)) &&
+        (exeName[strlen(test)] == '\0' || exeName[strlen(test)] == '.');
+}
+
 /*-************************************
 *  Command Line
 **************************************/
@@ -133,141 +143,174 @@ static void checkLibVersion(void)
  */
 static void usage(FILE* f, const char* programName)
 {
-    DISPLAY_F(f, "Usage : \n");
-    DISPLAY_F(f, "      %s [args] [FILE(s)] [-o file] \n", programName);
+    DISPLAY_F(f, "Compress or decompress the INPUT file(s); reads from STDIN if INPUT is `-` or not provided.\n\n");
+    DISPLAY_F(f, "Usage: %s [OPTIONS...] [INPUT... | -] [-o OUTPUT]\n\n", programName);
+    DISPLAY_F(f, "Options:\n");
+    DISPLAY_F(f, "  -o OUTPUT                     Write output to a single file, OUTPUT.\n");
+    DISPLAY_F(f, "  -k, --keep                    Preserve INPUT file(s). [Default] \n");
+    DISPLAY_F(f, "  --rm                          Remove INPUT file(s) after successful (de)compression.\n");
+#ifdef ZSTD_GZCOMPRESS
+    if (exeNameMatch(programName, ZSTD_GZ)) {     /* behave like gzip */
+        DISPLAY_F(f, "  -n, --no-name                 Do not store original filename when compressing.\n\n");
+    }
+#endif
     DISPLAY_F(f, "\n");
-    DISPLAY_F(f, "FILE    : a filename \n");
-    DISPLAY_F(f, "          with no FILE, or when FILE is - , read standard input\n");
-    DISPLAY_F(f, "Arguments : \n");
 #ifndef ZSTD_NOCOMPRESS
-    DISPLAY_F(f, " -#     : # compression level (1-%d, default: %d) \n", ZSTDCLI_CLEVEL_MAX, ZSTDCLI_CLEVEL_DEFAULT);
+    DISPLAY_F(f, "  -#                            Desired compression level, where `#` is a number between 1 and %d;\n", ZSTDCLI_CLEVEL_MAX);
+    DISPLAY_F(f, "                                lower numbers provide faster compression, higher numbers yield\n");
+    DISPLAY_F(f, "                                better compression ratios. [Default: %d]\n\n", ZSTDCLI_CLEVEL_DEFAULT);
 #endif
 #ifndef ZSTD_NODECOMPRESS
-    DISPLAY_F(f, " -d     : decompression \n");
+    DISPLAY_F(f, "  -d, --decompress              Perform decompression.\n");
 #endif
-    DISPLAY_F(f, " -D DICT: use DICT as Dictionary for compression or decompression \n");
-    DISPLAY_F(f, " -o file: result stored into `file` (only 1 output file) \n");
-    DISPLAY_F(f, " -f     : disable input and output checks. Allows overwriting existing files,\n");
-    DISPLAY_F(f, "          input from console, output to stdout, operating on links,\n");
-    DISPLAY_F(f, "          block devices, etc.\n");
-    DISPLAY_F(f, "--rm    : remove source file(s) after successful de/compression \n");
-    DISPLAY_F(f, " -k     : preserve source file(s) (default) \n");
-    DISPLAY_F(f, " -h/-H  : display help/long help and exit \n");
+    DISPLAY_F(f, "  -D DICT                       Use DICT as the dictionary for compression or decompression.\n\n");
+    DISPLAY_F(f, "  -f, --force                   Disable input and output checks. Allows overwriting existing files,\n");
+    DISPLAY_F(f, "                                receiving input from the console, printing ouput to STDOUT, and\n");
+    DISPLAY_F(f, "                                operating on links, block devices, etc. Unrecognized formats will be\n");
+    DISPLAY_F(f, "                                passed-through through as-is.\n\n");
+
+    DISPLAY_F(f, "  -h                            Display short usage and exit.\n");
+    DISPLAY_F(f, "  -H, --help                    Display full help and exit.\n");
+    DISPLAY_F(f, "  -V, --version                 Display the program version and exit.\n");
+    DISPLAY_F(f, "\n");
 }
 
 static void usage_advanced(const char* programName)
 {
     DISPLAYOUT(WELCOME_MESSAGE);
+    DISPLAYOUT("\n");
     usage(stdout, programName);
-    DISPLAYOUT( "\n");
-    DISPLAYOUT( "Advanced arguments : \n");
-    DISPLAYOUT( " -V     : display Version number and exit \n");
-
-    DISPLAYOUT( " -c     : force write to standard output, even if it is the console \n");
+    DISPLAYOUT("Advanced options:\n");
+    DISPLAYOUT("  -c, --stdout                  Write to STDOUT (even if it is a console) and keep the INPUT file(s).\n\n");
 
-    DISPLAYOUT( " -v     : verbose mode; specify multiple times to increase verbosity \n");
-    DISPLAYOUT( " -q     : suppress warnings; specify twice to suppress errors too \n");
-    DISPLAYOUT( "--[no-]progress : forcibly display, or never display the progress counter.\n");
-    DISPLAYOUT( "                  note: any (de)compressed output to terminal will mix with progress counter text. \n");
+    DISPLAYOUT("  -v, --verbose                 Enable verbose output; pass multiple times to increase verbosity.\n");
+    DISPLAYOUT("  -q, --quiet                   Suppress warnings; pass twice to suppress errors.\n");
+#ifndef ZSTD_NOTRACE
+    DISPLAYOUT("  --trace LOG                   Log tracing information to LOG.\n");
+#endif
+    DISPLAYOUT("\n");
+    DISPLAYOUT("  --[no-]progress               Forcibly show/hide the progress counter. NOTE: Any (de)compressed\n");
+    DISPLAYOUT("                                output to terminal will mix with progress counter text.\n\n");
 
 #ifdef UTIL_HAS_CREATEFILELIST
-    DISPLAYOUT( " -r     : operate recursively on directories \n");
-    DISPLAYOUT( "--filelist FILE : read list of files to operate upon from FILE \n");
-    DISPLAYOUT( "--output-dir-flat DIR : processed files are stored into DIR \n");
+    DISPLAYOUT("  -r                            Operate recursively on directories.\n");
+    DISPLAYOUT("  --filelist LIST               Read a list of files to operate on from LIST.\n");
+    DISPLAYOUT("  --output-dir-flat DIR         Store processed files in DIR.\n");
 #endif
 
 #ifdef UTIL_HAS_MIRRORFILELIST
-    DISPLAYOUT( "--output-dir-mirror DIR : processed files are stored into DIR respecting original directory structure \n");
+    DISPLAYOUT("  --output-dir-mirror DIR       Store processed files in DIR, respecting original directory structure.\n");
 #endif
+    if (AIO_supported())
+        DISPLAYOUT("  --[no-]asyncio                Use asynchronous IO. [Default: Enabled]\n");
 
-
+    DISPLAYOUT("\n");
 #ifndef ZSTD_NOCOMPRESS
-    DISPLAYOUT( "--[no-]check : during compression, add XXH64 integrity checksum to frame (default: enabled)");
+    DISPLAYOUT("  --[no-]check                  Add XXH64 integrity checksums during compression. [Default: Add, Validate]\n");
 #ifndef ZSTD_NODECOMPRESS
-    DISPLAYOUT( ". If specified with -d, decompressor will ignore/validate checksums in compressed frame (default: validate).");
+    DISPLAYOUT("                                If `-d` is present, ignore/validate checksums during decompression.\n");
 #endif
 #else
 #ifdef ZSTD_NOCOMPRESS
-    DISPLAYOUT( "--[no-]check : during decompression, ignore/validate checksums in compressed frame (default: validate).");
+    DISPLAYOUT("  --[no-]check                  Ignore/validate checksums during decompression. [Default: Validate]");
 #endif
 #endif /* ZSTD_NOCOMPRESS */
 
-#ifndef ZSTD_NOTRACE
-    DISPLAYOUT( "\n");
-    DISPLAYOUT( "--trace FILE : log tracing information to FILE.");
-#endif
-    DISPLAYOUT( "\n");
-
-    DISPLAYOUT( "--      : All arguments after \"--\" are treated as files \n");
+    DISPLAYOUT("\n");
+    DISPLAYOUT("  --                            Treat remaining arguments after `--` as files.\n");
 
 #ifndef ZSTD_NOCOMPRESS
-    DISPLAYOUT( "\n");
-    DISPLAYOUT( "Advanced compression arguments : \n");
-    DISPLAYOUT( "--ultra : enable levels beyond %i, up to %i (requires more memory) \n", ZSTDCLI_CLEVEL_MAX, ZSTD_maxCLevel());
-    DISPLAYOUT( "--long[=#]: enable long distance matching with given window log (default: %u) \n", g_defaultMaxWindowLog);
-    DISPLAYOUT( "--fast[=#]: switch to very fast compression levels (default: %u) \n", 1);
-    DISPLAYOUT( "--adapt : dynamically adapt compression level to I/O conditions \n");
-    DISPLAYOUT( "--[no-]row-match-finder : force enable/disable usage of fast row-based matchfinder for greedy, lazy, and lazy2 strategies \n");
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Advanced compression options:\n");
+    DISPLAYOUT("  --ultra                       Enable levels beyond %i, up to %i; requires more memory.\n", ZSTDCLI_CLEVEL_MAX, ZSTD_maxCLevel());
+    DISPLAYOUT("  --fast[=#]                    Use to very fast compression levels. [Default: %u]\n", 1);
+#ifdef ZSTD_GZCOMPRESS
+    if (exeNameMatch(programName, ZSTD_GZ)) {     /* behave like gzip */
+        DISPLAYOUT("  --best                        Compatibility alias for `-9`.\n");
+    }
+#endif
+    DISPLAYOUT("  --adapt                       Dynamically adapt compression level to I/O conditions.\n");
+    DISPLAYOUT("  --long[=#]                    Enable long distance matching with window log #. [Default: %u]\n", g_defaultMaxWindowLog);
+    DISPLAYOUT("  --patch-from=REF              Use REF as the reference point for Zstandard's diff engine. \n\n");
 # ifdef ZSTD_MULTITHREAD
-    DISPLAYOUT( " -T#    : spawns # compression threads (default: 1, 0==# cores) \n");
-    DISPLAYOUT( " -B#    : select size of each job (default: 0==automatic) \n");
-    DISPLAYOUT( "--single-thread : use a single thread for both I/O and compression (result slightly different than -T1) \n");
-    DISPLAYOUT( "--rsyncable : compress using a rsync-friendly method (-B sets block size) \n");
+    DISPLAYOUT("  -T#                           Spawn # compression threads. [Default: 1; pass 0 for core count.]\n");
+    DISPLAYOUT("  --single-thread               Share a single thread for I/O and compression (slightly different than `-T1`).\n");
+    DISPLAYOUT("  --auto-threads={physical|logical}\n");
+    DISPLAYOUT("                                Use physical/logical cores when using `-T0`. [Default: Physical]\n\n");
+    DISPLAYOUT("  -B#                           Set job size to #. [Default: 0 (automatic)]\n");
+    DISPLAYOUT("  --rsyncable                   Compress using a rsync-friendly method (`-B` sets block size). \n");
+    DISPLAYOUT("\n");
 # endif
-    DISPLAYOUT( "--exclude-compressed: only compress files that are not already compressed \n");
-    DISPLAYOUT( "--stream-size=# : specify size of streaming input from `stdin` \n");
-    DISPLAYOUT( "--size-hint=# optimize compression parameters for streaming input of approximately this size \n");
-    DISPLAYOUT( "--target-compressed-block-size=# : generate compressed block of approximately targeted size \n");
-    DISPLAYOUT( "--no-dictID : don't write dictID into header (dictionary compression only) \n");
-    DISPLAYOUT( "--[no-]compress-literals : force (un)compressed literals \n");
-
-    DISPLAYOUT( "--format=zstd : compress files to the .zst format (default) \n");
+    DISPLAYOUT("  --exclude-compressed          Only compress files that are not already compressed.\n\n");
+
+    DISPLAYOUT("  --stream-size=#               Specify size of streaming input from STDIN.\n");
+    DISPLAYOUT("  --size-hint=#                 Optimize compression parameters for streaming input of approximately size #.\n");
+    DISPLAYOUT("  --target-compressed-block-size=#\n");
+    DISPLAYOUT("                                Generate compressed blocks of approximately # size.\n\n");
+    DISPLAYOUT("  --no-dictID                   Don't write `dictID` into the header (dictionary compression only).\n");
+    DISPLAYOUT("  --[no-]compress-literals      Force (un)compressed literals.\n");
+    DISPLAYOUT("  --[no-]row-match-finder       Explicitly enable/disable the fast, row-based matchfinder for\n");
+    DISPLAYOUT("                                the 'greedy', 'lazy', and 'lazy2' strategies.\n");
+
+    DISPLAYOUT("\n");
+    DISPLAYOUT("  --format=zstd                 Compress files to the `.zst` format. [Default]\n");
 #ifdef ZSTD_GZCOMPRESS
-    DISPLAYOUT( "--format=gzip : compress files to the .gz format \n");
+    DISPLAYOUT("  --format=gzip                 Compress files to the `.gz` format.\n");
 #endif
 #ifdef ZSTD_LZMACOMPRESS
-    DISPLAYOUT( "--format=xz : compress files to the .xz format \n");
-    DISPLAYOUT( "--format=lzma : compress files to the .lzma format \n");
+    DISPLAYOUT("  --format=xz                   Compress files to the `.xz` format.\n");
+    DISPLAYOUT("  --format=lzma                 Compress files to the `.lzma` format.\n");
 #endif
 #ifdef ZSTD_LZ4COMPRESS
-    DISPLAYOUT( "--format=lz4 : compress files to the .lz4 format \n");
+    DISPLAYOUT( "  --format=lz4                 Compress files to the `.lz4` format.\n");
 #endif
 #endif  /* !ZSTD_NOCOMPRESS */
 
 #ifndef ZSTD_NODECOMPRESS
-    DISPLAYOUT( "\n");
-    DISPLAYOUT( "Advanced decompression arguments : \n");
-    DISPLAYOUT( " -l     : print information about zstd compressed files \n");
-    DISPLAYOUT( "--test  : test compressed file integrity \n");
-    DISPLAYOUT( " -M#    : Set a memory usage limit for decompression \n");
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Advanced decompression options:\n");
+    DISPLAYOUT("  -l                            Print information about Zstandard-compressed files.\n");
+    DISPLAYOUT("  --test                        Test compressed file integrity.\n");
+    DISPLAYOUT("  -M#                           Set the memory usage limit to # megabytes.\n");
 # if ZSTD_SPARSE_DEFAULT
-    DISPLAYOUT( "--[no-]sparse : sparse mode (default: enabled on file, disabled on stdout) \n");
+    DISPLAYOUT("  --[no-]sparse                 Enable sparse mode. [Default: Enabled for files, disabled for STDOUT.]\n");
 # else
-    DISPLAYOUT( "--[no-]sparse : sparse mode (default: disabled) \n");
+    DISPLAYOUT("  --[no-]sparse                 Enable sparse mode. [Default: Disabled]\n");
 # endif
+    {
+        char const* passThroughDefault = "Disabled";
+        if (exeNameMatch(programName, ZSTD_CAT) ||
+            exeNameMatch(programName, ZSTD_ZCAT) ||
+            exeNameMatch(programName, ZSTD_GZCAT)) {
+            passThroughDefault = "Enabled";
+        }
+        DISPLAYOUT("  --[no-]pass-through           Pass through uncompressed files as-is. [Default: %s]\n", passThroughDefault);
+    }
 #endif  /* ZSTD_NODECOMPRESS */
 
 #ifndef ZSTD_NODICT
-    DISPLAYOUT( "\n");
-    DISPLAYOUT( "Dictionary builder : \n");
-    DISPLAYOUT( "--train ## : create a dictionary from a training set of files \n");
-    DISPLAYOUT( "--train-cover[=k=#,d=#,steps=#,split=#,shrink[=#]] : use the cover algorithm with optional args \n");
-    DISPLAYOUT( "--train-fastcover[=k=#,d=#,f=#,steps=#,split=#,accel=#,shrink[=#]] : use the fast cover algorithm with optional args \n");
-    DISPLAYOUT( "--train-legacy[=s=#] : use the legacy algorithm with selectivity (default: %u) \n", g_defaultSelectivityLevel);
-    DISPLAYOUT( " -o DICT : DICT is dictionary name (default: %s) \n", g_defaultDictName);
-    DISPLAYOUT( "--maxdict=# : limit dictionary to specified size (default: %u) \n", g_defaultMaxDictSize);
-    DISPLAYOUT( "--dictID=# : force dictionary ID to specified value (default: random) \n");
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Dictionary builder:\n");
+    DISPLAYOUT("  --train                       Create a dictionary from a training set of files.\n\n");
+    DISPLAYOUT("  --train-cover[=k=#,d=#,steps=#,split=#,shrink[=#]]\n");
+    DISPLAYOUT("                                Use the cover algorithm (with optional arguments).\n");
+    DISPLAYOUT("  --train-fastcover[=k=#,d=#,f=#,steps=#,split=#,accel=#,shrink[=#]]\n");
+    DISPLAYOUT("                                Use the fast cover algorithm (with optional arguments).\n\n");
+    DISPLAYOUT("  --train-legacy[=s=#]          Use the legacy algorithm with selectivity #. [Default: %u]\n", g_defaultSelectivityLevel);
+    DISPLAYOUT("  -o NAME                       Use NAME as dictionary name. [Default: %s]\n", g_defaultDictName);
+    DISPLAYOUT("  --maxdict=#                   Limit dictionary to specified size #. [Default: %u]\n", g_defaultMaxDictSize);
+    DISPLAYOUT("  --dictID=#                    Force dictionary ID to #. [Default: Random]\n");
 #endif
 
 #ifndef ZSTD_NOBENCH
-    DISPLAYOUT( "\n");
-    DISPLAYOUT( "Benchmark arguments : \n");
-    DISPLAYOUT( " -b#    : benchmark file(s), using # compression level (default: %d) \n", ZSTDCLI_CLEVEL_DEFAULT);
-    DISPLAYOUT( " -e#    : test all compression levels successively from -b# to -e# (default: 1) \n");
-    DISPLAYOUT( " -i#    : minimum evaluation time in seconds (default: 3s) \n");
-    DISPLAYOUT( " -B#    : cut file into independent blocks of size # (default: no block) \n");
-    DISPLAYOUT( " -S     : output one benchmark result per input file (default: consolidated result) \n");
-    DISPLAYOUT( "--priority=rt : set process priority to real-time \n");
+    DISPLAYOUT("\n");
+    DISPLAYOUT("Benchmark options:\n");
+    DISPLAYOUT("  -b#                           Perform benchmarking with compression level #. [Default: %d]\n", ZSTDCLI_CLEVEL_DEFAULT);
+    DISPLAYOUT("  -e#                           Test all compression levels up to #; starting level is `-b#`. [Default: 1]\n");
+    DISPLAYOUT("  -i#                           Set the minimum evaluation to time # seconds. [Default: 3]\n");
+    DISPLAYOUT("  -B#                           Cut file into independent chunks of size #. [Default: No chunking]\n");
+    DISPLAYOUT("  -S                            Output one benchmark result per input file. [Default: Consolidated result]\n");
+    DISPLAYOUT("  --priority=rt                 Set process priority to real-time.\n");
 #endif
 
 }
@@ -294,18 +337,9 @@ static const char* lastNameFromPath(const char* path)
     return name;
 }
 
-/*! exeNameMatch() :
-    @return : a non-zero value if exeName matches test, excluding the extension
-   */
-static int exeNameMatch(const char* exeName, const char* test)
-{
-    return !strncmp(exeName, test, strlen(test)) &&
-        (exeName[strlen(test)] == '\0' || exeName[strlen(test)] == '.');
-}
-
 static void errorOut(const char* msg)
 {
-    DISPLAY("%s \n", msg); exit(1);
+    DISPLAYLEVEL(1, "%s \n", msg); exit(1);
 }
 
 /*! readU32FromCharChecked() :
@@ -353,6 +387,23 @@ static unsigned readU32FromChar(const char** stringPtr) {
     return result;
 }
 
+/*! readIntFromChar() :
+ * @return : signed integer value read from input in `char` format.
+ *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
+ *  Will also modify `*stringPtr`, advancing it to position where it stopped reading.
+ *  Note : function will exit() program if digit sequence overflows */
+static int readIntFromChar(const char** stringPtr) {
+    static const char errorMsg[] = "error: numeric value overflows 32-bit int";
+    int sign = 1;
+    unsigned result;
+    if (**stringPtr=='-') {
+        (*stringPtr)++;
+        sign = -1;
+    }
+    if (readU32FromCharChecked(stringPtr, &result)) { errorOut(errorMsg); }
+    return (int) result * sign;
+}
+
 /*! readSizeTFromCharChecked() :
  * @return 0 if success, and store the result in *value.
  *  allows and interprets K, KB, KiB, M, MB and MiB suffix.
@@ -546,8 +597,8 @@ static ZDICT_fastCover_params_t defaultFastCoverParams(void)
 static unsigned parseAdaptParameters(const char* stringPtr, int* adaptMinPtr, int* adaptMaxPtr)
 {
     for ( ; ;) {
-        if (longCommandWArg(&stringPtr, "min=")) { *adaptMinPtr = (int)readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
-        if (longCommandWArg(&stringPtr, "max=")) { *adaptMaxPtr = (int)readU32FromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "min=")) { *adaptMinPtr = readIntFromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
+        if (longCommandWArg(&stringPtr, "max=")) { *adaptMaxPtr = readIntFromChar(&stringPtr); if (stringPtr[0]==',') { stringPtr++; continue; } else break; }
         DISPLAYLEVEL(4, "invalid compression parameter \n");
         return 0;
     }
@@ -615,6 +666,11 @@ static void printVersion(void)
     #endif
         DISPLAYOUT("\n");
         if (g_displayLevel >= 4) {
+            /* library versions */
+            DISPLAYOUT("zlib version %s\n", FIO_zlibVersion());
+            DISPLAYOUT("lz4 version %s\n", FIO_lz4Version());
+            DISPLAYOUT("lzma version %s\n", FIO_lzmaVersion());
+
             /* posix support */
         #ifdef _POSIX_C_SOURCE
             DISPLAYOUT("_POSIX_C_SOURCE defined: %ldL\n", (long) _POSIX_C_SOURCE);
@@ -628,6 +684,48 @@ static void printVersion(void)
     }   }
 }
 
+#define ZSTD_NB_STRATEGIES 9
+static const char* ZSTD_strategyMap[ZSTD_NB_STRATEGIES + 1] = { "", "ZSTD_fast",
+                "ZSTD_dfast", "ZSTD_greedy", "ZSTD_lazy", "ZSTD_lazy2", "ZSTD_btlazy2",
+                "ZSTD_btopt", "ZSTD_btultra", "ZSTD_btultra2"};
+
+#ifndef ZSTD_NOCOMPRESS
+
+static void printDefaultCParams(const char* filename, const char* dictFileName, int cLevel) {
+    unsigned long long fileSize = UTIL_getFileSize(filename);
+    const size_t dictSize = dictFileName != NULL ? (size_t)UTIL_getFileSize(dictFileName) : 0;
+    const ZSTD_compressionParameters cParams = ZSTD_getCParams(cLevel, fileSize, dictSize);
+    if (fileSize != UTIL_FILESIZE_UNKNOWN) DISPLAY("%s (%u bytes)\n", filename, (unsigned)fileSize);
+    else DISPLAY("%s (src size unknown)\n", filename);
+    DISPLAY(" - windowLog     : %u\n", cParams.windowLog);
+    DISPLAY(" - chainLog      : %u\n", cParams.chainLog);
+    DISPLAY(" - hashLog       : %u\n", cParams.hashLog);
+    DISPLAY(" - searchLog     : %u\n", cParams.searchLog);
+    DISPLAY(" - minMatch      : %u\n", cParams.minMatch);
+    DISPLAY(" - targetLength  : %u\n", cParams.targetLength);
+    assert(cParams.strategy < ZSTD_NB_STRATEGIES + 1);
+    DISPLAY(" - strategy      : %s (%u)\n", ZSTD_strategyMap[(int)cParams.strategy], (unsigned)cParams.strategy);
+}
+
+static void printActualCParams(const char* filename, const char* dictFileName, int cLevel, const ZSTD_compressionParameters* cParams) {
+    unsigned long long fileSize = UTIL_getFileSize(filename);
+    const size_t dictSize = dictFileName != NULL ? (size_t)UTIL_getFileSize(dictFileName) : 0;
+    ZSTD_compressionParameters actualCParams = ZSTD_getCParams(cLevel, fileSize, dictSize);
+    assert(g_displayLevel >= 4);
+    actualCParams.windowLog = cParams->windowLog == 0 ? actualCParams.windowLog : cParams->windowLog;
+    actualCParams.chainLog = cParams->chainLog == 0 ? actualCParams.chainLog : cParams->chainLog;
+    actualCParams.hashLog = cParams->hashLog == 0 ? actualCParams.hashLog : cParams->hashLog;
+    actualCParams.searchLog = cParams->searchLog == 0 ? actualCParams.searchLog : cParams->searchLog;
+    actualCParams.minMatch = cParams->minMatch == 0 ? actualCParams.minMatch : cParams->minMatch;
+    actualCParams.targetLength = cParams->targetLength == 0 ? actualCParams.targetLength : cParams->targetLength;
+    actualCParams.strategy = cParams->strategy == 0 ? actualCParams.strategy : cParams->strategy;
+    DISPLAY("--zstd=wlog=%d,clog=%d,hlog=%d,slog=%d,mml=%d,tlen=%d,strat=%d\n",
+            actualCParams.windowLog, actualCParams.chainLog, actualCParams.hashLog, actualCParams.searchLog,
+            actualCParams.minMatch, actualCParams.targetLength, actualCParams.strategy);
+}
+
+#endif
+
 /* Environment variables for parameter setting */
 #define ENV_CLEVEL "ZSTD_CLEVEL"
 #define ENV_NBTHREADS "ZSTD_NBTHREADS"    /* takes lower precedence than directly specifying -T# in the CLI */
@@ -688,13 +786,13 @@ static unsigned init_nbThreads(void) {
     } else {                      \
         argNb++;                  \
         if (argNb >= argCount) {  \
-            DISPLAY("error: missing command argument \n"); \
+            DISPLAYLEVEL(1, "error: missing command argument \n"); \
             CLEAN_RETURN(1);      \
         }                         \
         ptr = argv[argNb];        \
         assert(ptr != NULL);      \
         if (ptr[0]=='-') {        \
-            DISPLAY("error: command cannot be separated from its argument by another command \n"); \
+            DISPLAYLEVEL(1, "error: command cannot be separated from its argument by another command \n"); \
             CLEAN_RETURN(1);      \
 }   }   }
 
@@ -702,12 +800,19 @@ static unsigned init_nbThreads(void) {
     const char* __nb;             \
     NEXT_FIELD(__nb);             \
     val32 = readU32FromChar(&__nb); \
+    if(*__nb != 0) {         \
+        errorOut("error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed"); \
+    }                             \
 }
 
-#define ZSTD_NB_STRATEGIES 9
-static const char* ZSTD_strategyMap[ZSTD_NB_STRATEGIES + 1] = { "", "ZSTD_fast",
-                "ZSTD_dfast", "ZSTD_greedy", "ZSTD_lazy", "ZSTD_lazy2", "ZSTD_btlazy2",
-                "ZSTD_btopt", "ZSTD_btultra", "ZSTD_btultra2"};
+#define NEXT_TSIZE(valTsize) {      \
+    const char* __nb;             \
+    NEXT_FIELD(__nb);             \
+    valTsize = readSizeTFromChar(&__nb); \
+    if(*__nb != 0) {         \
+        errorOut("error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed"); \
+    }                             \
+}
 
 typedef enum { zom_compress, zom_decompress, zom_test, zom_bench, zom_train, zom_list } zstd_operation_mode;
 
@@ -722,7 +827,7 @@ typedef enum { zom_compress, zom_decompress, zom_test, zom_bench, zom_train, zom
 # define MAXCLEVEL  ZSTD_maxCLevel()
 #endif
 
-int main(int const argCount, const char* argv[])
+int main(int argCount, const char* argv[])
 {
     int argNb,
         followLinks = 0,
@@ -732,9 +837,7 @@ int main(int const argCount, const char* argv[])
         hasStdout = 0,
         ldmFlag = 0,
         main_pause = 0,
-        nbWorkers = 0,
         adapt = 0,
-        useRowMatchFinder = 0,
         adaptMin = MINCLEVEL,
         adaptMax = MAXCLEVEL,
         rsyncable = 0,
@@ -743,15 +846,21 @@ int main(int const argCount, const char* argv[])
         separateFiles = 0,
         setRealTimePrio = 0,
         singleThread = 0,
+        defaultLogicalCores = 0,
         showDefaultCParams = 0,
         ultra=0,
-        contentSize=1;
+        contentSize=1,
+        removeSrcFile=0;
+    ZSTD_paramSwitch_e useRowMatchFinder = ZSTD_ps_auto;
+    FIO_compressionType_t cType = FIO_zstdCompression;
+    unsigned nbWorkers = 0;
     double compressibility = 0.5;
     unsigned bench_nbSeconds = 3;   /* would be better if this value was synchronized from bench */
     size_t blockSize = 0;
 
     FIO_prefs_t* const prefs = FIO_createPreferences();
     FIO_ctx_t* const fCtx = FIO_createContext();
+    FIO_progressSetting_e progress = FIO_ps_auto;
     zstd_operation_mode operation = zom_compress;
     ZSTD_compressionParameters compressionParams;
     int cLevel = init_cLevel();
@@ -772,6 +881,7 @@ int main(int const argCount, const char* argv[])
     size_t streamSrcSize = 0;
     size_t targetCBlockSize = 0;
     size_t srcSizeHint = 0;
+    size_t nbInputFileNames = 0;
     int dictCLevel = g_defaultDictCLevel;
     unsigned dictSelect = g_defaultSelectivityLevel;
 #ifndef ZSTD_NODICT
@@ -782,7 +892,7 @@ int main(int const argCount, const char* argv[])
 #ifndef ZSTD_NOBENCH
     BMK_advancedParams_t benchParams = BMK_initAdvancedParams();
 #endif
-    ZSTD_literalCompressionMode_e literalCompressionMode = ZSTD_lcm_auto;
+    ZSTD_paramSwitch_e literalCompressionMode = ZSTD_ps_auto;
 
 
     /* init */
@@ -790,7 +900,7 @@ int main(int const argCount, const char* argv[])
     (void)recursive; (void)cLevelLast;    /* not used when ZSTD_NOBENCH set */
     (void)memLimit;
     assert(argCount >= 1);
-    if ((filenames==NULL) || (file_of_names==NULL)) { DISPLAY("zstd: allocation error \n"); exit(1); }
+    if ((filenames==NULL) || (file_of_names==NULL)) { DISPLAYLEVEL(1, "zstd: allocation error \n"); exit(1); }
     programName = lastNameFromPath(programName);
 #ifdef ZSTD_MULTITHREAD
     nbWorkers = init_nbThreads();
@@ -799,17 +909,20 @@ int main(int const argCount, const char* argv[])
     /* preset behaviors */
     if (exeNameMatch(programName, ZSTD_ZSTDMT)) nbWorkers=0, singleThread=0;
     if (exeNameMatch(programName, ZSTD_UNZSTD)) operation=zom_decompress;
-    if (exeNameMatch(programName, ZSTD_CAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; outFileName=stdoutmark; g_displayLevel=1; }     /* supports multiple formats */
-    if (exeNameMatch(programName, ZSTD_ZCAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; outFileName=stdoutmark; g_displayLevel=1; }    /* behave like zcat, also supports multiple formats */
-    if (exeNameMatch(programName, ZSTD_GZ)) { suffix = GZ_EXTENSION; FIO_setCompressionType(prefs, FIO_gzipCompression); FIO_setRemoveSrcFile(prefs, 1); }        /* behave like gzip */
-    if (exeNameMatch(programName, ZSTD_GUNZIP)) { operation=zom_decompress; FIO_setRemoveSrcFile(prefs, 1); }                                                     /* behave like gunzip, also supports multiple formats */
-    if (exeNameMatch(programName, ZSTD_GZCAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; outFileName=stdoutmark; g_displayLevel=1; }   /* behave like gzcat, also supports multiple formats */
-    if (exeNameMatch(programName, ZSTD_LZMA)) { suffix = LZMA_EXTENSION; FIO_setCompressionType(prefs, FIO_lzmaCompression); FIO_setRemoveSrcFile(prefs, 1); }    /* behave like lzma */
-    if (exeNameMatch(programName, ZSTD_UNLZMA)) { operation=zom_decompress; FIO_setCompressionType(prefs, FIO_lzmaCompression); FIO_setRemoveSrcFile(prefs, 1); } /* behave like unlzma, also supports multiple formats */
-    if (exeNameMatch(programName, ZSTD_XZ)) { suffix = XZ_EXTENSION; FIO_setCompressionType(prefs, FIO_xzCompression); FIO_setRemoveSrcFile(prefs, 1); }          /* behave like xz */
-    if (exeNameMatch(programName, ZSTD_UNXZ)) { operation=zom_decompress; FIO_setCompressionType(prefs, FIO_xzCompression); FIO_setRemoveSrcFile(prefs, 1); }     /* behave like unxz, also supports multiple formats */
-    if (exeNameMatch(programName, ZSTD_LZ4)) { suffix = LZ4_EXTENSION; FIO_setCompressionType(prefs, FIO_lz4Compression); }                                       /* behave like lz4 */
-    if (exeNameMatch(programName, ZSTD_UNLZ4)) { operation=zom_decompress; FIO_setCompressionType(prefs, FIO_lz4Compression); }                                   /* behave like unlz4, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_CAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; FIO_setPassThroughFlag(prefs, 1); outFileName=stdoutmark; g_displayLevel=1; }     /* supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_ZCAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; FIO_setPassThroughFlag(prefs, 1); outFileName=stdoutmark; g_displayLevel=1; }    /* behave like zcat, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_GZ)) {   /* behave like gzip */
+        suffix = GZ_EXTENSION; cType = FIO_gzipCompression; removeSrcFile=1;
+        dictCLevel = cLevel = 6;  /* gzip default is -6 */
+    }
+    if (exeNameMatch(programName, ZSTD_GUNZIP)) { operation=zom_decompress; removeSrcFile=1; }                                                     /* behave like gunzip, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_GZCAT)) { operation=zom_decompress; FIO_overwriteMode(prefs); forceStdout=1; followLinks=1; FIO_setPassThroughFlag(prefs, 1); outFileName=stdoutmark; g_displayLevel=1; }   /* behave like gzcat, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_LZMA)) { suffix = LZMA_EXTENSION; cType = FIO_lzmaCompression; removeSrcFile=1; }    /* behave like lzma */
+    if (exeNameMatch(programName, ZSTD_UNLZMA)) { operation=zom_decompress; cType = FIO_lzmaCompression; removeSrcFile=1; } /* behave like unlzma, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_XZ)) { suffix = XZ_EXTENSION; cType = FIO_xzCompression; removeSrcFile=1; }          /* behave like xz */
+    if (exeNameMatch(programName, ZSTD_UNXZ)) { operation=zom_decompress; cType = FIO_xzCompression; removeSrcFile=1; }     /* behave like unxz, also supports multiple formats */
+    if (exeNameMatch(programName, ZSTD_LZ4)) { suffix = LZ4_EXTENSION; cType = FIO_lz4Compression; }                        /* behave like lz4 */
+    if (exeNameMatch(programName, ZSTD_UNLZ4)) { operation=zom_decompress; cType = FIO_lz4Compression; }                    /* behave like unlz4, also supports multiple formats */
     memset(&compressionParams, 0, sizeof(compressionParams));
 
     /* init crash handler */
@@ -846,43 +959,55 @@ int main(int const argCount, const char* argv[])
                 if (!strcmp(argument, "--help")) { usage_advanced(programName); CLEAN_RETURN(0); }
                 if (!strcmp(argument, "--verbose")) { g_displayLevel++; continue; }
                 if (!strcmp(argument, "--quiet")) { g_displayLevel--; continue; }
-                if (!strcmp(argument, "--stdout")) { forceStdout=1; outFileName=stdoutmark; g_displayLevel-=(g_displayLevel==2); continue; }
+                if (!strcmp(argument, "--stdout")) { forceStdout=1; outFileName=stdoutmark; removeSrcFile=0; continue; }
                 if (!strcmp(argument, "--ultra")) { ultra=1; continue; }
                 if (!strcmp(argument, "--check")) { FIO_setChecksumFlag(prefs, 2); continue; }
                 if (!strcmp(argument, "--no-check")) { FIO_setChecksumFlag(prefs, 0); continue; }
                 if (!strcmp(argument, "--sparse")) { FIO_setSparseWrite(prefs, 2); continue; }
                 if (!strcmp(argument, "--no-sparse")) { FIO_setSparseWrite(prefs, 0); continue; }
+                if (!strcmp(argument, "--pass-through")) { FIO_setPassThroughFlag(prefs, 1); continue; }
+                if (!strcmp(argument, "--no-pass-through")) { FIO_setPassThroughFlag(prefs, 0); continue; }
                 if (!strcmp(argument, "--test")) { operation=zom_test; continue; }
+                if (!strcmp(argument, "--asyncio")) { FIO_setAsyncIOFlag(prefs, 1); continue;}
+                if (!strcmp(argument, "--no-asyncio")) { FIO_setAsyncIOFlag(prefs, 0); continue;}
                 if (!strcmp(argument, "--train")) { operation=zom_train; if (outFileName==NULL) outFileName=g_defaultDictName; continue; }
                 if (!strcmp(argument, "--no-dictID")) { FIO_setDictIDFlag(prefs, 0); continue; }
-                if (!strcmp(argument, "--keep")) { FIO_setRemoveSrcFile(prefs, 0); continue; }
-                if (!strcmp(argument, "--rm")) { FIO_setRemoveSrcFile(prefs, 1); continue; }
+                if (!strcmp(argument, "--keep")) { removeSrcFile=0; continue; }
+                if (!strcmp(argument, "--rm")) { removeSrcFile=1; continue; }
                 if (!strcmp(argument, "--priority=rt")) { setRealTimePrio = 1; continue; }
                 if (!strcmp(argument, "--show-default-cparams")) { showDefaultCParams = 1; continue; }
                 if (!strcmp(argument, "--content-size")) { contentSize = 1; continue; }
                 if (!strcmp(argument, "--no-content-size")) { contentSize = 0; continue; }
                 if (!strcmp(argument, "--adapt")) { adapt = 1; continue; }
-                if (!strcmp(argument, "--no-row-match-finder")) { useRowMatchFinder = 1; continue; }
-                if (!strcmp(argument, "--row-match-finder")) { useRowMatchFinder = 2; continue; }
+                if (!strcmp(argument, "--no-row-match-finder")) { useRowMatchFinder = ZSTD_ps_disable; continue; }
+                if (!strcmp(argument, "--row-match-finder")) { useRowMatchFinder = ZSTD_ps_enable; continue; }
                 if (longCommandWArg(&argument, "--adapt=")) { adapt = 1; if (!parseAdaptParameters(argument, &adaptMin, &adaptMax)) { badusage(programName); CLEAN_RETURN(1); } continue; }
                 if (!strcmp(argument, "--single-thread")) { nbWorkers = 0; singleThread = 1; continue; }
-                if (!strcmp(argument, "--format=zstd")) { suffix = ZSTD_EXTENSION; FIO_setCompressionType(prefs, FIO_zstdCompression); continue; }
+                if (!strcmp(argument, "--format=zstd")) { suffix = ZSTD_EXTENSION; cType = FIO_zstdCompression; continue; }
 #ifdef ZSTD_GZCOMPRESS
-                if (!strcmp(argument, "--format=gzip")) { suffix = GZ_EXTENSION; FIO_setCompressionType(prefs, FIO_gzipCompression); continue; }
+                if (!strcmp(argument, "--format=gzip")) { suffix = GZ_EXTENSION; cType = FIO_gzipCompression; continue; }
+                if (exeNameMatch(programName, ZSTD_GZ)) {     /* behave like gzip */
+                    if (!strcmp(argument, "--best")) { dictCLevel = cLevel = 9; continue; }
+                    if (!strcmp(argument, "--no-name")) { /* ignore for now */; continue; }
+                }
 #endif
 #ifdef ZSTD_LZMACOMPRESS
-                if (!strcmp(argument, "--format=lzma")) { suffix = LZMA_EXTENSION; FIO_setCompressionType(prefs, FIO_lzmaCompression);  continue; }
-                if (!strcmp(argument, "--format=xz")) { suffix = XZ_EXTENSION; FIO_setCompressionType(prefs, FIO_xzCompression);  continue; }
+                if (!strcmp(argument, "--format=lzma")) { suffix = LZMA_EXTENSION; cType = FIO_lzmaCompression; continue; }
+                if (!strcmp(argument, "--format=xz")) { suffix = XZ_EXTENSION; cType = FIO_xzCompression; continue; }
 #endif
 #ifdef ZSTD_LZ4COMPRESS
-                if (!strcmp(argument, "--format=lz4")) { suffix = LZ4_EXTENSION; FIO_setCompressionType(prefs, FIO_lz4Compression);  continue; }
+                if (!strcmp(argument, "--format=lz4")) { suffix = LZ4_EXTENSION; cType = FIO_lz4Compression; continue; }
 #endif
                 if (!strcmp(argument, "--rsyncable")) { rsyncable = 1; continue; }
-                if (!strcmp(argument, "--compress-literals")) { literalCompressionMode = ZSTD_lcm_huffman; continue; }
-                if (!strcmp(argument, "--no-compress-literals")) { literalCompressionMode = ZSTD_lcm_uncompressed; continue; }
-                if (!strcmp(argument, "--no-progress")) { FIO_setProgressSetting(FIO_ps_never); continue; }
-                if (!strcmp(argument, "--progress")) { FIO_setProgressSetting(FIO_ps_always); continue; }
+                if (!strcmp(argument, "--compress-literals")) { literalCompressionMode = ZSTD_ps_enable; continue; }
+                if (!strcmp(argument, "--no-compress-literals")) { literalCompressionMode = ZSTD_ps_disable; continue; }
+                if (!strcmp(argument, "--no-progress")) { progress = FIO_ps_never; continue; }
+                if (!strcmp(argument, "--progress")) { progress = FIO_ps_always; continue; }
                 if (!strcmp(argument, "--exclude-compressed")) { FIO_setExcludeCompressedFile(prefs, 1); continue; }
+                if (!strcmp(argument, "--fake-stdin-is-console")) { UTIL_fakeStdinIsConsole(); continue; }
+                if (!strcmp(argument, "--fake-stdout-is-console")) { UTIL_fakeStdoutIsConsole(); continue; }
+                if (!strcmp(argument, "--fake-stderr-is-console")) { UTIL_fakeStderrIsConsole(); continue; }
+                if (!strcmp(argument, "--trace-file-stat")) { UTIL_traceFileStat(); continue; }
 
                 /* long commands with arguments */
 #ifndef ZSTD_NODICT
@@ -924,16 +1049,37 @@ int main(int const argCount, const char* argv[])
                 if (longCommandWArg(&argument, "--memlimit")) { NEXT_UINT32(memLimit); continue; }
                 if (longCommandWArg(&argument, "--memory")) { NEXT_UINT32(memLimit); continue; }
                 if (longCommandWArg(&argument, "--memlimit-decompress")) { NEXT_UINT32(memLimit); continue; }
-                if (longCommandWArg(&argument, "--block-size=")) { blockSize = readSizeTFromChar(&argument); continue; }
+                if (longCommandWArg(&argument, "--block-size")) { NEXT_TSIZE(blockSize); continue; }
                 if (longCommandWArg(&argument, "--maxdict")) { NEXT_UINT32(maxDictSize); continue; }
                 if (longCommandWArg(&argument, "--dictID")) { NEXT_UINT32(dictID); continue; }
-                if (longCommandWArg(&argument, "--zstd=")) { if (!parseCompressionParameters(argument, &compressionParams)) { badusage(programName); CLEAN_RETURN(1); } continue; }
-                if (longCommandWArg(&argument, "--stream-size=")) { streamSrcSize = readSizeTFromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--target-compressed-block-size=")) { targetCBlockSize = readSizeTFromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--size-hint=")) { srcSizeHint = readSizeTFromChar(&argument); continue; }
-                if (longCommandWArg(&argument, "--output-dir-flat")) { NEXT_FIELD(outDirName); continue; }
+                if (longCommandWArg(&argument, "--zstd=")) { if (!parseCompressionParameters(argument, &compressionParams)) { badusage(programName); CLEAN_RETURN(1); } ; cType = FIO_zstdCompression; continue; }
+                if (longCommandWArg(&argument, "--stream-size")) { NEXT_TSIZE(streamSrcSize); continue; }
+                if (longCommandWArg(&argument, "--target-compressed-block-size")) { NEXT_TSIZE(targetCBlockSize); continue; }
+                if (longCommandWArg(&argument, "--size-hint")) { NEXT_TSIZE(srcSizeHint); continue; }
+                if (longCommandWArg(&argument, "--output-dir-flat")) {
+                    NEXT_FIELD(outDirName);
+                    if (strlen(outDirName) == 0) {
+                        DISPLAYLEVEL(1, "error: output dir cannot be empty string (did you mean to pass '.' instead?)\n");
+                        CLEAN_RETURN(1);
+                    }
+                    continue;
+                }
+                if (longCommandWArg(&argument, "--auto-threads")) {
+                    const char* threadDefault = NULL;
+                    NEXT_FIELD(threadDefault);
+                    if (strcmp(threadDefault, "logical") == 0)
+                        defaultLogicalCores = 1;
+                    continue;
+                }
 #ifdef UTIL_HAS_MIRRORFILELIST
-                if (longCommandWArg(&argument, "--output-dir-mirror")) { NEXT_FIELD(outMirroredDirName); continue; }
+                if (longCommandWArg(&argument, "--output-dir-mirror")) {
+                    NEXT_FIELD(outMirroredDirName);
+                    if (strlen(outMirroredDirName) == 0) {
+                        DISPLAYLEVEL(1, "error: output dir cannot be empty string (did you mean to pass '.' instead?)\n");
+                        CLEAN_RETURN(1);
+                    }
+                    continue;
+                }
 #endif
 #ifndef ZSTD_NOTRACE
                 if (longCommandWArg(&argument, "--trace")) { char const* traceFile; NEXT_FIELD(traceFile); TRACE_enable(traceFile); continue; }
@@ -950,6 +1096,8 @@ int main(int const argCount, const char* argv[])
                         /* Invalid character following --long */
                         badusage(programName);
                         CLEAN_RETURN(1);
+                    } else {
+                        ldmWindowLog = g_defaultMaxWindowLog;
                     }
                     /* Only set windowLog if not already set by --zstd */
                     if (compressionParams.windowLog == 0)
@@ -1007,8 +1155,8 @@ int main(int const argCount, const char* argv[])
                 {
                     /* Display help */
                 case 'V': printVersion(); CLEAN_RETURN(0);   /* Version Only */
-                case 'H':
-                case 'h': usage_advanced(programName); CLEAN_RETURN(0);
+                case 'H': usage_advanced(programName); CLEAN_RETURN(0);
+                case 'h': usage(stdout, programName); CLEAN_RETURN(0);
 
                      /* Compress */
                 case 'z': operation=zom_compress; argument++; break;
@@ -1022,7 +1170,10 @@ int main(int const argCount, const char* argv[])
                         operation=zom_decompress; argument++; break;
 
                     /* Force stdout, even if stdout==console */
-                case 'c': forceStdout=1; outFileName=stdoutmark; argument++; break;
+                case 'c': forceStdout=1; outFileName=stdoutmark; removeSrcFile=0; argument++; break;
+
+                    /* do not store filename - gzip compatibility - nothing to do */
+                case 'n': argument++; break;
 
                     /* Use file content as dictionary */
                 case 'D': argument++; NEXT_FIELD(dictFileName); break;
@@ -1037,7 +1188,7 @@ int main(int const argCount, const char* argv[])
                 case 'q': g_displayLevel--; argument++; break;
 
                     /* keep source file (default) */
-                case 'k': FIO_setRemoveSrcFile(prefs, 0); argument++; break;
+                case 'k': removeSrcFile=0; argument++; break;
 
                     /* Checksum */
                 case 'C': FIO_setChecksumFlag(prefs, 2); argument++; break;
@@ -1096,7 +1247,7 @@ int main(int const argCount, const char* argv[])
                     /* nb of threads (hidden option) */
                 case 'T':
                     argument++;
-                    nbWorkers = (int)readU32FromChar(&argument);
+                    nbWorkers = readU32FromChar(&argument);
                     break;
 
                     /* Dictionary Selection level */
@@ -1136,17 +1287,26 @@ int main(int const argCount, const char* argv[])
     DISPLAYLEVEL(3, WELCOME_MESSAGE);
 
 #ifdef ZSTD_MULTITHREAD
+    if ((operation==zom_decompress) && (!singleThread) && (nbWorkers > 1)) {
+        DISPLAYLEVEL(2, "Warning : decompression does not support multi-threading\n");
+    }
     if ((nbWorkers==0) && (!singleThread)) {
         /* automatically set # workers based on # of reported cpus */
-        nbWorkers = UTIL_countPhysicalCores();
-        DISPLAYLEVEL(3, "Note: %d physical core(s) detected \n", nbWorkers);
+        if (defaultLogicalCores) {
+            nbWorkers = (unsigned)UTIL_countLogicalCores();
+            DISPLAYLEVEL(3, "Note: %d logical core(s) detected \n", nbWorkers);
+        } else {
+            nbWorkers = (unsigned)UTIL_countPhysicalCores();
+            DISPLAYLEVEL(3, "Note: %d physical core(s) detected \n", nbWorkers);
+        }
     }
 #else
-    (void)singleThread; (void)nbWorkers;
+    (void)singleThread; (void)nbWorkers; (void)defaultLogicalCores;
 #endif
 
-#ifdef UTIL_HAS_CREATEFILELIST
     g_utilDisplayLevel = g_displayLevel;
+
+#ifdef UTIL_HAS_CREATEFILELIST
     if (!followLinks) {
         unsigned u, fileNamesNb;
         unsigned const nbFilenames = (unsigned)filenames->tableSize;
@@ -1177,6 +1337,8 @@ int main(int const argCount, const char* argv[])
         }
     }
 
+    nbInputFileNames = filenames->tableSize; /* saving number of input files */
+
     if (recursive) {  /* at this stage, filenameTable is a list of paths, which can contain both files and directories */
         UTIL_expandFNT(&filenames, followLinks);
     }
@@ -1189,7 +1351,7 @@ int main(int const argCount, const char* argv[])
         int const ret = FIO_listMultipleFiles((unsigned)filenames->tableSize, filenames->fileNames, g_displayLevel);
         CLEAN_RETURN(ret);
 #else
-        DISPLAY("file information is not supported \n");
+        DISPLAYLEVEL(1, "file information is not supported \n");
         CLEAN_RETURN(1);
 #endif
     }
@@ -1197,14 +1359,18 @@ int main(int const argCount, const char* argv[])
     /* Check if benchmark is selected */
     if (operation==zom_bench) {
 #ifndef ZSTD_NOBENCH
+        if (cType != FIO_zstdCompression) {
+            DISPLAYLEVEL(1, "benchmark mode is only compatible with zstd format \n");
+            CLEAN_RETURN(1);
+        }
         benchParams.blockSize = blockSize;
-        benchParams.nbWorkers = nbWorkers;
+        benchParams.nbWorkers = (int)nbWorkers;
         benchParams.realTime = (unsigned)setRealTimePrio;
         benchParams.nbSeconds = bench_nbSeconds;
         benchParams.ldmFlag = ldmFlag;
         benchParams.ldmMinMatch = (int)g_ldmMinMatch;
         benchParams.ldmHashLog = (int)g_ldmHashLog;
-        benchParams.useRowMatchFinder = useRowMatchFinder;
+        benchParams.useRowMatchFinder = (int)useRowMatchFinder;
         if (g_ldmBucketSizeLog != LDM_PARAM_DEFAULT) {
             benchParams.ldmBucketSizeLog = (int)g_ldmBucketSizeLog;
         }
@@ -1225,15 +1391,18 @@ int main(int const argCount, const char* argv[])
                     int c;
                     DISPLAYLEVEL(3, "Benchmarking %s \n", filenames->fileNames[i]);
                     for(c = cLevel; c <= cLevelLast; c++) {
-                        BMK_benchFilesAdvanced(&filenames->fileNames[i], 1, dictFileName, c, &compressionParams, g_displayLevel, &benchParams);
+                        BMK_benchOutcome_t const bo = BMK_benchFilesAdvanced(&filenames->fileNames[i], 1, dictFileName, c, &compressionParams, g_displayLevel, &benchParams);
+                        if (!BMK_isSuccessful_benchOutcome(bo)) return 1;
                 }   }
             } else {
                 for(; cLevel <= cLevelLast; cLevel++) {
-                    BMK_benchFilesAdvanced(filenames->fileNames, (unsigned)filenames->tableSize, dictFileName, cLevel, &compressionParams, g_displayLevel, &benchParams);
+                    BMK_benchOutcome_t const bo = BMK_benchFilesAdvanced(filenames->fileNames, (unsigned)filenames->tableSize, dictFileName, cLevel, &compressionParams, g_displayLevel, &benchParams);
+                    if (!BMK_isSuccessful_benchOutcome(bo)) return 1;
             }   }
         } else {
             for(; cLevel <= cLevelLast; cLevel++) {
-                BMK_syntheticTest(cLevel, compressibility, &compressionParams, g_displayLevel, &benchParams);
+                BMK_benchOutcome_t const bo = BMK_syntheticTest(cLevel, compressibility, &compressionParams, g_displayLevel, &benchParams);
+                if (!BMK_isSuccessful_benchOutcome(bo)) return 1;
         }   }
 
 #else
@@ -1253,18 +1422,18 @@ int main(int const argCount, const char* argv[])
             int const optimize = !coverParams.k || !coverParams.d;
             coverParams.nbThreads = (unsigned)nbWorkers;
             coverParams.zParams = zParams;
-            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (unsigned)filenames->tableSize, blockSize, NULL, &coverParams, NULL, optimize);
+            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (int)filenames->tableSize, blockSize, NULL, &coverParams, NULL, optimize, memLimit);
         } else if (dict == fastCover) {
             int const optimize = !fastCoverParams.k || !fastCoverParams.d;
             fastCoverParams.nbThreads = (unsigned)nbWorkers;
             fastCoverParams.zParams = zParams;
-            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (unsigned)filenames->tableSize, blockSize, NULL, NULL, &fastCoverParams, optimize);
+            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (int)filenames->tableSize, blockSize, NULL, NULL, &fastCoverParams, optimize, memLimit);
         } else {
             ZDICT_legacy_params_t dictParams;
             memset(&dictParams, 0, sizeof(dictParams));
             dictParams.selectivityLevel = dictSelect;
             dictParams.zParams = zParams;
-            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (unsigned)filenames->tableSize, blockSize, &dictParams, NULL, NULL, 0);
+            operationResult = DiB_trainFromFiles(outFileName, maxDictSize, filenames->fileNames, (int)filenames->tableSize, blockSize, &dictParams, NULL, NULL, 0, memLimit);
         }
 #else
         (void)dictCLevel; (void)dictSelect; (void)dictID;  (void)maxDictSize; /* not used when ZSTD_NODICT set */
@@ -1275,24 +1444,34 @@ int main(int const argCount, const char* argv[])
     }
 
 #ifndef ZSTD_NODECOMPRESS
-    if (operation==zom_test) { FIO_setTestMode(prefs, 1); outFileName=nulmark; FIO_setRemoveSrcFile(prefs, 0); }  /* test mode */
+    if (operation==zom_test) { FIO_setTestMode(prefs, 1); outFileName=nulmark; removeSrcFile=0; }  /* test mode */
 #endif
 
     /* No input filename ==> use stdin and stdout */
-    if (filenames->tableSize == 0) UTIL_refFilename(filenames, stdinmark);
-    if (!strcmp(filenames->fileNames[0], stdinmark) && !outFileName)
+    if (filenames->tableSize == 0) {
+      /* It is possible that the input
+       was a number of empty directories. In this case
+       stdin and stdout should not be used */
+       if (nbInputFileNames > 0 ){
+        DISPLAYLEVEL(1, "please provide correct input file(s) or non-empty directories -- ignored \n");
+        CLEAN_RETURN(0);
+       }
+       UTIL_refFilename(filenames, stdinmark);
+    }
+
+    if (filenames->tableSize == 1 && !strcmp(filenames->fileNames[0], stdinmark) && !outFileName)
         outFileName = stdoutmark;  /* when input is stdin, default output is stdout */
 
     /* Check if input/output defined as console; trigger an error in this case */
     if (!forceStdin
-     && !strcmp(filenames->fileNames[0], stdinmark)
-     && IS_CONSOLE(stdin) ) {
+     && (UTIL_searchFileNamesTable(filenames, stdinmark) != -1)
+     && UTIL_isConsole(stdin) ) {
         DISPLAYLEVEL(1, "stdin is a console, aborting\n");
         CLEAN_RETURN(1);
     }
-    if ( outFileName && !strcmp(outFileName, stdoutmark)
-      && IS_CONSOLE(stdout)
-      && !strcmp(filenames->fileNames[0], stdinmark)
+    if ( (!outFileName || !strcmp(outFileName, stdoutmark))
+      && UTIL_isConsole(stdout)
+      && (UTIL_searchFileNamesTable(filenames, stdinmark) != -1)
       && !forceStdout
       && operation!=zom_decompress ) {
         DISPLAYLEVEL(1, "stdout is a console, aborting\n");
@@ -1310,26 +1489,36 @@ int main(int const argCount, const char* argv[])
 
     if (showDefaultCParams) {
         if (operation == zom_decompress) {
-            DISPLAY("error : can't use --show-default-cparams in decomrpession mode \n");
+            DISPLAYLEVEL(1, "error : can't use --show-default-cparams in decompression mode \n");
             CLEAN_RETURN(1);
         }
     }
 
     if (dictFileName != NULL && patchFromDictFileName != NULL) {
-        DISPLAY("error : can't use -D and --patch-from=# at the same time \n");
+        DISPLAYLEVEL(1, "error : can't use -D and --patch-from=# at the same time \n");
         CLEAN_RETURN(1);
     }
 
     if (patchFromDictFileName != NULL && filenames->tableSize > 1) {
-        DISPLAY("error : can't use --patch-from=# on multiple files \n");
+        DISPLAYLEVEL(1, "error : can't use --patch-from=# on multiple files \n");
         CLEAN_RETURN(1);
     }
 
-    /* No status message in pipe mode (stdin - stdout) */
+    /* No status message by default when output is stdout */
     hasStdout = outFileName && !strcmp(outFileName,stdoutmark);
-
     if (hasStdout && (g_displayLevel==2)) g_displayLevel=1;
 
+    /* when stderr is not the console, do not pollute it with progress updates (unless requested) */
+    if (!UTIL_isConsole(stderr) && (progress!=FIO_ps_always)) progress=FIO_ps_never;
+    FIO_setProgressSetting(progress);
+
+    /* don't remove source files when output is stdout */;
+    if (hasStdout && removeSrcFile) {
+        DISPLAYLEVEL(3, "Note: src files are not removed when output is stdout \n");
+        removeSrcFile = 0;
+    }
+    FIO_setRemoveSrcFile(prefs, removeSrcFile);
+
     /* IO Stream/File */
     FIO_setHasStdoutOutput(fCtx, hasStdout);
     FIO_setNbFilesTotal(fCtx, (int)filenames->tableSize);
@@ -1348,8 +1537,9 @@ int main(int const argCount, const char* argv[])
     FIO_setMemLimit(prefs, memLimit);
     if (operation==zom_compress) {
 #ifndef ZSTD_NOCOMPRESS
+        FIO_setCompressionType(prefs, cType);
         FIO_setContentSize(prefs, contentSize);
-        FIO_setNbWorkers(prefs, nbWorkers);
+        FIO_setNbWorkers(prefs, (int)nbWorkers);
         FIO_setBlockSize(prefs, (int)blockSize);
         if (g_overlapLog!=OVERLAP_LOG_DEFAULT) FIO_setOverlapLog(prefs, (int)g_overlapLog);
         FIO_setLdmFlag(prefs, (unsigned)ldmFlag);
@@ -1357,8 +1547,8 @@ int main(int const argCount, const char* argv[])
         FIO_setLdmMinMatch(prefs, (int)g_ldmMinMatch);
         if (g_ldmBucketSizeLog != LDM_PARAM_DEFAULT) FIO_setLdmBucketSizeLog(prefs, (int)g_ldmBucketSizeLog);
         if (g_ldmHashRateLog != LDM_PARAM_DEFAULT) FIO_setLdmHashRateLog(prefs, (int)g_ldmHashRateLog);
-        FIO_setAdaptiveMode(prefs, (unsigned)adapt);
-        FIO_setUseRowMatchFinder(prefs, useRowMatchFinder);
+        FIO_setAdaptiveMode(prefs, adapt);
+        FIO_setUseRowMatchFinder(prefs, (int)useRowMatchFinder);
         FIO_setAdaptMin(prefs, adaptMin);
         FIO_setAdaptMax(prefs, adaptMax);
         FIO_setRsyncable(prefs, rsyncable);
@@ -1366,6 +1556,7 @@ int main(int const argCount, const char* argv[])
         FIO_setTargetCBlockSize(prefs, targetCBlockSize);
         FIO_setSrcSizeHint(prefs, srcSizeHint);
         FIO_setLiteralCompressionMode(prefs, literalCompressionMode);
+        FIO_setSparseWrite(prefs, 0);
         if (adaptMin > cLevel) cLevel = adaptMin;
         if (adaptMax < cLevel) cLevel = adaptMax;
 
@@ -1374,32 +1565,29 @@ int main(int const argCount, const char* argv[])
           assert(ZSTD_NB_STRATEGIES == strategyBounds.upperBound);
           (void)strategyBounds; }
 
-        if (showDefaultCParams) {
+        if (showDefaultCParams || g_displayLevel >= 4) {
             size_t fileNb;
             for (fileNb = 0; fileNb < (size_t)filenames->tableSize; fileNb++) {
-                unsigned long long fileSize = UTIL_getFileSize(filenames->fileNames[fileNb]);
-                const size_t dictSize = dictFileName != NULL ? (size_t)UTIL_getFileSize(dictFileName) : 0;
-                const ZSTD_compressionParameters cParams = ZSTD_getCParams(cLevel, fileSize, dictSize);
-                if (fileSize != UTIL_FILESIZE_UNKNOWN) DISPLAY("%s (%u bytes)\n", filenames->fileNames[fileNb], (unsigned)fileSize);
-                else DISPLAY("%s (src size unknown)\n", filenames->fileNames[fileNb]);
-                DISPLAY(" - windowLog     : %u\n", cParams.windowLog);
-                DISPLAY(" - chainLog      : %u\n", cParams.chainLog);
-                DISPLAY(" - hashLog       : %u\n", cParams.hashLog);
-                DISPLAY(" - searchLog     : %u\n", cParams.searchLog);
-                DISPLAY(" - minMatch      : %u\n", cParams.minMatch);
-                DISPLAY(" - targetLength  : %u\n", cParams.targetLength);
-                assert(cParams.strategy < ZSTD_NB_STRATEGIES + 1);
-                DISPLAY(" - strategy      : %s (%u)\n", ZSTD_strategyMap[(int)cParams.strategy], (unsigned)cParams.strategy);
+                if (showDefaultCParams)
+                    printDefaultCParams(filenames->fileNames[fileNb], dictFileName, cLevel);
+                if (g_displayLevel >= 4)
+                    printActualCParams(filenames->fileNames[fileNb], dictFileName, cLevel, &compressionParams);
             }
         }
 
+        if (g_displayLevel >= 4)
+            FIO_displayCompressionParameters(prefs);
         if ((filenames->tableSize==1) && outFileName)
             operationResult = FIO_compressFilename(fCtx, prefs, outFileName, filenames->fileNames[0], dictFileName, cLevel, compressionParams);
         else
             operationResult = FIO_compressMultipleFilenames(fCtx, prefs, filenames->fileNames, outMirroredDirName, outDirName, outFileName, suffix, dictFileName, cLevel, compressionParams);
 #else
-        (void)contentSize; (void)suffix; (void)adapt; (void)rsyncable; (void)ultra; (void)cLevel; (void)ldmFlag; (void)literalCompressionMode; (void)targetCBlockSize; (void)streamSrcSize; (void)srcSizeHint; (void)ZSTD_strategyMap; (void)useRowMatchFinder; /* not used when ZSTD_NOCOMPRESS set */
-        DISPLAY("Compression not supported \n");
+        /* these variables are only used when compression mode is enabled */
+        (void)contentSize; (void)suffix; (void)adapt; (void)rsyncable;
+        (void)ultra; (void)cLevel; (void)ldmFlag; (void)literalCompressionMode;
+        (void)targetCBlockSize; (void)streamSrcSize; (void)srcSizeHint;
+        (void)ZSTD_strategyMap; (void)useRowMatchFinder; (void)cType;
+        DISPLAYLEVEL(1, "Compression not supported \n");
 #endif
     } else {  /* decompression or test */
 #ifndef ZSTD_NODECOMPRESS
@@ -1409,7 +1597,7 @@ int main(int const argCount, const char* argv[])
             operationResult = FIO_decompressMultipleFilenames(fCtx, prefs, filenames->fileNames, outMirroredDirName, outDirName, outFileName, dictFileName);
         }
 #else
-        DISPLAY("Decompression not supported \n");
+        DISPLAYLEVEL(1, "Decompression not supported \n");
 #endif
     }
 
diff --git a/priv/zstd/programs/zstdcli_trace.c b/priv/zstd/programs/zstdcli_trace.c
index b3b977f..35075a5 100644
--- a/priv/zstd/programs/zstdcli_trace.c
+++ b/priv/zstd/programs/zstdcli_trace.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/programs/zstdcli_trace.h b/priv/zstd/programs/zstdcli_trace.h
index 38c27dc..9c135d3 100644
--- a/priv/zstd/programs/zstdcli_trace.h
+++ b/priv/zstd/programs/zstdcli_trace.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/programs/zstdgrep.1 b/priv/zstd/programs/zstdgrep.1
index bf96185..77d29b4 100644
--- a/priv/zstd/programs/zstdgrep.1
+++ b/priv/zstd/programs/zstdgrep.1
@@ -1,23 +1,17 @@
-.
-.TH "ZSTDGREP" "1" "May 2021" "zstd 1.5.0" "User Commands"
-.
+.TH "ZSTDGREP" "1" "February 2023" "zstd 1.5.4" "User Commands"
 .SH "NAME"
 \fBzstdgrep\fR \- print lines matching a pattern in zstandard\-compressed files
-.
 .SH "SYNOPSIS"
-\fBzstdgrep\fR [\fIgrep\-flags\fR] [\-\-] \fIpattern\fR [\fIfiles\fR \.\.\.]
-.
+\fBzstdgrep\fR [\fIgrep\-flags\fR] [\-\-] \fIpattern\fR [\fIfiles\fR \|\.\|\.\|\.]
 .SH "DESCRIPTION"
-\fBzstdgrep\fR runs \fBgrep (1)\fR on files or stdin, if no files argument is given, after decompressing them with \fBzstdcat (1)\fR\.
-.
+\fBzstdgrep\fR runs \fBgrep\fR(1) on files, or \fBstdin\fR if no files argument is given, after decompressing them with \fBzstdcat\fR(1)\.
 .P
-The grep\-flags and pattern arguments are passed on to \fBgrep (1)\fR\. If an \fB\-e\fR flag is found in the \fBgrep\-flags\fR, \fBzstdgrep\fR will not look for a pattern argument\.
-.
+The \fIgrep\-flags\fR and \fIpattern\fR arguments are passed on to \fBgrep\fR(1)\. If an \fB\-e\fR flag is found in the \fIgrep\-flags\fR, \fBzstdgrep\fR will not look for a \fIpattern\fR argument\.
+.P
+Note that modern \fBgrep\fR alternatives such as \fBripgrep\fR (\fBrg\fR(1)) support \fBzstd\fR\-compressed files out of the box, and can prove better alternatives than \fBzstdgrep\fR notably for unsupported complex pattern searches\. Note though that such alternatives may also feature some minor command line differences\.
 .SH "EXIT STATUS"
 In case of missing arguments or missing pattern, 1 will be returned, otherwise 0\.
-.
 .SH "SEE ALSO"
-\fBzstd (1)\fR
-.
+\fBzstd\fR(1)
 .SH "AUTHORS"
 Thomas Klausner \fIwiz@NetBSD\.org\fR
diff --git a/priv/zstd/programs/zstdgrep.1.md b/priv/zstd/programs/zstdgrep.1.md
index 363ad4f..6370a81 100644
--- a/priv/zstd/programs/zstdgrep.1.md
+++ b/priv/zstd/programs/zstdgrep.1.md
@@ -4,14 +4,18 @@ zstdgrep(1) -- print lines matching a pattern in zstandard-compressed files
 SYNOPSIS
 --------
 
-`zstdgrep` [*grep-flags*] [--] _pattern_ [_files_ ...]
+`zstdgrep` [] [--]  [ ...]
 
 
 DESCRIPTION
 -----------
-`zstdgrep` runs `grep (1)` on files or stdin, if no files argument is given, after decompressing them with `zstdcat (1)`.
+`zstdgrep` runs `grep`(1) on files, or `stdin` if no files argument is given, after decompressing them with `zstdcat`(1).
 
-The grep-flags and pattern arguments are passed on to `grep (1)`.  If an `-e` flag is found in the `grep-flags`, `zstdgrep` will not look for a pattern argument.
+The  and  arguments are passed on to `grep`(1).  If an `-e` flag is found in the , `zstdgrep` will not look for a  argument.
+
+Note that modern `grep` alternatives such as `ripgrep` (`rg`(1)) support `zstd`-compressed files out of the box,
+and can prove better alternatives than `zstdgrep` notably for unsupported complex pattern searches.
+Note though that such alternatives may also feature some minor command line differences.
 
 EXIT STATUS
 -----------
@@ -19,7 +23,7 @@ In case of missing arguments or missing pattern, 1 will be returned, otherwise 0
 
 SEE ALSO
 --------
-`zstd (1)`
+`zstd`(1)
 
 AUTHORS
 -------
diff --git a/priv/zstd/programs/zstdless b/priv/zstd/programs/zstdless
index 893799e..17726a4 100755
--- a/priv/zstd/programs/zstdless
+++ b/priv/zstd/programs/zstdless
@@ -1,2 +1,8 @@
 #!/bin/sh
-zstdcat "$@" | less
+
+zstd=${ZSTD:-zstd}
+
+# TODO: Address quirks and bugs tied to old versions of less, provide a mechanism to pass flags directly to zstd
+
+export LESSOPEN="|-${zstd} -cdfq %s"
+exec less "$@"
diff --git a/priv/zstd/programs/zstdless.1 b/priv/zstd/programs/zstdless.1
index f08bc19..6c82b9c 100644
--- a/priv/zstd/programs/zstdless.1
+++ b/priv/zstd/programs/zstdless.1
@@ -1,14 +1,9 @@
-.
-.TH "ZSTDLESS" "1" "May 2021" "zstd 1.5.0" "User Commands"
-.
+.TH "ZSTDLESS" "1" "February 2023" "zstd 1.5.4" "User Commands"
 .SH "NAME"
 \fBzstdless\fR \- view zstandard\-compressed files
-.
 .SH "SYNOPSIS"
-\fBzstdless\fR [\fIflags\fR] [\fIfile\fR \.\.\.]
-.
+\fBzstdless\fR [\fIflags\fR] [\fIfile\fR \|\.\|\.\|\.]
 .SH "DESCRIPTION"
-\fBzstdless\fR runs \fBless (1)\fR on files or stdin, if no files argument is given, after decompressing them with \fBzstdcat (1)\fR\.
-.
+\fBzstdless\fR runs \fBless\fR(1) on files or stdin, if no \fIfile\fR argument is given, after decompressing them with \fBzstdcat\fR(1)\.
 .SH "SEE ALSO"
-\fBzstd (1)\fR
+\fBzstd\fR(1)
diff --git a/priv/zstd/programs/zstdless.1.md b/priv/zstd/programs/zstdless.1.md
index d91d48a..67c1c76 100644
--- a/priv/zstd/programs/zstdless.1.md
+++ b/priv/zstd/programs/zstdless.1.md
@@ -4,13 +4,13 @@ zstdless(1) -- view zstandard-compressed files
 SYNOPSIS
 --------
 
-`zstdless` [*flags*] [_file_ ...]
+`zstdless` [] [ ...]
 
 
 DESCRIPTION
 -----------
-`zstdless` runs `less (1)` on files or stdin, if no files argument is given, after decompressing them with `zstdcat (1)`.
+`zstdless` runs `less`(1) on files or stdin, if no  argument is given, after decompressing them with `zstdcat`(1).
 
 SEE ALSO
 --------
-`zstd (1)`
+`zstd`(1)
diff --git a/priv/zstd/tests/.gitignore b/priv/zstd/tests/.gitignore
index 9a6939a..fcb865d 100644
--- a/priv/zstd/tests/.gitignore
+++ b/priv/zstd/tests/.gitignore
@@ -12,6 +12,7 @@ zstreamtest
 zstreamtest32
 zstreamtest_asan
 zstreamtest_tsan
+zstreamtest_ubsan
 zstreamtest-dll
 datagen
 paramgrill
diff --git a/priv/zstd/tests/DEPRECATED-test-zstd-speed.py b/priv/zstd/tests/DEPRECATED-test-zstd-speed.py
index 665e0a7..71d75b8 100755
--- a/priv/zstd/tests/DEPRECATED-test-zstd-speed.py
+++ b/priv/zstd/tests/DEPRECATED-test-zstd-speed.py
@@ -2,7 +2,7 @@
 # THIS BENCHMARK IS BEING REPLACED BY automated-bencmarking.py
 
 # ################################################################
-# Copyright (c) Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/Makefile b/priv/zstd/tests/Makefile
index 8555300..3eed19e 100644
--- a/priv/zstd/tests/Makefile
+++ b/priv/zstd/tests/Makefile
@@ -1,5 +1,6 @@
+
 # ################################################################
-# Copyright (c) Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -19,46 +20,43 @@
 # zstreamtest32: Same as zstreamtest, but forced to compile in 32-bits mode
 # ##########################################################################
 
-ZSTDDIR = ../lib
+LIBZSTD = ../lib
+
+ZSTD_LEGACY_SUPPORT ?= 0
+
+DEBUGLEVEL ?= 2
+export DEBUGLEVEL  # transmit value to sub-makefiles
+
+include $(LIBZSTD)/libzstd.mk
+
+ZSTDDIR = $(LIBZSTD)
 PRGDIR  = ../programs
 PYTHON ?= python3
 TESTARTEFACT := versionsTest
 
-DEBUGLEVEL ?= 2
-export DEBUGLEVEL  # transmit value to sub-makefiles
-DEBUGFLAGS  = -g -DDEBUGLEVEL=$(DEBUGLEVEL)
+DEBUGFLAGS += -g -Wno-c++-compat
 CPPFLAGS   += -I$(ZSTDDIR) -I$(ZSTDDIR)/common -I$(ZSTDDIR)/compress \
               -I$(ZSTDDIR)/dictBuilder -I$(ZSTDDIR)/deprecated -I$(PRGDIR) \
-			  -DZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY=1
-ifeq ($(OS),Windows_NT)   # MinGW assumed
-CPPFLAGS   += -D__USE_MINGW_ANSI_STDIO   # compatibility with %zu formatting
-endif
-CFLAGS     ?= -O3
-CFLAGS     += -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow                 \
-              -Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
-              -Wstrict-prototypes -Wundef                                     \
-              -Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings      \
-              -Wredundant-decls -Wmissing-prototypes -Wno-deprecated-declarations
-CFLAGS     += $(DEBUGFLAGS)
-CPPFLAGS   += $(MOREFLAGS)
-
-
-ZSTDCOMMON_FILES := $(ZSTDDIR)/common/*.c
-ZSTDCOMP_FILES   := $(ZSTDDIR)/compress/*.c
-ZSTDDECOMP_FILES := $(ZSTDDIR)/decompress/*.c
+              -DZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY=1
+
+ZSTDCOMMON_FILES := $(sort $(ZSTD_COMMON_FILES))
+ZSTDCOMP_FILES   := $(sort $(ZSTD_COMPRESS_FILES))
+ZSTDDECOMP_FILES := $(sort $(ZSTD_DECOMPRESS_FILES))
 ZSTD_FILES  := $(ZSTDDECOMP_FILES) $(ZSTDCOMMON_FILES) $(ZSTDCOMP_FILES)
-ZDICT_FILES := $(ZSTDDIR)/dictBuilder/*.c
+ZDICT_FILES := $(sort $(ZSTD_DICTBUILDER_FILES))
 
-ZSTD_F1 := $(wildcard $(ZSTD_FILES))
+ZSTD_F1 := $(sort $(wildcard $(ZSTD_FILES)))
 ZSTD_OBJ1 := $(subst $(ZSTDDIR)/common/,zstdm_,$(ZSTD_F1))
 ZSTD_OBJ2 := $(subst $(ZSTDDIR)/compress/,zstdc_,$(ZSTD_OBJ1))
 ZSTD_OBJ3 := $(subst $(ZSTDDIR)/decompress/,zstdd_,$(ZSTD_OBJ2))
-ZSTD_OBJECTS := $(ZSTD_OBJ3:.c=.o)
+ZSTD_OBJ4 := $(ZSTD_OBJ3:.c=.o)
+ZSTD_OBJECTS := $(ZSTD_OBJ4:.S=.o)
 
 ZSTDMT_OBJ1 := $(subst $(ZSTDDIR)/common/,zstdmt_m_,$(ZSTD_F1))
 ZSTDMT_OBJ2 := $(subst $(ZSTDDIR)/compress/,zstdmt_c_,$(ZSTDMT_OBJ1))
 ZSTDMT_OBJ3 := $(subst $(ZSTDDIR)/decompress/,zstdmt_d_,$(ZSTDMT_OBJ2))
-ZSTDMT_OBJECTS := $(ZSTDMT_OBJ3:.c=.o)
+ZSTDMT_OBJ4 := $(ZSTDMT_OBJ3:.c=.o)
+ZSTDMT_OBJECTS := $(ZSTDMT_OBJ4:.S=.o)
 
 # Define *.exe as extension for Windows systems
 ifneq (,$(filter Windows%,$(OS)))
@@ -119,6 +117,9 @@ zstdc_%.o : $(ZSTDDIR)/compress/%.c
 zstdd_%.o : $(ZSTDDIR)/decompress/%.c
 	$(CC) -c $(CPPFLAGS) $(CFLAGS) $< -o $@
 
+zstdd_%.o : $(ZSTDDIR)/decompress/%.S
+	$(CC) -c $(CPPFLAGS) $(ASFLAGS) $< -o $@
+
 zstdmt%.o : CPPFLAGS += $(MULTITHREAD_CPP)
 
 zstdmt_m_%.o : $(ZSTDDIR)/common/%.c
@@ -130,14 +131,20 @@ zstdmt_c_%.o : $(ZSTDDIR)/compress/%.c
 zstdmt_d_%.o : $(ZSTDDIR)/decompress/%.c
 	$(CC) -c $(CPPFLAGS) $(CFLAGS) $< -o $@
 
+zstdmt_d_%.o : $(ZSTDDIR)/decompress/%.S
+	$(CC) -c $(CPPFLAGS) $(ASFLAGS) $< -o $@
+
+FULLBENCHS := fullbench fullbench32
+CLEAN += $(FULLBENCHS)
 fullbench32: CPPFLAGS += -m32
-fullbench fullbench32 : CPPFLAGS += $(MULTITHREAD_CPP) -Wno-deprecated-declarations
-fullbench fullbench32 : LDFLAGS += $(MULTITHREAD_LD)
-fullbench fullbench32 : DEBUGFLAGS = -DNDEBUG  # turn off assert() for speed measurements
-fullbench fullbench32 : $(ZSTD_FILES)
-fullbench fullbench32 : $(PRGDIR)/datagen.c $(PRGDIR)/util.c $(PRGDIR)/timefn.c $(PRGDIR)/benchfn.c fullbench.c
+$(FULLBENCHS) : CPPFLAGS += $(MULTITHREAD_CPP) -Wno-deprecated-declarations
+$(FULLBENCHS) : LDFLAGS += $(MULTITHREAD_LD)
+$(FULLBENCHS) : DEBUGFLAGS = -DNDEBUG  # turn off assert() for speed measurements
+$(FULLBENCHS) : $(ZSTD_FILES)
+$(FULLBENCHS) : $(PRGDIR)/datagen.c $(PRGDIR)/util.c $(PRGDIR)/timefn.c $(PRGDIR)/benchfn.c fullbench.c
 	$(LINK.c) $^ -o $@$(EXT)
 
+CLEAN += fullbench-lib
 fullbench-lib : CPPFLAGS += -DXXH_NAMESPACE=ZSTD_
 fullbench-lib : $(PRGDIR)/datagen.c $(PRGDIR)/util.c $(PRGDIR)/timefn.c $(PRGDIR)/benchfn.c $(ZSTDDIR)/libzstd.a fullbench.c
 	$(LINK.c) $^ -o $@$(EXT)
@@ -147,6 +154,7 @@ fullbench-dll: $(PRGDIR)/datagen.c $(PRGDIR)/util.c $(PRGDIR)/benchfn.c $(PRGDIR
 #	$(CC) $(FLAGS) $(filter %.c,$^) -o $@$(EXT) -DZSTD_DLL_IMPORT=1 $(ZSTDDIR)/dll/libzstd.dll
 	$(LINK.c) $^ $(LDLIBS) -o $@$(EXT)
 
+CLEAN += fuzzer fuzzer32
 fuzzer : CPPFLAGS += $(MULTITHREAD_CPP) -Wno-deprecated-declarations
 fuzzer : LDFLAGS += $(MULTITHREAD_LD)
 fuzzer : $(ZSTDMT_OBJECTS)
@@ -160,7 +168,8 @@ fuzzer32 : $(ZSTD_FILES)
 fuzzer-dll : $(ZSTDDIR)/common/xxhash.c $(PRGDIR)/util.c $(PRGDIR)/timefn.c $(PRGDIR)/datagen.c fuzzer.c
 	$(CC) $(CPPFLAGS) $(CFLAGS) $(filter %.c,$^) $(LDFLAGS) -o $@$(EXT)
 
-ZSTREAM_LOCAL_FILES := $(PRGDIR)/datagen.c $(PRGDIR)/util.c $(PRGDIR)/timefn.c seqgen.c zstreamtest.c
+CLEAN += zstreamtest zstreamtest32
+ZSTREAM_LOCAL_FILES := $(PRGDIR)/datagen.c $(PRGDIR)/util.c $(PRGDIR)/timefn.c seqgen.c zstreamtest.c external_matchfinder.c
 ZSTREAM_PROPER_FILES := $(ZDICT_FILES) $(ZSTREAM_LOCAL_FILES)
 ZSTREAMFILES := $(ZSTD_FILES) $(ZSTREAM_PROPER_FILES)
 zstreamtest32 : CFLAGS += -m32
@@ -171,42 +180,58 @@ zstreamtest32 : $(ZSTREAMFILES)
 zstreamtest zstreamtest32 :
 	$(LINK.c) $^ -o $@$(EXT)
 
+CLEAN += zstreamtest_asan
 zstreamtest_asan : CFLAGS += -fsanitize=address
 zstreamtest_asan : $(ZSTREAMFILES)
 	$(LINK.c) $(MULTITHREAD) $^ -o $@$(EXT)
 
+CLEAN += zstreamtest_tsan
 zstreamtest_tsan : CFLAGS += -fsanitize=thread
 zstreamtest_tsan : $(ZSTREAMFILES)
 	$(LINK.c) $(MULTITHREAD) $^ -o $@$(EXT)
 
+CLEAN += zstreamtest_ubsan
+zstreamtest_ubsan : CFLAGS += -fsanitize=undefined
+zstreamtest_ubsan : $(ZSTREAMFILES)
+	$(LINK.c) $(MULTITHREAD) $^ -o $@$(EXT)
+
 # note : broken : requires symbols unavailable from dynamic library
 zstreamtest-dll : $(ZSTDDIR)/common/xxhash.c  # xxh symbols not exposed from dll
 zstreamtest-dll : $(ZSTREAM_LOCAL_FILES)
 	$(CC) $(CPPFLAGS) $(CFLAGS) $(filter %.c,$^) $(LDFLAGS) -o $@$(EXT)
 
+CLEAN += paramgrill
 paramgrill : DEBUGFLAGS =   # turn off debug for speed measurements
 paramgrill : LDLIBS += -lm
 paramgrill : $(ZSTD_FILES) $(PRGDIR)/util.c $(PRGDIR)/timefn.c $(PRGDIR)/benchfn.c $(PRGDIR)/benchzstd.c $(PRGDIR)/datagen.c paramgrill.c
 
+CLEAN += datagen
 datagen : $(PRGDIR)/datagen.c datagencli.c
 	$(LINK.c) $^ -o $@$(EXT)
 
+CLEAN += roundTripCrash
 roundTripCrash: CFLAGS += $(MULTITHREAD)
 roundTripCrash : $(ZSTD_OBJECTS) roundTripCrash.c
 
+CLEAN += longmatch
 longmatch : $(ZSTD_OBJECTS) longmatch.c
 
+CLEAN += bigdict
 bigdict: CFLAGS += $(MULTITHREAD)
 bigdict: $(ZSTDMT_OBJECTS) $(PRGDIR)/datagen.c bigdict.c
 
+CLEAN += invalidDictionaries
 invalidDictionaries : $(ZSTD_OBJECTS) invalidDictionaries.c
 
-legacy : CPPFLAGS += -I$(ZSTDDIR)/legacy -DZSTD_LEGACY_SUPPORT=4
-legacy : $(ZSTD_FILES) $(wildcard $(ZSTDDIR)/legacy/*.c) legacy.c
+CLEAN += legacy
+legacy : CPPFLAGS += -I$(ZSTDDIR)/legacy -UZSTD_LEGACY_SUPPORT -DZSTD_LEGACY_SUPPORT=4
+legacy : $(ZSTD_FILES) $(sort $(wildcard $(ZSTDDIR)/legacy/*.c)) legacy.c
 
+CLEAN += decodecorpus
 decodecorpus : LDLIBS += -lm
 decodecorpus : $(filter-out zstdc_zstd_compress.o, $(ZSTD_OBJECTS)) $(ZDICT_FILES) $(PRGDIR)/util.c $(PRGDIR)/timefn.c decodecorpus.c
 
+CLEAN += poolTests
 poolTests : $(PRGDIR)/util.c $(PRGDIR)/timefn.c poolTests.c $(ZSTDDIR)/common/pool.c $(ZSTDDIR)/common/threading.c $(ZSTDDIR)/common/zstd_common.c $(ZSTDDIR)/common/error_private.c
 	$(LINK.c) $(MULTITHREAD) $^ -o $@$(EXT)
 
@@ -218,7 +243,8 @@ versionsTest: clean
 automated_benchmarking: clean
 	$(PYTHON) automated_benchmarking.py
 
-# make checkTag
+# make checkTag : check that release tag corresponds to release version
+CLEAN += checkTag
 checkTag.o : $(ZSTDDIR)/zstd.h
 
 .PHONY: clean
@@ -227,28 +253,22 @@ clean:
 	$(MAKE) -C $(PRGDIR) clean
 	$(RM) -fR $(TESTARTEFACT)
 	$(RM) -rf tmp*  # some test directories are named tmp*
-	$(RM) core *.o *.tmp result* *.gcda dictionary *.zst \
+	$(RM) $(CLEAN) core *.o *.tmp result* *.gcda dictionary *.zst \
         $(PRGDIR)/zstd$(EXT) $(PRGDIR)/zstd32$(EXT) \
-        fullbench$(EXT) fullbench32$(EXT) \
-        fullbench-lib$(EXT) fullbench-dll$(EXT) \
-        fuzzer$(EXT) fuzzer32$(EXT) \
-        fuzzer-dll$(EXT) zstreamtest-dll$(EXT) \
-        zstreamtest$(EXT) zstreamtest32$(EXT) \
-        datagen$(EXT) paramgrill$(EXT) roundTripCrash$(EXT) longmatch$(EXT) \
-        symbols$(EXT) invalidDictionaries$(EXT) legacy$(EXT) poolTests$(EXT) \
-        decodecorpus$(EXT) checkTag$(EXT) bigdict$(EXT)
+        fullbench-dll$(EXT) fuzzer-dll$(EXT) zstreamtest-dll$(EXT)
 	@echo Cleaning completed
 
 
 #----------------------------------------------------------------------------------
-# valgrind tests are validated only for some posix platforms
+# valgrind tests validated only for some posix platforms
 #----------------------------------------------------------------------------------
 UNAME := $(shell uname)
-ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS))
+ifneq (,$(filter $(UNAME),Linux Darwin GNU/kFreeBSD GNU OpenBSD FreeBSD NetBSD DragonFly SunOS AIX))
 HOST_OS = POSIX
 
-valgrindTest: VALGRIND = valgrind --leak-check=full --show-leak-kinds=all --error-exitcode=1
-valgrindTest: zstd datagen fuzzer fullbench
+.PHONY: test-valgrind
+test-valgrind: VALGRIND = valgrind --leak-check=full --show-leak-kinds=all --error-exitcode=1
+test-valgrind: zstd datagen fuzzer fullbench
 	@echo "\n ---- valgrind tests : memory analyzer ----"
 	$(VALGRIND) ./datagen -g50M > $(VOID)
 	$(VALGRIND) $(PRGDIR)/zstd ; if [ $$? -eq 0 ] ; then echo "zstd without argument should have failed"; false; fi
@@ -257,13 +277,12 @@ valgrindTest: zstd datagen fuzzer fullbench
 	./datagen -g2930KB | $(VALGRIND) $(PRGDIR)/zstd -5 -vf - -o tmp
 	$(VALGRIND) $(PRGDIR)/zstd -vdf tmp -c > $(VOID)
 	./datagen -g64MB | $(VALGRIND) $(PRGDIR)/zstd -vf - -c > $(VOID)
-	@rm tmp
+	$(RM) tmp
 	$(VALGRIND) ./fuzzer -T1mn -t1
 	$(VALGRIND) ./fullbench -i1
 
 endif
 
-
 ifneq (,$(filter MINGW% MSYS%,$(UNAME)))
   HOST_OS = MSYS
 endif
@@ -294,7 +313,7 @@ check: shortest
 fuzztest: test-fuzzer test-zstream test-decodecorpus
 
 .PHONY: test
-test: test-zstd test-fullbench test-fuzzer test-zstream test-invalidDictionaries test-legacy test-decodecorpus
+test: test-zstd test-fullbench test-fuzzer test-zstream test-invalidDictionaries test-legacy test-decodecorpus test-cli-tests
 ifeq ($(QEMU_SYS),)
 test: test-pool
 endif
@@ -303,9 +322,9 @@ endif
 test32: test-zstd32 test-fullbench32 test-fuzzer32 test-zstream32
 
 .PHONY: test-all
-test-all: test test32 valgrindTest test-decodecorpus-cli
+test-all: test test32 test-decodecorpus-cli
 
-.PHONY: test-zstd test-zstd32 test-zstd-nolegacy test-zstdgrep
+.PHONY: test-zstd test-zstd32 test-zstd-nolegacy
 test-zstd: ZSTD = $(PRGDIR)/zstd
 test-zstd: zstd
 
@@ -319,29 +338,42 @@ test-zstd test-zstd32 test-zstd-nolegacy: datagen
 	file $(ZSTD)
 	EXE_PREFIX="$(QEMU_SYS)" ZSTD_BIN="$(ZSTD)" DATAGEN_BIN=./datagen ./playTests.sh $(ZSTDRTTEST)
 
+.PHONY: test-cli-tests
+test-cli-tests: ZSTD = $(PRGDIR)/zstd
+test-cli-tests: zstd datagen
+	file $(ZSTD)
+	./cli-tests/run.py --exec-prefix="$(QEMU_SYS)" --zstd="$(ZSTD)" --datagen=./datagen
+
+.PHONY: test-fullbench
 test-fullbench: fullbench datagen
 	$(QEMU_SYS) ./fullbench -i1
 	$(QEMU_SYS) ./fullbench -i1 -P0
 
+.PHONY: test-fullbench32
 test-fullbench32: fullbench32 datagen
 	$(QEMU_SYS) ./fullbench32 -i1
 	$(QEMU_SYS) ./fullbench32 -i1 -P0
 
+.PHONY: test-fuzzer
 test-fuzzer: fuzzer
 	$(QEMU_SYS) ./fuzzer -v $(FUZZERTEST) $(FUZZER_FLAGS)
 
+# Note : this test presumes `fuzzer` will be built
+.PHONY: test-fuzzer-stackmode
 test-fuzzer-stackmode: MOREFLAGS += -DZSTD_HEAPMODE=0
 test-fuzzer-stackmode: test-fuzzer
 
+.PHONY: test-fuzzer32
 test-fuzzer32: fuzzer32
 	$(QEMU_SYS) ./fuzzer32 -v $(FUZZERTEST) $(FUZZER_FLAGS)
 
+.PHONY: test-zstream
 test-zstream: zstreamtest
 	$(QEMU_SYS) ./zstreamtest -v $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS)
 	$(QEMU_SYS) ./zstreamtest --newapi -t1 $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS)
 
 test-zstream32: zstreamtest32
-	$(QEMU_SYS) ./zstreamtest32 $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS)
+	$(QEMU_SYS) ./zstreamtest32 -v $(ZSTREAM_TESTTIME) $(FUZZER_FLAGS)
 
 test-longmatch: longmatch
 	$(QEMU_SYS) ./longmatch
diff --git a/priv/zstd/tests/README.md b/priv/zstd/tests/README.md
index cd255e9..2cf0e76 100644
--- a/priv/zstd/tests/README.md
+++ b/priv/zstd/tests/README.md
@@ -45,7 +45,7 @@ optional arguments:
   -h, --help            show this help message and exit
   --directory DIRECTORY
                         directory with files to benchmark
-  --levels LEVELS       levels to test eg ('1,2,3')
+  --levels LEVELS       levels to test e.g. ('1,2,3')
   --iterations ITERATIONS
                         number of benchmark iterations to run
   --emails EMAILS       email addresses of people who will be alerted upon
@@ -56,7 +56,7 @@ optional arguments:
   --mode MODE           'fastmode', 'onetime', 'current', or 'continuous' (see
                         README.md for details)
   --dict DICT           filename of dictionary to use (when set, this
-                        dictioanry will be used to compress the files provided
+                        dictionary will be used to compress the files provided
                         inside --directory)
 ```
 
@@ -70,7 +70,7 @@ After `sleepTime` (an optional parameter, default 300 seconds) seconds the scrip
 If a new commit is found it is compiled and a speed benchmark for this commit is performed.
 The results of the speed benchmark are compared to the previous results.
 If compression or decompression speed for one of zstd levels is lower than `lowerLimit` (an optional parameter, default 0.98) the speed benchmark is restarted.
-If second results are also lower than `lowerLimit` the warning e-mail is send to recipients from the list (the `emails` parameter).
+If second results are also lower than `lowerLimit` the warning e-mail is sent to recipients from the list (the `emails` parameter).
 
 Additional remarks:
 - To be sure that speed results are accurate the script should be run on a "stable" target system with no other jobs running in parallel
@@ -168,7 +168,7 @@ Full list of arguments
                     can use all --zstd parameter names and 'cParams' as a shorthand for all parameters used in ZSTD_compressionParameters
                     (Default: display all params available)
  -P#          : generated sample compressibility (when no file is provided)
- -t#          : Caps runtime of operation in seconds (default : 99999 seconds (about 27 hours ))
+ -t#          : Caps runtime of operation in seconds (default: 99999 seconds (about 27 hours))
  -v           : Prints Benchmarking output
  -D           : Next argument dictionary file
  -s           : Benchmark all files separately
diff --git a/priv/zstd/tests/automated_benchmarking.py b/priv/zstd/tests/automated_benchmarking.py
index 458bda4..153e7db 100644
--- a/priv/zstd/tests/automated_benchmarking.py
+++ b/priv/zstd/tests/automated_benchmarking.py
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -87,7 +87,7 @@ def clone_and_build(build):
             git clone {github_url} zstd-{user}-{sha} &&
             cd zstd-{user}-{sha} &&
             {checkout_command}
-            make &&
+            make -j &&
             cd ../
         """.format(
                 user=build["user"],
@@ -100,7 +100,7 @@ def clone_and_build(build):
         )
         return "zstd-{user}-{sha}/zstd".format(user=build["user"], sha=build["hash"])
     else:
-        os.system("cd ../ && make && cd tests")
+        os.system("cd ../ && make -j && cd tests")
         return "../zstd"
 
 
@@ -112,9 +112,9 @@ def parse_benchmark_output(output):
 def benchmark_single(executable, level, filename):
     return parse_benchmark_output((
         subprocess.run(
-            [executable, "-qb{}".format(level), filename], stderr=subprocess.PIPE
+            [executable, "-qb{}".format(level), filename], stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
         )
-        .stderr.decode("utf-8")
+        .stdout.decode("utf-8")
         .split(" ")
     ))
 
@@ -145,7 +145,7 @@ def benchmark(build, filenames, levels, iterations):
 def benchmark_dictionary_single(executable, filenames_directory, dictionary_filename, level, iterations):
     cspeeds, dspeeds = [], []
     for _ in range(iterations):
-        output = subprocess.run([executable, "-qb{}".format(level), "-D", dictionary_filename, "-r", filenames_directory], stderr=subprocess.PIPE).stderr.decode("utf-8").split(" ")
+        output = subprocess.run([executable, "-qb{}".format(level), "-D", dictionary_filename, "-r", filenames_directory], stdout=subprocess.PIPE).stdout.decode("utf-8").split(" ")
         cspeed, dspeed = parse_benchmark_output(output)
         cspeeds.append(cspeed)
         dspeeds.append(dspeed)
@@ -291,12 +291,12 @@ def main(filenames, levels, iterations, builds=None, emails=None, continuous=Fal
     parser = argparse.ArgumentParser()
 
     parser.add_argument("--directory", help="directory with files to benchmark", default="golden-compression")
-    parser.add_argument("--levels", help="levels to test eg ('1,2,3')", default="1")
+    parser.add_argument("--levels", help="levels to test e.g. ('1,2,3')", default="1")
     parser.add_argument("--iterations", help="number of benchmark iterations to run", default="1")
     parser.add_argument("--emails", help="email addresses of people who will be alerted upon regression. Only for continuous mode", default=None)
     parser.add_argument("--frequency", help="specifies the number of seconds to wait before each successive check for new PRs in continuous mode", default=DEFAULT_MAX_API_CALL_FREQUENCY_SEC)
     parser.add_argument("--mode", help="'fastmode', 'onetime', 'current', or 'continuous' (see README.md for details)", default="current")
-    parser.add_argument("--dict", help="filename of dictionary to use (when set, this dictioanry will be used to compress the files provided inside --directory)", default=None)
+    parser.add_argument("--dict", help="filename of dictionary to use (when set, this dictionary will be used to compress the files provided inside --directory)", default=None)
 
     args = parser.parse_args()
     filenames = args.directory
diff --git a/priv/zstd/tests/bigdict.c b/priv/zstd/tests/bigdict.c
index fb08925..ff2bb2d 100644
--- a/priv/zstd/tests/bigdict.c
+++ b/priv/zstd/tests/bigdict.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/checkTag.c b/priv/zstd/tests/checkTag.c
index f6c5e97..26871ed 100644
--- a/priv/zstd/tests/checkTag.c
+++ b/priv/zstd/tests/checkTag.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/check_size.py b/priv/zstd/tests/check_size.py
new file mode 100755
index 0000000..028b0a9
--- /dev/null
+++ b/priv/zstd/tests/check_size.py
@@ -0,0 +1,31 @@
+#!/usr/bin/env python3
+# ################################################################
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under both the BSD-style license (found in the
+# LICENSE file in the root directory of this source tree) and the GPLv2 (found
+# in the COPYING file in the root directory of this source tree).
+# You may select, at your option, one of the above-listed licenses.
+# ################################################################
+
+import os
+import subprocess
+import sys
+
+if len(sys.argv) != 3:
+	print(f"Usage: {sys.argv[0]} FILE SIZE_LIMIT")
+	sys.exit(1)
+
+file = sys.argv[1]
+limit = int(sys.argv[2])
+
+if not os.path.exists(file):
+	print(f"{file} does not exist")
+	sys.exit(1)
+
+size = os.path.getsize(file)
+
+if size > limit:
+	print(f"file {file} is {size} bytes, which is greater than the limit of {limit} bytes")
+	sys.exit(1)
diff --git a/priv/zstd/tests/cli-tests/.gitignore b/priv/zstd/tests/cli-tests/.gitignore
new file mode 100644
index 0000000..0ad01b2
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/.gitignore
@@ -0,0 +1,6 @@
+!bin/
+!datagen
+!zstdcat
+
+scratch/
+bin/symlinks
diff --git a/priv/zstd/tests/cli-tests/README.md b/priv/zstd/tests/cli-tests/README.md
new file mode 100644
index 0000000..7ca07c3
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/README.md
@@ -0,0 +1,258 @@
+# CLI tests
+
+The CLI tests are focused on testing the zstd CLI.
+They are intended to be simple tests that the CLI and arguments work as advertised.
+They are not intended to test the library, only the code in `programs/`.
+The library will get incidental coverage, but if you find yourself trying to trigger a specific condition in the library, this is the wrong tool.
+
+## Test runner usage
+
+The test runner `run.py` will run tests against the in-tree build of `zstd` and `datagen` by default. Which means that `zstd` and `datagen` must be built.
+
+The `zstd` binary used can be passed with `--zstd /path/to/zstd`.
+Additionally, to run `zstd` through a tool like `valgrind` or `qemu`, set the `--exec-prefix 'valgrind -q'` flag.
+
+Similarly, the `--datagen`, and `--zstdgrep` flags can be set to specify
+the paths to their respective binaries. However, these tools do not use
+the `EXEC_PREFIX`.
+
+Each test executes in its own scratch directory under `scratch/test/name`. E.g. `scratch/basic/help.sh/`. Normally these directories are removed after the test executes. However, the `--preserve` flag will preserve these directories after execution, and save the tests exit code, stdout, and stderr in the scratch directory to `exit`, `stderr`, and `stdout` respectively. This can be useful for debugging/editing a test and updating the expected output.
+
+### Running all the tests
+
+By default the test runner `run.py` will run all the tests, and report the results.
+
+Examples:
+
+```
+./run.py
+./run.py --preserve
+./run.py --zstd ../../build/programs/zstd --datagen ../../build/tests/datagen
+```
+
+### Running specific tests
+
+A set of test names can be passed to the test runner `run.py` to only execute those tests.
+This can be useful for writing or debugging a test, especially with `--preserve`.
+
+The test name can either be the path to the test file, or the test name, which is the path relative to the test directory.
+
+Examples:
+
+```
+./run.py basic/help.sh
+./run.py --preserve basic/help.sh basic/version.sh
+./run.py --preserve --verbose basic/help.sh
+```
+
+### Updating exact output
+
+If a test is failing because a `.stderr.exact` or `.stdout.exact` no longer matches, you can re-run the tests with `--set-exact-output` and the correct output will be written.
+
+Example:
+```
+./run.py --set-exact-output
+./run.py basic/help.sh --set-exact-output
+```
+
+## Writing a test
+
+Test cases are arbitrary executables, and can be written in any language, but are generally shell scripts.
+After the script executes, the exit code, stderr, and stdout are compared against the expectations.
+
+Each test is run in a clean directory that the test can use for intermediate files. This directory will be cleaned up at the end of the test, unless `--preserve` is passed to the test runner. Additionally, the `setup` script can prepare the directory before the test runs.
+
+### Calling zstd, utilities, and environment variables
+
+The `$PATH` for tests is prepended with the `bin/` sub-directory, which contains helper scripts for ease of testing.
+The `zstd` binary will call the zstd binary specified by `run.py` with the correct `$EXEC_PREFIX`.
+Similarly, `datagen`, `unzstd`, `zstdgrep`, `zstdcat`, etc, are provided.
+
+Helper utilities like `cmp_size`, `println`, and `die` are provided here too. See their scripts for details.
+
+Common shell script libraries are provided under `common/`, with helper variables and functions. They can be sourced with `source "$COMMON/library.sh`.
+
+Lastly, environment variables are provided for testing, which can be listed when calling `run.py` with `--verbose`.
+They are generally used by the helper scripts in `bin/` to coordinate everything.
+
+### Basic test case
+
+When executing your `$TEST` executable, by default the exit code is expected to be `0`. However, you can provide an alternate expected exit code in a `$TEST.exit` file.
+
+When executing your `$TEST` executable, by default the expected stderr and stdout are empty. However, you can override the default by providing one of three files:
+
+* `$TEST.{stdout,stderr}.exact`
+* `$TEST.{stdout,stderr}.glob`
+* `$TEST.{stdout,stderr}.ignore`
+
+If you provide a `.exact` file, the output is expected to exactly match, byte-for-byte.
+
+If you provide a `.glob` file, the output is expected to match the expected file, where each line is interpreted as a glob syntax. Additionally, a line containing only `...` matches all lines until the next expected line matches.
+
+If you provide a `.ignore` file, the output is ignored.
+
+#### Passing examples
+
+All these examples pass.
+
+Exit 1, and change the expectation to be 1.
+
+```
+exit-1.sh
+---
+#!/bin/sh
+exit 1
+---
+
+exit-1.sh.exit
+---
+1
+---
+```
+
+Check the stdout output exactly matches.
+
+```
+echo.sh
+---
+#!/bin/sh
+echo "hello world"
+---
+
+echo.sh.stdout.exact
+---
+hello world
+---
+```
+
+Check the stderr output using a glob.
+
+```
+random.sh
+---
+#!/bin/sh
+head -c 10 < /dev/urandom | xxd >&2
+---
+
+random.sh.stderr.glob
+---
+00000000: * * * * *                 *
+```
+
+Multiple lines can be matched with ...
+
+```
+random-num-lines.sh
+---
+#!/bin/sh
+echo hello
+seq 0 $RANDOM
+echo world
+---
+
+random-num-lines.sh.stdout.glob
+---
+hello
+0
+...
+world
+---
+```
+
+#### Failing examples
+
+Exit code is expected to be 0, but is 1.
+
+```
+exit-1.sh
+---
+#!/bin/sh
+exit 1
+---
+```
+
+Stdout is expected to be empty, but isn't.
+
+```
+echo.sh
+---
+#!/bin/sh
+echo hello world
+```
+
+Stderr is expected to be hello but is world.
+
+```
+hello.sh
+---
+#!/bin/sh
+echo world >&2
+---
+
+hello.sh.stderr.exact
+---
+hello
+---
+```
+
+### Setup & teardown scripts
+
+Finally, test writing can be eased with setup and teardown scripts.
+Each directory in the test directory is a test-suite consisting of all tests within that directory (but not sub-directories).
+This test suite can come with 4 scripts to help test writing:
+
+* `setup_once`
+* `teardown_once`
+* `setup`
+* `teardown`
+
+The `setup_once` and `teardown_once` are run once before and after all the tests in the suite respectively.
+They operate in the scratch directory for the test suite, which is the parent directory of each scratch directory for each test case.
+They can do work that is shared between tests to improve test efficiency.
+For example, the `dictionaries/setup_once` script builds several dictionaries, for use in the `dictionaries` tests.
+
+The `setup` and `teardown` scripts run before and after each test case respectively, in the test case's scratch directory.
+These scripts can do work that is shared between test cases to make tests more succinct.
+For example, the `dictionaries/setup` script copies the dictionaries built by the `dictionaries/setup_once` script into the test's scratch directory, to make them easier to use, and make sure they aren't accidentally modified.
+
+#### Examples
+
+```
+basic/setup
+---
+#!/bin/sh
+# Create some files for testing with
+datagen > file
+datagen > file0
+datagen > file1
+---
+
+basic/test.sh
+---
+#!/bin/sh
+zstd file file0 file1
+---
+
+dictionaries/setup_once
+---
+#!/bin/sh
+set -e
+
+mkdir files/ dicts/
+for i in $(seq 10); do
+	datagen -g1000 > files/$i
+done
+
+zstd --train -r files/ -o dicts/0
+---
+
+dictionaries/setup
+---
+#!/bin/sh
+
+# Runs in the test case's scratch directory.
+# The test suite's scratch directory that
+# `setup_once` operates in is the parent directory.
+cp -r ../files ../dicts .
+---
+```
diff --git a/priv/zstd/tests/cli-tests/basic/help.sh b/priv/zstd/tests/cli-tests/basic/help.sh
new file mode 100755
index 0000000..927c3ff
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/help.sh
@@ -0,0 +1,10 @@
+#!/bin/sh
+
+set -e
+
+println "+ zstd -h"
+zstd -h
+println "+ zstd -H"
+zstd -H
+println "+ zstd --help"
+zstd --help
diff --git a/priv/zstd/tests/cli-tests/basic/help.sh.stdout.glob b/priv/zstd/tests/cli-tests/basic/help.sh.stdout.glob
new file mode 100644
index 0000000..5580dc6
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/help.sh.stdout.glob
@@ -0,0 +1,34 @@
++ zstd -h
+Compress or decompress the INPUT file(s); reads from STDIN if INPUT is `-` or not provided.
+
+Usage: zstd *OPTIONS...* *INPUT... | -* *-o OUTPUT*
+
+Options:
+  -o OUTPUT                     Write output to a single file, OUTPUT.
+  -k, --keep                    Preserve INPUT file(s). *Default*
+  --rm                          Remove INPUT file(s) after successful (de)compression.
+
+  -#                            Desired compression level, where `#` is a number between 1 and 19;
+                                lower numbers provide faster compression, higher numbers yield
+                                better compression ratios. *Default: 3*
+
+  -d, --decompress              Perform decompression.
+  -D DICT                       Use DICT as the dictionary for compression or decompression.
+
+  -f, --force                   Disable input and output checks. Allows overwriting existing files,
+                                receiving input from the console, printing ouput to STDOUT, and
+                                operating on links, block devices, etc. Unrecognized formats will be
+                                passed-through through as-is.
+
+  -h                            Display short usage and exit.
+  -H, --help                    Display full help and exit.
+  -V, --version                 Display the program version and exit.
+
++ zstd -H
+...
+Advanced options:
+...
++ zstd --help
+...
+Advanced options:
+...
diff --git a/priv/zstd/tests/cli-tests/basic/memlimit.sh b/priv/zstd/tests/cli-tests/basic/memlimit.sh
new file mode 100755
index 0000000..88d734d
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/memlimit.sh
@@ -0,0 +1,40 @@
+#!/bin/sh
+
+echo "some data" > file
+
+println "+ zstd --memory=32LB file"
+zstd --memory=32LB file && die "Should not allow bogus suffix"
+println "+ zstd --memory=32LiB file"
+zstd --memory=32LiB file && die "Should not allow bogus suffix"
+println "+ zstd --memory=32A file"
+zstd --memory=32A file && die "Should not allow bogus suffix"
+println "+ zstd --memory=32r82347dn83 file"
+zstd --memory=32r82347dn83 file && die "Should not allow bogus suffix"
+println "+ zstd --memory=32asbdf file"
+zstd --memory=32asbdf file && die "Should not allow bogus suffix"
+println "+ zstd --memory=hello file"
+zstd --memory=hello file && die "Should not allow non-numeric parameter"
+println "+ zstd --memory=1 file"
+zstd -q --memory=1 file && die "Should allow numeric parameter without suffix"
+rm file.zst
+println "+ zstd --memory=1K file"
+zstd -q --memory=1K file && die "Should allow numeric parameter with expected suffix"
+rm file.zst
+println "+ zstd --memory=1KB file"
+zstd -q --memory=1KB file && die "Should allow numeric parameter with expected suffix"
+rm file.zst
+println "+ zstd --memory=1KiB file"
+zstd -q --memory=1KiB file && die "Should allow numeric parameter with expected suffix"
+rm file.zst
+println "+ zstd --memory=1M file"
+zstd -q --memory=1M file && die "Should allow numeric parameter with expected suffix"
+rm file.zst
+println "+ zstd --memory=1MB file"
+zstd -q --memory=1MB file && die "Should allow numeric parameter with expected suffix"
+rm file.zst
+println "+ zstd --memory=1MiB file"
+zstd -q --memory=1MiB file && die "Should allow numeric parameter with expected suffix"
+rm file.zst
+
+rm file
+exit 0
diff --git a/priv/zstd/tests/cli-tests/basic/memlimit.sh.stderr.exact b/priv/zstd/tests/cli-tests/basic/memlimit.sh.stderr.exact
new file mode 100644
index 0000000..3785b0f
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/memlimit.sh.stderr.exact
@@ -0,0 +1,13 @@
+error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed 
+error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed 
+error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed 
+error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed 
+error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed 
+error: only numeric values with optional suffixes K, KB, KiB, M, MB, MiB are allowed 
+Should allow numeric parameter without suffix
+Should allow numeric parameter with expected suffix
+Should allow numeric parameter with expected suffix
+Should allow numeric parameter with expected suffix
+Should allow numeric parameter with expected suffix
+Should allow numeric parameter with expected suffix
+Should allow numeric parameter with expected suffix
diff --git a/priv/zstd/tests/cli-tests/basic/memlimit.sh.stdout.exact b/priv/zstd/tests/cli-tests/basic/memlimit.sh.stdout.exact
new file mode 100644
index 0000000..1821648
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/memlimit.sh.stdout.exact
@@ -0,0 +1,13 @@
++ zstd --memory=32LB file
++ zstd --memory=32LiB file
++ zstd --memory=32A file
++ zstd --memory=32r82347dn83 file
++ zstd --memory=32asbdf file
++ zstd --memory=hello file
++ zstd --memory=1 file
++ zstd --memory=1K file
++ zstd --memory=1KB file
++ zstd --memory=1KiB file
++ zstd --memory=1M file
++ zstd --memory=1MB file
++ zstd --memory=1MiB file
diff --git a/priv/zstd/tests/cli-tests/basic/output_dir.sh b/priv/zstd/tests/cli-tests/basic/output_dir.sh
new file mode 100755
index 0000000..a8819d2
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/output_dir.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+println "+ zstd -r * --output-dir-mirror=\"\""
+zstd -r * --output-dir-mirror="" && die "Should not allow empty output dir!"
+println "+ zstd -r * --output-dir-flat=\"\""
+zstd -r * --output-dir-flat="" && die "Should not allow empty output dir!"
+exit 0
diff --git a/priv/zstd/tests/cli-tests/basic/output_dir.sh.stderr.exact b/priv/zstd/tests/cli-tests/basic/output_dir.sh.stderr.exact
new file mode 100644
index 0000000..e12b504
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/output_dir.sh.stderr.exact
@@ -0,0 +1,2 @@
+error: output dir cannot be empty string (did you mean to pass '.' instead?)
+error: output dir cannot be empty string (did you mean to pass '.' instead?)
diff --git a/priv/zstd/tests/cli-tests/basic/output_dir.sh.stdout.exact b/priv/zstd/tests/cli-tests/basic/output_dir.sh.stdout.exact
new file mode 100644
index 0000000..1e478cd
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/output_dir.sh.stdout.exact
@@ -0,0 +1,2 @@
++ zstd -r * --output-dir-mirror=""
++ zstd -r * --output-dir-flat=""
diff --git a/priv/zstd/tests/cli-tests/basic/version.sh b/priv/zstd/tests/cli-tests/basic/version.sh
new file mode 100755
index 0000000..f75eaa8
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/version.sh
@@ -0,0 +1,6 @@
+#!/bin/sh
+
+set -e
+
+zstd -V
+zstd --version
diff --git a/priv/zstd/tests/cli-tests/basic/version.sh.stdout.glob b/priv/zstd/tests/cli-tests/basic/version.sh.stdout.glob
new file mode 100644
index 0000000..4cc9fb9
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/basic/version.sh.stdout.glob
@@ -0,0 +1,2 @@
+*** Zstandard CLI (*-bit) v1.*.*, by Yann Collet ***
+*** Zstandard CLI (*-bit) v1.*.*, by Yann Collet ***
diff --git a/priv/zstd/tests/cli-tests/bin/cmp_size b/priv/zstd/tests/cli-tests/bin/cmp_size
new file mode 100755
index 0000000..8e4bef8
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/cmp_size
@@ -0,0 +1,44 @@
+#!/bin/sh
+
+set -e
+
+usage()
+{
+	printf "USAGE:\n\t$0 [-eq|-ne|-lt|-le|-gt|-ge] FILE1 FILE2\n"
+}
+
+help()
+{
+	printf "Small utility to compare file sizes without printing them with set -x.\n\n"
+	usage
+}
+
+case "$1" in
+	-h) help; exit 0  ;;
+	--help) help; exit 0 ;;
+esac
+
+if ! test -f $2; then
+	printf "FILE1='%b' is not a file\n\n" "$2"
+	usage
+	exit 1
+fi
+
+if ! test -f $3; then
+	printf "FILE2='%b' is not a file\n\n" "$3"
+	usage
+	exit 1
+fi
+
+
+size1=$(wc -c < $2)
+size2=$(wc -c < $3)
+
+case "$1" in
+	-eq) [ "$size1" -eq "$size2" ] ;;
+	-ne) [ "$size1" -ne "$size2" ] ;;
+	-lt) [ "$size1" -lt "$size2" ] ;;
+	-le) [ "$size1" -le "$size2" ] ;;
+	-gt) [ "$size1" -gt "$size2" ] ;;
+	-ge) [ "$size1" -ge "$size2" ] ;;
+esac
diff --git a/priv/zstd/tests/cli-tests/bin/datagen b/priv/zstd/tests/cli-tests/bin/datagen
new file mode 100755
index 0000000..8c60cbc
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/datagen
@@ -0,0 +1,3 @@
+#!/bin/sh
+
+"$DATAGEN_BIN" $@
diff --git a/priv/zstd/tests/cli-tests/bin/die b/priv/zstd/tests/cli-tests/bin/die
new file mode 100755
index 0000000..8633bc9
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/die
@@ -0,0 +1,4 @@
+#!/bin/sh
+
+println "${*}" 1>&2
+exit 1
diff --git a/priv/zstd/tests/cli-tests/bin/println b/priv/zstd/tests/cli-tests/bin/println
new file mode 100755
index 0000000..494eb18
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/println
@@ -0,0 +1,2 @@
+#!/bin/sh
+printf '%b\n' "${*}"
diff --git a/priv/zstd/tests/cli-tests/bin/unzstd b/priv/zstd/tests/cli-tests/bin/unzstd
new file mode 120000
index 0000000..613f917
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/unzstd
@@ -0,0 +1 @@
+zstd
\ No newline at end of file
diff --git a/priv/zstd/tests/cli-tests/bin/zstd b/priv/zstd/tests/cli-tests/bin/zstd
new file mode 100755
index 0000000..7a40aec
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/zstd
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+zstdname=$(basename $0)
+
+if [ -z "$EXEC_PREFIX" ]; then
+    "$ZSTD_SYMLINK_DIR/$zstdname" $@
+else
+    $EXEC_PREFIX "$ZSTD_SYMLINK_DIR/$zstdname" $@
+fi
diff --git a/priv/zstd/tests/cli-tests/bin/zstdcat b/priv/zstd/tests/cli-tests/bin/zstdcat
new file mode 120000
index 0000000..613f917
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/zstdcat
@@ -0,0 +1 @@
+zstd
\ No newline at end of file
diff --git a/priv/zstd/tests/cli-tests/bin/zstdgrep b/priv/zstd/tests/cli-tests/bin/zstdgrep
new file mode 100755
index 0000000..8821ebb
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/zstdgrep
@@ -0,0 +1,2 @@
+#!/bin/sh
+"$ZSTDGREP_BIN" $@
diff --git a/priv/zstd/tests/cli-tests/bin/zstdless b/priv/zstd/tests/cli-tests/bin/zstdless
new file mode 100755
index 0000000..d1d6f82
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/bin/zstdless
@@ -0,0 +1,2 @@
+#!/bin/sh
+"$ZSTDLESS_BIN" $@
diff --git a/priv/zstd/tests/cli-tests/cltools/setup b/priv/zstd/tests/cli-tests/cltools/setup
new file mode 100755
index 0000000..3009bd5
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/setup
@@ -0,0 +1,6 @@
+#!/bin/sh
+
+set -e
+
+echo "1234" > file
+zstd file
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh
new file mode 100755
index 0000000..6cd68b7
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh
@@ -0,0 +1,8 @@
+#!/bin/sh
+
+set -e
+
+println "+ good path"
+zstdgrep "1234" file file.zst
+println "+ bad path"
+zstdgrep "1234" bad.zst
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.exit b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.exit
new file mode 100644
index 0000000..56a6051
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.exit
@@ -0,0 +1 @@
+1
\ No newline at end of file
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stderr.exact b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stderr.exact
new file mode 100644
index 0000000..f147f28
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stderr.exact
@@ -0,0 +1 @@
+zstd: can't stat bad.zst : No such file or directory -- ignored 
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stdout.glob b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stdout.glob
new file mode 100644
index 0000000..96d4fa2
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdgrep.sh.stdout.glob
@@ -0,0 +1,4 @@
++ good path
+file:1234
+file.zst:1234
++ bad path
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdless.sh b/priv/zstd/tests/cli-tests/cltools/zstdless.sh
new file mode 100755
index 0000000..a0697bd
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdless.sh
@@ -0,0 +1,10 @@
+#!/bin/sh
+
+set -e
+
+println "+ good path"
+zstdless file.zst
+println "+ pass parameters"
+zstdless -N file.zst # This parameter does not produce line #s when piped, but still serves to test that the flag went to less and not zstd
+println "+ bad path"
+zstdless bad.zst >&2
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdless.sh.stderr.exact b/priv/zstd/tests/cli-tests/cltools/zstdless.sh.stderr.exact
new file mode 100644
index 0000000..5a726f1
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdless.sh.stderr.exact
@@ -0,0 +1,2 @@
+zstd: can't stat bad.zst : No such file or directory -- ignored 
+bad.zst: No such file or directory
diff --git a/priv/zstd/tests/cli-tests/cltools/zstdless.sh.stdout.glob b/priv/zstd/tests/cli-tests/cltools/zstdless.sh.stdout.glob
new file mode 100644
index 0000000..2784ddd
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/cltools/zstdless.sh.stdout.glob
@@ -0,0 +1,5 @@
++ good path
+1234
++ pass parameters
+1234
++ bad path
diff --git a/priv/zstd/tests/cli-tests/common/format.sh b/priv/zstd/tests/cli-tests/common/format.sh
new file mode 100644
index 0000000..e574e97
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/common/format.sh
@@ -0,0 +1,19 @@
+#!/bin/sh
+
+. "$COMMON/platform.sh"
+
+zstd_supports_format()
+{
+	zstd -h | grep > $INTOVOID -- "--format=$1"
+}
+
+format_extension()
+{
+	if [ "$1" = "zstd" ]; then
+		printf "zst"
+	elif [ "$1" = "gzip" ]; then
+		printf "gz"
+	else
+		printf "$1"
+	fi
+}
diff --git a/priv/zstd/tests/cli-tests/common/mtime.sh b/priv/zstd/tests/cli-tests/common/mtime.sh
new file mode 100644
index 0000000..344074d
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/common/mtime.sh
@@ -0,0 +1,13 @@
+. "$COMMON/platform.sh"
+
+MTIME="stat -c %Y"
+case "$UNAME" in
+    Darwin | FreeBSD | OpenBSD | NetBSD) MTIME="stat -f %m" ;;
+esac
+
+assertSameMTime() {
+    MT1=$($MTIME "$1")
+    MT2=$($MTIME "$2")
+    echo MTIME $MT1 $MT2
+    [ "$MT1" = "$MT2" ] || die "mtime on $1 doesn't match mtime on $2 ($MT1 != $MT2)"
+}
diff --git a/priv/zstd/tests/cli-tests/common/permissions.sh b/priv/zstd/tests/cli-tests/common/permissions.sh
new file mode 100644
index 0000000..6bce1f0
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/common/permissions.sh
@@ -0,0 +1,18 @@
+. "$COMMON/platform.sh"
+
+GET_PERMS="stat -c %a"
+case "$UNAME" in
+    Darwin | FreeBSD | OpenBSD | NetBSD) GET_PERMS="stat -f %Lp" ;;
+esac
+
+assertFilePermissions() {
+    STAT1=$($GET_PERMS "$1")
+    STAT2=$2
+    [ "$STAT1" = "$STAT2" ] || die "permissions on $1 don't match expected ($STAT1 != $STAT2)"
+}
+
+assertSamePermissions() {
+    STAT1=$($GET_PERMS "$1")
+    STAT2=$($GET_PERMS "$2")
+    [ "$STAT1" = "$STAT2" ] || die "permissions on $1 don't match those on $2 ($STAT1 != $STAT2)"
+}
diff --git a/priv/zstd/tests/cli-tests/common/platform.sh b/priv/zstd/tests/cli-tests/common/platform.sh
new file mode 100644
index 0000000..6eb45ea
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/common/platform.sh
@@ -0,0 +1,37 @@
+#!/bin/sh
+
+UNAME=$(uname)
+
+isWindows=false
+INTOVOID="/dev/null"
+case "$UNAME" in
+  GNU) DEVDEVICE="/dev/random" ;;
+  *) DEVDEVICE="/dev/zero" ;;
+esac
+case "$OS" in
+  Windows*)
+    isWindows=true
+    INTOVOID="NUL"
+    DEVDEVICE="NUL"
+    ;;
+esac
+
+case "$UNAME" in
+  Darwin) MD5SUM="md5 -r" ;;
+  FreeBSD) MD5SUM="gmd5sum" ;;
+  NetBSD) MD5SUM="md5 -n" ;;
+  OpenBSD) MD5SUM="md5" ;;
+  *) MD5SUM="md5sum" ;;
+esac
+
+DIFF="diff"
+case "$UNAME" in
+  SunOS) DIFF="gdiff" ;;
+esac
+
+if echo hello | zstd -v -T2 2>&1 > $INTOVOID | grep -q 'multi-threading is disabled'
+then
+    hasMT=""
+else
+    hasMT="true"
+fi
diff --git a/priv/zstd/tests/cli-tests/compression/adapt.sh b/priv/zstd/tests/cli-tests/compression/adapt.sh
new file mode 100755
index 0000000..30b9afa
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/adapt.sh
@@ -0,0 +1,14 @@
+#!/bin/sh
+
+set -e
+
+# Test --adapt
+zstd -f file --adapt -c | zstd -t
+
+datagen -g100M > file100M
+
+# Pick parameters to force fast adaptation, even on slow systems
+zstd --adapt -vvvv -19 --zstd=wlog=10 file100M -o /dev/null 2>&1 | grep -q "faster speed , lighter compression"
+
+# Adaption still happens with --no-progress
+zstd --no-progress --adapt -vvvv -19 --zstd=wlog=10 file100M -o /dev/null 2>&1 | grep -q "faster speed , lighter compression"
diff --git a/priv/zstd/tests/cli-tests/compression/basic.sh b/priv/zstd/tests/cli-tests/compression/basic.sh
new file mode 100755
index 0000000..b6e2aa0
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/basic.sh
@@ -0,0 +1,34 @@
+#!/bin/sh
+
+set -e
+
+# Uncomment the set -v line for debugging
+# set -v
+
+# Test compression flags and check that they work
+zstd file                 ; zstd -t file.zst
+zstd -f file              ; zstd -t file.zst
+zstd -f -z file           ; zstd -t file.zst
+zstd -f -k file           ; zstd -t file.zst
+zstd -f -C file           ; zstd -t file.zst
+zstd -f --check file      ; zstd -t file.zst
+zstd -f --no-check file   ; zstd -t file.zst
+zstd -f -- file           ; zstd -t file.zst
+
+# Test output file compression
+zstd -o file-out.zst ; zstd -t file-out.zst
+zstd -fo file-out.zst; zstd -t file-out.zst
+
+# Test compression to stdout
+zstd -c file       | zstd -t
+zstd --stdout file | zstd -t
+println bob | zstd | zstd -t
+
+# Test keeping input file when compressing to stdout in gzip mode
+$ZSTD_SYMLINK_DIR/gzip -c file       | zstd -t ; test -f file
+$ZSTD_SYMLINK_DIR/gzip --stdout file | zstd -t ; test -f file
+
+# Test --rm
+cp file file-rm
+zstd --rm file-rm; zstd -t file-rm.zst
+test ! -f file-rm
diff --git a/priv/zstd/tests/cli-tests/compression/compress-literals.sh b/priv/zstd/tests/cli-tests/compression/compress-literals.sh
new file mode 100755
index 0000000..573481a
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/compress-literals.sh
@@ -0,0 +1,10 @@
+#!/bin/sh
+
+set -e
+
+# Test --[no-]compress-literals
+zstd file --no-compress-literals -1 -c       | zstd -t
+zstd file --no-compress-literals -19 -c      | zstd -t
+zstd file --no-compress-literals --fast=1 -c | zstd -t
+zstd file --compress-literals -1 -c          | zstd -t
+zstd file --compress-literals --fast=1 -c    | zstd -t
diff --git a/priv/zstd/tests/cli-tests/compression/format.sh b/priv/zstd/tests/cli-tests/compression/format.sh
new file mode 100755
index 0000000..192fa2c
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/format.sh
@@ -0,0 +1,16 @@
+#!/bin/sh
+
+. "$COMMON/format.sh"
+
+set -e
+
+# Test --format
+zstd --format=zstd file -f
+zstd -t file.zst
+for format in "gzip" "lz4" "xz" "lzma"; do
+	if zstd_supports_format $format; then
+		zstd --format=$format file
+		zstd -t file.$(format_extension $format)
+		zstd -c --format=$format file | zstd -t --format=$format
+	fi
+done
diff --git a/priv/zstd/tests/cli-tests/compression/golden.sh b/priv/zstd/tests/cli-tests/compression/golden.sh
new file mode 100755
index 0000000..85dd3fd
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/golden.sh
@@ -0,0 +1,12 @@
+#!/bin/sh
+
+set -e
+
+GOLDEN_DIR="$ZSTD_REPO_DIR/tests/golden-compression/"
+cp -r "$GOLDEN_DIR" golden/
+
+zstd -rf golden/ --output-dir-mirror golden-compressed/
+zstd -r -t golden-compressed/
+
+zstd --target-compressed-block-size=1024 -rf golden/ --output-dir-mirror golden-compressed/
+zstd -r -t golden-compressed/
diff --git a/priv/zstd/tests/cli-tests/compression/gzip-compat.sh b/priv/zstd/tests/cli-tests/compression/gzip-compat.sh
new file mode 100755
index 0000000..bb72e05
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/gzip-compat.sh
@@ -0,0 +1,15 @@
+#!/bin/sh
+
+set -e
+
+# Uncomment the set -v line for debugging
+# set -v
+
+# Test gzip specific compression option
+$ZSTD_SYMLINK_DIR/gzip --fast file ; $ZSTD_SYMLINK_DIR/gzip -d file.gz
+$ZSTD_SYMLINK_DIR/gzip --best file ; $ZSTD_SYMLINK_DIR/gzip -d file.gz
+
+# Test -n / --no-name: do not embed original filename in archive
+$ZSTD_SYMLINK_DIR/gzip -n file           ; grep -qv file file.gz  ; $ZSTD_SYMLINK_DIR/gzip -d file.gz
+$ZSTD_SYMLINK_DIR/gzip --no-name file    ; grep -qv file file.gz  ; $ZSTD_SYMLINK_DIR/gzip -d file.gz
+$ZSTD_SYMLINK_DIR/gzip -c --no-name file | grep -qv file
diff --git a/priv/zstd/tests/cli-tests/compression/levels.sh b/priv/zstd/tests/cli-tests/compression/levels.sh
new file mode 100755
index 0000000..cc2700a
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/levels.sh
@@ -0,0 +1,62 @@
+#!/bin/sh
+
+set -e
+set -v
+
+datagen > file
+
+# Compress with various levels and ensure that their sizes are ordered
+zstd --fast=10 file -o file-f10.zst -q
+zstd --fast=1 file -o file-f1.zst -q
+zstd -1 file -o file-1.zst -q
+zstd -19 file -o file-19.zst -q
+
+zstd -t file-f10.zst file-f1.zst file-1.zst file-19.zst
+
+cmp_size -lt file-19.zst file-1.zst
+cmp_size -lt file-1.zst file-f1.zst
+cmp_size -lt file-f1.zst file-f10.zst
+
+# Test default levels
+zstd --fast file -f -q
+cmp file.zst file-f1.zst || die "--fast is not level -1"
+
+zstd -0 file -o file-0.zst -q
+zstd -f file -q
+cmp file.zst file-0.zst || die "Level 0 is not the default level"
+
+# Test level clamping
+zstd -99 file -o file-99.zst -q
+cmp file-19.zst file-99.zst || die "Level 99 is clamped to 19"
+zstd --fast=200000 file -c | zstd -t
+
+zstd -5000000000 -f file       && die "Level too large, must fail" ||:
+zstd --fast=5000000000 -f file && die "Level too large, must fail" ||:
+
+# Test setting a level through the environment variable
+ZSTD_CLEVEL=-10 zstd file -o file-f10-env.zst -q
+ZSTD_CLEVEL=1 zstd file -o file-1-env.zst -q
+ZSTD_CLEVEL=+19 zstd file -o file-19-env.zst -q
+ZSTD_CLEVEL=+99 zstd file -o file-99-env.zst -q
+
+cmp file-f10.zst file-f10-env.zst || die "Environment variable failed to set level"
+cmp file-1.zst file-1-env.zst || die "Environment variable failed to set level"
+cmp file-19.zst file-19-env.zst || die "Environment variable failed to set level"
+cmp file-99.zst file-99-env.zst || die "Environment variable failed to set level"
+
+# Test invalid environment clevel is the default level
+zstd -f file -q
+ZSTD_CLEVEL=- zstd -f file -o file-env.zst -q      ; cmp file.zst file-env.zst
+ZSTD_CLEVEL=+ zstd -f file -o file-env.zst -q      ; cmp file.zst file-env.zst
+ZSTD_CLEVEL=a zstd -f file -o file-env.zst -q      ; cmp file.zst file-env.zst
+ZSTD_CLEVEL=-a zstd -f file -o file-env.zst -q     ; cmp file.zst file-env.zst
+ZSTD_CLEVEL=+a zstd -f file -o file-env.zst -q     ; cmp file.zst file-env.zst
+ZSTD_CLEVEL=3a7 zstd -f file -o file-env.zst -q    ; cmp file.zst file-env.zst
+ZSTD_CLEVEL=5000000000 zstd -f file -o file-env.zst -q ; cmp file.zst file-env.zst
+
+# Test environment clevel is overridden by command line
+ZSTD_CLEVEL=10 zstd -f file -1 -o file-1-env.zst -q
+ZSTD_CLEVEL=10 zstd -f file --fast=1 -o file-f1-env.zst -q
+
+cmp file-1.zst file-1-env.zst  || die "Environment variable not overridden"
+cmp file-f1.zst file-f1-env.zst || die "Environment variable not overridden"
diff --git a/priv/zstd/tests/cli-tests/compression/levels.sh.stderr.exact b/priv/zstd/tests/cli-tests/compression/levels.sh.stderr.exact
new file mode 100644
index 0000000..c8fb79c
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/levels.sh.stderr.exact
@@ -0,0 +1,69 @@
+
+datagen > file
+
+# Compress with various levels and ensure that their sizes are ordered
+zstd --fast=10 file -o file-f10.zst -q
+zstd --fast=1 file -o file-f1.zst -q
+zstd -1 file -o file-1.zst -q
+zstd -19 file -o file-19.zst -q
+
+zstd -t file-f10.zst file-f1.zst file-1.zst file-19.zst
+4 files decompressed : 262148 bytes total 
+
+cmp_size -lt file-19.zst file-1.zst
+cmp_size -lt file-1.zst file-f1.zst
+cmp_size -lt file-f1.zst file-f10.zst
+
+# Test default levels
+zstd --fast file -f -q
+cmp file.zst file-f1.zst || die "--fast is not level -1"
+
+zstd -0 file -o file-0.zst -q
+zstd -f file -q
+cmp file.zst file-0.zst || die "Level 0 is not the default level"
+
+# Test level clamping
+zstd -99 file -o file-99.zst -q
+cmp file-19.zst file-99.zst || die "Level 99 is clamped to 19"
+zstd --fast=200000 file -c | zstd -t
+/*stdin*\           : 65537 bytes 
+
+zstd -5000000000 -f file       && die "Level too large, must fail" ||:
+error: numeric value overflows 32-bit unsigned int 
+zstd --fast=5000000000 -f file && die "Level too large, must fail" ||:
+error: numeric value overflows 32-bit unsigned int 
+
+# Test setting a level through the environment variable
+ZSTD_CLEVEL=-10 zstd file -o file-f10-env.zst -q
+ZSTD_CLEVEL=1 zstd file -o file-1-env.zst -q
+ZSTD_CLEVEL=+19 zstd file -o file-19-env.zst -q
+ZSTD_CLEVEL=+99 zstd file -o file-99-env.zst -q
+
+cmp file-f10.zst file-f10-env.zst || die "Environment variable failed to set level"
+cmp file-1.zst file-1-env.zst || die "Environment variable failed to set level"
+cmp file-19.zst file-19-env.zst || die "Environment variable failed to set level"
+cmp file-99.zst file-99-env.zst || die "Environment variable failed to set level"
+
+# Test invalid environment clevel is the default level
+zstd -f file -q
+ZSTD_CLEVEL=- zstd -f file -o file-env.zst -q      ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=-: not a valid integer value 
+ZSTD_CLEVEL=+ zstd -f file -o file-env.zst -q      ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=+: not a valid integer value 
+ZSTD_CLEVEL=a zstd -f file -o file-env.zst -q      ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=a: not a valid integer value 
+ZSTD_CLEVEL=-a zstd -f file -o file-env.zst -q     ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=-a: not a valid integer value 
+ZSTD_CLEVEL=+a zstd -f file -o file-env.zst -q     ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=+a: not a valid integer value 
+ZSTD_CLEVEL=3a7 zstd -f file -o file-env.zst -q    ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=3a7: not a valid integer value 
+ZSTD_CLEVEL=5000000000 zstd -f file -o file-env.zst -q ; cmp file.zst file-env.zst
+Ignore environment variable setting ZSTD_CLEVEL=5000000000: numeric value too large 
+
+# Test environment clevel is overridden by command line
+ZSTD_CLEVEL=10 zstd -f file -1 -o file-1-env.zst -q
+ZSTD_CLEVEL=10 zstd -f file --fast=1 -o file-f1-env.zst -q
+
+cmp file-1.zst file-1-env.zst  || die "Environment variable not overridden"
+cmp file-f1.zst file-f1-env.zst || die "Environment variable not overridden"
diff --git a/priv/zstd/tests/cli-tests/compression/long-distance-matcher.sh b/priv/zstd/tests/cli-tests/compression/long-distance-matcher.sh
new file mode 100755
index 0000000..8f2c61b
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/long-distance-matcher.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+# Test --long
+zstd -f file --long   ; zstd -t file.zst
+zstd -f file --long=20; zstd -t file.zst
diff --git a/priv/zstd/tests/cli-tests/compression/multi-threaded.sh b/priv/zstd/tests/cli-tests/compression/multi-threaded.sh
new file mode 100755
index 0000000..17a5eb5
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/multi-threaded.sh
@@ -0,0 +1,15 @@
+#!/bin/sh
+
+set -e
+
+# Test multi-threaded flags
+zstd --single-thread file -f -q         ; zstd -t file.zst
+zstd -T2 -f file -q                     ; zstd -t file.zst
+zstd --rsyncable -f file -q             ; zstd -t file.zst
+zstd -T0 -f file -q                     ; zstd -t file.zst
+zstd -T0 --auto-threads=logical -f file -q ; zstd -t file.zst
+zstd -T0 --auto-threads=physical -f file -q ; zstd -t file.zst
+
+# multi-thread decompression warning test
+zstd -T0 -f file -q                     ; zstd -t file.zst; zstd -T0 -d file.zst -o file3
+zstd -T0 -f file -q                     ; zstd -t file.zst; zstd -T2 -d file.zst -o file4
diff --git a/priv/zstd/tests/cli-tests/compression/multi-threaded.sh.stderr.exact b/priv/zstd/tests/cli-tests/compression/multi-threaded.sh.stderr.exact
new file mode 100644
index 0000000..11daff6
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/multi-threaded.sh.stderr.exact
@@ -0,0 +1,11 @@
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+file.zst            : 65537 bytes 
+Warning : decompression does not support multi-threading
+file.zst            : 65537 bytes 
diff --git a/priv/zstd/tests/cli-tests/compression/multiple-files.sh b/priv/zstd/tests/cli-tests/compression/multiple-files.sh
new file mode 100755
index 0000000..aeb74cf
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/multiple-files.sh
@@ -0,0 +1,21 @@
+#!/bin/sh
+set -e
+
+# setup
+echo "file1" > file1
+echo "file2" > file2
+
+echo "Test zstd ./file1 - file2"
+rm -f ./file*.zst
+echo "stdin" | zstd ./file1 - ./file2 | zstd -d
+cat file1.zst | zstd -d
+cat file2.zst | zstd -d
+
+echo "Test zstd -d ./file1.zst - file2.zst"
+rm ./file1 ./file2
+echo "stdin" | zstd - | zstd -d ./file1.zst - file2.zst
+cat file1
+cat file2
+
+echo "zstd -d ./file1.zst - file2.zst -c"
+echo "stdin" | zstd | zstd -d ./file1.zst - file2.zst -c
diff --git a/priv/zstd/tests/cli-tests/compression/multiple-files.sh.stdout.exact b/priv/zstd/tests/cli-tests/compression/multiple-files.sh.stdout.exact
new file mode 100644
index 0000000..aad61d6
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/multiple-files.sh.stdout.exact
@@ -0,0 +1,12 @@
+Test zstd ./file1 - file2
+stdin
+file1
+file2
+Test zstd -d ./file1.zst - file2.zst
+stdin
+file1
+file2
+zstd -d ./file1.zst - file2.zst -c
+file1
+stdin
+file2
diff --git a/priv/zstd/tests/cli-tests/compression/row-match-finder.sh b/priv/zstd/tests/cli-tests/compression/row-match-finder.sh
new file mode 100755
index 0000000..5b36017
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/row-match-finder.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+# Test --[no-]row-match-finder
+zstd file -7f --row-match-finder
+zstd file -7f --no-row-match-finder
diff --git a/priv/zstd/tests/cli-tests/compression/setup b/priv/zstd/tests/cli-tests/compression/setup
new file mode 100755
index 0000000..96e2309
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/setup
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+datagen > file
+datagen > file0
+datagen > file1
diff --git a/priv/zstd/tests/cli-tests/compression/stream-size.sh b/priv/zstd/tests/cli-tests/compression/stream-size.sh
new file mode 100755
index 0000000..7344769
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/stream-size.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+# Test stream size & hint
+datagen -g7654 | zstd --stream-size=7654 | zstd -t
+datagen -g7654 | zstd --size-hint=7000   | zstd -t
diff --git a/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh b/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh
new file mode 100755
index 0000000..88ee11a
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh
@@ -0,0 +1,11 @@
+#!/bin/sh
+
+set -e
+
+. "$COMMON/platform.sh"
+
+zstd < file -vv -19 -o file.19.zst
+zstd -vv -l file.19.zst
+
+zstd < file -vv -19 --long -o file.19.long.zst
+zstd -vv -l file.19.long.zst
diff --git a/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stderr.glob b/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stderr.glob
new file mode 100644
index 0000000..1353471
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stderr.glob
@@ -0,0 +1,5 @@
+...
+*wlog=23*
+...
+*wlog=27*
+...
diff --git a/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stdout.glob b/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stdout.glob
new file mode 100644
index 0000000..19913a1
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/verbose-wlog.sh.stdout.glob
@@ -0,0 +1,5 @@
+...
+*Window Size: 8388608 B*
+...
+*Window Size: 134217728 B*
+...
diff --git a/priv/zstd/tests/cli-tests/compression/window-resize.sh b/priv/zstd/tests/cli-tests/compression/window-resize.sh
new file mode 100755
index 0000000..3b5e6fe
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/window-resize.sh
@@ -0,0 +1,9 @@
+#!/bin/sh
+datagen -g1G > file
+zstd --long=31 -1 --single-thread --no-content-size -f file
+zstd -l -v file.zst
+
+# We want to ignore stderr (its outputting "*** zstd command line interface
+# 64-bits v1.5.3, by Yann Collet ***")
+
+rm file file.zst
diff --git a/priv/zstd/tests/cli-tests/compression/window-resize.sh.stderr.ignore b/priv/zstd/tests/cli-tests/compression/window-resize.sh.stderr.ignore
new file mode 100644
index 0000000..e69de29
diff --git a/priv/zstd/tests/cli-tests/compression/window-resize.sh.stdout.glob b/priv/zstd/tests/cli-tests/compression/window-resize.sh.stdout.glob
new file mode 100644
index 0000000..313d216
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/compression/window-resize.sh.stdout.glob
@@ -0,0 +1,3 @@
+...
+Window Size: 1.000 GiB (1073741824 B)
+...
diff --git a/priv/zstd/tests/cli-tests/decompression/golden.sh b/priv/zstd/tests/cli-tests/decompression/golden.sh
new file mode 100755
index 0000000..36919e6
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/decompression/golden.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+GOLDEN_DIR="$ZSTD_REPO_DIR/tests/golden-decompression/"
+
+zstd -r -t "$GOLDEN_DIR"
diff --git a/priv/zstd/tests/cli-tests/decompression/pass-through.sh b/priv/zstd/tests/cli-tests/decompression/pass-through.sh
new file mode 100755
index 0000000..2cab463
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/decompression/pass-through.sh
@@ -0,0 +1,57 @@
+#!/bin/sh
+
+set -e
+
+. "$COMMON/platform.sh"
+
+echo "" > 1
+echo "2" > 2
+echo "23" > 3
+echo "234" > 4
+echo "some data" > file
+
+println "+ passthrough enabled"
+
+zstd file
+
+# Test short files
+zstd -dc --pass-through 1 2 3 4
+
+# Test *cat symlinks
+zstdcat file
+"$ZSTD_SYMLINK_DIR/zcat" file
+"$ZSTD_SYMLINK_DIR/gzcat" file
+
+# Test multiple files with mix of compressed & not
+zstdcat file file.zst
+zstdcat file.zst file
+
+# Test --pass-through
+zstd -dc --pass-through file
+zstd -d --pass-through file -o pass-through-file
+
+# Test legacy implicit passthrough with -fc
+zstd -dcf file
+zstd -dcf file file.zst
+zstd -df < file
+zstd -dcf < file file.zst -
+zstd -dcf < file.zst file -
+
+$DIFF file pass-through-file
+
+println "+ passthrough disabled"
+
+# Test *cat
+zstdcat --no-pass-through file && die "should fail"
+"$ZSTD_SYMLINK_DIR/zcat" --no-pass-through file && die "should fail"
+"$ZSTD_SYMLINK_DIR/gzcat" --no-pass-through file && die "should fail"
+# Test zstd without implicit passthrough
+zstd -d file -o no-pass-through-file && die "should fail"
+zstd -d < file && die "should fail"
+
+# Test legacy implicit passthrough with -fc
+zstd --no-pass-through -dcf file && die "should fail"
+zstd --no-pass-through -dcf file file.zst && die "should fail"
+zstd --no-pass-through -df < file && die "should fail"
+zstd --no-pass-through -dcf < file file.zst - && die "should fail"
+zstd --no-pass-through -dcf < file.zst file - && die "should fail" ||:
diff --git a/priv/zstd/tests/cli-tests/decompression/pass-through.sh.stderr.exact b/priv/zstd/tests/cli-tests/decompression/pass-through.sh.stderr.exact
new file mode 100644
index 0000000..62f96ae
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/decompression/pass-through.sh.stderr.exact
@@ -0,0 +1,11 @@
+file                 :230.00%   (    10 B =>     23 B, file.zst) 
+zstd: file: unsupported format 
+zstd: file: unsupported format 
+zstd: file: unsupported format 
+zstd: file: unsupported format 
+zstd: /*stdin*\: unsupported format 
+zstd: file: unsupported format 
+zstd: file: unsupported format 
+zstd: /*stdin*\: unsupported format 
+zstd: /*stdin*\: unsupported format 
+zstd: file: unsupported format 
diff --git a/priv/zstd/tests/cli-tests/decompression/pass-through.sh.stdout.exact b/priv/zstd/tests/cli-tests/decompression/pass-through.sh.stdout.exact
new file mode 100644
index 0000000..b0d494c
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/decompression/pass-through.sh.stdout.exact
@@ -0,0 +1,25 @@
++ passthrough enabled
+
+2
+23
+234
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
+some data
++ passthrough disabled
+some data
+some data
+some data
diff --git a/priv/zstd/tests/cli-tests/dict-builder/empty-input.sh b/priv/zstd/tests/cli-tests/dict-builder/empty-input.sh
new file mode 100755
index 0000000..b500bfe
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dict-builder/empty-input.sh
@@ -0,0 +1,9 @@
+#!/bin/sh
+set -e
+for i in $(seq 50); do
+    datagen -s$i > file$i
+done
+touch empty
+
+set -v
+zstd -q --train empty file*
diff --git a/priv/zstd/tests/cli-tests/dict-builder/empty-input.sh.stderr.exact b/priv/zstd/tests/cli-tests/dict-builder/empty-input.sh.stderr.exact
new file mode 100644
index 0000000..2747e76
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dict-builder/empty-input.sh.stderr.exact
@@ -0,0 +1 @@
+zstd -q --train empty file*
diff --git a/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh b/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh
new file mode 100755
index 0000000..416b837
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh
@@ -0,0 +1,3 @@
+#!/bin/sh
+set -v
+zstd --train
diff --git a/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.exit b/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.exit
new file mode 100644
index 0000000..8351c19
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.exit
@@ -0,0 +1 @@
+14
diff --git a/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.stderr.exact b/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.stderr.exact
new file mode 100644
index 0000000..d7b3ea0
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dict-builder/no-inputs.sh.stderr.exact
@@ -0,0 +1,5 @@
+zstd --train
+!  Warning : nb of samples too low for proper processing ! 
+!  Please provide _one file per sample_. 
+!  Alternatively, split files into fixed-size blocks representative of samples, with -B# 
+Error 14 : nb of samples too low
diff --git a/priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh b/priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh
new file mode 100755
index 0000000..885cac2
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh
@@ -0,0 +1,29 @@
+#!/bin/sh
+
+. "$COMMON/platform.sh"
+
+set -e
+
+if [ false ]; then
+	for seed in $(seq 100); do
+		datagen -g1000 -s$seed > file$seed
+	done
+
+	zstd --train -r . -o dict0 -qq
+
+	for seed in $(seq 101 200); do
+		datagen -g1000 -s$seed > file$seed
+	done
+
+	zstd --train -r . -o dict1 -qq
+
+	[ "$($MD5SUM < dict0)" != "$($MD5SUM < dict1)" ] || die "dictionaries must not match"
+
+	datagen -g1000 -s0 > file0
+fi
+
+set -v
+zstd files/0 -D dicts/0 -q
+zstd -t files/0.zst -D dicts/0
+zstd -t files/0.zst -D dicts/1 && die "Must fail" ||:
+zstd -t files/0.zst            && die "Must fail" ||:
diff --git a/priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh.stderr.exact b/priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh.stderr.exact
new file mode 100644
index 0000000..8896763
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dictionaries/dictionary-mismatch.sh.stderr.exact
@@ -0,0 +1,7 @@
+zstd files/0 -D dicts/0 -q
+zstd -t files/0.zst -D dicts/0
+files/0.zst         : 1000 bytes 
+zstd -t files/0.zst -D dicts/1 && die "Must fail" ||:
+files/0.zst : Decoding error (36) : Dictionary mismatch 
+zstd -t files/0.zst            && die "Must fail" ||:
+files/0.zst : Decoding error (36) : Dictionary mismatch 
diff --git a/priv/zstd/tests/cli-tests/dictionaries/golden.sh b/priv/zstd/tests/cli-tests/dictionaries/golden.sh
new file mode 100755
index 0000000..85da2ee
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dictionaries/golden.sh
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+set -e
+
+GOLDEN_COMP_DIR="$ZSTD_REPO_DIR/tests/golden-compression/"
+GOLDEN_DICT_DIR="$ZSTD_REPO_DIR/tests/golden-dictionaries/"
+
+zstd -D "$GOLDEN_DICT_DIR/http-dict-missing-symbols" "$GOLDEN_COMP_DIR/http" -o http.zst
+zstd -D "$GOLDEN_DICT_DIR/http-dict-missing-symbols" -t http.zst
diff --git a/priv/zstd/tests/cli-tests/dictionaries/setup b/priv/zstd/tests/cli-tests/dictionaries/setup
new file mode 100755
index 0000000..616c73e
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dictionaries/setup
@@ -0,0 +1,6 @@
+#!/bin/sh
+
+set -e
+
+cp -r ../files .
+cp -r ../dicts .
diff --git a/priv/zstd/tests/cli-tests/dictionaries/setup_once b/priv/zstd/tests/cli-tests/dictionaries/setup_once
new file mode 100755
index 0000000..1241c57
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/dictionaries/setup_once
@@ -0,0 +1,24 @@
+#!/bin/sh
+
+set -e
+
+. "$COMMON/platform.sh"
+
+
+mkdir files/ dicts/
+
+for seed in $(seq 50); do
+	datagen -g1000 -s$seed > files/$seed
+done
+
+zstd --train -r files -o dicts/0 -qq
+
+for seed in $(seq 51 100); do
+	datagen -g1000 -s$seed > files/$seed
+done
+
+zstd --train -r files -o dicts/1 -qq
+
+cmp dicts/0 dicts/1 && die "dictionaries must not match!"
+
+datagen -g1000 > files/0
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh
new file mode 100755
index 0000000..c5f5900
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+set -e
+
+datagen > file
+chmod 642 file
+
+zstd file -q --trace-file-stat -o file.zst
+zstd -tq file.zst
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh.stderr.exact
new file mode 100644
index 0000000..a1ad09e
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-file.sh.stderr.exact
@@ -0,0 +1,42 @@
+Trace:FileStat: > UTIL_isLink(file)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_getFileSize(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat: < 65537
+Trace:FileStat: > UTIL_stat(file)
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_isDirectoryStat()
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_stat(file)
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_isSameFile(file, file.zst)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 1
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_getFileSize(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat: < 65537
+Trace:FileStat: > UTIL_setFileStat(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 1
+Trace:FileStat:  > UTIL_utime(file.zst)
+Trace:FileStat:  < 0
+Trace:FileStat:  > UTIL_chmod(file.zst, 0642)
+Trace:FileStat:   > chmod
+Trace:FileStat:   < 0
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh
new file mode 100755
index 0000000..99ebfc4
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh
@@ -0,0 +1,8 @@
+#!/bin/sh
+
+set -e
+
+datagen > file
+
+zstd file -cq --trace-file-stat > file.zst
+zstd -tq file.zst
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh.stderr.exact
new file mode 100644
index 0000000..7c690d2
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-file-to-stdout.sh.stderr.exact
@@ -0,0 +1,24 @@
+Trace:FileStat: > UTIL_isLink(file)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(1)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_getFileSize(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat: < 65537
+Trace:FileStat: > UTIL_stat(file)
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_isDirectoryStat()
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_stat(file)
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_isRegularFile(/*stdout*\)
+Trace:FileStat:  > UTIL_stat(/*stdout*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_getFileSize(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat: < 65537
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh
new file mode 100755
index 0000000..8379461
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh
@@ -0,0 +1,8 @@
+#!/bin/sh
+
+set -e
+
+datagen > file
+
+zstd < file -q --trace-file-stat -o file.zst
+zstd -tq file.zst
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh.stderr.exact
new file mode 100644
index 0000000..00afd97
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-file.sh.stderr.exact
@@ -0,0 +1,24 @@
+Trace:FileStat: > UTIL_isConsole(0)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_getFileSize(/*stdin*\)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < -1
+Trace:FileStat: > UTIL_isSameFile(/*stdin*\, file.zst)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 1
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_getFileSize(/*stdin*\)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < -1
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh
new file mode 100755
index 0000000..64f4b03
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh
@@ -0,0 +1,8 @@
+#!/bin/sh
+
+set -e
+
+datagen > file
+
+zstd < file -cq --trace-file-stat > file.zst
+zstd -tq file.zst
diff --git a/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh.stderr.exact
new file mode 100644
index 0000000..8bf05e6
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/compress-stdin-to-stdout.sh.stderr.exact
@@ -0,0 +1,18 @@
+Trace:FileStat: > UTIL_isConsole(0)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(1)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_getFileSize(/*stdin*\)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < -1
+Trace:FileStat: > UTIL_isRegularFile(/*stdout*\)
+Trace:FileStat:  > UTIL_stat(/*stdout*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_getFileSize(/*stdin*\)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < -1
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh
new file mode 100755
index 0000000..9e68f8f
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh
@@ -0,0 +1,8 @@
+#!/bin/sh
+
+set -e
+
+datagen | zstd -q > file.zst
+chmod 642 file.zst
+
+zstd -dq --trace-file-stat file.zst
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh.stderr.exact
new file mode 100644
index 0000000..d264c63
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-file.sh.stderr.exact
@@ -0,0 +1,38 @@
+Trace:FileStat: > UTIL_isLink(file.zst)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(1)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isDirectory(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 1
+Trace:FileStat:  > UTIL_isDirectoryStat()
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_stat(file.zst)
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_isSameFile(file.zst, file)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 1
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_setFileStat(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat:  > UTIL_utime(file)
+Trace:FileStat:  < 0
+Trace:FileStat:  > UTIL_chmod(file, 0642)
+Trace:FileStat:   > chmod
+Trace:FileStat:   < 0
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh
new file mode 100755
index 0000000..518c2a9
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+datagen | zstd -q > file.zst
+
+zstd -dcq --trace-file-stat file.zst > file
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh.stderr.exact
new file mode 100644
index 0000000..7fe6dda
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-file-to-stdout.sh.stderr.exact
@@ -0,0 +1,18 @@
+Trace:FileStat: > UTIL_isLink(file.zst)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(1)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isDirectory(file.zst)
+Trace:FileStat:  > UTIL_stat(file.zst)
+Trace:FileStat:  < 1
+Trace:FileStat:  > UTIL_isDirectoryStat()
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_stat(file.zst)
+Trace:FileStat: < 1
+Trace:FileStat: > UTIL_isRegularFile(/*stdout*\)
+Trace:FileStat:  > UTIL_stat(/*stdout*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh
new file mode 100755
index 0000000..135d755
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+datagen | zstd -q > file.zst
+
+zstd -dcq --trace-file-stat < file.zst -o file
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh.stderr.exact
new file mode 100644
index 0000000..749fd39
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-file.sh.stderr.exact
@@ -0,0 +1,20 @@
+Trace:FileStat: > UTIL_isConsole(0)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isDirectory(/*stdin*\)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isSameFile(/*stdin*\, file)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(file)
+Trace:FileStat:  > UTIL_stat(file)
+Trace:FileStat:  < 1
+Trace:FileStat: < 1
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh
new file mode 100755
index 0000000..495f07b
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh
@@ -0,0 +1,7 @@
+#!/bin/sh
+
+set -e
+
+datagen | zstd -q > file.zst
+
+zstd -dcq --trace-file-stat < file.zst > file
diff --git a/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh.stderr.exact b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh.stderr.exact
new file mode 100644
index 0000000..e36cb9d
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/file-stat/decompress-stdin-to-stdout.sh.stderr.exact
@@ -0,0 +1,14 @@
+Trace:FileStat: > UTIL_isConsole(0)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(1)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isConsole(2)
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isDirectory(/*stdin*\)
+Trace:FileStat:  > UTIL_stat(/*stdin*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
+Trace:FileStat: > UTIL_isRegularFile(/*stdout*\)
+Trace:FileStat:  > UTIL_stat(/*stdout*\)
+Trace:FileStat:  < 0
+Trace:FileStat: < 0
diff --git a/priv/zstd/tests/cli-tests/progress/no-progress.sh b/priv/zstd/tests/cli-tests/progress/no-progress.sh
new file mode 100755
index 0000000..708878f
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/progress/no-progress.sh
@@ -0,0 +1,46 @@
+#!/bin/sh
+
+#!/bin/sh
+
+. "$COMMON/platform.sh"
+
+set -e
+
+echo hello > hello
+echo world > world
+
+zstd -q hello world
+
+println >&2 "Tests cases where progress information should not be printed"
+
+for args in \
+	"" \
+	"--fake-stderr-is-console -q" \
+	"--fake-stderr-is-console -qq --progress" \
+	"--no-progress --fake-stderr-is-console" \
+	"--no-progress --fake-stderr-is-console -v"
+do
+	println >&2 "args = $args"
+	println >&2 "compress file to file"
+	zstd $args -f hello
+	println >&2 "compress pipe to pipe"
+	zstd $args < hello > $INTOVOID
+	println >&2 "compress pipe to file"
+	zstd $args < hello -fo hello.zst
+	println >&2 "compress file to pipe"
+	zstd $args hello -c > $INTOVOID
+	println >&2 "compress 2 files"
+	zstd $args -f hello world
+
+	println >&2 "decompress file to file"
+	zstd $args -d -f hello.zst
+	println >&2 "decompress pipe to pipe"
+	zstd $args -d < hello.zst > $INTOVOID
+	println >&2 "decompress pipe to file"
+	zstd $args -d < hello.zst -fo hello
+	println >&2 "decompress file to pipe"
+	zstd $args -d hello.zst -c > $INTOVOID
+	println >&2 "decompress 2 files"
+	zstd $args -d -f hello.zst world.zst
+	println >&2 ""
+done
diff --git a/priv/zstd/tests/cli-tests/progress/no-progress.sh.stderr.glob b/priv/zstd/tests/cli-tests/progress/no-progress.sh.stderr.glob
new file mode 100644
index 0000000..d0f9112
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/progress/no-progress.sh.stderr.glob
@@ -0,0 +1,96 @@
+Tests cases where progress information should not be printed
+args =
+compress file to file
+hello*hello.zst*
+compress pipe to pipe
+compress pipe to file
+*stdin*hello.zst*
+compress file to pipe
+compress 2 files
+2 files compressed*
+decompress file to file
+hello.zst*
+decompress pipe to pipe
+decompress pipe to file
+*stdin*
+decompress file to pipe
+decompress 2 files
+2 files decompressed*
+
+args = --fake-stderr-is-console -q
+compress file to file
+compress pipe to pipe
+compress pipe to file
+compress file to pipe
+compress 2 files
+decompress file to file
+decompress pipe to pipe
+decompress pipe to file
+decompress file to pipe
+decompress 2 files
+
+args = --fake-stderr-is-console -qq --progress
+compress file to file
+compress pipe to pipe
+compress pipe to file
+compress file to pipe
+compress 2 files
+decompress file to file
+decompress pipe to pipe
+decompress pipe to file
+decompress file to pipe
+decompress 2 files
+
+args = --no-progress --fake-stderr-is-console
+compress file to file
+hello*hello.zst*
+compress pipe to pipe
+compress pipe to file
+*stdin*hello.zst*
+compress file to pipe
+compress 2 files
+2 files compressed*
+decompress file to file
+hello.zst*
+decompress pipe to pipe
+decompress pipe to file
+*stdin*
+decompress file to pipe
+decompress 2 files
+2 files decompressed*
+
+args = --no-progress --fake-stderr-is-console -v
+compress file to file
+*Zstandard CLI*
+hello*hello.zst*
+compress pipe to pipe
+*Zstandard CLI*
+*stdin*stdout*
+compress pipe to file
+*Zstandard CLI*
+*stdin*hello.zst*
+compress file to pipe
+*Zstandard CLI*
+*hello*stdout*
+compress 2 files
+*Zstandard CLI*
+*hello*hello.zst*
+*world*world.zst*
+2 files compressed*
+decompress file to file
+*Zstandard CLI*
+hello.zst*
+decompress pipe to pipe
+*Zstandard CLI*
+*stdin*
+decompress pipe to file
+*Zstandard CLI*
+*stdin*
+decompress file to pipe
+*Zstandard CLI*
+hello.zst*
+decompress 2 files
+*Zstandard CLI*
+hello.zst*
+world.zst*
+2 files decompressed*
diff --git a/priv/zstd/tests/cli-tests/progress/progress.sh b/priv/zstd/tests/cli-tests/progress/progress.sh
new file mode 100755
index 0000000..eb46499
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/progress/progress.sh
@@ -0,0 +1,41 @@
+#!/bin/sh
+
+. "$COMMON/platform.sh"
+
+set -e
+
+println >&2 "Tests cases where progress information should be printed"
+
+echo hello > hello
+echo world > world
+
+zstd -q hello world
+
+for args in \
+	"--progress" \
+	"--fake-stderr-is-console" \
+	"--progress --fake-stderr-is-console -q"; do
+	println >&2 "args = $args"
+	println >&2 "compress file to file"
+	zstd $args -f hello
+	println >&2 "compress pipe to pipe"
+	zstd $args < hello > $INTOVOID
+	println >&2 "compress pipe to file"
+	zstd $args < hello -fo hello.zst
+	println >&2 "compress file to pipe"
+	zstd $args hello -c > $INTOVOID
+	println >&2 "compress 2 files"
+	zstd $args -f hello world
+
+	println >&2 "decompress file to file"
+	zstd $args -d -f hello.zst
+	println >&2 "decompress pipe to pipe"
+	zstd $args -d < hello.zst > $INTOVOID
+	println >&2 "decompress pipe to file"
+	zstd $args -d < hello.zst -fo hello
+	println >&2 "decompress file to pipe"
+	zstd $args -d hello.zst -c > $INTOVOID
+	println >&2 "decompress 2 files"
+	zstd $args -d -f hello.zst world.zst
+	println >&2 ""
+done
diff --git a/priv/zstd/tests/cli-tests/progress/progress.sh.stderr.glob b/priv/zstd/tests/cli-tests/progress/progress.sh.stderr.glob
new file mode 100644
index 0000000..ca620d3
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/progress/progress.sh.stderr.glob
@@ -0,0 +1,62 @@
+Tests cases where progress information should be printed
+args = --progress
+compress file to file
+*Read:*hello*hello.zst*
+compress pipe to pipe
+*Read:*stdin*stdout*
+compress pipe to file
+*Read:*stdin*hello.zst*
+compress file to pipe
+*Read:*hello*stdout*
+compress 2 files
+*Read*2 files compressed*
+decompress file to file
+*hello.zst*hello.zst*
+decompress pipe to pipe
+*stdin*stdin*
+decompress pipe to file
+*stdin*stdin*
+decompress file to pipe
+*hello.zst*hello.zst*
+decompress 2 files
+*hello.zst*2 files decompressed*
+
+args = --fake-stderr-is-console
+compress file to file
+*Read:*hello*hello.zst*
+compress pipe to pipe
+compress pipe to file
+*Read:*stdin*hello.zst*
+compress file to pipe
+compress 2 files
+*Read*2 files compressed*
+decompress file to file
+*hello.zst*hello.zst*
+decompress pipe to pipe
+decompress pipe to file
+*stdin*stdin*
+decompress file to pipe
+decompress 2 files
+*hello.zst*2 files decompressed*
+
+args = --progress --fake-stderr-is-console -q
+compress file to file
+*Read:*hello*hello.zst*
+compress pipe to pipe
+*Read:*stdin*stdout*
+compress pipe to file
+*Read:*stdin*hello.zst*
+compress file to pipe
+*Read:*hello*stdout*
+compress 2 files
+*Read*2 files compressed*
+decompress file to file
+*hello.zst*hello.zst*
+decompress pipe to pipe
+*stdin*stdin*
+decompress pipe to file
+*stdin*stdin*
+decompress file to pipe
+*hello.zst*hello.zst*
+decompress 2 files
+*hello.zst*2 files decompressed*
diff --git a/priv/zstd/tests/cli-tests/run.py b/priv/zstd/tests/cli-tests/run.py
new file mode 100755
index 0000000..45af512
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/run.py
@@ -0,0 +1,738 @@
+#!/usr/bin/env python3
+# ################################################################
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under both the BSD-style license (found in the
+# LICENSE file in the root directory of this source tree) and the GPLv2 (found
+# in the COPYING file in the root directory of this source tree).
+# You may select, at your option, one of the above-listed licenses.
+# ##########################################################################
+
+import argparse
+import contextlib
+import copy
+import fnmatch
+import os
+import shutil
+import subprocess
+import sys
+import tempfile
+import typing
+
+
+ZSTD_SYMLINKS = [
+    "zstd",
+    "zstdmt",
+    "unzstd",
+    "zstdcat",
+    "zcat",
+    "gzip",
+    "gunzip",
+    "gzcat",
+    "lzma",
+    "unlzma",
+    "xz",
+    "unxz",
+    "lz4",
+    "unlz4",
+]
+
+
+EXCLUDED_DIRS = {
+    "bin",
+    "common",
+    "scratch",
+}
+
+
+EXCLUDED_BASENAMES = {
+    "setup",
+    "setup_once",
+    "teardown",
+    "teardown_once",
+    "README.md",
+    "run.py",
+    ".gitignore",
+}
+
+EXCLUDED_SUFFIXES = [
+    ".exact",
+    ".glob",
+    ".ignore",
+    ".exit",
+]
+
+
+def exclude_dir(dirname: str) -> bool:
+    """
+    Should files under the directory :dirname: be excluded from the test runner?
+    """
+    if dirname in EXCLUDED_DIRS:
+        return True
+    return False
+
+
+def exclude_file(filename: str) -> bool:
+    """Should the file :filename: be excluded from the test runner?"""
+    if filename in EXCLUDED_BASENAMES:
+        return True
+    for suffix in EXCLUDED_SUFFIXES:
+        if filename.endswith(suffix):
+            return True
+    return False
+
+def read_file(filename: str) -> bytes:
+    """Reads the file :filename: and returns the contents as bytes."""
+    with open(filename, "rb") as f:
+        return f.read()
+
+
+def diff(a: bytes, b: bytes) -> str:
+    """Returns a diff between two different byte-strings :a: and :b:."""
+    assert a != b
+    with tempfile.NamedTemporaryFile("wb") as fa:
+        fa.write(a)
+        fa.flush()
+        with tempfile.NamedTemporaryFile("wb") as fb:
+            fb.write(b)
+            fb.flush()
+
+            diff_bytes = subprocess.run(["diff", fa.name, fb.name], stdout=subprocess.PIPE, stderr=subprocess.DEVNULL).stdout
+            return diff_bytes.decode("utf8")
+
+
+def pop_line(data: bytes) -> typing.Tuple[typing.Optional[bytes], bytes]:
+    """
+    Pop the first line from :data: and returns the first line and the remainder
+    of the data as a tuple. If :data: is empty, returns :(None, data):. Otherwise
+    the first line always ends in a :\n:, even if it is the last line and :data:
+    doesn't end in :\n:.
+    """
+    NEWLINE = b"\n"[0]
+
+    if data == b'':
+        return (None, data)
+
+    newline_idx = data.find(b"\n")
+    if newline_idx == -1:
+        end_idx = len(data)
+    else:
+        end_idx = newline_idx + 1
+
+    line = data[:end_idx]
+    data = data[end_idx:]
+
+    assert len(line) != 0
+    if line[-1] != NEWLINE:
+        line += NEWLINE
+
+    return (line, data)
+
+
+def glob_line_matches(actual: bytes, expect: bytes) -> bool:
+    """
+    Does the `actual` line match the expected glob line `expect`?
+    """
+    return fnmatch.fnmatchcase(actual.strip(), expect.strip())
+
+
+def glob_diff(actual: bytes, expect: bytes) -> bytes:
+    """
+    Returns None if the :actual: content matches the expected glob :expect:,
+    otherwise returns the diff bytes.
+    """
+    diff = b''
+    actual_line, actual = pop_line(actual)
+    expect_line, expect = pop_line(expect)
+    while True:
+        # Handle end of file conditions - allow extra newlines
+        while expect_line is None and actual_line == b"\n":
+            actual_line, actual = pop_line(actual)
+        while actual_line is None and expect_line == b"\n":
+            expect_line, expect = pop_line(expect)
+
+        if expect_line is None and actual_line is None:
+            if diff == b'':
+                return None
+            return diff
+        elif expect_line is None:
+            diff += b"---\n"
+            while actual_line != None:
+                diff += b"> "
+                diff += actual_line
+                actual_line, actual = pop_line(actual)
+            return diff
+        elif actual_line is None:
+            diff += b"---\n"
+            while expect_line != None:
+                diff += b"< "
+                diff += expect_line
+                expect_line, expect = pop_line(expect)
+            return diff
+
+        assert expect_line is not None
+        assert actual_line is not None
+
+        if expect_line == b'...\n':
+            next_expect_line, expect = pop_line(expect)
+            if next_expect_line is None:
+                if diff == b'':
+                    return None
+                return diff
+            while not glob_line_matches(actual_line, next_expect_line):
+                actual_line, actual = pop_line(actual)
+                if actual_line is None:
+                    diff += b"---\n"
+                    diff += b"< "
+                    diff += next_expect_line
+                    return diff
+            expect_line = next_expect_line
+            continue
+
+        if not glob_line_matches(actual_line, expect_line):
+            diff += b'---\n'
+            diff += b'< ' + expect_line
+            diff += b'> ' + actual_line
+
+        actual_line, actual = pop_line(actual)
+        expect_line, expect = pop_line(expect)
+
+
+class Options:
+    """Options configuring how to run a :TestCase:."""
+    def __init__(
+        self,
+        env: typing.Dict[str, str],
+        timeout: typing.Optional[int],
+        verbose: bool,
+        preserve: bool,
+        scratch_dir: str,
+        test_dir: str,
+        set_exact_output: bool,
+    ) -> None:
+        self.env = env
+        self.timeout = timeout
+        self.verbose = verbose
+        self.preserve = preserve
+        self.scratch_dir = scratch_dir
+        self.test_dir = test_dir
+        self.set_exact_output = set_exact_output
+
+
+class TestCase:
+    """
+    Logic and state related to running a single test case.
+
+    1. Initialize the test case.
+    2. Launch the test case with :TestCase.launch():.
+       This will start the test execution in a subprocess, but
+       not wait for completion. So you could launch multiple test
+       cases in parallel. This will now print any test output.
+    3. Analyze the results with :TestCase.analyze():. This will
+       join the test subprocess, check the results against the
+       expectations, and print the results to stdout.
+
+    :TestCase.run(): is also provided which combines the launch & analyze
+    steps for single-threaded use-cases.
+
+    All other methods, prefixed with _, are private helper functions.
+    """
+    def __init__(self, test_filename: str, options: Options) -> None:
+        """
+        Initialize the :TestCase: for the test located in :test_filename:
+        with the given :options:.
+        """
+        self._opts = options
+        self._test_file = test_filename
+        self._test_name = os.path.normpath(
+            os.path.relpath(test_filename, start=self._opts.test_dir)
+        )
+        self._success = {}
+        self._message = {}
+        self._test_stdin = None
+        self._scratch_dir = os.path.abspath(os.path.join(self._opts.scratch_dir, self._test_name))
+
+    @property
+    def name(self) -> str:
+        """Returns the unique name for the test."""
+        return self._test_name
+
+    def launch(self) -> None:
+        """
+        Launch the test case as a subprocess, but do not block on completion.
+        This allows users to run multiple tests in parallel. Results aren't yet
+        printed out.
+        """
+        self._launch_test()
+
+    def analyze(self) -> bool:
+        """
+        Must be called after :TestCase.launch():. Joins the test subprocess and
+        checks the results against expectations. Finally prints the results to
+        stdout and returns the success.
+        """
+        self._join_test()
+        self._check_exit()
+        self._check_stderr()
+        self._check_stdout()
+        self._analyze_results()
+        return self._succeeded
+
+    def run(self) -> bool:
+        """Shorthand for combining both :TestCase.launch(): and :TestCase.analyze():."""
+        self.launch()
+        return self.analyze()
+
+    def _log(self, *args, **kwargs) -> None:
+        """Logs test output."""
+        print(file=sys.stdout, *args, **kwargs)
+
+    def _vlog(self, *args, **kwargs) -> None:
+        """Logs verbose test output."""
+        if self._opts.verbose:
+            print(file=sys.stdout, *args, **kwargs)
+
+    def _test_environment(self) -> typing.Dict[str, str]:
+        """
+        Returns the environment to be used for the
+        test subprocess.
+        """
+        # We want to omit ZSTD cli flags so tests will be consistent across environments
+        env = {k: v for k, v in os.environ.items() if not k.startswith("ZSTD")}
+        for k, v in self._opts.env.items():
+            self._vlog(f"${k}='{v}'")
+            env[k] = v
+        return env
+
+    def _launch_test(self) -> None:
+        """Launch the test subprocess, but do not join it."""
+        args = [os.path.abspath(self._test_file)]
+        stdin_name = f"{self._test_file}.stdin"
+        if os.path.exists(stdin_name):
+            self._test_stdin = open(stdin_name, "rb")
+            stdin = self._test_stdin
+        else:
+            stdin = subprocess.DEVNULL
+        cwd = self._scratch_dir
+        env = self._test_environment()
+        self._test_process = subprocess.Popen(
+            args=args,
+            stdin=stdin,
+            cwd=cwd,
+            env=env,
+            stderr=subprocess.PIPE,
+            stdout=subprocess.PIPE
+        )
+
+    def _join_test(self) -> None:
+        """Join the test process and save stderr, stdout, and the exit code."""
+        (stdout, stderr) = self._test_process.communicate(timeout=self._opts.timeout)
+        self._output = {}
+        self._output["stdout"] = stdout
+        self._output["stderr"] = stderr
+        self._exit_code = self._test_process.returncode
+        self._test_process = None
+        if self._test_stdin is not None:
+            self._test_stdin.close()
+            self._test_stdin = None
+
+    def _check_output_exact(self, out_name: str, expected: bytes, exact_name: str) -> None:
+        """
+        Check the output named :out_name: for an exact match against the :expected: content.
+        Saves the success and message.
+        """
+        check_name = f"check_{out_name}"
+        actual = self._output[out_name]
+        if actual == expected:
+            self._success[check_name] = True
+            self._message[check_name] = f"{out_name} matches!"
+        else:
+            self._success[check_name] = False
+            self._message[check_name] = f"{out_name} does not match!\n> diff expected actual\n{diff(expected, actual)}"
+
+            if self._opts.set_exact_output:
+                with open(exact_name, "wb") as f:
+                    f.write(actual)
+
+    def _check_output_glob(self, out_name: str, expected: bytes) -> None:
+        """
+        Check the output named :out_name: for a glob match against the :expected: glob.
+        Saves the success and message.
+        """
+        check_name = f"check_{out_name}"
+        actual = self._output[out_name]
+        diff = glob_diff(actual, expected)
+        if diff is None:
+            self._success[check_name] = True
+            self._message[check_name] = f"{out_name} matches!"
+        else:
+            utf8_diff = diff.decode('utf8')
+            self._success[check_name] = False
+            self._message[check_name] = f"{out_name} does not match!\n> diff expected actual\n{utf8_diff}"
+
+    def _check_output(self, out_name: str) -> None:
+        """
+        Checks the output named :out_name: for a match against the expectation.
+        We check for a .exact, .glob, and a .ignore file. If none are found we
+        expect that the output should be empty.
+
+        If :Options.preserve: was set then we save the scratch directory and
+        save the stderr, stdout, and exit code to the scratch directory for
+        debugging.
+        """
+        if self._opts.preserve:
+            # Save the output to the scratch directory
+            actual_name = os.path.join(self._scratch_dir, f"{out_name}")
+            with open(actual_name, "wb") as f:
+                    f.write(self._output[out_name])
+
+        exact_name = f"{self._test_file}.{out_name}.exact"
+        glob_name = f"{self._test_file}.{out_name}.glob"
+        ignore_name = f"{self._test_file}.{out_name}.ignore"
+
+        if os.path.exists(exact_name):
+            return self._check_output_exact(out_name, read_file(exact_name), exact_name)
+        elif os.path.exists(glob_name):
+            return self._check_output_glob(out_name, read_file(glob_name))
+        else:
+            check_name = f"check_{out_name}"
+            self._success[check_name] = True
+            self._message[check_name] = f"{out_name} ignored!"
+
+    def _check_stderr(self) -> None:
+        """Checks the stderr output against the expectation."""
+        self._check_output("stderr")
+
+    def _check_stdout(self) -> None:
+        """Checks the stdout output against the expectation."""
+        self._check_output("stdout")
+
+    def _check_exit(self) -> None:
+        """
+        Checks the exit code against expectations. If a .exit file
+        exists, we expect that the exit code matches the contents.
+        Otherwise we expect the exit code to be zero.
+
+        If :Options.preserve: is set we save the exit code to the
+        scratch directory under the filename "exit".
+        """
+        if self._opts.preserve:
+            exit_name = os.path.join(self._scratch_dir, "exit")
+            with open(exit_name, "w") as f:
+                f.write(str(self._exit_code) + "\n")
+        exit_name = f"{self._test_file}.exit"
+        if os.path.exists(exit_name):
+            exit_code: int = int(read_file(exit_name))
+        else:
+            exit_code: int = 0
+        if exit_code == self._exit_code:
+            self._success["check_exit"] = True
+            self._message["check_exit"] = "Exit code matches!"
+        else:
+            self._success["check_exit"] = False
+            self._message["check_exit"] = f"Exit code mismatch! Expected {exit_code} but got {self._exit_code}"
+
+    def _analyze_results(self) -> None:
+        """
+        After all tests have been checked, collect all the successes
+        and messages, and print the results to stdout.
+        """
+        STATUS = {True: "PASS", False: "FAIL"}
+        checks = sorted(self._success.keys())
+        self._succeeded = all(self._success.values())
+        self._log(f"{STATUS[self._succeeded]}: {self._test_name}")
+
+        if not self._succeeded or self._opts.verbose:
+            for check in checks:
+                if self._opts.verbose or not self._success[check]:
+                    self._log(f"{STATUS[self._success[check]]}: {self._test_name}.{check}")
+                    self._log(self._message[check])
+
+        self._log("----------------------------------------")
+
+
+class TestSuite:
+    """
+    Setup & teardown test suite & cases.
+    This class is intended to be used as a context manager.
+
+    TODO: Make setup/teardown failure emit messages, not throw exceptions.
+    """
+    def __init__(self, test_directory: str, options: Options) -> None:
+        self._opts = options
+        self._test_dir = os.path.abspath(test_directory)
+        rel_test_dir = os.path.relpath(test_directory, start=self._opts.test_dir)
+        assert not rel_test_dir.startswith(os.path.sep)
+        self._scratch_dir = os.path.normpath(os.path.join(self._opts.scratch_dir, rel_test_dir))
+
+    def __enter__(self) -> 'TestSuite':
+        self._setup_once()
+        return self
+
+    def __exit__(self, _exc_type, _exc_value, _traceback) -> None:
+        self._teardown_once()
+
+    @contextlib.contextmanager
+    def test_case(self, test_basename: str) -> TestCase:
+        """
+        Context manager for a test case in the test suite.
+        Pass the basename of the test relative to the :test_directory:.
+        """
+        assert os.path.dirname(test_basename) == ""
+        try:
+            self._setup(test_basename)
+            test_filename = os.path.join(self._test_dir, test_basename)
+            yield TestCase(test_filename, self._opts)
+        finally:
+            self._teardown(test_basename)
+
+    def _remove_scratch_dir(self, dir: str) -> None:
+        """Helper to remove a scratch directory with sanity checks"""
+        assert "scratch" in dir
+        assert dir.startswith(self._scratch_dir)
+        assert os.path.exists(dir)
+        shutil.rmtree(dir)
+
+    def _setup_once(self) -> None:
+        if os.path.exists(self._scratch_dir):
+            self._remove_scratch_dir(self._scratch_dir)
+        os.makedirs(self._scratch_dir)
+        setup_script = os.path.join(self._test_dir, "setup_once")
+        if os.path.exists(setup_script):
+            self._run_script(setup_script, cwd=self._scratch_dir)
+
+    def _teardown_once(self) -> None:
+        assert os.path.exists(self._scratch_dir)
+        teardown_script = os.path.join(self._test_dir, "teardown_once")
+        if os.path.exists(teardown_script):
+            self._run_script(teardown_script, cwd=self._scratch_dir)
+        if not self._opts.preserve:
+            self._remove_scratch_dir(self._scratch_dir)
+
+    def _setup(self, test_basename: str) -> None:
+        test_scratch_dir = os.path.join(self._scratch_dir, test_basename)
+        assert not os.path.exists(test_scratch_dir)
+        os.makedirs(test_scratch_dir)
+        setup_script = os.path.join(self._test_dir, "setup")
+        if os.path.exists(setup_script):
+            self._run_script(setup_script, cwd=test_scratch_dir)
+
+    def _teardown(self, test_basename: str) -> None:
+        test_scratch_dir = os.path.join(self._scratch_dir, test_basename)
+        assert os.path.exists(test_scratch_dir)
+        teardown_script = os.path.join(self._test_dir, "teardown")
+        if os.path.exists(teardown_script):
+            self._run_script(teardown_script, cwd=test_scratch_dir)
+        if not self._opts.preserve:
+            self._remove_scratch_dir(test_scratch_dir)
+
+    def _run_script(self, script: str, cwd: str) -> None:
+        env = copy.copy(os.environ)
+        for k, v in self._opts.env.items():
+            env[k] = v
+        try:
+            subprocess.run(
+                args=[script],
+                stdin=subprocess.DEVNULL,
+                capture_output=True,
+                cwd=cwd,
+                env=env,
+                check=True,
+            )
+        except subprocess.CalledProcessError as e:
+            print(f"{script} failed with exit code {e.returncode}!")
+            print(f"stderr:\n{e.stderr}")
+            print(f"stdout:\n{e.stdout}")
+            raise
+
+TestSuites = typing.Dict[str, typing.List[str]]
+
+def get_all_tests(options: Options) -> TestSuites:
+    """
+    Find all the test in the test directory and return the test suites.
+    """
+    test_suites = {}
+    for root, dirs, files in os.walk(options.test_dir, topdown=True):
+        dirs[:] = [d for d in dirs if not exclude_dir(d)]
+        test_cases = []
+        for file in files:
+            if not exclude_file(file):
+                test_cases.append(file)
+        assert root == os.path.normpath(root)
+        test_suites[root] = test_cases
+    return test_suites
+
+
+def resolve_listed_tests(
+    tests: typing.List[str], options: Options
+) -> TestSuites:
+    """
+    Resolve the list of tests passed on the command line into their
+    respective test suites. Tests can either be paths, or test names
+    relative to the test directory.
+    """
+    test_suites = {}
+    for test in tests:
+        if not os.path.exists(test):
+            test = os.path.join(options.test_dir, test)
+            if not os.path.exists(test):
+                raise RuntimeError(f"Test {test} does not exist!")
+
+        test = os.path.normpath(os.path.abspath(test))
+        assert test.startswith(options.test_dir)
+        test_suite = os.path.dirname(test)
+        test_case = os.path.basename(test)
+        test_suites.setdefault(test_suite, []).append(test_case)
+
+    return test_suites
+
+def run_tests(test_suites: TestSuites, options: Options) -> bool:
+    """
+    Runs all the test in the :test_suites: with the given :options:.
+    Prints the results to stdout.
+    """
+    tests = {}
+    for test_dir, test_files in test_suites.items():
+        with TestSuite(test_dir, options) as test_suite:
+            test_files = sorted(set(test_files))
+            for test_file in test_files:
+                with test_suite.test_case(test_file) as test_case:
+                    tests[test_case.name] = test_case.run()
+
+    successes = 0
+    for test, status in tests.items():
+        if status:
+            successes += 1
+        else:
+            print(f"FAIL: {test}")
+    if successes == len(tests):
+        print(f"PASSED all {len(tests)} tests!")
+        return True
+    else:
+        print(f"FAILED {len(tests) - successes} / {len(tests)} tests!")
+        return False
+
+
+def setup_zstd_symlink_dir(zstd_symlink_dir: str, zstd: str) -> None:
+    assert os.path.join("bin", "symlinks") in zstd_symlink_dir
+    if not os.path.exists(zstd_symlink_dir):
+        os.makedirs(zstd_symlink_dir)
+    for symlink in ZSTD_SYMLINKS:
+        path = os.path.join(zstd_symlink_dir, symlink)
+        if os.path.exists(path):
+            os.remove(path)
+        os.symlink(zstd, path)
+
+if __name__ == "__main__":
+    CLI_TEST_DIR = os.path.dirname(sys.argv[0])
+    REPO_DIR = os.path.join(CLI_TEST_DIR, "..", "..")
+    PROGRAMS_DIR = os.path.join(REPO_DIR, "programs")
+    TESTS_DIR = os.path.join(REPO_DIR, "tests")
+    ZSTD_PATH = os.path.join(PROGRAMS_DIR, "zstd")
+    ZSTDGREP_PATH = os.path.join(PROGRAMS_DIR, "zstdgrep")
+    ZSTDLESS_PATH = os.path.join(PROGRAMS_DIR, "zstdless")
+    DATAGEN_PATH = os.path.join(TESTS_DIR, "datagen")
+
+    parser = argparse.ArgumentParser(
+        (
+            "Runs the zstd CLI tests. Exits nonzero on failure. Default arguments are\n"
+            "generally correct. Pass --preserve to preserve test output for debugging,\n"
+            "and --verbose to get verbose test output.\n"
+        )
+    )
+    parser.add_argument(
+        "--preserve",
+        action="store_true",
+        help="Preserve the scratch directory TEST_DIR/scratch/ for debugging purposes."
+    )
+    parser.add_argument("--verbose", action="store_true", help="Verbose test output.")
+    parser.add_argument("--timeout", default=60, type=int, help="Test case timeout in seconds. Set to 0 to disable timeouts.")
+    parser.add_argument(
+        "--exec-prefix",
+        default=None,
+        help="Sets the EXEC_PREFIX environment variable. Prefix to invocations of the zstd CLI."
+    )
+    parser.add_argument(
+        "--zstd",
+        default=ZSTD_PATH,
+        help="Sets the ZSTD_BIN environment variable. Path of the zstd CLI."
+    )
+    parser.add_argument(
+        "--zstdgrep",
+        default=ZSTDGREP_PATH,
+        help="Sets the ZSTDGREP_BIN environment variable. Path of the zstdgrep CLI."
+    )
+    parser.add_argument(
+        "--zstdless",
+        default=ZSTDLESS_PATH,
+        help="Sets the ZSTDLESS_BIN environment variable. Path of the zstdless CLI."
+    )
+    parser.add_argument(
+        "--datagen",
+        default=DATAGEN_PATH,
+        help="Sets the DATAGEN_BIN environment variable. Path to the datagen CLI."
+    )
+    parser.add_argument(
+        "--test-dir",
+        default=CLI_TEST_DIR,
+        help=(
+            "Runs the tests under this directory. "
+            "Adds TEST_DIR/bin/ to path. "
+            "Scratch directory located in TEST_DIR/scratch/."
+        )
+    )
+    parser.add_argument(
+        "--set-exact-output",
+        action="store_true",
+        help="Set stderr.exact and stdout.exact for all failing tests, unless .ignore or .glob already exists"
+    )
+    parser.add_argument(
+        "tests",
+        nargs="*",
+        help="Run only these test cases. Can either be paths or test names relative to TEST_DIR/"
+    )
+    args = parser.parse_args()
+
+    if args.timeout <= 0:
+        args.timeout = None
+
+    args.test_dir = os.path.normpath(os.path.abspath(args.test_dir))
+    bin_dir = os.path.abspath(os.path.join(args.test_dir, "bin"))
+    zstd_symlink_dir = os.path.join(bin_dir, "symlinks")
+    scratch_dir = os.path.join(args.test_dir, "scratch")
+
+    setup_zstd_symlink_dir(zstd_symlink_dir, os.path.abspath(args.zstd))
+
+    env = {}
+    if args.exec_prefix is not None:
+        env["EXEC_PREFIX"] = args.exec_prefix
+    env["ZSTD_SYMLINK_DIR"] = zstd_symlink_dir
+    env["ZSTD_REPO_DIR"] = os.path.abspath(REPO_DIR)
+    env["DATAGEN_BIN"] = os.path.abspath(args.datagen)
+    env["ZSTDGREP_BIN"] = os.path.abspath(args.zstdgrep)
+    env["ZSTDLESS_BIN"] = os.path.abspath(args.zstdless)
+    env["COMMON"] = os.path.abspath(os.path.join(args.test_dir, "common"))
+    env["PATH"] = bin_dir + ":" + os.getenv("PATH", "")
+    env["LC_ALL"] = "C"
+
+    opts = Options(
+        env=env,
+        timeout=args.timeout,
+        verbose=args.verbose,
+        preserve=args.preserve,
+        test_dir=args.test_dir,
+        scratch_dir=scratch_dir,
+        set_exact_output=args.set_exact_output,
+    )
+
+    if len(args.tests) == 0:
+        tests = get_all_tests(opts)
+    else:
+        tests = resolve_listed_tests(args.tests, opts)
+
+    success = run_tests(tests, opts)
+    if success:
+        sys.exit(0)
+    else:
+        sys.exit(1)
diff --git a/priv/zstd/tests/cli-tests/zstd-symlinks/setup b/priv/zstd/tests/cli-tests/zstd-symlinks/setup
new file mode 100755
index 0000000..cf391ed
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/zstd-symlinks/setup
@@ -0,0 +1,6 @@
+#!/bin/sh
+set -e
+
+println "hello" > hello
+println "world" > world
+zstd hello world
diff --git a/priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh b/priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh
new file mode 100755
index 0000000..74ec063
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh
@@ -0,0 +1,12 @@
+#!/bin/sh
+set -e
+
+# Test zstdcat symlink in bin/
+zstdcat hello.zst
+zstdcat hello.zst world
+zstdcat hello world.zst
+zstdcat hello.zst world.zst
+
+# Test local zstdcat symlink
+ln -s $(which zstd) ./zstdcat
+./zstdcat hello.zst
diff --git a/priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh.stdout.exact b/priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh.stdout.exact
new file mode 100644
index 0000000..3205b05
--- /dev/null
+++ b/priv/zstd/tests/cli-tests/zstd-symlinks/zstdcat.sh.stdout.exact
@@ -0,0 +1,8 @@
+hello
+hello
+world
+hello
+world
+hello
+world
+hello
diff --git a/priv/zstd/tests/datagencli.c b/priv/zstd/tests/datagencli.c
index ecc05f9..09ec5e9 100644
--- a/priv/zstd/tests/datagencli.c
+++ b/priv/zstd/tests/datagencli.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -124,7 +124,7 @@ int main(int argc, const char** argv)
     DISPLAYLEVEL(3, "Seed = %u \n", (unsigned)seed);
 
     RDG_genStdout(size, (double)probaU32/100, litProba, seed);
-    DISPLAYLEVEL(1, "\n");
+    DISPLAYLEVEL(3, "\n");
 
     return 0;
 }
diff --git a/priv/zstd/tests/decodecorpus.c b/priv/zstd/tests/decodecorpus.c
index fa6a2d6..148088a 100644
--- a/priv/zstd/tests/decodecorpus.c
+++ b/priv/zstd/tests/decodecorpus.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,6 +14,7 @@
 #include 
 #include 
 #include 
+#include   /* time(), for seed random initialization */
 
 #include "util.h"
 #include "timefn.h"   /* UTIL_clockSpanMicro, SEC_TO_MICRO, UTIL_TIME_INITIALIZER */
@@ -24,21 +25,13 @@
 #include "zdict.h"
 
 /* Direct access to internal compression functions is required */
-#include "zstd_compress.c"
+#include "compress/zstd_compress.c" /* ZSTD_resetSeqStore, ZSTD_storeSeq, *_TO_OFFBASE, HIST_countFast_wksp, HIST_isError */
 
 #define XXH_STATIC_LINKING_ONLY
 #include "xxhash.h"     /* XXH64 */
 
-#ifndef MIN
-    #define MIN(a, b) ((a) < (b) ? (a) : (b))
-#endif
-
-#ifndef MAX_PATH
-    #ifdef PATH_MAX
-        #define MAX_PATH PATH_MAX
-    #else
-        #define MAX_PATH 256
-    #endif
+#if !(defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */))
+# define inline  /* disable */
 #endif
 
 /*-************************************
@@ -70,6 +63,7 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
         }                                                                      \
     } while (0)
 
+
 /*-*******************************************************
 *  Random function
 *********************************************************/
@@ -136,7 +130,7 @@ static void RAND_genDist(U32* seed, BYTE* dist, double weight)
     BYTE step = (BYTE) ((RAND(seed) % 256) | 1); /* force it to be odd so it's relatively prime to 256 */
 
     while (i < DISTSIZE) {
-        size_t states = ((size_t)(weight * statesLeft)) + 1;
+        size_t states = ((size_t)(weight * (double)statesLeft)) + 1;
         size_t j;
         for (j = 0; j < states && i < DISTSIZE; j++, i++) {
             dist[i] = symb;
@@ -165,7 +159,7 @@ static double RAND_exp(U32* seed, double mean)
 /*-*******************************************************
 *  Constants and Structs
 *********************************************************/
-const char *BLOCK_TYPES[] = {"raw", "rle", "compressed"};
+const char* BLOCK_TYPES[] = {"raw", "rle", "compressed"};
 
 #define MAX_DECOMPRESSED_SIZE_LOG 20
 #define MAX_DECOMPRESSED_SIZE (1ULL << MAX_DECOMPRESSED_SIZE_LOG)
@@ -175,6 +169,14 @@ const char *BLOCK_TYPES[] = {"raw", "rle", "compressed"};
 #define MIN_SEQ_LEN (3)
 #define MAX_NB_SEQ ((ZSTD_BLOCKSIZE_MAX + MIN_SEQ_LEN - 1) / MIN_SEQ_LEN)
 
+#ifndef MAX_PATH
+    #ifdef PATH_MAX
+        #define MAX_PATH PATH_MAX
+    #else
+        #define MAX_PATH 256
+    #endif
+#endif
+
 BYTE CONTENT_BUFFER[MAX_DECOMPRESSED_SIZE];
 BYTE FRAME_BUFFER[MAX_DECOMPRESSED_SIZE * 2];
 BYTE LITERAL_BUFFER[ZSTD_BLOCKSIZE_MAX];
@@ -185,7 +187,7 @@ BYTE SEQUENCE_LLCODE[ZSTD_BLOCKSIZE_MAX];
 BYTE SEQUENCE_MLCODE[ZSTD_BLOCKSIZE_MAX];
 BYTE SEQUENCE_OFCODE[ZSTD_BLOCKSIZE_MAX];
 
-unsigned WKSP[HUF_WORKSPACE_SIZE_U32];
+U64 WKSP[HUF_WORKSPACE_SIZE_U64];
 
 typedef struct {
     size_t contentSize; /* 0 means unknown (unless contentSize == windowSize == 0) */
@@ -199,7 +201,7 @@ typedef struct {
     int hufInit;
     /* the distribution used in the previous block for repeat mode */
     BYTE hufDist[DISTSIZE];
-    HUF_CElt hufTable [256];
+    HUF_CElt hufTable [HUF_CTABLE_SIZE_ST(255)];
 
     int fseInit;
     FSE_CTable offcodeCTable  [FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
@@ -240,6 +242,10 @@ typedef enum {
   gt_block,      /* generate compressed blocks without block/frame headers */
 } genType_e;
 
+#ifndef MIN
+    #define MIN(a, b) ((a) < (b) ? (a) : (b))
+#endif
+
 /*-*******************************************************
 *  Global variables (set from command line)
 *********************************************************/
@@ -454,7 +460,7 @@ static size_t writeHufHeader(U32* seed, HUF_CElt* hufTable, void* dst, size_t ds
     }
 
     /* Write table description header */
-    {   size_t const hSize = HUF_writeCTable (op, dstSize, hufTable, maxSymbolValue, huffLog);
+    {   size_t const hSize = HUF_writeCTable_wksp (op, dstSize, hufTable, maxSymbolValue, huffLog, WKSP, sizeof(WKSP));
         if (hSize + 12 >= srcSize) return 0;   /* not useful to try compression */
         op += hSize;
     }
@@ -558,10 +564,10 @@ static size_t writeLiteralsBlockCompressed(U32* seed, frame_t* frame, size_t con
                     sizeFormat == 0
                             ? HUF_compress1X_usingCTable(
                                       op, opend - op, LITERAL_BUFFER, litSize,
-                                      frame->stats.hufTable)
+                                      frame->stats.hufTable, /* flags */ 0)
                             : HUF_compress4X_usingCTable(
                                       op, opend - op, LITERAL_BUFFER, litSize,
-                                      frame->stats.hufTable);
+                                      frame->stats.hufTable, /* flags */ 0);
             CHECKERR(compressedSize);
             /* this only occurs when it could not compress or similar */
         } while (compressedSize <= 0);
@@ -633,17 +639,16 @@ static inline void initSeqStore(seqStore_t *seqStore) {
 }
 
 /* Randomly generate sequence commands */
-static U32 generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
-                                size_t contentSize, size_t literalsSize, dictInfo info)
+static U32
+generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
+                  size_t contentSize, size_t literalsSize, dictInfo info)
 {
     /* The total length of all the matches */
     size_t const remainingMatch = contentSize - literalsSize;
     size_t excessMatch = 0;
     U32 numSequences = 0;
-
     U32 i;
 
-
     const BYTE* literals = LITERAL_BUFFER;
     BYTE* srcPtr = frame->src;
 
@@ -663,16 +668,16 @@ static U32 generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
          * ensure nice numbers */
         U32 matchLen =
                 MIN_SEQ_LEN +
-                ROUND(RAND_exp(seed, excessMatch / (double)(numSequences - i)));
+                ROUND(RAND_exp(seed, (double)excessMatch / (double)(numSequences - i)));
         U32 literalLen =
                 (RAND(seed) & 7)
                         ? ROUND(RAND_exp(seed,
-                                         literalsSize /
+                                         (double)literalsSize /
                                                  (double)(numSequences - i)))
                         : 0;
         /* actual offset, code to send, and point to copy up to when shifting
          * codes in the repeat offsets history */
-        U32 offset, offsetCode, repIndex;
+        U32 offset, offBase, repIndex;
 
         /* bounds checks */
         matchLen = (U32) MIN(matchLen, excessMatch + MIN_SEQ_LEN);
@@ -703,32 +708,31 @@ static U32 generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
                             lenPastStart = MIN(lenPastStart+MIN_SEQ_LEN, (U32)info.dictContentSize);
                             offset = (U32)((BYTE*)srcPtr - (BYTE*)frame->srcStart) + lenPastStart;
                         }
-                        {
-                            U32 const matchLenBound = MIN(frame->header.windowSize, lenPastStart);
+                        {   U32 const matchLenBound = MIN(frame->header.windowSize, lenPastStart);
                             matchLen = MIN(matchLen, matchLenBound);
                         }
                     }
                 }
-                offsetCode = offset + ZSTD_REP_MOVE;
+                offBase = OFFSET_TO_OFFBASE(offset);
                 repIndex = 2;
             } else {
                 /* do a repeat offset */
-                offsetCode = RAND(seed) % 3;
+                U32 const randomRepIndex = RAND(seed) % 3;
+                offBase = REPCODE_TO_OFFBASE(randomRepIndex + 1);  /* expects values between 1 & 3 */
                 if (literalLen > 0) {
-                    offset = frame->stats.rep[offsetCode];
-                    repIndex = offsetCode;
+                    offset = frame->stats.rep[randomRepIndex];
+                    repIndex = randomRepIndex;
                 } else {
-                    /* special case */
-                    offset = offsetCode == 2 ? frame->stats.rep[0] - 1
-                                           : frame->stats.rep[offsetCode + 1];
-                    repIndex = MIN(2, offsetCode + 1);
+                    /* special case : literalLen == 0 */
+                    offset = randomRepIndex == 2 ? frame->stats.rep[0] - 1
+                                           : frame->stats.rep[randomRepIndex + 1];
+                    repIndex = MIN(2, randomRepIndex + 1);
                 }
             }
         } while (((!info.useDict) && (offset > (size_t)((BYTE*)srcPtr - (BYTE*)frame->srcStart))) || offset == 0);
 
-        {
+        {   BYTE* const dictEnd = info.dictContent + info.dictContentSize;
             size_t j;
-            BYTE* const dictEnd = info.dictContent + info.dictContentSize;
             for (j = 0; j < matchLen; j++) {
                 if ((U32)((BYTE*)srcPtr - (BYTE*)frame->srcStart) < offset) {
                     /* copy from dictionary instead of literals */
@@ -739,8 +743,7 @@ static U32 generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
                     *srcPtr = *(srcPtr-offset);
                 }
                 srcPtr++;
-            }
-        }
+        }   }
 
         {   int r;
             for (r = repIndex; r > 0; r--) {
@@ -754,12 +757,12 @@ static U32 generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
         DISPLAYLEVEL(7, " srcPos: %8u seqNb: %3u",
                      (unsigned)((BYTE*)srcPtr - (BYTE*)frame->srcStart), (unsigned)i);
         DISPLAYLEVEL(6, "\n");
-        if (offsetCode < 3) {
+        if (OFFBASE_IS_REPCODE(offBase)) {  /* expects sumtype numeric representation of ZSTD_storeSeq() */
             DISPLAYLEVEL(7, "        repeat offset: %d\n", (int)repIndex);
         }
         /* use libzstd sequence handling */
         ZSTD_storeSeq(seqStore, literalLen, literals, literals + literalLen,
-                      offsetCode, matchLen - MINMATCH);
+                      offBase, matchLen);
 
         literalsSize -= literalLen;
         excessMatch -= (matchLen - MIN_SEQ_LEN);
@@ -768,8 +771,8 @@ static U32 generateSequences(U32* seed, frame_t* frame, seqStore_t* seqStore,
 
     memcpy(srcPtr, literals, literalsSize);
     srcPtr += literalsSize;
-    DISPLAYLEVEL(6, "      excess literals: %5u", (unsigned)literalsSize);
-    DISPLAYLEVEL(7, " srcPos: %8u", (unsigned)((BYTE*)srcPtr - (BYTE*)frame->srcStart));
+    DISPLAYLEVEL(6, "      excess literals: %5u ", (unsigned)literalsSize);
+    DISPLAYLEVEL(7, "srcPos: %8u ", (unsigned)((BYTE*)srcPtr - (BYTE*)frame->srcStart));
     DISPLAYLEVEL(6, "\n");
 
     return numSequences;
@@ -949,9 +952,9 @@ static size_t writeSequences(U32* seed, frame_t* frame, seqStore_t* seqStorePtr,
         FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
         BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
         if (MEM_32bits()) BIT_flushBits(&blockStream);
-        BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
+        BIT_addBits(&blockStream, sequences[nbSeq-1].mlBase, ML_bits[mlCodeTable[nbSeq-1]]);
         if (MEM_32bits()) BIT_flushBits(&blockStream);
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offBase, ofCodeTable[nbSeq-1]);
         BIT_flushBits(&blockStream);
 
         {   size_t n;
@@ -971,9 +974,9 @@ static size_t writeSequences(U32* seed, frame_t* frame, seqStore_t* seqStorePtr,
                     BIT_flushBits(&blockStream);                                /* (7)*/
                 BIT_addBits(&blockStream, sequences[n].litLength, llBits);
                 if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
-                BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
+                BIT_addBits(&blockStream, sequences[n].mlBase, mlBits);
                 if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
-                BIT_addBits(&blockStream, sequences[n].offset, ofBits);         /* 31 */
+                BIT_addBits(&blockStream, sequences[n].offBase, ofBits);         /* 31 */
                 BIT_flushBits(&blockStream);                                    /* (7)*/
         }   }
 
diff --git a/priv/zstd/tests/external_matchfinder.c b/priv/zstd/tests/external_matchfinder.c
new file mode 100644
index 0000000..76ad411
--- /dev/null
+++ b/priv/zstd/tests/external_matchfinder.c
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "external_matchfinder.h"
+#include 
+#include "zstd_compress_internal.h"
+
+#define HSIZE 1024
+static U32 const HLOG = 10;
+static U32 const MLS = 4;
+static U32 const BADIDX = 0xffffffff;
+
+static size_t simpleSequenceProducer(
+  void* sequenceProducerState,
+  ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
+  const void* src, size_t srcSize,
+  const void* dict, size_t dictSize,
+  int compressionLevel,
+  size_t windowSize
+) {
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    size_t seqCount = 0;
+    U32 hashTable[HSIZE];
+
+    (void)sequenceProducerState;
+    (void)dict;
+    (void)dictSize;
+    (void)outSeqsCapacity;
+    (void)compressionLevel;
+
+    {   int i;
+        for (i=0; i < HSIZE; i++) {
+            hashTable[i] = BADIDX;
+    }   }
+
+    while (ip + MLS < iend) {
+        size_t const hash = ZSTD_hashPtr(ip, HLOG, MLS);
+        U32 const matchIndex = hashTable[hash];
+        hashTable[hash] = (U32)(ip - istart);
+
+        if (matchIndex != BADIDX) {
+            const BYTE* const match = istart + matchIndex;
+            U32 const matchLen = (U32)ZSTD_count(ip, match, iend);
+            if (matchLen >= ZSTD_MINMATCH_MIN) {
+                U32 const litLen = (U32)(ip - anchor);
+                U32 const offset = (U32)(ip - match);
+                ZSTD_Sequence const seq = {
+                    offset, litLen, matchLen, 0
+                };
+
+                /* Note: it's crucial to stay within the window size! */
+                if (offset <= windowSize) {
+                    outSeqs[seqCount++] = seq;
+                    ip += matchLen;
+                    anchor = ip;
+                    continue;
+                }
+            }
+        }
+
+        ip++;
+    }
+
+    {   ZSTD_Sequence const finalSeq = {
+            0, (U32)(iend - anchor), 0, 0
+        };
+        outSeqs[seqCount++] = finalSeq;
+    }
+
+    return seqCount;
+}
+
+size_t zstreamSequenceProducer(
+  void* sequenceProducerState,
+  ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
+  const void* src, size_t srcSize,
+  const void* dict, size_t dictSize,
+  int compressionLevel,
+  size_t windowSize
+) {
+    EMF_testCase const testCase = *((EMF_testCase*)sequenceProducerState);
+    memset(outSeqs, 0, outSeqsCapacity);
+
+    switch (testCase) {
+        case EMF_ZERO_SEQS:
+            return 0;
+        case EMF_ONE_BIG_SEQ:
+            outSeqs[0].offset = 0;
+            outSeqs[0].matchLength = 0;
+            outSeqs[0].litLength = (U32)(srcSize);
+            return 1;
+         case EMF_LOTS_OF_SEQS:
+            return simpleSequenceProducer(
+                sequenceProducerState,
+                outSeqs, outSeqsCapacity,
+                src, srcSize,
+                dict, dictSize,
+                compressionLevel,
+                windowSize
+            );
+        case EMF_INVALID_OFFSET:
+            outSeqs[0].offset = 1 << 20;
+            outSeqs[0].matchLength = 4;
+            outSeqs[0].litLength = (U32)(srcSize - 4);
+            return 1;
+        case EMF_INVALID_MATCHLEN:
+            outSeqs[0].offset = 1;
+            outSeqs[0].matchLength = (U32)(srcSize);
+            outSeqs[0].litLength = 1;
+            return 1;
+        case EMF_INVALID_LITLEN:
+            outSeqs[0].offset = 0;
+            outSeqs[0].matchLength = 0;
+            outSeqs[0].litLength = (U32)(srcSize + 1);
+            return 1;
+        case EMF_INVALID_LAST_LITS:
+            outSeqs[0].offset = 1;
+            outSeqs[0].matchLength = 1;
+            outSeqs[0].litLength = 1;
+            outSeqs[1].offset = 0;
+            outSeqs[1].matchLength = 0;
+            outSeqs[1].litLength = (U32)(srcSize - 1);
+            return 2;
+        case EMF_SMALL_ERROR:
+            return outSeqsCapacity + 1;
+        case EMF_BIG_ERROR:
+        default:
+            return ZSTD_SEQUENCE_PRODUCER_ERROR;
+    }
+}
diff --git a/priv/zstd/tests/external_matchfinder.h b/priv/zstd/tests/external_matchfinder.h
new file mode 100644
index 0000000..e38dc25
--- /dev/null
+++ b/priv/zstd/tests/external_matchfinder.h
@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) Yann Collet, Meta Platforms, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef EXTERNAL_MATCHFINDER
+#define EXTERNAL_MATCHFINDER
+
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+
+/* See external_matchfinder.c for details on each test case */
+typedef enum {
+    EMF_ZERO_SEQS = 0,
+    EMF_ONE_BIG_SEQ = 1,
+    EMF_LOTS_OF_SEQS = 2,
+    EMF_BIG_ERROR = 3,
+    EMF_SMALL_ERROR = 4,
+    EMF_INVALID_OFFSET = 5,
+    EMF_INVALID_MATCHLEN = 6,
+    EMF_INVALID_LITLEN = 7,
+    EMF_INVALID_LAST_LITS = 8
+} EMF_testCase;
+
+size_t zstreamSequenceProducer(
+  void* sequenceProducerState,
+  ZSTD_Sequence* outSeqs, size_t outSeqsCapacity,
+  const void* src, size_t srcSize,
+  const void* dict, size_t dictSize,
+  int compressionLevel,
+  size_t windowSize
+);
+
+#endif /* EXTERNAL_MATCHFINDER */
diff --git a/priv/zstd/tests/fullbench.c b/priv/zstd/tests/fullbench.c
index a71117e..3a72d89 100644
--- a/priv/zstd/tests/fullbench.c
+++ b/priv/zstd/tests/fullbench.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -108,7 +108,25 @@ local_ZSTD_compress(const void* src, size_t srcSize,
     p.fParams = f;
     p.cParams = *(ZSTD_compressionParameters*)payload;
     return ZSTD_compress_advanced (g_zcc, dst, dstSize, src, srcSize, NULL ,0, p);
-    //return ZSTD_compress(dst, dstSize, src, srcSize, cLevel);
+}
+
+static size_t
+local_ZSTD_compress_freshCCtx(const void* src, size_t srcSize,
+                    void* dst, size_t dstSize,
+                    void* payload)
+{
+    ZSTD_parameters p;
+    ZSTD_frameParameters f = { 1 /* contentSizeHeader*/, 0, 0 };
+    p.fParams = f;
+    p.cParams = *(ZSTD_compressionParameters*)payload;
+    if (g_zcc != NULL) ZSTD_freeCCtx(g_zcc);
+    g_zcc = ZSTD_createCCtx();
+    assert(g_zcc != NULL);
+    {   size_t const r = ZSTD_compress_advanced (g_zcc, dst, dstSize, src, srcSize, NULL ,0, p);
+        ZSTD_freeCCtx(g_zcc);
+        g_zcc = NULL;
+        return r;
+    }
 }
 
 static size_t g_cSize = 0;
@@ -123,11 +141,15 @@ static size_t local_ZSTD_decompress(const void* src, size_t srcSize,
 static ZSTD_DCtx* g_zdc = NULL;
 
 #ifndef ZSTD_DLL_IMPORT
-extern size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* ctx, const void* src, size_t srcSize);
+typedef enum {
+    not_streaming = 0,
+    is_streaming = 1
+} streaming_operation;
+extern size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* ctx, const void* src, size_t srcSize, void* dst, size_t dstCapacity, const streaming_operation streaming);
 static size_t local_ZSTD_decodeLiteralsBlock(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2)
 {
     (void)src; (void)srcSize; (void)dst; (void)dstSize;
-    return ZSTD_decodeLiteralsBlock(g_zdc, buff2, g_cSize);
+    return ZSTD_decodeLiteralsBlock(g_zdc, buff2, g_cSize, dst, dstSize, not_streaming);
 }
 
 static size_t local_ZSTD_decodeSeqHeaders(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2)
@@ -149,6 +171,7 @@ FORCE_NOINLINE size_t ZSTD_decodeLiteralsHeader(ZSTD_DCtx* dctx, void const* src
                 size_t lhSize, litSize, litCSize;
                 U32 const lhlCode = (istart[0] >> 2) & 3;
                 U32 const lhc = MEM_readLE32(istart);
+                int const flags = ZSTD_DCtx_get_bmi2(dctx) ? HUF_flags_bmi2 : 0;
                 switch(lhlCode)
                 {
                 case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
@@ -173,16 +196,16 @@ FORCE_NOINLINE size_t ZSTD_decodeLiteralsHeader(ZSTD_DCtx* dctx, void const* src
                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
                 RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
 #ifndef HUF_FORCE_DECOMPRESS_X2
-                return HUF_readDTableX1_wksp_bmi2(
+                return HUF_readDTableX1_wksp(
                         dctx->entropy.hufTable,
                         istart+lhSize, litCSize,
                         dctx->workspace, sizeof(dctx->workspace),
-                        dctx->bmi2);
+                        flags);
 #else
                 return HUF_readDTableX2_wksp(
                         dctx->entropy.hufTable,
                         istart+lhSize, litCSize,
-                        dctx->workspace, sizeof(dctx->workspace));
+                        dctx->workspace, sizeof(dctx->workspace), flags);
 #endif
             }
         }
@@ -226,14 +249,15 @@ local_ZSTD_compressStream_freshCCtx(const void* src, size_t srcSize,
                           void* dst, size_t dstCapacity,
                           void* payload)
 {
-    ZSTD_CCtx* const cctx = ZSTD_createCCtx();
-    size_t r;
-    assert(cctx != NULL);
-
-    r = local_ZSTD_compressStream(src, srcSize, dst, dstCapacity, payload);
-
-    ZSTD_freeCCtx(cctx);
-    return r;
+    if (g_cstream != NULL) ZSTD_freeCCtx(g_cstream);
+    g_cstream = ZSTD_createCCtx();
+    assert(g_cstream != NULL);
+
+    {   size_t const r = local_ZSTD_compressStream(src, srcSize, dst, dstCapacity, payload);
+        ZSTD_freeCCtx(g_cstream);
+        g_cstream = NULL;
+        return r;
+    }
 }
 
 static size_t
@@ -426,6 +450,9 @@ static int benchMem(unsigned benchNb,
     case 2:
         benchFunction = local_ZSTD_decompress; benchName = "decompress";
         break;
+    case 3:
+        benchFunction = local_ZSTD_compress_freshCCtx; benchName = "compress_freshCCtx";
+        break;
 #ifndef ZSTD_DLL_IMPORT
     case 11:
         benchFunction = local_ZSTD_compressContinue; benchName = "compressContinue";
@@ -504,7 +531,6 @@ static int benchMem(unsigned benchNb,
     ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_targetLength, (int)cparams.targetLength);
     ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_strategy, cparams.strategy);
 
-
     ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_compressionLevel, cLevel);
     ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_windowLog, (int)cparams.windowLog);
     ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_hashLog, (int)cparams.hashLog);
@@ -523,6 +549,9 @@ static int benchMem(unsigned benchNb,
     case 2:
         g_cSize = ZSTD_compress(dstBuff2, dstBuffSize, src, srcSize, cLevel);
         break;
+    case 3:
+        payload = &cparams;
+        break;
 #ifndef ZSTD_DLL_IMPORT
     case 11:
         payload = &cparams;
@@ -577,7 +606,7 @@ static int benchMem(unsigned benchNb,
             ip += ZSTD_blockHeaderSize;    /* skip block header */
             ZSTD_decompressBegin(g_zdc);
             CONTROL(iend > ip);
-            ip += ZSTD_decodeLiteralsBlock(g_zdc, ip, (size_t)(iend-ip));   /* skip literal segment */
+            ip += ZSTD_decodeLiteralsBlock(g_zdc, ip, (size_t)(iend-ip), dstBuff, dstBuffSize, not_streaming);   /* skip literal segment */
             g_cSize = (size_t)(iend-ip);
             memcpy(dstBuff2, ip, g_cSize);   /* copy rest of block (it starts by SeqHeader) */
             srcSize = srcSize > 128 KB ? 128 KB : srcSize;   /* speed relative to block */
diff --git a/priv/zstd/tests/fuzz/.gitignore b/priv/zstd/tests/fuzz/.gitignore
index 93d935a..28f2b8d 100644
--- a/priv/zstd/tests/fuzz/.gitignore
+++ b/priv/zstd/tests/fuzz/.gitignore
@@ -17,6 +17,8 @@ decompress_dstSize_tooSmall
 fse_read_ncount
 sequence_compression_api
 seekable_roundtrip
+huf_decompress
+huf_round_trip
 fuzz-*.log
 rt_lib_*
 d_lib_*
diff --git a/priv/zstd/tests/fuzz/Makefile b/priv/zstd/tests/fuzz/Makefile
index ccb574b..bbb262a 100644
--- a/priv/zstd/tests/fuzz/Makefile
+++ b/priv/zstd/tests/fuzz/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -12,6 +12,7 @@
 CFLAGS ?= -O3
 CXXFLAGS ?= -O3
 CPPFLAGS ?=
+ASFLAGS ?=
 LDFLAGS ?=
 ARFLAGS ?=
 LIB_FUZZING_ENGINE ?= libregression.a
@@ -23,20 +24,32 @@ else
 endif
 CORPORA_URL_PREFIX:=https://github.com/facebook/zstd/releases/download/fuzz-corpora/
 
+LIBZSTD = ../../lib
+DEBUGLEVEL ?= 2
+ZSTD_LEGACY_SUPPORT ?= 1
+
+include $(LIBZSTD)/libzstd.mk
+
 ZSTDDIR = ../../lib
 PRGDIR = ../../programs
 CONTRIBDIR = ../../contrib
 
+# TODO(embg) make it possible to plug in an arbitrary matchfinder as a .o file
+MATCHFINDER_DIR = $(CONTRIBDIR)/externalSequenceProducer
+MATCHFINDER_SRC = $(MATCHFINDER_DIR)/sequence_producer.c
+
 FUZZ_CPPFLAGS := -I$(ZSTDDIR) -I$(ZSTDDIR)/common -I$(ZSTDDIR)/compress \
 	-I$(ZSTDDIR)/dictBuilder -I$(ZSTDDIR)/deprecated -I$(ZSTDDIR)/legacy \
-	-I$(CONTRIBDIR)/seekable_format -I$(PRGDIR) -DZSTD_MULTITHREAD -DZSTD_LEGACY_SUPPORT=1 $(CPPFLAGS)
+	-I$(CONTRIBDIR)/seekable_format -I$(PRGDIR) -I$(MATCHFINDER_DIR) \
+	-DZSTD_MULTITHREAD -DZSTD_LEGACY_SUPPORT=1 $(CPPFLAGS)
 FUZZ_EXTRA_FLAGS := -Wall -Wextra -Wcast-qual -Wcast-align -Wshadow \
 	-Wstrict-aliasing=1 -Wswitch-enum -Wdeclaration-after-statement \
 	-Wstrict-prototypes -Wundef \
 	-Wvla -Wformat=2 -Winit-self -Wfloat-equal -Wwrite-strings \
-	-Wredundant-decls \
+	-Wredundant-decls -Wno-deprecated-declarations \
 	-g -fno-omit-frame-pointer
 FUZZ_CFLAGS := $(FUZZ_EXTRA_FLAGS) $(CFLAGS)
+FUZZ_ASFLAGS := $(FUZZ_EXTRA_FLAGS) $(ASFLAGS)
 FUZZ_CXXFLAGS := $(FUZZ_EXTRA_FLAGS) -std=c++11 $(CXXFLAGS)
 FUZZ_LDFLAGS := -pthread $(LDFLAGS)
 FUZZ_ARFLAGS := $(ARFLAGS)
@@ -50,19 +63,20 @@ FUZZ_SRC := $(PRGDIR)/util.c ./fuzz_helpers.c ./zstd_helpers.c ./fuzz_data_produ
 SEEKABLE_HEADERS = $(CONTRIBDIR)/seekable_format/zstd_seekable.h
 SEEKABLE_OBJS = $(CONTRIBDIR)/seekable_format/zstdseek_compress.c $(CONTRIBDIR)/seekable_format/zstdseek_decompress.c
 
-ZSTDCOMMON_SRC := $(ZSTDDIR)/common/*.c
-ZSTDCOMP_SRC   := $(ZSTDDIR)/compress/*.c
-ZSTDDECOMP_SRC := $(ZSTDDIR)/decompress/*.c
-ZSTDDICT_SRC := $(ZSTDDIR)/dictBuilder/*.c
-ZSTDLEGACY_SRC := $(ZSTDDIR)/legacy/*.c
+ZSTDCOMMON_SRC := $(ZSTD_COMMON_FILES)
+ZSTDCOMP_SRC   := $(ZSTD_COMPRESS_FILES)
+ZSTDDECOMP_SRC := $(ZSTD_DECOMPRESS_FILES)
+ZSTDDICT_SRC := $(ZSTD_DICTBUILDER_FILES)
+ZSTDLEGACY_SRC := $(ZSTD_LEGACY_FILES)
 FUZZ_SRC       := \
 	$(FUZZ_SRC) \
 	$(ZSTDDECOMP_SRC) \
 	$(ZSTDCOMMON_SRC) \
 	$(ZSTDCOMP_SRC) \
 	$(ZSTDDICT_SRC) \
-	$(ZSTDLEGACY_SRC)
-FUZZ_SRC := $(wildcard $(FUZZ_SRC))
+	$(ZSTDLEGACY_SRC) \
+	$(MATCHFINDER_SRC)
+FUZZ_SRC := $(sort $(wildcard $(FUZZ_SRC)))
 
 FUZZ_D_OBJ1 := $(subst $(ZSTDDIR)/common/,d_lib_common_,$(FUZZ_SRC))
 FUZZ_D_OBJ2 := $(subst $(ZSTDDIR)/compress/,d_lib_compress_,$(FUZZ_D_OBJ1))
@@ -70,8 +84,10 @@ FUZZ_D_OBJ3 := $(subst $(ZSTDDIR)/decompress/,d_lib_decompress_,$(FUZZ_D_OBJ2))
 FUZZ_D_OBJ4 := $(subst $(ZSTDDIR)/dictBuilder/,d_lib_dictBuilder_,$(FUZZ_D_OBJ3))
 FUZZ_D_OBJ5 := $(subst $(ZSTDDIR)/legacy/,d_lib_legacy_,$(FUZZ_D_OBJ4))
 FUZZ_D_OBJ6 := $(subst $(PRGDIR)/,d_prg_,$(FUZZ_D_OBJ5))
-FUZZ_D_OBJ7 := $(subst $\./,d_fuzz_,$(FUZZ_D_OBJ6))
-FUZZ_DECOMPRESS_OBJ := $(FUZZ_D_OBJ7:.c=.o)
+FUZZ_D_OBJ7 := $(subst $(MATCHFINDER_DIR)/,d_matchfinder_,$(FUZZ_D_OBJ6))
+FUZZ_D_OBJ8 := $(subst $\./,d_fuzz_,$(FUZZ_D_OBJ7))
+FUZZ_D_OBJ9 := $(FUZZ_D_OBJ8:.c=.o)
+FUZZ_DECOMPRESS_OBJ := $(FUZZ_D_OBJ9:.S=.o)
 
 FUZZ_RT_OBJ1 := $(subst $(ZSTDDIR)/common/,rt_lib_common_,$(FUZZ_SRC))
 FUZZ_RT_OBJ2 := $(subst $(ZSTDDIR)/compress/,rt_lib_compress_,$(FUZZ_RT_OBJ1))
@@ -79,8 +95,10 @@ FUZZ_RT_OBJ3 := $(subst $(ZSTDDIR)/decompress/,rt_lib_decompress_,$(FUZZ_RT_OBJ2
 FUZZ_RT_OBJ4 := $(subst $(ZSTDDIR)/dictBuilder/,rt_lib_dictBuilder_,$(FUZZ_RT_OBJ3))
 FUZZ_RT_OBJ5 := $(subst $(ZSTDDIR)/legacy/,rt_lib_legacy_,$(FUZZ_RT_OBJ4))
 FUZZ_RT_OBJ6 := $(subst $(PRGDIR)/,rt_prg_,$(FUZZ_RT_OBJ5))
-FUZZ_RT_OBJ7 := $(subst $\./,rt_fuzz_,$(FUZZ_RT_OBJ6))
-FUZZ_ROUND_TRIP_OBJ := $(FUZZ_RT_OBJ7:.c=.o)
+FUZZ_RT_OBJ7 := $(subst $(MATCHFINDER_DIR)/,rt_matchfinder_,$(FUZZ_RT_OBJ6))
+FUZZ_RT_OBJ8 := $(subst $\./,rt_fuzz_,$(FUZZ_RT_OBJ7))
+FUZZ_RT_OBJ9 := $(FUZZ_RT_OBJ8:.c=.o)
+FUZZ_ROUND_TRIP_OBJ := $(FUZZ_RT_OBJ9:.S=.o)
 
 .PHONY: default all clean cleanall
 
@@ -103,7 +121,9 @@ FUZZ_TARGETS :=       \
 	decompress_dstSize_tooSmall \
 	fse_read_ncount \
 	sequence_compression_api \
-	seekable_roundtrip
+	seekable_roundtrip \
+	huf_round_trip \
+	huf_decompress
 
 all: libregression.a $(FUZZ_TARGETS)
 
@@ -116,6 +136,9 @@ rt_lib_compress_%.o: $(ZSTDDIR)/compress/%.c
 rt_lib_decompress_%.o: $(ZSTDDIR)/decompress/%.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $(FUZZ_ROUND_TRIP_FLAGS) $< -c -o $@
 
+rt_lib_decompress_%.o: $(ZSTDDIR)/decompress/%.S
+	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_ASFLAGS) $(FUZZ_ROUND_TRIP_FLAGS) $< -c -o $@
+
 rt_lib_dictBuilder_%.o: $(ZSTDDIR)/dictBuilder/%.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $(FUZZ_ROUND_TRIP_FLAGS) $< -c -o $@
 
@@ -128,6 +151,9 @@ rt_prg_%.o: $(PRGDIR)/%.c
 rt_fuzz_%.o: %.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $(FUZZ_ROUND_TRIP_FLAGS) $< -c -o $@
 
+rt_matchfinder_%.o: $(MATCHFINDER_DIR)/%.c
+	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $(FUZZ_ROUND_TRIP_FLAGS) $< -c -o $@
+
 d_lib_common_%.o: $(ZSTDDIR)/common/%.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $< -c -o $@
 
@@ -137,6 +163,9 @@ d_lib_compress_%.o: $(ZSTDDIR)/compress/%.c
 d_lib_decompress_%.o: $(ZSTDDIR)/decompress/%.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $< -c -o $@
 
+d_lib_decompress_%.o: $(ZSTDDIR)/decompress/%.S
+	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_ASFLAGS) $< -c -o $@
+
 d_lib_dictBuilder_%.o: $(ZSTDDIR)/dictBuilder/%.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $< -c -o $@
 
@@ -149,6 +178,9 @@ d_prg_%.o: $(PRGDIR)/%.c
 d_fuzz_%.o: %.c
 	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $< -c -o $@
 
+d_matchfinder_%.o: $(MATCHFINDER_DIR)/%.c
+	$(CC) $(FUZZ_CPPFLAGS) $(FUZZ_CFLAGS) $< -c -o $@
+
 simple_round_trip: $(FUZZ_HEADERS) $(FUZZ_ROUND_TRIP_OBJ) rt_fuzz_simple_round_trip.o
 	$(CXX) $(FUZZ_TARGET_FLAGS) $(FUZZ_ROUND_TRIP_OBJ) rt_fuzz_simple_round_trip.o $(LIB_FUZZING_ENGINE) -o $@
 
@@ -200,6 +232,12 @@ sequence_compression_api: $(FUZZ_HEADERS) $(FUZZ_ROUND_TRIP_OBJ) rt_fuzz_sequenc
 seekable_roundtrip: $(FUZZ_HEADERS) $(SEEKABLE_HEADERS) $(FUZZ_ROUND_TRIP_OBJ) $(SEEKABLE_OBJS)  rt_fuzz_seekable_roundtrip.o
 	$(CXX) $(FUZZ_TARGET_FLAGS) $(FUZZ_ROUND_TRIP_OBJ) $(SEEKABLE_OBJS) rt_fuzz_seekable_roundtrip.o $(LIB_FUZZING_ENGINE) -o $@
 
+huf_round_trip: $(FUZZ_HEADERS) $(FUZZ_ROUND_TRIP_OBJ) rt_fuzz_huf_round_trip.o
+	$(CXX) $(FUZZ_TARGET_FLAGS) $(FUZZ_ROUND_TRIP_OBJ) rt_fuzz_huf_round_trip.o $(LIB_FUZZING_ENGINE) -o $@
+
+huf_decompress: $(FUZZ_HEADERS) $(FUZZ_DECOMPRESS_OBJ) d_fuzz_huf_decompress.o
+	$(CXX) $(FUZZ_TARGET_FLAGS) $(FUZZ_DECOMPRESS_OBJ) d_fuzz_huf_decompress.o $(LIB_FUZZING_ENGINE) -o $@
+
 libregression.a: $(FUZZ_HEADERS) $(PRGDIR)/util.h $(PRGDIR)/util.c d_fuzz_regression_driver.o
 	$(AR) $(FUZZ_ARFLAGS) $@ d_fuzz_regression_driver.o
 
diff --git a/priv/zstd/tests/fuzz/README.md b/priv/zstd/tests/fuzz/README.md
index 71afa40..4ff7fe3 100644
--- a/priv/zstd/tests/fuzz/README.md
+++ b/priv/zstd/tests/fuzz/README.md
@@ -50,6 +50,20 @@ See the help of the relevant command for options.
 Flags not parsed by `fuzz.py` are passed to the fuzzing engine.
 The command used to run the fuzzer is printed for debugging.
 
+Here's a helpful command to fuzz each target across all cores,
+stopping only if a bug is found:
+```
+for target in $(./fuzz.py list); do
+    ./fuzz.py libfuzzer $target -jobs=10 -workers=10 -max_total_time=1000 || break;
+done
+```
+Alternatively, you can fuzz all targets in parallel, using one core per target:
+```
+python3 ./fuzz.py list | xargs -P$(python3 ./fuzz.py list | wc -l) -I__ sh -c "python3 ./fuzz.py libfuzzer __ 2>&1 | tee __.log"
+```
+Either way, to double-check that no crashes were found, run `ls corpora/*crash`. 
+If any crashes were found, you can use the hashes to reproduce them.
+
 ## LibFuzzer
 
 ```
diff --git a/priv/zstd/tests/fuzz/block_decompress.c b/priv/zstd/tests/fuzz/block_decompress.c
index bdbf769..e4767b3 100644
--- a/priv/zstd/tests/fuzz/block_decompress.c
+++ b/priv/zstd/tests/fuzz/block_decompress.c
@@ -1,5 +1,5 @@
 /**
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/block_round_trip.c b/priv/zstd/tests/fuzz/block_round_trip.c
index 46a84c7..54012a1 100644
--- a/priv/zstd/tests/fuzz/block_round_trip.c
+++ b/priv/zstd/tests/fuzz/block_round_trip.c
@@ -1,5 +1,5 @@
 /**
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/decompress_dstSize_tooSmall.c b/priv/zstd/tests/fuzz/decompress_dstSize_tooSmall.c
index 3f7607b..8ad3ea1 100644
--- a/priv/zstd/tests/fuzz/decompress_dstSize_tooSmall.c
+++ b/priv/zstd/tests/fuzz/decompress_dstSize_tooSmall.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/dictionary_decompress.c b/priv/zstd/tests/fuzz/dictionary_decompress.c
index 33c58c8..8dea3e5 100644
--- a/priv/zstd/tests/fuzz/dictionary_decompress.c
+++ b/priv/zstd/tests/fuzz/dictionary_decompress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/dictionary_loader.c b/priv/zstd/tests/fuzz/dictionary_loader.c
index 5b60bc4..1ac2274 100644
--- a/priv/zstd/tests/fuzz/dictionary_loader.c
+++ b/priv/zstd/tests/fuzz/dictionary_loader.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/dictionary_round_trip.c b/priv/zstd/tests/fuzz/dictionary_round_trip.c
index 7bff4bd..6dd78c3 100644
--- a/priv/zstd/tests/fuzz/dictionary_round_trip.c
+++ b/priv/zstd/tests/fuzz/dictionary_round_trip.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -42,8 +42,23 @@ static size_t roundTripTest(void *result, size_t resultCapacity,
                 src, srcSize,
                 dict.buff, dict.size,
                 cLevel);
+        FUZZ_ZASSERT(cSize);
+        // Compress a second time and check for determinism
+        {
+            size_t const cSize0 = cSize;
+            XXH64_hash_t const hash0 = XXH64(compressed, cSize, 0);
+            cSize = ZSTD_compress_usingDict(cctx,
+                    compressed, compressedCapacity,
+                    src, srcSize,
+                    dict.buff, dict.size,
+                    cLevel);
+            FUZZ_ASSERT(cSize == cSize0);
+            FUZZ_ASSERT(XXH64(compressed, cSize, 0) == hash0);
+        }
     } else {
+        size_t remainingBytes;
         dictContentType = FUZZ_dataProducer_uint32Range(producer, 0, 2);
+        remainingBytes = FUZZ_dataProducer_remainingBytes(producer);
         FUZZ_setRandomParameters(cctx, srcSize, producer);
         /* Disable checksum so we can use sizes smaller than compress bound. */
         FUZZ_ZASSERT(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 0));
@@ -51,14 +66,29 @@ static size_t roundTripTest(void *result, size_t resultCapacity,
             FUZZ_ZASSERT(ZSTD_CCtx_refPrefix_advanced(
                 cctx, dict.buff, dict.size,
                 dictContentType));
-        else 
+        else
             FUZZ_ZASSERT(ZSTD_CCtx_loadDictionary_advanced(
                 cctx, dict.buff, dict.size,
                 (ZSTD_dictLoadMethod_e)FUZZ_dataProducer_uint32Range(producer, 0, 1),
                 dictContentType));
         cSize = ZSTD_compress2(cctx, compressed, compressedCapacity, src, srcSize);
+        FUZZ_ZASSERT(cSize);
+        // Compress a second time and check for determinism
+        {
+            size_t const cSize0 = cSize;
+            XXH64_hash_t const hash0 = XXH64(compressed, cSize, 0);
+            FUZZ_dataProducer_rollBack(producer, remainingBytes);
+            FUZZ_setRandomParameters(cctx, srcSize, producer);
+            FUZZ_ZASSERT(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 0));
+            if (refPrefix)
+                FUZZ_ZASSERT(ZSTD_CCtx_refPrefix_advanced(
+                    cctx, dict.buff, dict.size,
+                    dictContentType));
+            cSize = ZSTD_compress2(cctx, compressed, compressedCapacity, src, srcSize);
+            FUZZ_ASSERT(cSize == cSize0);
+            FUZZ_ASSERT(XXH64(compressed, cSize, 0) == hash0);
+        }
     }
-    FUZZ_ZASSERT(cSize);
     if (refPrefix)
         FUZZ_ZASSERT(ZSTD_DCtx_refPrefix_advanced(
             dctx, dict.buff, dict.size,
diff --git a/priv/zstd/tests/fuzz/dictionary_stream_round_trip.c b/priv/zstd/tests/fuzz/dictionary_stream_round_trip.c
index 9af712f..3ebbd84 100644
--- a/priv/zstd/tests/fuzz/dictionary_stream_round_trip.c
+++ b/priv/zstd/tests/fuzz/dictionary_stream_round_trip.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/fse_read_ncount.c b/priv/zstd/tests/fuzz/fse_read_ncount.c
index c323860..29e1944 100644
--- a/priv/zstd/tests/fuzz/fse_read_ncount.c
+++ b/priv/zstd/tests/fuzz/fse_read_ncount.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/fuzz.h b/priv/zstd/tests/fuzz/fuzz.h
index 810daa2..d1e439f 100644
--- a/priv/zstd/tests/fuzz/fuzz.h
+++ b/priv/zstd/tests/fuzz/fuzz.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -26,8 +26,7 @@
  * @param MEM_FORCE_MEMORY_ACCESS:
  *        This flag controls how the zstd library accesses unaligned memory.
  *        It can be undefined, or 0 through 2. If it is undefined, it selects
- *        the method to use based on the compiler. If testing with UBSAN set
- *        MEM_FORCE_MEMORY_ACCESS=0 to use the standard compliant method.
+ *        the method to use based on the compiler.
  * @param FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
  *        This is the canonical flag to enable deterministic builds for fuzzing.
  *        Changes to zstd for fuzzing are gated behind this define.
diff --git a/priv/zstd/tests/fuzz/fuzz.py b/priv/zstd/tests/fuzz/fuzz.py
index d8dfa77..03ffeee 100755
--- a/priv/zstd/tests/fuzz/fuzz.py
+++ b/priv/zstd/tests/fuzz/fuzz.py
@@ -1,7 +1,7 @@
 #!/usr/bin/env python
 
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -63,6 +63,8 @@ def __init__(self, input_type, frame_type=FrameType.ZSTD):
     'fse_read_ncount': TargetInfo(InputType.RAW_DATA),
     'sequence_compression_api': TargetInfo(InputType.RAW_DATA),
     'seekable_roundtrip': TargetInfo(InputType.RAW_DATA),
+    'huf_round_trip': TargetInfo(InputType.RAW_DATA),
+    'huf_decompress': TargetInfo(InputType.RAW_DATA),
 }
 TARGETS = list(TARGET_INFO.keys())
 ALL_TARGETS = TARGETS + ['all']
@@ -634,7 +636,7 @@ def regression(args):
     try:
         description = """
         Runs one or more regression tests.
-        The fuzzer should have been built with with
+        The fuzzer should have been built with
         LIB_FUZZING_ENGINE='libregression.a'.
         Takes input from CORPORA.
         """
diff --git a/priv/zstd/tests/fuzz/fuzz_data_producer.c b/priv/zstd/tests/fuzz/fuzz_data_producer.c
index beb0155..a93e8ba 100644
--- a/priv/zstd/tests/fuzz/fuzz_data_producer.c
+++ b/priv/zstd/tests/fuzz/fuzz_data_producer.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -66,6 +66,12 @@ size_t FUZZ_dataProducer_remainingBytes(FUZZ_dataProducer_t *producer){
     return producer->size;
 }
 
+void FUZZ_dataProducer_rollBack(FUZZ_dataProducer_t *producer, size_t remainingBytes)
+{
+    FUZZ_ASSERT(remainingBytes >= producer->size);
+    producer->size = remainingBytes;
+}
+
 int FUZZ_dataProducer_empty(FUZZ_dataProducer_t *producer) {
     return producer->size == 0;
 }
diff --git a/priv/zstd/tests/fuzz/fuzz_data_producer.h b/priv/zstd/tests/fuzz/fuzz_data_producer.h
index 045aaff..8ca501f 100644
--- a/priv/zstd/tests/fuzz/fuzz_data_producer.h
+++ b/priv/zstd/tests/fuzz/fuzz_data_producer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -49,6 +49,9 @@ int32_t FUZZ_dataProducer_int32Range(FUZZ_dataProducer_t *producer,
 /* Returns the size of the remaining bytes of data in the producer */
 size_t FUZZ_dataProducer_remainingBytes(FUZZ_dataProducer_t *producer);
 
+/* Rolls back the data producer state to have remainingBytes remaining */
+void FUZZ_dataProducer_rollBack(FUZZ_dataProducer_t *producer, size_t remainingBytes);
+
 /* Returns true if the data producer is out of bytes */
 int FUZZ_dataProducer_empty(FUZZ_dataProducer_t *producer);
 
diff --git a/priv/zstd/tests/fuzz/fuzz_helpers.c b/priv/zstd/tests/fuzz/fuzz_helpers.c
index 61c0deb..1b6ad97 100644
--- a/priv/zstd/tests/fuzz/fuzz_helpers.c
+++ b/priv/zstd/tests/fuzz/fuzz_helpers.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/fuzz_helpers.h b/priv/zstd/tests/fuzz/fuzz_helpers.h
index c180478..aaf4c1d 100644
--- a/priv/zstd/tests/fuzz/fuzz_helpers.h
+++ b/priv/zstd/tests/fuzz/fuzz_helpers.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/huf_decompress.c b/priv/zstd/tests/fuzz/huf_decompress.c
new file mode 100644
index 0000000..fcd4b1a
--- /dev/null
+++ b/priv/zstd/tests/fuzz/huf_decompress.c
@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/**
+ * This fuzz target performs a zstd round-trip test (compress & decompress),
+ * compares the result with the original, and calls abort() on corruption.
+ */
+
+#include 
+#include 
+#include 
+#include 
+#include "common/cpu.h"
+#include "common/huf.h"
+#include "fuzz_helpers.h"
+#include "fuzz_data_producer.h"
+
+int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
+{
+    FUZZ_dataProducer_t *producer = FUZZ_dataProducer_create(src, size);
+    /* Select random parameters: #streams, X1 or X2 decoding, bmi2 */
+    int const streams = FUZZ_dataProducer_int32Range(producer, 0, 1);
+    int const symbols = FUZZ_dataProducer_int32Range(producer, 0, 1);
+    int const flags = 0
+        | (ZSTD_cpuid_bmi2(ZSTD_cpuid()) && FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_bmi2 : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_optimalDepth : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_preferRepeat : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_suspectUncompressible : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_disableAsm : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_disableFast : 0);
+    /* Select a random cBufSize - it may be too small */
+    size_t const dBufSize = FUZZ_dataProducer_uint32Range(producer, 0, 8 * size + 500);
+    size_t const maxTableLog = FUZZ_dataProducer_uint32Range(producer, 1, HUF_TABLELOG_MAX);
+    HUF_DTable* dt = (HUF_DTable*)FUZZ_malloc(HUF_DTABLE_SIZE(maxTableLog) * sizeof(HUF_DTable));
+    size_t const wkspSize = HUF_WORKSPACE_SIZE;
+    void* wksp = FUZZ_malloc(wkspSize);
+    void* dBuf = FUZZ_malloc(dBufSize);
+    dt[0] = maxTableLog * 0x01000001;
+    size = FUZZ_dataProducer_remainingBytes(producer);
+
+    if (symbols == 0) {
+        size_t const err = HUF_readDTableX1_wksp(dt, src, size, wksp, wkspSize, flags);
+        if (ZSTD_isError(err))
+            goto _out;
+    } else {
+        size_t const err = HUF_readDTableX2_wksp(dt, src, size, wksp, wkspSize, flags);
+        if (ZSTD_isError(err))
+            goto _out;
+    }
+    if (streams == 0)
+        HUF_decompress1X_usingDTable(dBuf, dBufSize, src, size, dt, flags);
+    else
+        HUF_decompress4X_usingDTable(dBuf, dBufSize, src, size, dt, flags);
+
+_out:
+    free(dt);
+    free(wksp);
+    free(dBuf);
+    FUZZ_dataProducer_free(producer);
+    return 0;
+}
diff --git a/priv/zstd/tests/fuzz/huf_round_trip.c b/priv/zstd/tests/fuzz/huf_round_trip.c
new file mode 100644
index 0000000..4d0f8de
--- /dev/null
+++ b/priv/zstd/tests/fuzz/huf_round_trip.c
@@ -0,0 +1,137 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/**
+ * This fuzz target performs a zstd round-trip test (compress & decompress),
+ * compares the result with the original, and calls abort() on corruption.
+ */
+
+#include 
+#include 
+#include 
+#include 
+#include "common/cpu.h"
+#include "compress/hist.h"
+#include "common/huf.h"
+#include "fuzz_helpers.h"
+#include "fuzz_data_producer.h"
+#include "common/bits.h"
+
+static size_t adjustTableLog(size_t tableLog, size_t maxSymbol)
+{
+    size_t const alphabetSize = maxSymbol + 1;
+    size_t minTableLog = ZSTD_highbit32(alphabetSize) + 1;
+    if ((alphabetSize & (alphabetSize - 1)) != 0) {
+        ++minTableLog;
+    }
+    assert(minTableLog <= 9);
+    if (tableLog < minTableLog)
+        return minTableLog;
+    else
+        return tableLog;
+}
+
+int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
+{
+    FUZZ_dataProducer_t *producer = FUZZ_dataProducer_create(src, size);
+    /* Select random parameters: #streams, X1 or X2 decoding, bmi2 */
+    int const streams = FUZZ_dataProducer_int32Range(producer, 0, 1);
+    int const symbols = FUZZ_dataProducer_int32Range(producer, 0, 1);
+    int const flags = 0
+        | (ZSTD_cpuid_bmi2(ZSTD_cpuid()) && FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_bmi2 : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_optimalDepth : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_preferRepeat : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_suspectUncompressible : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_disableAsm : 0)
+        | (FUZZ_dataProducer_int32Range(producer, 0, 1) ? HUF_flags_disableFast : 0);
+    /* Select a random cBufSize - it may be too small */
+    size_t const cBufSize = FUZZ_dataProducer_uint32Range(producer, 0, 4 * size);
+    /* Select a random tableLog - we'll adjust it up later */
+    size_t tableLog = FUZZ_dataProducer_uint32Range(producer, 1, 12);
+    size_t const kMaxSize = 256 * 1024;
+    size = FUZZ_dataProducer_remainingBytes(producer);
+    if (size > kMaxSize)
+        size = kMaxSize;
+
+    if (size <= 1) {
+        FUZZ_dataProducer_free(producer);
+        return 0;
+    }
+
+    uint32_t maxSymbol = 255;
+
+    U32 count[256];
+    size_t const mostFrequent = HIST_count(count, &maxSymbol, src, size);
+    FUZZ_ZASSERT(mostFrequent);
+    if (mostFrequent == size) {
+        /* RLE */
+        FUZZ_dataProducer_free(producer);
+        return 0;
+
+    }
+    FUZZ_ASSERT(maxSymbol <= 255);
+    tableLog = adjustTableLog(tableLog, maxSymbol);
+
+    size_t const wkspSize = HUF_WORKSPACE_SIZE;
+    void* wksp = FUZZ_malloc(wkspSize);
+    void* rBuf = FUZZ_malloc(size);
+    void* cBuf = FUZZ_malloc(cBufSize);
+    HUF_CElt* ct = (HUF_CElt*)FUZZ_malloc(HUF_CTABLE_SIZE(maxSymbol));
+    HUF_DTable* dt = (HUF_DTable*)FUZZ_malloc(HUF_DTABLE_SIZE(tableLog) * sizeof(HUF_DTable));
+    dt[0] = tableLog * 0x01000001;
+
+    tableLog = HUF_optimalTableLog(tableLog, size, maxSymbol, wksp, wkspSize, ct, count, flags);
+    FUZZ_ASSERT(tableLog <= 12);
+    tableLog = HUF_buildCTable_wksp(ct, count, maxSymbol, tableLog, wksp, wkspSize);
+    FUZZ_ZASSERT(tableLog);
+    size_t const tableSize = HUF_writeCTable_wksp(cBuf, cBufSize, ct, maxSymbol, tableLog, wksp, wkspSize);
+    if (ERR_isError(tableSize)) {
+        /* Errors on uncompressible data or cBufSize too small */
+        goto _out;
+    }
+    FUZZ_ZASSERT(tableSize);
+    if (symbols == 0) {
+        FUZZ_ZASSERT(HUF_readDTableX1_wksp(dt, cBuf, tableSize, wksp, wkspSize, flags));
+    } else {
+        size_t const ret = HUF_readDTableX2_wksp(dt, cBuf, tableSize, wksp, wkspSize, flags);
+        if (ERR_getErrorCode(ret) == ZSTD_error_tableLog_tooLarge) {
+            FUZZ_ZASSERT(HUF_readDTableX1_wksp(dt, cBuf, tableSize, wksp, wkspSize, flags));
+        } else {
+            FUZZ_ZASSERT(ret);
+        }
+    }
+
+    size_t cSize;
+    size_t rSize;
+    if (streams == 0) {
+        cSize = HUF_compress1X_usingCTable(cBuf, cBufSize, src, size, ct, flags);
+        FUZZ_ZASSERT(cSize);
+        if (cSize != 0)
+            rSize = HUF_decompress1X_usingDTable(rBuf, size, cBuf, cSize, dt, flags);
+    } else {
+        cSize = HUF_compress4X_usingCTable(cBuf, cBufSize, src, size, ct, flags);
+        FUZZ_ZASSERT(cSize);
+        if (cSize != 0)
+            rSize = HUF_decompress4X_usingDTable(rBuf, size, cBuf, cSize, dt, flags);
+    }
+    if (cSize != 0) {
+        FUZZ_ZASSERT(rSize);
+        FUZZ_ASSERT_MSG(rSize == size, "Incorrect regenerated size");
+        FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, rBuf, size), "Corruption!");
+    }
+_out:
+    free(rBuf);
+    free(cBuf);
+    free(ct);
+    free(dt);
+    free(wksp);
+    FUZZ_dataProducer_free(producer);
+    return 0;
+}
diff --git a/priv/zstd/tests/fuzz/raw_dictionary_round_trip.c b/priv/zstd/tests/fuzz/raw_dictionary_round_trip.c
index 0e65176..7ceab2b 100644
--- a/priv/zstd/tests/fuzz/raw_dictionary_round_trip.c
+++ b/priv/zstd/tests/fuzz/raw_dictionary_round_trip.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/regression_driver.c b/priv/zstd/tests/fuzz/regression_driver.c
index e6d2dec..550c65d 100644
--- a/priv/zstd/tests/fuzz/regression_driver.c
+++ b/priv/zstd/tests/fuzz/regression_driver.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/seekable_roundtrip.c b/priv/zstd/tests/fuzz/seekable_roundtrip.c
index dcdcaae..6f0aa28 100644
--- a/priv/zstd/tests/fuzz/seekable_roundtrip.c
+++ b/priv/zstd/tests/fuzz/seekable_roundtrip.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/sequence_compression_api.c b/priv/zstd/tests/fuzz/sequence_compression_api.c
index cc840bf..9d3f0a1 100644
--- a/priv/zstd/tests/fuzz/sequence_compression_api.c
+++ b/priv/zstd/tests/fuzz/sequence_compression_api.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -26,16 +26,20 @@
 #include "zstd_helpers.h"
 #include "fuzz_data_producer.h"
 
-static ZSTD_CCtx *cctx = NULL;
-static ZSTD_DCtx *dctx = NULL;
+static ZSTD_CCtx* cctx = NULL;
+static ZSTD_DCtx* dctx = NULL;
 static void* literalsBuffer = NULL;
 static void* generatedSrc = NULL;
 static ZSTD_Sequence* generatedSequences = NULL;
 
+static void* dictBuffer = NULL;
+static ZSTD_CDict* cdict = NULL;
+static ZSTD_DDict* ddict = NULL;
+
 #define ZSTD_FUZZ_GENERATED_SRC_MAXSIZE (1 << 20) /* Allow up to 1MB generated data */
+#define ZSTD_FUZZ_GENERATED_LITERALS_SIZE (1 << 20) /* Fixed size 1MB literals buffer */
 #define ZSTD_FUZZ_MATCHLENGTH_MAXSIZE (1 << 18) /* Allow up to 256KB matches */
-#define ZSTD_FUZZ_GENERATED_DICT_MAXSIZE (1 << 18) /* Allow up to a 256KB dict */
-#define ZSTD_FUZZ_GENERATED_LITERALS_SIZE (1 << 18) /* Fixed size 256KB literals buffer */
+#define ZSTD_FUZZ_GENERATED_DICT_MAXSIZE (1 << ZSTD_WINDOWLOG_MAX_32) /* Allow up to 1 << ZSTD_WINDOWLOG_MAX_32 dictionary */
 #define ZSTD_FUZZ_MAX_NBSEQ (1 << 17) /* Maximum of 128K sequences */
 
 /* Deterministic random number generator */
@@ -55,9 +59,9 @@ static uint32_t FUZZ_RDG_rand(uint32_t* src)
 /* Make a pseudorandom string - this simple function exists to avoid
  * taking a dependency on datagen.h to have RDG_genBuffer().
  */
-static char *generatePseudoRandomString(char *str, size_t size) {
+static char* generatePseudoRandomString(char* str, size_t size, FUZZ_dataProducer_t* producer) {
     const char charset[] = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJK1234567890!@#$^&*()_";
-    uint32_t seed = 0;
+    uint32_t seed = FUZZ_dataProducer_uint32(producer);
     if (size) {
         for (size_t n = 0; n < size; n++) {
             int key = FUZZ_RDG_rand(&seed) % (int) (sizeof charset - 1);
@@ -69,47 +73,49 @@ static char *generatePseudoRandomString(char *str, size_t size) {
 
 /* Returns size of source buffer */
 static size_t decodeSequences(void* dst, size_t nbSequences,
-                              size_t literalsSize, const void* dict, size_t dictSize) {
+                              size_t literalsSize,
+                              const void* dict, size_t dictSize,
+                              ZSTD_sequenceFormat_e mode)
+{
     const uint8_t* litPtr = literalsBuffer;
     const uint8_t* const litBegin = literalsBuffer;
-    const uint8_t* const litEnd = literalsBuffer + literalsSize;
+    const uint8_t* const litEnd = litBegin + literalsSize;
     const uint8_t* dictPtr = dict;
     uint8_t* op = dst;
-    const uint8_t* const oend = dst + ZSTD_FUZZ_GENERATED_SRC_MAXSIZE;
+    const uint8_t* const oend = (uint8_t*)dst + ZSTD_FUZZ_GENERATED_SRC_MAXSIZE;
     size_t generatedSrcBufferSize = 0;
     size_t bytesWritten = 0;
-    uint32_t lastLLSize;
 
     for (size_t i = 0; i < nbSequences; ++i) {
-        FUZZ_ASSERT(generatedSequences[i].matchLength != 0);
-        FUZZ_ASSERT(generatedSequences[i].offset != 0);
+        /* block boundary */
+        if (generatedSequences[i].offset == 0)
+            FUZZ_ASSERT(generatedSequences[i].matchLength == 0);
 
         if (litPtr + generatedSequences[i].litLength > litEnd) {
             litPtr = litBegin;
         }
-        ZSTD_memcpy(op, litPtr, generatedSequences[i].litLength);
+        memcpy(op, litPtr, generatedSequences[i].litLength);
         bytesWritten += generatedSequences[i].litLength;
         op += generatedSequences[i].litLength;
         litPtr += generatedSequences[i].litLength;
 
-        FUZZ_ASSERT(generatedSequences[i].offset != 0);
         /* Copy over the match */
         {   size_t matchLength = generatedSequences[i].matchLength;
             size_t j = 0;
             size_t k = 0;
             if (dictSize != 0) {
-                if (generatedSequences[i].offset > bytesWritten) {
-                    /* Offset goes into the dictionary */
-                    size_t offsetFromEndOfDict = generatedSequences[i].offset - bytesWritten;
-                    for (; k < offsetFromEndOfDict && k < matchLength; ++k) {
-                        op[k] = dictPtr[dictSize - offsetFromEndOfDict + k];
+                if (generatedSequences[i].offset > bytesWritten) { /* Offset goes into the dictionary */
+                    size_t dictOffset = generatedSequences[i].offset - bytesWritten;
+                    size_t matchInDict = MIN(matchLength, dictOffset);
+                    for (; k < matchInDict; ++k) {
+                        op[k] = dictPtr[dictSize - dictOffset + k];
                     }
-                    matchLength -= k;
-                    op += k;
+                    matchLength -= matchInDict;
+                    op += matchInDict;
                 }
             }
             for (; j < matchLength; ++j) {
-                op[j] = op[j-(int)generatedSequences[i].offset];
+                op[j] = op[j - generatedSequences[i].offset];
             }
             op += j;
             FUZZ_ASSERT(generatedSequences[i].matchLength == j + k);
@@ -118,55 +124,65 @@ static size_t decodeSequences(void* dst, size_t nbSequences,
     }
     generatedSrcBufferSize = bytesWritten;
     FUZZ_ASSERT(litPtr <= litEnd);
-    lastLLSize = (uint32_t)(litEnd - litPtr);
-    if (lastLLSize <= oend - op) {
-        ZSTD_memcpy(op, litPtr, lastLLSize);
-        generatedSrcBufferSize += lastLLSize;
-    }
+    if (mode == ZSTD_sf_noBlockDelimiters) {
+        const uint32_t lastLLSize = (uint32_t)(litEnd - litPtr);
+        if (lastLLSize <= oend - op) {
+            memcpy(op, litPtr, lastLLSize);
+            generatedSrcBufferSize += lastLLSize;
+    }   }
     return generatedSrcBufferSize;
 }
 
 /* Returns nb sequences generated
- * TODO: Add repcode fuzzing once we support repcode match splits
+ * Note : random sequences are always valid in ZSTD_sf_noBlockDelimiters mode.
+ * However, it can fail with ZSTD_sf_explicitBlockDelimiters,
+ * due to potential lack of space in
  */
 static size_t generateRandomSequences(FUZZ_dataProducer_t* producer,
                                       size_t literalsSizeLimit, size_t dictSize,
-                                      size_t windowLog) {
+                                      size_t windowLog, ZSTD_sequenceFormat_e mode)
+{
+    const uint32_t repCode = 0;  /* not used by sequence ingestion api */
+    size_t windowSize = 1ULL << windowLog;
+    size_t blockSizeMax = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
+    uint32_t matchLengthMax = ZSTD_FUZZ_MATCHLENGTH_MAXSIZE;
     uint32_t bytesGenerated = 0;
     uint32_t nbSeqGenerated = 0;
-    uint32_t litLength;
-    uint32_t matchLength;
-    uint32_t matchBound;
-    uint32_t offset;
-    uint32_t offsetBound;
-    uint32_t repCode = 0;
     uint32_t isFirstSequence = 1;
-    uint32_t windowSize = 1 << windowLog;
+    uint32_t blockSize = 0;
 
-    while (nbSeqGenerated < ZSTD_FUZZ_MAX_NBSEQ
+    if (mode == ZSTD_sf_explicitBlockDelimiters) {
+        /* ensure that no sequence can be larger than one block */
+        literalsSizeLimit = MIN(literalsSizeLimit, blockSizeMax/2);
+        matchLengthMax = MIN(matchLengthMax, blockSizeMax/2);
+    }
+
+    while ( nbSeqGenerated < ZSTD_FUZZ_MAX_NBSEQ - 3 /* extra room for explicit delimiters */
          && bytesGenerated < ZSTD_FUZZ_GENERATED_SRC_MAXSIZE
          && !FUZZ_dataProducer_empty(producer)) {
-        matchBound = ZSTD_FUZZ_MATCHLENGTH_MAXSIZE;
-        litLength = isFirstSequence && dictSize == 0 ? FUZZ_dataProducer_uint32Range(producer, 1, literalsSizeLimit)
-                                                     : FUZZ_dataProducer_uint32Range(producer, 0, literalsSizeLimit);
+        uint32_t matchLength;
+        uint32_t matchBound = matchLengthMax;
+        uint32_t offset;
+        uint32_t offsetBound;
+        const uint32_t minLitLength = (isFirstSequence && (dictSize == 0));
+        const uint32_t litLength = FUZZ_dataProducer_uint32Range(producer, minLitLength, (uint32_t)literalsSizeLimit);
         bytesGenerated += litLength;
         if (bytesGenerated > ZSTD_FUZZ_GENERATED_SRC_MAXSIZE) {
             break;
         }
-        offsetBound = bytesGenerated > windowSize ? windowSize : bytesGenerated + dictSize;
+        offsetBound = (bytesGenerated > windowSize) ? windowSize : bytesGenerated + (uint32_t)dictSize;
         offset = FUZZ_dataProducer_uint32Range(producer, 1, offsetBound);
         if (dictSize > 0 && bytesGenerated <= windowSize) {
             /* Prevent match length from being such that it would be associated with an offset too large
              * from the decoder's perspective. If not possible (match would be too small),
              * then reduce the offset if necessary.
              */
-            size_t bytesToReachWindowSize = windowSize - bytesGenerated;
+            const size_t bytesToReachWindowSize = windowSize - bytesGenerated;
             if (bytesToReachWindowSize < ZSTD_MINMATCH_MIN) {
-                uint32_t newOffsetBound = offsetBound > windowSize ? windowSize : offsetBound;
+                const uint32_t newOffsetBound = offsetBound > windowSize ? windowSize : offsetBound;
                 offset = FUZZ_dataProducer_uint32Range(producer, 1, newOffsetBound);
             } else {
-                matchBound = bytesToReachWindowSize > ZSTD_FUZZ_MATCHLENGTH_MAXSIZE ?
-                             ZSTD_FUZZ_MATCHLENGTH_MAXSIZE : bytesToReachWindowSize;
+                matchBound = MIN(matchLengthMax, (uint32_t)bytesToReachWindowSize);
             }
         }
         matchLength = FUZZ_dataProducer_uint32Range(producer, ZSTD_MINMATCH_MIN, matchBound);
@@ -174,56 +190,78 @@ static size_t generateRandomSequences(FUZZ_dataProducer_t* producer,
         if (bytesGenerated > ZSTD_FUZZ_GENERATED_SRC_MAXSIZE) {
             break;
         }
-        ZSTD_Sequence seq = {offset, litLength, matchLength, repCode};
-        generatedSequences[nbSeqGenerated++] = seq;
-        isFirstSequence = 0;
+        {   ZSTD_Sequence seq = {offset, litLength, matchLength, repCode};
+            const uint32_t lastLits = FUZZ_dataProducer_uint32Range(producer, 0, litLength);
+            #define SPLITPROB 6000
+            #define SPLITMARK 5234
+            const int split = (FUZZ_dataProducer_uint32Range(producer, 0, SPLITPROB) == SPLITMARK);
+            if (mode == ZSTD_sf_explicitBlockDelimiters) {
+                const size_t seqSize = seq.litLength + seq.matchLength;
+                if (blockSize + seqSize > blockSizeMax) {  /* reaching limit : must end block now */
+                    const ZSTD_Sequence endBlock = {0, 0, 0, 0};
+                    generatedSequences[nbSeqGenerated++] = endBlock;
+                    blockSize = seqSize;
+                }
+                if (split) {
+                    const ZSTD_Sequence endBlock = {0, lastLits, 0, 0};
+                    generatedSequences[nbSeqGenerated++] = endBlock;
+                    assert(lastLits <= seq.litLength);
+                    seq.litLength -= lastLits;
+                    blockSize = seqSize - lastLits;
+                } else {
+                    blockSize += seqSize;
+                }
+            }
+            generatedSequences[nbSeqGenerated++] = seq;
+            isFirstSequence = 0;
+        }
     }
 
+    if (mode == ZSTD_sf_explicitBlockDelimiters) {
+        /* always end sequences with a block delimiter */
+        const ZSTD_Sequence endBlock = {0, 0, 0, 0};
+        assert(nbSeqGenerated < ZSTD_FUZZ_MAX_NBSEQ);
+        generatedSequences[nbSeqGenerated++] = endBlock;
+    }
     return nbSeqGenerated;
 }
 
-static size_t roundTripTest(void *result, size_t resultCapacity,
-                            void *compressed, size_t compressedCapacity,
-                            size_t srcSize,
-                            const void *dict, size_t dictSize,
-                            size_t generatedSequencesSize,
-                            size_t wLog, unsigned cLevel, unsigned hasDict)
+static size_t roundTripTest(void* result, size_t resultCapacity,
+                            void* compressed, size_t compressedCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_Sequence* seqs, size_t seqSize,
+                            unsigned hasDict,
+                            ZSTD_sequenceFormat_e mode)
 {
     size_t cSize;
     size_t dSize;
-    ZSTD_CDict* cdict = NULL;
-    ZSTD_DDict* ddict = NULL;
 
-    ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 0);
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, cLevel);
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, wLog);
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, ZSTD_MINMATCH_MIN);
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 1);
-    /* TODO: Add block delim mode fuzzing */
-    ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_noBlockDelimiters);
     if (hasDict) {
-        FUZZ_ZASSERT(ZSTD_CCtx_loadDictionary(cctx, dict, dictSize));
-        FUZZ_ZASSERT(ZSTD_DCtx_loadDictionary(dctx, dict, dictSize));
+        FUZZ_ZASSERT(ZSTD_CCtx_refCDict(cctx, cdict));
+        FUZZ_ZASSERT(ZSTD_DCtx_refDDict(dctx, ddict));
     }
 
     cSize = ZSTD_compressSequences(cctx, compressed, compressedCapacity,
-                                   generatedSequences, generatedSequencesSize,
-                                   generatedSrc, srcSize);
+                                   seqs, seqSize,
+                                   src, srcSize);
+    if ( (ZSTD_getErrorCode(cSize) == ZSTD_error_dstSize_tooSmall)
+      && (mode == ZSTD_sf_explicitBlockDelimiters) ) {
+        /* Valid scenario : in explicit delimiter mode,
+         * it might be possible for the compressed size to outgrow dstCapacity.
+         * In which case, it's still a valid fuzzer scenario,
+         * but no roundtrip shall be possible */
+        return 0;
+    }
+    /* round-trip */
     FUZZ_ZASSERT(cSize);
     dSize = ZSTD_decompressDCtx(dctx, result, resultCapacity, compressed, cSize);
     FUZZ_ZASSERT(dSize);
-
-    if (cdict) {
-        ZSTD_freeCDict(cdict);
-    }
-    if (ddict) {
-        ZSTD_freeDDict(ddict);
-    }
+    FUZZ_ASSERT_MSG(dSize == srcSize, "Incorrect regenerated size");
+    FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, result, srcSize), "Corruption!");
     return dSize;
 }
 
-int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
+int LLVMFuzzerTestOneInput(const uint8_t* src, size_t size)
 {
     void* rBuf;
     size_t rBufSize;
@@ -231,27 +269,70 @@ int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
     size_t cBufSize;
     size_t generatedSrcSize;
     size_t nbSequences;
-    void* dictBuffer;
     size_t dictSize = 0;
     unsigned hasDict;
     unsigned wLog;
     int cLevel;
+    ZSTD_sequenceFormat_e mode;
+
+    FUZZ_dataProducer_t* const producer = FUZZ_dataProducer_create(src, size);
+    FUZZ_ASSERT(producer);
+
+    if (!cctx) {
+        cctx = ZSTD_createCCtx();
+        FUZZ_ASSERT(cctx);
+    }
+    if (!dctx) {
+        dctx = ZSTD_createDCtx();
+        FUZZ_ASSERT(dctx);
+    }
 
-    FUZZ_dataProducer_t *producer = FUZZ_dataProducer_create(src, size);
-    if (literalsBuffer == NULL) {
+    /* Generate window log first so we don't generate offsets too large */
+    wLog = FUZZ_dataProducer_uint32Range(producer, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+    cLevel = FUZZ_dataProducer_int32Range(producer, -3, 22);
+    mode = (ZSTD_sequenceFormat_e)FUZZ_dataProducer_int32Range(producer, 0, 1);
+
+    ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 0);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, cLevel);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, wLog);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, ZSTD_MINMATCH_MIN);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 1);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, mode);
+    ZSTD_CCtx_setParameter(cctx, ZSTD_c_forceAttachDict, ZSTD_dictForceAttach);
+
+    if (!literalsBuffer) {
         literalsBuffer = FUZZ_malloc(ZSTD_FUZZ_GENERATED_LITERALS_SIZE);
-        literalsBuffer = generatePseudoRandomString(literalsBuffer, ZSTD_FUZZ_GENERATED_LITERALS_SIZE);
+        FUZZ_ASSERT(literalsBuffer);
+        literalsBuffer = generatePseudoRandomString(literalsBuffer, ZSTD_FUZZ_GENERATED_LITERALS_SIZE, producer);
     }
 
+    if (!dictBuffer) { /* Generate global dictionary buffer */
+        ZSTD_compressionParameters cParams;
+
+        /* Generate a large dictionary buffer */
+        dictBuffer = calloc(ZSTD_FUZZ_GENERATED_DICT_MAXSIZE, 1);
+        FUZZ_ASSERT(dictBuffer);
+
+        /* Create global cdict and ddict */
+        cParams = ZSTD_getCParams(1, ZSTD_FUZZ_GENERATED_SRC_MAXSIZE, ZSTD_FUZZ_GENERATED_DICT_MAXSIZE);
+        cParams.minMatch = ZSTD_MINMATCH_MIN;
+        cParams.hashLog = ZSTD_HASHLOG_MIN;
+        cParams.chainLog = ZSTD_CHAINLOG_MIN;
+
+        cdict = ZSTD_createCDict_advanced(dictBuffer, ZSTD_FUZZ_GENERATED_DICT_MAXSIZE, ZSTD_dlm_byRef, ZSTD_dct_rawContent, cParams, ZSTD_defaultCMem);
+        ddict = ZSTD_createDDict_advanced(dictBuffer, ZSTD_FUZZ_GENERATED_DICT_MAXSIZE, ZSTD_dlm_byRef, ZSTD_dct_rawContent, ZSTD_defaultCMem);
+        FUZZ_ASSERT(cdict);
+        FUZZ_ASSERT(ddict);
+    }
+
+    FUZZ_ASSERT(cdict);
+    FUZZ_ASSERT(ddict);
+
     hasDict = FUZZ_dataProducer_uint32Range(producer, 0, 1);
     if (hasDict) {
-        dictSize = FUZZ_dataProducer_uint32Range(producer, 1, ZSTD_FUZZ_GENERATED_DICT_MAXSIZE);
-        dictBuffer = FUZZ_malloc(dictSize);
-        dictBuffer = generatePseudoRandomString(dictBuffer, dictSize);
+        dictSize = ZSTD_FUZZ_GENERATED_DICT_MAXSIZE;
     }
-    /* Generate window log first so we dont generate offsets too large */
-    wLog = FUZZ_dataProducer_uint32Range(producer, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX_32);
-    cLevel = FUZZ_dataProducer_int32Range(producer, -3, 22);
 
     if (!generatedSequences) {
         generatedSequences = FUZZ_malloc(sizeof(ZSTD_Sequence)*ZSTD_FUZZ_MAX_NBSEQ);
@@ -259,39 +340,32 @@ int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
     if (!generatedSrc) {
         generatedSrc = FUZZ_malloc(ZSTD_FUZZ_GENERATED_SRC_MAXSIZE);
     }
-    nbSequences = generateRandomSequences(producer, ZSTD_FUZZ_GENERATED_LITERALS_SIZE, dictSize, wLog);
-    generatedSrcSize = decodeSequences(generatedSrc, nbSequences, ZSTD_FUZZ_GENERATED_LITERALS_SIZE, dictBuffer, dictSize);
+
+    nbSequences = generateRandomSequences(producer, ZSTD_FUZZ_GENERATED_LITERALS_SIZE, dictSize, wLog, mode);
+    generatedSrcSize = decodeSequences(generatedSrc, nbSequences, ZSTD_FUZZ_GENERATED_LITERALS_SIZE, dictBuffer, dictSize, mode);
+
+    /* Note : in explicit block delimiters mode,
+     * the fuzzer might generate a lot of small blocks.
+     * In which case, the final compressed size might be > ZSTD_compressBound().
+     * This is still a valid scenario fuzzer though, which makes it possible to check under-sized dstCapacity.
+     * The test just doesn't roundtrip. */
     cBufSize = ZSTD_compressBound(generatedSrcSize);
     cBuf = FUZZ_malloc(cBufSize);
 
     rBufSize = generatedSrcSize;
     rBuf = FUZZ_malloc(rBufSize);
 
-    if (!cctx) {
-        cctx = ZSTD_createCCtx();
-        FUZZ_ASSERT(cctx);
-    }
-    if (!dctx) {
-        dctx = ZSTD_createDCtx();
-        FUZZ_ASSERT(dctx);
-    }
-
-    size_t const result = roundTripTest(rBuf, rBufSize,
+    {   const size_t result = roundTripTest(rBuf, rBufSize,
                                         cBuf, cBufSize,
-                                        generatedSrcSize,
-                                        dictBuffer, dictSize,
-                                        nbSequences,
-                                        wLog, cLevel, hasDict);
-    FUZZ_ZASSERT(result);
-    FUZZ_ASSERT_MSG(result == generatedSrcSize, "Incorrect regenerated size");
-    FUZZ_ASSERT_MSG(!FUZZ_memcmp(generatedSrc, rBuf, generatedSrcSize), "Corruption!");
+                                        generatedSrc, generatedSrcSize,
+                                        generatedSequences, nbSequences,
+                                        hasDict, mode);
+        FUZZ_ASSERT(result <= generatedSrcSize);  /* can be 0 when no round-trip */
+    }
 
     free(rBuf);
     free(cBuf);
     FUZZ_dataProducer_free(producer);
-    if (hasDict) {
-        free(dictBuffer);
-    }
 #ifndef STATEFUL_FUZZING
     ZSTD_freeCCtx(cctx); cctx = NULL;
     ZSTD_freeDCtx(dctx); dctx = NULL;
diff --git a/priv/zstd/tests/fuzz/simple_compress.c b/priv/zstd/tests/fuzz/simple_compress.c
index 3716d0d..c9fea22 100644
--- a/priv/zstd/tests/fuzz/simple_compress.c
+++ b/priv/zstd/tests/fuzz/simple_compress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -9,7 +9,7 @@
  */
 
 /**
- * This fuzz target attempts to comprss the fuzzed data with the simple
+ * This fuzz target attempts to compress the fuzzed data with the simple
  * compression function with an output buffer that may be too small to
  * ensure that the compressor never crashes.
  */
diff --git a/priv/zstd/tests/fuzz/simple_decompress.c b/priv/zstd/tests/fuzz/simple_decompress.c
index dfff11c..ce5f9f0 100644
--- a/priv/zstd/tests/fuzz/simple_decompress.c
+++ b/priv/zstd/tests/fuzz/simple_decompress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/simple_round_trip.c b/priv/zstd/tests/fuzz/simple_round_trip.c
index c9fac26..c2c69d9 100644
--- a/priv/zstd/tests/fuzz/simple_round_trip.c
+++ b/priv/zstd/tests/fuzz/simple_round_trip.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -26,6 +26,23 @@
 static ZSTD_CCtx *cctx = NULL;
 static ZSTD_DCtx *dctx = NULL;
 
+static size_t getDecompressionMargin(void const* compressed, size_t cSize, size_t srcSize, int hasSmallBlocks)
+{
+    size_t margin = ZSTD_decompressionMargin(compressed, cSize);
+    if (!hasSmallBlocks) {
+        /* The macro should be correct in this case, but it may be smaller
+         * because of e.g. block splitting, so take the smaller of the two.
+         */
+        ZSTD_frameHeader zfh;
+        size_t marginM;
+        FUZZ_ZASSERT(ZSTD_getFrameHeader(&zfh, compressed, cSize));
+        marginM = ZSTD_DECOMPRESSION_MARGIN(srcSize, zfh.blockSizeMax);
+        if (marginM < margin)
+            margin = marginM;
+    }
+    return margin;
+}
+
 static size_t roundTripTest(void *result, size_t resultCapacity,
                             void *compressed, size_t compressedCapacity,
                             const void *src, size_t srcSize,
@@ -35,18 +52,57 @@ static size_t roundTripTest(void *result, size_t resultCapacity,
     size_t dSize;
     int targetCBlockSize = 0;
     if (FUZZ_dataProducer_uint32Range(producer, 0, 1)) {
+        size_t const remainingBytes = FUZZ_dataProducer_remainingBytes(producer);
         FUZZ_setRandomParameters(cctx, srcSize, producer);
         cSize = ZSTD_compress2(cctx, compressed, compressedCapacity, src, srcSize);
+        FUZZ_ZASSERT(cSize);
         FUZZ_ZASSERT(ZSTD_CCtx_getParameter(cctx, ZSTD_c_targetCBlockSize, &targetCBlockSize));
+        // Compress a second time and check for determinism
+        {
+            size_t const cSize0 = cSize;
+            XXH64_hash_t const hash0 = XXH64(compressed, cSize, 0);
+            FUZZ_dataProducer_rollBack(producer, remainingBytes);
+            FUZZ_setRandomParameters(cctx, srcSize, producer);
+            cSize = ZSTD_compress2(cctx, compressed, compressedCapacity, src, srcSize);
+            FUZZ_ASSERT(cSize == cSize0);
+            FUZZ_ASSERT(XXH64(compressed, cSize, 0) == hash0);
+        }
     } else {
-      int const cLevel = FUZZ_dataProducer_int32Range(producer, kMinClevel, kMaxClevel);
-
+        int const cLevel = FUZZ_dataProducer_int32Range(producer, kMinClevel, kMaxClevel);
         cSize = ZSTD_compressCCtx(
             cctx, compressed, compressedCapacity, src, srcSize, cLevel);
+        FUZZ_ZASSERT(cSize);
+        // Compress a second time and check for determinism
+        {
+            size_t const cSize0 = cSize;
+            XXH64_hash_t const hash0 = XXH64(compressed, cSize, 0);
+            cSize = ZSTD_compressCCtx(
+                cctx, compressed, compressedCapacity, src, srcSize, cLevel);
+            FUZZ_ASSERT(cSize == cSize0);
+            FUZZ_ASSERT(XXH64(compressed, cSize, 0) == hash0);
+        }
     }
-    FUZZ_ZASSERT(cSize);
     dSize = ZSTD_decompressDCtx(dctx, result, resultCapacity, compressed, cSize);
     FUZZ_ZASSERT(dSize);
+    FUZZ_ASSERT_MSG(dSize == srcSize, "Incorrect regenerated size");
+    FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, result, dSize), "Corruption!");
+
+    {
+        size_t margin = getDecompressionMargin(compressed, cSize, srcSize, targetCBlockSize);
+        size_t const outputSize = srcSize + margin;
+        char* const output = (char*)FUZZ_malloc(outputSize);
+        char* const input = output + outputSize - cSize;
+        FUZZ_ASSERT(outputSize >= cSize);
+        memcpy(input, compressed, cSize);
+
+        dSize = ZSTD_decompressDCtx(dctx, output, outputSize, input, cSize);
+        FUZZ_ZASSERT(dSize);
+        FUZZ_ASSERT_MSG(dSize == srcSize, "Incorrect regenerated size");
+        FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, output, srcSize), "Corruption!");
+
+        free(output);
+    }
+
     /* When superblock is enabled make sure we don't expand the block more than expected.
      * NOTE: This test is currently disabled because superblock mode can arbitrarily
      * expand the block in the worst case. Once superblock mode has been improved we can
@@ -100,13 +156,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
         FUZZ_ASSERT(dctx);
     }
 
-    {
-        size_t const result =
-            roundTripTest(rBuf, rBufSize, cBuf, cBufSize, src, size, producer);
-        FUZZ_ZASSERT(result);
-        FUZZ_ASSERT_MSG(result == size, "Incorrect regenerated size");
-        FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, rBuf, size), "Corruption!");
-    }
+    roundTripTest(rBuf, rBufSize, cBuf, cBufSize, src, size, producer);
     free(rBuf);
     free(cBuf);
     FUZZ_dataProducer_free(producer);
diff --git a/priv/zstd/tests/fuzz/stream_decompress.c b/priv/zstd/tests/fuzz/stream_decompress.c
index e0cdd34..0254d06 100644
--- a/priv/zstd/tests/fuzz/stream_decompress.c
+++ b/priv/zstd/tests/fuzz/stream_decompress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -99,14 +99,14 @@ int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
 
     while (size > 0) {
         ZSTD_inBuffer in = makeInBuffer(&src, &size, producer);
-        while (in.pos != in.size) {
+        do {
+            size_t const rc = ZSTD_decompressStream(dstream, &out, &in);
+            if (ZSTD_isError(rc)) goto error;
             if (out.pos == out.size) {
                 if (stableOutBuffer) goto error;
                 out = makeOutBuffer(producer, buf, bufSize);
             }
-            size_t const rc = ZSTD_decompressStream(dstream, &out, &in);
-            if (ZSTD_isError(rc)) goto error;
-        }
+        } while (in.pos != in.size);
     }
 
 error:
diff --git a/priv/zstd/tests/fuzz/stream_round_trip.c b/priv/zstd/tests/fuzz/stream_round_trip.c
index 719eac4..fae9ccb 100644
--- a/priv/zstd/tests/fuzz/stream_round_trip.c
+++ b/priv/zstd/tests/fuzz/stream_round_trip.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -166,6 +166,24 @@ int LLVMFuzzerTestOneInput(const uint8_t *src, size_t size)
         FUZZ_ZASSERT(rSize);
         FUZZ_ASSERT_MSG(rSize == size, "Incorrect regenerated size");
         FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, rBuf, size), "Corruption!");
+
+        /* Test in-place decompression (note the macro doesn't work in this case) */
+        {
+            size_t const margin = ZSTD_decompressionMargin(cBuf, cSize);
+            size_t const outputSize = size + margin;
+            char* const output = (char*)FUZZ_malloc(outputSize);
+            char* const input = output + outputSize - cSize;
+            size_t dSize;
+            FUZZ_ASSERT(outputSize >= cSize);
+            memcpy(input, cBuf, cSize);
+
+            dSize = ZSTD_decompressDCtx(dctx, output, outputSize, input, cSize);
+            FUZZ_ZASSERT(dSize);
+            FUZZ_ASSERT_MSG(dSize == size, "Incorrect regenerated size");
+            FUZZ_ASSERT_MSG(!FUZZ_memcmp(src, output, size), "Corruption!");
+
+            free(output);
+        }
     }
 
     FUZZ_dataProducer_free(producer);
diff --git a/priv/zstd/tests/fuzz/zstd_frame_info.c b/priv/zstd/tests/fuzz/zstd_frame_info.c
index 9ce645d..95dbdd4 100644
--- a/priv/zstd/tests/fuzz/zstd_frame_info.c
+++ b/priv/zstd/tests/fuzz/zstd_frame_info.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/fuzz/zstd_helpers.c b/priv/zstd/tests/fuzz/zstd_helpers.c
index 4d889de..411b639 100644
--- a/priv/zstd/tests/fuzz/zstd_helpers.c
+++ b/priv/zstd/tests/fuzz/zstd_helpers.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -17,6 +17,7 @@
 #include "fuzz_helpers.h"
 #include "zstd.h"
 #include "zdict.h"
+#include "sequence_producer.h"
 
 const int kMinClevel = -3;
 const int kMaxClevel = 19;
@@ -26,9 +27,14 @@ static void set(ZSTD_CCtx *cctx, ZSTD_cParameter param, int value)
     FUZZ_ZASSERT(ZSTD_CCtx_setParameter(cctx, param, value));
 }
 
+static unsigned produceParamValue(unsigned min, unsigned max,
+                                  FUZZ_dataProducer_t *producer) {
+    return FUZZ_dataProducer_uint32Range(producer, min, max);
+}
+
 static void setRand(ZSTD_CCtx *cctx, ZSTD_cParameter param, unsigned min,
                     unsigned max, FUZZ_dataProducer_t *producer) {
-    unsigned const value = FUZZ_dataProducer_uint32Range(producer, min, max);
+    unsigned const value = produceParamValue(min, max, producer);
     set(cctx, param, value);
 }
 
@@ -65,6 +71,17 @@ ZSTD_parameters FUZZ_randomParams(size_t srcSize, FUZZ_dataProducer_t *producer)
     return params;
 }
 
+static void setSequenceProducerParams(ZSTD_CCtx *cctx, FUZZ_dataProducer_t *producer) {
+    ZSTD_registerSequenceProducer(
+        cctx,
+        NULL,
+        simpleSequenceProducer
+    );
+    setRand(cctx, ZSTD_c_enableSeqProducerFallback, 0, 1, producer);
+    FUZZ_ZASSERT(ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 0));
+    FUZZ_ZASSERT(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, ZSTD_ps_disable));
+}
+
 void FUZZ_setRandomParameters(ZSTD_CCtx *cctx, size_t srcSize, FUZZ_dataProducer_t *producer)
 {
     ZSTD_compressionParameters cParams = FUZZ_randomCParams(srcSize, producer);
@@ -80,7 +97,7 @@ void FUZZ_setRandomParameters(ZSTD_CCtx *cctx, size_t srcSize, FUZZ_dataProducer
     setRand(cctx, ZSTD_c_checksumFlag, 0, 1, producer);
     setRand(cctx, ZSTD_c_dictIDFlag, 0, 1, producer);
     /* Select long distance matching parameters */
-    setRand(cctx, ZSTD_c_enableLongDistanceMatching, 0, 1, producer);
+    setRand(cctx, ZSTD_c_enableLongDistanceMatching, ZSTD_ps_auto, ZSTD_ps_disable, producer);
     setRand(cctx, ZSTD_c_ldmHashLog, ZSTD_HASHLOG_MIN, 16, producer);
     setRand(cctx, ZSTD_c_ldmMinMatch, ZSTD_LDM_MINMATCH_MIN,
             ZSTD_LDM_MINMATCH_MAX, producer);
@@ -89,21 +106,42 @@ void FUZZ_setRandomParameters(ZSTD_CCtx *cctx, size_t srcSize, FUZZ_dataProducer
     setRand(cctx, ZSTD_c_ldmHashRateLog, ZSTD_LDM_HASHRATELOG_MIN,
             ZSTD_LDM_HASHRATELOG_MAX, producer);
     /* Set misc parameters */
+#ifndef ZSTD_MULTITHREAD
+    // To reproduce with or without ZSTD_MULTITHREAD, we are going to use
+    // the same amount of entropy.
+    unsigned const nbWorkers_value = produceParamValue(0, 2, producer);
+    unsigned const rsyncable_value = produceParamValue(0, 1, producer);
+    (void)nbWorkers_value;
+    (void)rsyncable_value;
+    set(cctx, ZSTD_c_nbWorkers, 0);
+    set(cctx, ZSTD_c_rsyncable, 0);
+#else
     setRand(cctx, ZSTD_c_nbWorkers, 0, 2, producer);
     setRand(cctx, ZSTD_c_rsyncable, 0, 1, producer);
+#endif
     setRand(cctx, ZSTD_c_useRowMatchFinder, 0, 2, producer);
     setRand(cctx, ZSTD_c_enableDedicatedDictSearch, 0, 1, producer);
     setRand(cctx, ZSTD_c_forceMaxWindow, 0, 1, producer);
     setRand(cctx, ZSTD_c_literalCompressionMode, 0, 2, producer);
     setRand(cctx, ZSTD_c_forceAttachDict, 0, 2, producer);
-    setRand(cctx, ZSTD_c_splitBlocks, 0, 1, producer);
+    setRand(cctx, ZSTD_c_useBlockSplitter, 0, 2, producer);
     setRand(cctx, ZSTD_c_deterministicRefPrefix, 0, 1, producer);
+    setRand(cctx, ZSTD_c_prefetchCDictTables, 0, 2, producer);
+    setRand(cctx, ZSTD_c_maxBlockSize, ZSTD_BLOCKSIZE_MAX_MIN, ZSTD_BLOCKSIZE_MAX, producer);
+    setRand(cctx, ZSTD_c_validateSequences, 0, 1, producer);
+    setRand(cctx, ZSTD_c_searchForExternalRepcodes, 0, 2, producer);
     if (FUZZ_dataProducer_uint32Range(producer, 0, 1) == 0) {
       setRand(cctx, ZSTD_c_srcSizeHint, ZSTD_SRCSIZEHINT_MIN, 2 * srcSize, producer);
     }
     if (FUZZ_dataProducer_uint32Range(producer, 0, 1) == 0) {
       setRand(cctx, ZSTD_c_targetCBlockSize, ZSTD_TARGETCBLOCKSIZE_MIN, ZSTD_TARGETCBLOCKSIZE_MAX, producer);
     }
+
+    if (FUZZ_dataProducer_uint32Range(producer, 0, 10) == 1) {
+        setSequenceProducerParams(cctx, producer);
+    } else {
+        ZSTD_registerSequenceProducer(cctx, NULL, NULL);
+    }
 }
 
 FUZZ_dict_t FUZZ_train(void const* src, size_t srcSize, FUZZ_dataProducer_t *producer)
@@ -123,7 +161,7 @@ FUZZ_dict_t FUZZ_train(void const* src, size_t srcSize, FUZZ_dataProducer_t *pro
       size_t const offset = FUZZ_dataProducer_uint32Range(producer, 0, MAX(srcSize, 1) - 1);
       size_t const limit = MIN(srcSize - offset, remaining);
       size_t const toCopy = MIN(limit, remaining / (nbSamples - sample));
-      memcpy(samples + pos, src + offset, toCopy);
+      memcpy(samples + pos, (const char*)src + offset, toCopy);
       pos += toCopy;
       samplesSizes[sample] = toCopy;
     }
diff --git a/priv/zstd/tests/fuzz/zstd_helpers.h b/priv/zstd/tests/fuzz/zstd_helpers.h
index 7813884..a4cfe32 100644
--- a/priv/zstd/tests/fuzz/zstd_helpers.h
+++ b/priv/zstd/tests/fuzz/zstd_helpers.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -17,6 +17,7 @@
 #define ZSTD_STATIC_LINKING_ONLY
 
 #include "zstd.h"
+#include "zstd_errors.h"
 #include "fuzz_data_producer.h"
 #include 
 
diff --git a/priv/zstd/tests/fuzzer.c b/priv/zstd/tests/fuzzer.c
index 1ea6521..fc78c7f 100644
--- a/priv/zstd/tests/fuzzer.c
+++ b/priv/zstd/tests/fuzzer.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -25,7 +25,8 @@
 #include        /* free */
 #include         /* fgets, sscanf */
 #include        /* strcmp */
-#undef NDEBUG
+#include          /* time(), time_t */
+#undef NDEBUG             /* always enable assert() */
 #include 
 #define ZSTD_STATIC_LINKING_ONLY  /* ZSTD_compressContinue, ZSTD_compressBlock */
 #include "debug.h"        /* DEBUG_STATIC_ASSERT */
@@ -42,7 +43,7 @@
 #include "util.h"
 #include "timefn.h"       /* SEC_TO_MICRO, UTIL_time_t, UTIL_TIME_INITIALIZER, UTIL_clockSpanMicro, UTIL_getTime */
 /* must be included after util.h, due to ERROR macro redefinition issue on Visual Studio */
-#include "zstd_internal.h"  /* ZSTD_WORKSPACETOOLARGE_MAXDURATION, ZSTD_WORKSPACETOOLARGE_FACTOR, KB, MB */
+#include "zstd_internal.h" /* ZSTD_WORKSPACETOOLARGE_MAXDURATION, ZSTD_WORKSPACETOOLARGE_FACTOR, KB, MB */
 #include "threading.h"    /* ZSTD_pthread_create, ZSTD_pthread_join */
 
 
@@ -82,8 +83,8 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
 void FUZ_bug976(void);
 void FUZ_bug976(void)
 {   /* these constants shall not depend on MIN() macro */
-    assert(ZSTD_HASHLOG_MAX < 31);
-    assert(ZSTD_CHAINLOG_MAX < 31);
+    DEBUG_STATIC_ASSERT(ZSTD_HASHLOG_MAX < 31);
+    DEBUG_STATIC_ASSERT(ZSTD_CHAINLOG_MAX < 31);
 }
 
 
@@ -118,23 +119,24 @@ static U32 FUZ_highbit32(U32 v32)
 /*=============================================
 *   Test macros
 =============================================*/
-#define CHECK_Z(f) {                               \
-    size_t const err = f;                          \
-    if (ZSTD_isError(err)) {                       \
-        DISPLAY("Error => %s : %s ",               \
-                #f, ZSTD_getErrorName(err));       \
-        exit(1);                                   \
+#define CHECK(fn)  { if(!(fn)) { DISPLAYLEVEL(1, "Error : test (%s) failed \n", #fn); exit(1); } }
+
+#define CHECK_Z(f) {                          \
+    size_t const err = f;                     \
+    if (ZSTD_isError(err)) {                  \
+        DISPLAY("Error => %s : %s ",          \
+                #f, ZSTD_getErrorName(err));  \
+        exit(1);                              \
 }   }
 
-#define CHECK_VAR(var, fn)  var = fn; if (ZSTD_isError(var)) { DISPLAYLEVEL(1, "%s : fails : %s \n", #fn, ZSTD_getErrorName(var)); goto _output_error; }
+#define CHECK_VAR(var, fn)  var = fn; if (ZSTD_isError(var)) { DISPLAYLEVEL(1, "%s : fails : %s \n", #fn, ZSTD_getErrorName(var)); exit(1); }
 #define CHECK_NEWV(var, fn)  size_t const CHECK_VAR(var, fn)
-#define CHECK(fn)  { CHECK_NEWV(err, fn); }
 #define CHECKPLUS(var, fn, more)  { CHECK_NEWV(var, fn); more; }
 
 #define CHECK_OP(op, lhs, rhs) {                                  \
     if (!((lhs) op (rhs))) {                                      \
         DISPLAY("Error L%u => FAILED %s %s %s ", __LINE__, #lhs, #op, #rhs);  \
-        goto _output_error;                                       \
+         exit(1);                                                 \
     }                                                             \
 }
 #define CHECK_EQ(lhs, rhs) CHECK_OP(==, lhs, rhs)
@@ -338,6 +340,7 @@ static void FUZ_decodeSequences(BYTE* dst, ZSTD_Sequence* seqs, size_t seqsSize,
 }
 
 #ifdef ZSTD_MULTITHREAD
+
 typedef struct {
     ZSTD_CCtx* cctx;
     ZSTD_threadPool* pool;
@@ -429,8 +432,8 @@ static int threadPoolTests(void) {
 
         ZSTD_pthread_create(&t1, NULL, threadPoolTests_compressionJob, &p1);
         ZSTD_pthread_create(&t2, NULL, threadPoolTests_compressionJob, &p2);
-        ZSTD_pthread_join(t1, NULL);
-        ZSTD_pthread_join(t2, NULL);
+        ZSTD_pthread_join(t1);
+        ZSTD_pthread_join(t2);
 
         assert(!memcmp(decodedBuffer, decodedBuffer2, CNBuffSize));
         free(decodedBuffer2);
@@ -461,6 +464,101 @@ static int threadPoolTests(void) {
 *   Unit tests
 =============================================*/
 
+static void test_compressBound(unsigned tnb)
+{
+    DISPLAYLEVEL(3, "test%3u : compressBound : ", tnb);
+
+    /* check ZSTD_compressBound == ZSTD_COMPRESSBOUND
+     * for a large range of known valid values */
+    DEBUG_STATIC_ASSERT(sizeof(size_t) >= 4);
+    {   int s;
+        for (s=0; s<30; s++) {
+            size_t const w = (size_t)1 << s;
+            CHECK_EQ(ZSTD_compressBound(w), ZSTD_COMPRESSBOUND(w));
+    }   }
+
+    /* Ensure error if srcSize too big */
+    {   size_t const w = ZSTD_MAX_INPUT_SIZE + 1;
+        CHECK(ZSTD_isError(ZSTD_compressBound(w))); /* must fail */
+        CHECK_EQ(ZSTD_COMPRESSBOUND(w), 0);
+    }
+
+    DISPLAYLEVEL(3, "OK \n");
+}
+
+static void test_decompressBound(unsigned tnb)
+{
+    DISPLAYLEVEL(3, "test%3u : decompressBound : ", tnb);
+
+    /* Simple compression, with size : should provide size; */
+    {   const char example[] = "abcd";
+        char cBuffer[ZSTD_COMPRESSBOUND(sizeof(example))];
+        size_t const cSize = ZSTD_compress(cBuffer, sizeof(cBuffer), example, sizeof(example), 0);
+        CHECK_Z(cSize);
+        CHECK_EQ(ZSTD_decompressBound(cBuffer, cSize), (unsigned long long)sizeof(example));
+    }
+
+    /* Simple small compression without size : should provide 1 block size */
+    {   char cBuffer[ZSTD_COMPRESSBOUND(0)];
+        ZSTD_outBuffer out = { cBuffer, sizeof(cBuffer), 0 };
+        ZSTD_inBuffer in = { NULL, 0, 0 };
+        ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+        assert(cctx);
+        CHECK_Z( ZSTD_initCStream(cctx, 0) );
+        CHECK_Z( ZSTD_compressStream(cctx, &out, &in) );
+        CHECK_EQ( ZSTD_endStream(cctx, &out), 0 );
+        CHECK_EQ( ZSTD_decompressBound(cBuffer, out.pos), ZSTD_BLOCKSIZE_MAX );
+        ZSTD_freeCCtx(cctx);
+    }
+
+    /* Attempt to overflow 32-bit intermediate multiplication result
+     * This requires dBound >= 4 GB, aka 2^32.
+     * This requires 2^32 / 2^17 = 2^15 blocks
+     * => create 2^15 blocks (can be empty, or just 1 byte). */
+    {   const char input[] = "a";
+        size_t const nbBlocks = (1 << 15) + 1;
+        size_t blockNb;
+        size_t const outCapacity = 1 << 18; /* large margin */
+        char* const outBuffer = malloc (outCapacity);
+        ZSTD_outBuffer out = { outBuffer, outCapacity, 0 };
+        ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+        assert(cctx);
+        assert(outBuffer);
+        CHECK_Z( ZSTD_initCStream(cctx, 0) );
+        for (blockNb=0; blockNb 0x100000000LLU /* 4 GB */ );
+        ZSTD_freeCCtx(cctx);
+        free(outBuffer);
+    }
+
+    DISPLAYLEVEL(3, "OK \n");
+}
+
+static void test_setCParams(unsigned tnb)
+{
+    ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+    ZSTD_compressionParameters cparams;
+    assert(cctx);
+
+    DISPLAYLEVEL(3, "test%3u : ZSTD_CCtx_setCParams : ", tnb);
+
+    /* valid cparams */
+    cparams = ZSTD_getCParams(1, 0, 0);
+    CHECK_Z(ZSTD_CCtx_setCParams(cctx, cparams));
+
+    /* invalid cparams (must fail) */
+    cparams.windowLog = 99;
+    CHECK(ZSTD_isError(ZSTD_CCtx_setCParams(cctx, cparams)));
+
+    free(cctx);
+    DISPLAYLEVEL(3, "OK \n");
+}
+
 static int basicUnitTests(U32 const seed, double compressibility)
 {
     size_t const CNBuffSize = 5 MB;
@@ -507,6 +605,12 @@ static int basicUnitTests(U32 const seed, double compressibility)
         DISPLAYLEVEL(3, "%u (OK) \n", vn);
     }
 
+    test_compressBound(testNb++);
+
+    test_decompressBound(testNb++);
+
+    test_setCParams(testNb++);
+
     DISPLAYLEVEL(3, "test%3u : ZSTD_adjustCParams : ", testNb++);
     {
         ZSTD_compressionParameters params;
@@ -580,6 +684,17 @@ static int basicUnitTests(U32 const seed, double compressibility)
       if (r != CNBuffSize) goto _output_error; }
     DISPLAYLEVEL(3, "OK \n");
 
+    DISPLAYLEVEL(3, "test%3i : decompress %u bytes with Huffman assembly disabled : ", testNb++, (unsigned)CNBuffSize);
+    {
+        ZSTD_DCtx* dctx = ZSTD_createDCtx();
+        size_t r;
+        CHECK_Z(ZSTD_DCtx_setParameter(dctx, ZSTD_d_disableHuffmanAssembly, 1));
+        r = ZSTD_decompress(decodedBuffer, CNBuffSize, compressedBuffer, cSize);
+        if (r != CNBuffSize || memcmp(decodedBuffer, CNBuffer, CNBuffSize)) goto _output_error;
+        ZSTD_freeDCtx(dctx);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
     DISPLAYLEVEL(3, "test%3i : check decompressed result : ", testNb++);
     {   size_t u;
         for (u=0; u 10 * refPrefixLdmComrpessedSize);
-        /* make sure the ldm comrpessed size is less than 1% of original */
-        assert((double)refPrefixLdmComrpessedSize / (double)size < 0.01);
+        assert(refPrefixCompressedSize > 10 * refPrefixLdmCompressedSize);
+        /* make sure the ldm compressed size is less than 1% of original */
+        assert((double)refPrefixLdmCompressedSize / (double)size < 0.01);
 
         ZSTD_freeDCtx(dctx);
         ZSTD_freeCCtx(cctx);
@@ -1117,6 +1232,60 @@ static int basicUnitTests(U32 const seed, double compressibility)
     }
     DISPLAYLEVEL(3, "OK \n");
 
+    DISPLAYLEVEL(3, "test%3i : in-place decompression : ", testNb++);
+    cSize = ZSTD_compress(compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, -ZSTD_BLOCKSIZE_MAX);
+    CHECK_Z(cSize);
+    CHECK_LT(CNBuffSize, cSize);
+    {
+        size_t const margin = ZSTD_decompressionMargin(compressedBuffer, cSize);
+        size_t const outputSize = (CNBuffSize + margin);
+        char* output = malloc(outputSize);
+        char* input = output + outputSize - cSize;
+        CHECK_LT(cSize, CNBuffSize + margin);
+        CHECK(output != NULL);
+        CHECK_Z(margin);
+        CHECK(margin <= ZSTD_DECOMPRESSION_MARGIN(CNBuffSize, ZSTD_BLOCKSIZE_MAX));
+        memcpy(input, compressedBuffer, cSize);
+
+        {
+            size_t const dSize = ZSTD_decompress(output, outputSize, input, cSize);
+            CHECK_Z(dSize);
+            CHECK_EQ(dSize, CNBuffSize);
+        }
+        CHECK(!memcmp(output, CNBuffer, CNBuffSize));
+        free(output);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+    DISPLAYLEVEL(3, "test%3i : in-place decompression with 2 frames : ", testNb++);
+    cSize = ZSTD_compress(compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize / 3, -ZSTD_BLOCKSIZE_MAX);
+    CHECK_Z(cSize);
+    {
+        size_t const cSize2 = ZSTD_compress((char*)compressedBuffer + cSize, compressedBufferSize - cSize, (char const*)CNBuffer + (CNBuffSize / 3), CNBuffSize / 3, -ZSTD_BLOCKSIZE_MAX);
+        CHECK_Z(cSize2);
+        cSize += cSize2;
+    }
+    {
+        size_t const srcSize = (CNBuffSize / 3) * 2;
+        size_t const margin = ZSTD_decompressionMargin(compressedBuffer, cSize);
+        size_t const outputSize = (CNBuffSize + margin);
+        char* output = malloc(outputSize);
+        char* input = output + outputSize - cSize;
+        CHECK_LT(cSize, CNBuffSize + margin);
+        CHECK(output != NULL);
+        CHECK_Z(margin);
+        memcpy(input, compressedBuffer, cSize);
+
+        {
+            size_t const dSize = ZSTD_decompress(output, outputSize, input, cSize);
+            CHECK_Z(dSize);
+            CHECK_EQ(dSize, srcSize);
+        }
+        CHECK(!memcmp(output, CNBuffer, srcSize));
+        free(output);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
     DISPLAYLEVEL(3, "test%3d: superblock uncompressible data, too many nocompress superblocks : ", testNb++);
     {
         ZSTD_CCtx* const cctx = ZSTD_createCCtx();
@@ -1206,7 +1375,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
     DISPLAYLEVEL(3, "test%3i : compress a NULL input with each level : ", testNb++);
     {   int level = -1;
-        ZSTD_CCtx* cctx = ZSTD_createCCtx();
+        ZSTD_CCtx* const cctx = ZSTD_createCCtx();
         if (!cctx) goto _output_error;
         for (level = -1; level <= ZSTD_maxCLevel(); ++level) {
           CHECK_Z( ZSTD_compress(compressedBuffer, compressedBufferSize, NULL, 0, level) );
@@ -1528,7 +1697,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
             DISPLAYLEVEL(3, "test%3i : resize context to full CCtx size : ", testNb++);
             staticCCtx = ZSTD_initStaticCStream(staticCCtxBuffer, staticCCtxSize);
-            DISPLAYLEVEL(4, "staticCCtxBuffer = %p,  staticCCtx = %p , ", staticCCtxBuffer, staticCCtx);
+            DISPLAYLEVEL(4, "staticCCtxBuffer = %p,  staticCCtx = %p , ", staticCCtxBuffer, (void*)staticCCtx);
             if (staticCCtx == NULL) goto _output_error;
             DISPLAYLEVEL(3, "OK \n");
 
@@ -1579,7 +1748,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
             DISPLAYLEVEL(3, "OK \n");
 
             DISPLAYLEVEL(3, "test%3i : init CCtx for small level %u (should work again) : ", testNb++, 1);
-            CHECK( ZSTD_compressBegin(staticCCtx, 1) );
+            CHECK_Z( ZSTD_compressBegin(staticCCtx, 1) );
             DISPLAYLEVEL(3, "OK \n");
 
             DISPLAYLEVEL(3, "test%3i : use CStream on CCtx-sized static context (should fail) : ", testNb++);
@@ -1622,6 +1791,94 @@ static int basicUnitTests(U32 const seed, double compressibility)
                 if (!ZSTD_isError(r)) goto _output_error;
             }
             DISPLAYLEVEL(3, "OK \n");
+
+            DISPLAYLEVEL(3, "test%3i : test estimation functions with default cctx params : ", testNb++);
+            {
+                // Test ZSTD_estimateCCtxSize_usingCCtxParams
+                {
+                    ZSTD_CCtx_params* params = ZSTD_createCCtxParams();
+                    size_t const cctxSizeDefault = ZSTD_estimateCCtxSize_usingCCtxParams(params);
+                    staticCCtx = ZSTD_initStaticCCtx(staticCCtxBuffer, cctxSizeDefault);
+                    CHECK_VAR(cSize, ZSTD_compressCCtx(staticCCtx,
+                                    compressedBuffer, compressedBufferSize,
+                                    CNBuffer, CNBuffSize, 3));
+
+                    {
+                        size_t const r = ZSTD_decompressDCtx(staticDCtx,
+                                                    decodedBuffer, CNBuffSize,
+                                                    compressedBuffer, cSize);
+                                                                        if (r != CNBuffSize) goto _output_error;
+                        if (memcmp(decodedBuffer, CNBuffer, CNBuffSize)) goto _output_error;
+                    }
+                    ZSTD_freeCCtxParams(params);
+                }
+
+                // Test ZSTD_estimateCStreamSize_usingCCtxParams
+                  {
+                    ZSTD_CCtx_params* params = ZSTD_createCCtxParams();
+                    size_t const cctxSizeDefault = ZSTD_estimateCStreamSize_usingCCtxParams(params);
+                    staticCCtx = ZSTD_initStaticCCtx(staticCCtxBuffer, cctxSizeDefault);
+                    CHECK_VAR(cSize, ZSTD_compressCCtx(staticCCtx,
+                                    compressedBuffer, compressedBufferSize,
+                                    CNBuffer, CNBuffSize, 3) );
+
+                    {
+                        size_t const r = ZSTD_decompressDCtx(staticDCtx,
+                                                    decodedBuffer, CNBuffSize,
+                                                    compressedBuffer, cSize);
+                                                                        if (r != CNBuffSize) goto _output_error;
+                        if (memcmp(decodedBuffer, CNBuffer, CNBuffSize)) goto _output_error;
+                    }
+                    ZSTD_freeCCtxParams(params);
+                }
+            }
+            DISPLAYLEVEL(3, "OK \n");
+
+            DISPLAYLEVEL(3, "test%3i : test estimation functions with maxBlockSize = 0 : ", testNb++);
+            {
+                // Test ZSTD_estimateCCtxSize_usingCCtxParams
+                {
+                    ZSTD_CCtx_params* params = ZSTD_createCCtxParams();
+                    size_t cctxSizeDefault;
+                    CHECK_Z(ZSTD_CCtxParams_setParameter(params, ZSTD_c_maxBlockSize, 0));
+                    cctxSizeDefault = ZSTD_estimateCCtxSize_usingCCtxParams(params);
+                    staticCCtx = ZSTD_initStaticCCtx(staticCCtxBuffer, cctxSizeDefault);
+                    CHECK_VAR(cSize, ZSTD_compressCCtx(staticCCtx,
+                                    compressedBuffer, compressedBufferSize,
+                                    CNBuffer, CNBuffSize, 3) );
+
+                    {
+                        size_t const r = ZSTD_decompressDCtx(staticDCtx,
+                                                    decodedBuffer, CNBuffSize,
+                                                    compressedBuffer, cSize);
+                                                                        if (r != CNBuffSize) goto _output_error;
+                        if (memcmp(decodedBuffer, CNBuffer, CNBuffSize)) goto _output_error;
+                    }
+                    ZSTD_freeCCtxParams(params);
+                }
+
+                // Test ZSTD_estimateCStreamSize_usingCCtxParams
+                  {
+                    ZSTD_CCtx_params* params = ZSTD_createCCtxParams();
+                    size_t cctxSizeDefault;
+                    CHECK_Z(ZSTD_CCtxParams_setParameter(params, ZSTD_c_maxBlockSize, 0));
+                    cctxSizeDefault = ZSTD_estimateCStreamSize_usingCCtxParams(params);
+                    staticCCtx = ZSTD_initStaticCCtx(staticCCtxBuffer, cctxSizeDefault);
+                    CHECK_VAR(cSize, ZSTD_compressCCtx(staticCCtx,
+                                    compressedBuffer, compressedBufferSize,
+                                    CNBuffer, CNBuffSize, 3) );
+
+                    {
+                        size_t const r = ZSTD_decompressDCtx(staticDCtx,
+                                                    decodedBuffer, CNBuffSize,
+                                                    compressedBuffer, cSize);
+                                                                        if (r != CNBuffSize) goto _output_error;
+                        if (memcmp(decodedBuffer, CNBuffer, CNBuffSize)) goto _output_error;
+                    }
+                    ZSTD_freeCCtxParams(params);
+                }
+            }
+            DISPLAYLEVEL(3, "OK \n");
         }
         free(staticCCtxBuffer);
         free(staticDCtxBuffer);
@@ -1647,8 +1904,8 @@ static int basicUnitTests(U32 const seed, double compressibility)
             testResult = 1;
             goto _end;
         }
-        CHECK( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_nbWorkers, 2) );
-        CHECK( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_compressionLevel, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_nbWorkers, 2) );
+        CHECK_Z( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_compressionLevel, 1) );
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3u : compress %u bytes with 2 threads : ", testNb++, (unsigned)CNBuffSize);
@@ -1678,9 +1935,9 @@ static int basicUnitTests(U32 const seed, double compressibility)
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : compress -T2 with checksum : ", testNb++);
-        CHECK( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_checksumFlag, 1) );
-        CHECK( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_contentSizeFlag, 1) );
-        CHECK( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_overlapLog, 3) );
+        CHECK_Z( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_checksumFlag, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_contentSizeFlag, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(mtctx, ZSTD_c_overlapLog, 3) );
         CHECK_VAR(cSize, ZSTD_compress2(mtctx,
                                 compressedBuffer, compressedBufferSize,
                                 CNBuffer, CNBuffSize) );
@@ -1699,11 +1956,11 @@ static int basicUnitTests(U32 const seed, double compressibility)
         ZSTD_DCtx* const dctx = ZSTD_createDCtx();
         char out[32];
         if (cctx == NULL || dctx == NULL) goto _output_error;
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_contentSizeFlag, 0) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_contentSizeFlag, 0) );
         CHECK_VAR(cSize, ZSTD_compress2(cctx, out, sizeof(out), NULL, 0) );
         DISPLAYLEVEL(3, "OK (%u bytes)\n", (unsigned)cSize);
 
-        CHECK( ZSTD_DCtx_setParameter(dctx, ZSTD_d_windowLogMax, 10) );
+        CHECK_Z( ZSTD_DCtx_setParameter(dctx, ZSTD_d_windowLogMax, 10) );
         {   char const* outPtr = out;
             ZSTD_inBuffer inBuffer = { outPtr, cSize, 0 };
             ZSTD_outBuffer outBuffer = { NULL, 0, 0 };
@@ -1718,9 +1975,9 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
     DISPLAYLEVEL(3, "test%3i : compress with block splitting : ", testNb++)
     {   ZSTD_CCtx* cctx = ZSTD_createCCtx();
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_splitBlocks, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_useBlockSplitter, ZSTD_ps_enable) );
         cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
-        CHECK(cSize);
+        CHECK_Z(cSize);
         ZSTD_freeCCtx(cctx);
     }
     DISPLAYLEVEL(3, "OK \n");
@@ -1728,13 +1985,13 @@ static int basicUnitTests(U32 const seed, double compressibility)
     DISPLAYLEVEL(3, "test%3i : compress -T2 with/without literals compression : ", testNb++)
     {   ZSTD_CCtx* cctx = ZSTD_createCCtx();
         size_t cSize1, cSize2;
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 1) );
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 2) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 2) );
         cSize1 = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
-        CHECK(cSize1);
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_literalCompressionMode, ZSTD_lcm_uncompressed) );
+        CHECK_Z(cSize1);
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_literalCompressionMode, ZSTD_ps_disable) );
         cSize2 = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
-        CHECK(cSize2);
+        CHECK_Z(cSize2);
         CHECK_LT(cSize1, cSize2);
         ZSTD_freeCCtx(cctx);
     }
@@ -1745,10 +2002,10 @@ static int basicUnitTests(U32 const seed, double compressibility)
         /* Set rsyncable and don't give the ZSTD_compressBound(CNBuffSize) so
          * ZSTDMT is forced to not take the shortcut.
          */
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 1) );
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 1) );
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_rsyncable, 1) );
-        CHECK( ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize - 1, CNBuffer, CNBuffSize) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_rsyncable, 1) );
+        CHECK_Z( ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize - 1, CNBuffer, CNBuffSize) );
         ZSTD_freeCCtx(cctx);
     }
     DISPLAYLEVEL(3, "OK \n");
@@ -1758,22 +2015,22 @@ static int basicUnitTests(U32 const seed, double compressibility)
         int const jobSize = 512 KB;
         int value;
         /* Check that the overlap log and job size are unset. */
-        CHECK( ZSTD_CCtxParams_getParameter(params, ZSTD_c_overlapLog, &value) );
+        CHECK_Z( ZSTD_CCtxParams_getParameter(params, ZSTD_c_overlapLog, &value) );
         CHECK_EQ(value, 0);
-        CHECK( ZSTD_CCtxParams_getParameter(params, ZSTD_c_jobSize, &value) );
+        CHECK_Z( ZSTD_CCtxParams_getParameter(params, ZSTD_c_jobSize, &value) );
         CHECK_EQ(value, 0);
         /* Set and check the overlap log and job size. */
-        CHECK( ZSTD_CCtxParams_setParameter(params, ZSTD_c_overlapLog, 5) );
-        CHECK( ZSTD_CCtxParams_setParameter(params, ZSTD_c_jobSize, jobSize) );
-        CHECK( ZSTD_CCtxParams_getParameter(params, ZSTD_c_overlapLog, &value) );
+        CHECK_Z( ZSTD_CCtxParams_setParameter(params, ZSTD_c_overlapLog, 5) );
+        CHECK_Z( ZSTD_CCtxParams_setParameter(params, ZSTD_c_jobSize, jobSize) );
+        CHECK_Z( ZSTD_CCtxParams_getParameter(params, ZSTD_c_overlapLog, &value) );
         CHECK_EQ(value, 5);
-        CHECK( ZSTD_CCtxParams_getParameter(params, ZSTD_c_jobSize, &value) );
+        CHECK_Z( ZSTD_CCtxParams_getParameter(params, ZSTD_c_jobSize, &value) );
         CHECK_EQ(value, jobSize);
         /* Set the number of workers and check the overlap log and job size. */
-        CHECK( ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, 2) );
-        CHECK( ZSTD_CCtxParams_getParameter(params, ZSTD_c_overlapLog, &value) );
+        CHECK_Z( ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, 2) );
+        CHECK_Z( ZSTD_CCtxParams_getParameter(params, ZSTD_c_overlapLog, &value) );
         CHECK_EQ(value, 5);
-        CHECK( ZSTD_CCtxParams_getParameter(params, ZSTD_c_jobSize, &value) );
+        CHECK_Z( ZSTD_CCtxParams_getParameter(params, ZSTD_c_jobSize, &value) );
         CHECK_EQ(value, jobSize);
         ZSTD_freeCCtxParams(params);
     }
@@ -1830,6 +2087,47 @@ static int basicUnitTests(U32 const seed, double compressibility)
     if (memcmp(decodedBuffer, CNBuffer, CNBuffSize / 2) != 0) goto _output_error;
     DISPLAYLEVEL(3, "OK \n");
 
+    /* Simple API skippable frame test */
+    DISPLAYLEVEL(3, "test%3i : read/write a skippable frame : ", testNb++);
+    {   U32 i;
+        unsigned readMagic;
+        unsigned long long receivedSize;
+        size_t skippableSize;
+        const U32 skipLen = 129 KB;
+        char* const skipBuff = (char*)malloc(skipLen);
+        assert(skipBuff != NULL);
+        for (i = 0; i < skipLen; i++)
+            skipBuff[i] = (char) ((seed + i) % 256);
+        skippableSize = ZSTD_writeSkippableFrame(
+                                compressedBuffer, compressedBufferSize,
+                                skipBuff, skipLen, seed % 15);
+        CHECK_Z(skippableSize);
+        CHECK_EQ(1, ZSTD_isSkippableFrame(compressedBuffer, skippableSize));
+        receivedSize = ZSTD_readSkippableFrame(decodedBuffer, CNBuffSize, &readMagic, compressedBuffer, skippableSize);
+        CHECK_EQ(skippableSize, receivedSize + ZSTD_SKIPPABLEHEADERSIZE);
+        CHECK_EQ(seed % 15, readMagic);
+        if (memcmp(decodedBuffer, skipBuff, skipLen) != 0) goto _output_error;
+
+        free(skipBuff);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+    DISPLAYLEVEL(3, "test%3i : read/write an empty skippable frame : ", testNb++);
+    {
+        unsigned readMagic;
+        unsigned long long receivedSize;
+        size_t skippableSize;
+        skippableSize = ZSTD_writeSkippableFrame(
+                                compressedBuffer, compressedBufferSize,
+                                CNBuffer, 0, seed % 15);
+        CHECK_EQ(ZSTD_SKIPPABLEHEADERSIZE, skippableSize);
+        CHECK_EQ(1, ZSTD_isSkippableFrame(compressedBuffer, skippableSize));
+        receivedSize = ZSTD_readSkippableFrame(NULL, 0, &readMagic, compressedBuffer, skippableSize);
+        CHECK_EQ(skippableSize, receivedSize + ZSTD_SKIPPABLEHEADERSIZE);
+        CHECK_EQ(seed % 15, readMagic);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
     /* Dictionary and CCtx Duplication tests */
     {   ZSTD_CCtx* const ctxOrig = ZSTD_createCCtx();
         ZSTD_CCtx* const ctxDuplicated = ZSTD_createCCtx();
@@ -1843,8 +2141,8 @@ static int basicUnitTests(U32 const seed, double compressibility)
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : load dictionary into context : ", testNb++);
-        CHECK( ZSTD_compressBegin_usingDict(ctxOrig, CNBuffer, dictSize, 2) );
-        CHECK( ZSTD_copyCCtx(ctxDuplicated, ctxOrig, 0) ); /* Begin_usingDict implies unknown srcSize, so match that */
+        CHECK_Z( ZSTD_compressBegin_usingDict(ctxOrig, CNBuffer, dictSize, 2) );
+        CHECK_Z( ZSTD_copyCCtx(ctxDuplicated, ctxOrig, 0) ); /* Begin_usingDict implies unknown srcSize, so match that */
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : compress with flat dictionary : ", testNb++);
@@ -1904,8 +2202,8 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         DISPLAYLEVEL(3, "test%3i : check content size on duplicated context : ", testNb++);
         {   size_t const testSize = CNBuffSize / 3;
-            CHECK( ZSTD_compressBegin(ctxOrig, ZSTD_defaultCLevel()) );
-            CHECK( ZSTD_copyCCtx(ctxDuplicated, ctxOrig, testSize) );
+            CHECK_Z( ZSTD_compressBegin(ctxOrig, ZSTD_defaultCLevel()) );
+            CHECK_Z( ZSTD_copyCCtx(ctxDuplicated, ctxOrig, testSize) );
 
             CHECK_VAR(cSize, ZSTD_compressEnd(ctxDuplicated, compressedBuffer, ZSTD_compressBound(testSize),
                                           (const char*)CNBuffer + dictSize, testSize) );
@@ -1915,29 +2213,37 @@ static int basicUnitTests(U32 const seed, double compressibility)
         }   }
         DISPLAYLEVEL(3, "OK \n");
 
+        /* Note : these tests should be replaced by proper regression tests,
+         *         but existing ones do not focus on small data + dictionary + all levels.
+         */
         if ((int)(compressibility * 100 + 0.1) == FUZ_compressibility_default) { /* test only valid with known input */
             size_t const flatdictSize = 22 KB;
             size_t const contentSize = 9 KB;
             const void* const dict = (const char*)CNBuffer;
             const void* const contentStart = (const char*)dict + flatdictSize;
             /* These upper bounds are generally within a few bytes of the compressed size */
-            size_t const target_nodict_cSize[22+1] = { 3840, 3770, 3870, 3830, 3770,
-                                                       3770, 3770, 3770, 3750, 3750,
-                                                       3742, 3670, 3670, 3660, 3660,
-                                                       3660, 3660, 3660, 3660, 3660,
-                                                       3660, 3660, 3660 };
-            size_t const target_wdict_cSize[22+1] =  { 2830, 2890, 2890, 2820, 2940,
+            size_t target_nodict_cSize[22+1] = { 3840, 3770, 3870, 3830, 3770,
+                                                 3770, 3770, 3770, 3750, 3750,
+                                                 3742, 3675, 3674, 3665, 3664,
+                                                 3663, 3662, 3661, 3660, 3660,
+                                                 3660, 3660, 3660 };
+            size_t const target_wdict_cSize[22+1] =  { 2830, 2896, 2893, 2820, 2940,
                                                        2950, 2950, 2925, 2900, 2891,
-                                                       2910, 2910, 2910, 2770, 2760,
-                                                       2750, 2750, 2750, 2750, 2750,
-                                                       2750, 2750, 2750 };
+                                                       2910, 2910, 2910, 2780, 2775,
+                                                       2765, 2760, 2755, 2754, 2753,
+                                                       2753, 2753, 2753 };
             int l = 1;
             int const maxLevel = ZSTD_maxCLevel();
+            /* clevels with strategies that support rowhash on small inputs */
+            int rowLevel = 4;
+            int const rowLevelEnd = 8;
 
             DISPLAYLEVEL(3, "test%3i : flat-dictionary efficiency test : \n", testNb++);
             assert(maxLevel == 22);
             RDG_genBuffer(CNBuffer, flatdictSize + contentSize, compressibility, 0., seed);
-            DISPLAYLEVEL(4, "content hash : %016llx;  dict hash : %016llx \n", XXH64(contentStart, contentSize, 0), XXH64(dict, flatdictSize, 0));
+            DISPLAYLEVEL(4, "content hash : %016llx;  dict hash : %016llx \n",
+                        (unsigned long long)XXH64(contentStart, contentSize, 0),
+                        (unsigned long long)XXH64(dict, flatdictSize, 0));
 
             for ( ; l <= maxLevel; l++) {
                 size_t const nodict_cSize = ZSTD_compress(compressedBuffer, compressedBufferSize,
@@ -1964,6 +2270,27 @@ static int basicUnitTests(U32 const seed, double compressibility)
                 DISPLAYLEVEL(4, "level %i with dictionary : max expected %u >= reached %u \n",
                                 l, (unsigned)target_wdict_cSize[l], (unsigned)wdict_cSize);
             }
+            /* Compression with ZSTD_compress2 and row match finder force enabled.
+             * Give some slack for force-enabled row matchfinder since we're on a small input (9KB)
+             */
+            for ( ; rowLevel <= rowLevelEnd; ++rowLevel) target_nodict_cSize[rowLevel] += 5;
+            for (l=1 ; l <= maxLevel; l++) {
+                ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+                size_t nodict_cSize;
+                ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, l);
+                ZSTD_CCtx_setParameter(cctx, ZSTD_c_useRowMatchFinder, ZSTD_ps_enable);
+                nodict_cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize,
+                                                           contentStart, contentSize);
+                if (nodict_cSize > target_nodict_cSize[l]) {
+                    DISPLAYLEVEL(1, "error : compression with compress2 at level %i worse than expected (%u > %u) \n",
+                                    l, (unsigned)nodict_cSize, (unsigned)target_nodict_cSize[l]);
+                    ZSTD_freeCCtx(cctx);
+                    goto _output_error;
+                }
+                DISPLAYLEVEL(4, "level %i with compress2 : max expected %u >= reached %u \n",
+                                l, (unsigned)target_nodict_cSize[l], (unsigned)nodict_cSize);
+                ZSTD_freeCCtx(cctx);
+            }
             /* Dict compression with DMS */
             for ( l=1 ; l <= maxLevel; l++) {
                 size_t wdict_cSize;
@@ -1971,6 +2298,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
                 CHECK_Z( ZSTD_CCtx_setParameter(ctxOrig, ZSTD_c_compressionLevel, l) );
                 CHECK_Z( ZSTD_CCtx_setParameter(ctxOrig, ZSTD_c_enableDedicatedDictSearch, 0) );
                 CHECK_Z( ZSTD_CCtx_setParameter(ctxOrig, ZSTD_c_forceAttachDict, ZSTD_dictForceAttach) );
+                CHECK_Z( ZSTD_CCtx_setParameter(ctxOrig, ZSTD_c_prefetchCDictTables, seed % 3) );
                 wdict_cSize = ZSTD_compress2(ctxOrig, compressedBuffer, compressedBufferSize, contentStart, contentSize);
                 if (wdict_cSize > target_wdict_cSize[l]) {
                     DISPLAYLEVEL(1, "error : compression with dictionary and compress2 at level %i worse than expected (%u > %u) \n",
@@ -2275,6 +2603,27 @@ static int basicUnitTests(U32 const seed, double compressibility)
         }
         DISPLAYLEVEL(3, "OK \n");
 
+        DISPLAYLEVEL(3, "test%3d : bufferless api with cdict : ", testNb++);
+        {   ZSTD_CDict* const cdict = ZSTD_createCDict(dictBuffer, dictSize, 1);
+            ZSTD_DCtx* const dctx = ZSTD_createDCtx();
+            ZSTD_frameParameters const fParams = { 0, 1, 0 };
+            size_t cBlockSize;
+            cSize = 0;
+            CHECK_Z(ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN));
+            cBlockSize = ZSTD_compressContinue(cctx, (char*)compressedBuffer + cSize, compressedBufferSize - cSize, CNBuffer, 1000);
+            CHECK_Z(cBlockSize);
+            cSize += cBlockSize;
+            cBlockSize = ZSTD_compressEnd(cctx, (char*)compressedBuffer + cSize, compressedBufferSize - cSize, (char const*)CNBuffer + 2000, 1000);
+            CHECK_Z(cBlockSize);
+            cSize += cBlockSize;
+
+            CHECK_Z(ZSTD_decompress_usingDict(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize, dictBuffer, dictSize));
+
+            ZSTD_freeCDict(cdict);
+            ZSTD_freeDCtx(dctx);
+        }
+        DISPLAYLEVEL(3, "OK \n");
+
         DISPLAYLEVEL(3, "test%3i : Building cdict w/ ZSTD_dct_fullDict on a good dictionary : ", testNb++);
         {   ZSTD_compressionParameters const cParams = ZSTD_getCParams(1, CNBuffSize, dictSize);
             ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_fullDict, cParams, ZSTD_defaultCMem);
@@ -2470,7 +2819,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
             ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters);
             CHECK_Z( ZSTD_DCtx_loadDictionary(dctx, dictBuffer, dictSize) );
             CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize) );
-            /* The dictionary should presist across calls. */
+            /* The dictionary should persist across calls. */
             CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize) );
             /* When we reset the context the dictionary is cleared. */
             ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters);
@@ -2489,7 +2838,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
             ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters);
             CHECK_Z( ZSTD_DCtx_refDDict(dctx, ddict) );
             CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize) );
-            /* The ddict should presist across calls. */
+            /* The ddict should persist across calls. */
             CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize) );
             /* When we reset the context the ddict is cleared. */
             ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters);
@@ -2515,6 +2864,90 @@ static int basicUnitTests(U32 const seed, double compressibility)
         }
         DISPLAYLEVEL(3, "OK \n");
 
+        DISPLAYLEVEL(3, "test%3i : ZSTD_fast attach dictionary with hashLog = 25 and chainLog = 25 : ", testNb++);
+        {
+            ZSTD_CCtx_params* cctxParams = ZSTD_createCCtxParams();
+            ZSTD_customMem customMem = {NULL, NULL, NULL};
+            ZSTD_DCtx* dctx = ZSTD_createDCtx();
+            ZSTD_CDict* cdict;
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_strategy, ZSTD_fast));
+            /* Set windowLog to 25 so hash/chain logs don't get sized down */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_windowLog, 25));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_hashLog, 25));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_chainLog, 25));
+            /* Set srcSizeHint to 2^25 so hash/chain logs don't get sized down */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_srcSizeHint, 1u << 25));
+            cdict = ZSTD_createCDict_advanced2(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, cctxParams, customMem);
+            CHECK_Z(ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_forceAttachDict, ZSTD_dictForceAttach));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1));
+            CHECK_Z(ZSTD_CCtx_refCDict(cctx, cdict));
+            cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
+            CHECK_Z(cSize);
+            CHECK_Z(ZSTD_decompress_usingDict(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize, dictBuffer, dictSize));
+            ZSTD_freeCDict(cdict);
+            ZSTD_freeDCtx(dctx);
+            ZSTD_freeCCtxParams(cctxParams);
+        }
+        DISPLAYLEVEL(3, "OK \n");
+
+        DISPLAYLEVEL(3, "test%3i : ZSTD_dfast attach dictionary with hashLog = 25 and chainLog = 25 : ", testNb++);
+        {
+            ZSTD_CCtx_params* cctxParams = ZSTD_createCCtxParams();
+            ZSTD_customMem customMem = {NULL, NULL, NULL};
+            ZSTD_DCtx* dctx = ZSTD_createDCtx();
+            ZSTD_CDict* cdict;
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_strategy, ZSTD_dfast));
+            /* Set windowLog to 25 so hash/chain logs don't get sized down */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_windowLog, 25));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_hashLog, 25));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_chainLog, 25));
+            /* Set srcSizeHint to 2^25 so hash/chain logs don't get sized down */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_srcSizeHint, 1u << 25));
+            cdict = ZSTD_createCDict_advanced2(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, cctxParams, customMem);
+            CHECK_Z(ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_forceAttachDict, ZSTD_dictForceAttach));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1));
+            CHECK_Z(ZSTD_CCtx_refCDict(cctx, cdict));
+            cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
+            CHECK_Z(cSize);
+            CHECK_Z(ZSTD_decompress_usingDict(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize, dictBuffer, dictSize));
+            ZSTD_freeCDict(cdict);
+            ZSTD_freeDCtx(dctx);
+            ZSTD_freeCCtxParams(cctxParams);
+        }
+        DISPLAYLEVEL(3, "OK \n");
+
+        DISPLAYLEVEL(3, "test%3i : ZSTD_lazy attach dictionary with hashLog = 29 and searchLog = 4 : ", testNb++);
+        if (MEM_64bits()) {
+            ZSTD_CCtx_params* cctxParams = ZSTD_createCCtxParams();
+            ZSTD_customMem customMem = {NULL, NULL, NULL};
+            ZSTD_DCtx* dctx = ZSTD_createDCtx();
+            ZSTD_CDict* cdict;
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_strategy, ZSTD_lazy));
+            /* Force enable row based match finder, and disable dedicated dict search. */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_useRowMatchFinder, ZSTD_ps_enable));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_enableDedicatedDictSearch, 0));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_searchLog, 4));
+            /* Set windowLog to 29 so hash/chain logs don't get sized down */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_windowLog, 29));
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_hashLog, 29));
+            /* Set srcSizeHint to 2^29 so hash/chain logs don't get sized down */
+            CHECK_Z(ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_srcSizeHint, 1u << 29));
+            cdict = ZSTD_createCDict_advanced2(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, cctxParams, customMem);
+            CHECK_Z(ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_forceAttachDict, ZSTD_dictForceAttach));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1));
+            CHECK_Z(ZSTD_CCtx_refCDict(cctx, cdict));
+            cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
+            CHECK_Z(cSize);
+            CHECK_Z(ZSTD_decompress_usingDict(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize, dictBuffer, dictSize));
+            ZSTD_freeCDict(cdict);
+            ZSTD_freeDCtx(dctx);
+            ZSTD_freeCCtxParams(cctxParams);
+        }
+        DISPLAYLEVEL(3, "OK \n");
+
         DISPLAYLEVEL(3, "test%3i : Dictionary with non-default repcodes : ", testNb++);
         { U32 u; for (u=0; u %u bytes)\n", (unsigned)inputSize, (unsigned)result);
@@ -2785,7 +3218,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         DISPLAYLEVEL(3, "test%3i : get dParameter bounds ", testNb++);
         {   ZSTD_bounds const bounds = ZSTD_dParam_getBounds(ZSTD_d_windowLogMax);
-            CHECK(bounds.error);
+            CHECK_Z(bounds.error);
         }
         DISPLAYLEVEL(3, "OK \n");
 
@@ -2819,7 +3252,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         /* basic block compression */
         DISPLAYLEVEL(3, "test%3i : magic-less format test : ", testNb++);
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_format, ZSTD_f_zstd1_magicless) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_format, ZSTD_f_zstd1_magicless) );
         {   ZSTD_inBuffer in = { CNBuffer, inputSize, 0 };
             ZSTD_outBuffer out = { compressedBuffer, ZSTD_compressBound(inputSize), 0 };
             size_t const result = ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_end);
@@ -2837,7 +3270,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         DISPLAYLEVEL(3, "test%3i : decompress of magic-less frame : ", testNb++);
         ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters);
-        CHECK( ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, ZSTD_f_zstd1_magicless) );
+        CHECK_Z( ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, ZSTD_f_zstd1_magicless) );
         {   ZSTD_frameHeader zfh;
             size_t const zfhrt = ZSTD_getFrameHeader_advanced(&zfh, compressedBuffer, cSize, ZSTD_f_zstd1_magicless);
             if (zfhrt != 0) goto _output_error;
@@ -2859,7 +3292,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         /* basic block compression */
         DISPLAYLEVEL(3, "test%3i : empty magic-less format test : ", testNb++);
-        CHECK( ZSTD_CCtx_setParameter(cctx, ZSTD_c_format, ZSTD_f_zstd1_magicless) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_format, ZSTD_f_zstd1_magicless) );
         {   ZSTD_inBuffer in = { CNBuffer, 0, 0 };
             ZSTD_outBuffer out = { compressedBuffer, ZSTD_compressBound(0), 0 };
             size_t const result = ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_end);
@@ -2871,7 +3304,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         DISPLAYLEVEL(3, "test%3i : decompress of empty magic-less frame : ", testNb++);
         ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters);
-        CHECK( ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, ZSTD_f_zstd1_magicless) );
+        CHECK_Z( ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, ZSTD_f_zstd1_magicless) );
         /* one shot */
         {   size_t const result = ZSTD_decompressDCtx(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize);
             if (result != 0) goto _output_error;
@@ -2906,13 +3339,13 @@ static int basicUnitTests(U32 const seed, double compressibility)
             int check;
             if (ZSTD_isError(bounds.error))
                 continue;
-            CHECK(ZSTD_DCtx_getParameter(dctx, dParam, &value1));
+            CHECK_Z(ZSTD_DCtx_getParameter(dctx, dParam, &value1));
             value2 = (value1 != bounds.lowerBound) ? bounds.lowerBound : bounds.upperBound;
-            CHECK(ZSTD_DCtx_setParameter(dctx, dParam, value2));
-            CHECK(ZSTD_DCtx_getParameter(dctx, dParam, &check));
+            CHECK_Z(ZSTD_DCtx_setParameter(dctx, dParam, value2));
+            CHECK_Z(ZSTD_DCtx_getParameter(dctx, dParam, &check));
             if (check != value2) goto _output_error;
-            CHECK(ZSTD_DCtx_reset(dctx, ZSTD_reset_parameters));
-            CHECK(ZSTD_DCtx_getParameter(dctx, dParam, &check));
+            CHECK_Z(ZSTD_DCtx_reset(dctx, ZSTD_reset_parameters));
+            CHECK_Z(ZSTD_DCtx_getParameter(dctx, dParam, &check));
             if (check != value1) goto _output_error;
         }
         ZSTD_freeDCtx(dctx);
@@ -2929,21 +3362,21 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         /* basic block compression */
         DISPLAYLEVEL(3, "test%3i : Block compression test : ", testNb++);
-        CHECK( ZSTD_compressBegin(cctx, 5) );
-        CHECK( ZSTD_getBlockSize(cctx) >= blockSize);
+        CHECK_Z( ZSTD_compressBegin(cctx, 5) );
+        CHECK_Z( ZSTD_getBlockSize(cctx) >= blockSize);
         CHECK_VAR(cSize, ZSTD_compressBlock(cctx, compressedBuffer, ZSTD_compressBound(blockSize), CNBuffer, blockSize) );
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : Block decompression test : ", testNb++);
-        CHECK( ZSTD_decompressBegin(dctx) );
+        CHECK_Z( ZSTD_decompressBegin(dctx) );
         { CHECK_NEWV(r, ZSTD_decompressBlock(dctx, decodedBuffer, CNBuffSize, compressedBuffer, cSize) );
           if (r != blockSize) goto _output_error; }
         DISPLAYLEVEL(3, "OK \n");
 
         /* very long stream of block compression */
         DISPLAYLEVEL(3, "test%3i : Huge block streaming compression test : ", testNb++);
-        CHECK( ZSTD_compressBegin(cctx, -199) );  /* we just want to quickly overflow internal U32 index */
-        CHECK( ZSTD_getBlockSize(cctx) >= blockSize);
+        CHECK_Z( ZSTD_compressBegin(cctx, -199) );  /* we just want to quickly overflow internal U32 index */
+        CHECK_Z( ZSTD_getBlockSize(cctx) >= blockSize);
         {   U64 const toCompress = 5000000000ULL;   /* > 4 GB */
             U64 compressed = 0;
             while (compressed < toCompress) {
@@ -2956,7 +3389,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         /* dictionary block compression */
         DISPLAYLEVEL(3, "test%3i : Dictionary Block compression test : ", testNb++);
-        CHECK( ZSTD_compressBegin_usingDict(cctx, CNBuffer, dictSize, 5) );
+        CHECK_Z( ZSTD_compressBegin_usingDict(cctx, CNBuffer, dictSize, 5) );
         CHECK_VAR(cSize,  ZSTD_compressBlock(cctx, compressedBuffer, ZSTD_compressBound(blockSize), (char*)CNBuffer+dictSize, blockSize));
         RDG_genBuffer((char*)CNBuffer+dictSize+blockSize, blockSize, 0.0, 0.0, seed);  /* create a non-compressible second block */
         { CHECK_NEWV(r, ZSTD_compressBlock(cctx, (char*)compressedBuffer+cSize, ZSTD_compressBound(blockSize), (char*)CNBuffer+dictSize+blockSize, blockSize) );  /* for cctx history consistency */
@@ -2967,7 +3400,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : Dictionary Block decompression test : ", testNb++);
-        CHECK( ZSTD_decompressBegin_usingDict(dctx, CNBuffer, dictSize) );
+        CHECK_Z( ZSTD_decompressBegin_usingDict(dctx, CNBuffer, dictSize) );
         {   CHECK_NEWV( r, ZSTD_decompressBlock(dctx, decodedBuffer, blockSize, compressedBuffer, cSize) );
             if (r != blockSize) {
                 DISPLAYLEVEL(1, "ZSTD_decompressBlock() with _usingDict() fails : %u, instead of %u expected \n", (unsigned)r, (unsigned)blockSize);
@@ -2986,8 +3419,8 @@ static int basicUnitTests(U32 const seed, double compressibility)
         DISPLAYLEVEL(3, "test%3i : Block compression with CDict : ", testNb++);
         {   ZSTD_CDict* const cdict = ZSTD_createCDict(CNBuffer, dictSize, 3);
             if (cdict==NULL) goto _output_error;
-            CHECK( ZSTD_compressBegin_usingCDict(cctx, cdict) );
-            CHECK( ZSTD_compressBlock(cctx, compressedBuffer, ZSTD_compressBound(blockSize), (char*)CNBuffer+dictSize, blockSize) );
+            CHECK_Z( ZSTD_compressBegin_usingCDict(cctx, cdict) );
+            CHECK_Z( ZSTD_compressBlock(cctx, compressedBuffer, ZSTD_compressBound(blockSize), (char*)CNBuffer+dictSize, blockSize) );
             ZSTD_freeCDict(cdict);
         }
         DISPLAYLEVEL(3, "OK \n");
@@ -3044,18 +3477,17 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
     DISPLAYLEVEL(3, "test%3i : ZSTD_getSequences followed by ZSTD_compressSequences : ", testNb++);
     {
-        size_t srcSize = 500 KB;
-        BYTE* src = (BYTE*)CNBuffer;
-        BYTE* dst = (BYTE*)compressedBuffer;
-        size_t dstSize = ZSTD_compressBound(srcSize);
-        size_t decompressSize = srcSize;
-        char* decompressBuffer = (char*)malloc(decompressSize);
+        const size_t srcSize = 500 KB;
+        const BYTE* const src = (BYTE*)CNBuffer;
+        BYTE* const dst = (BYTE*)compressedBuffer;
+        const size_t dstCapacity = ZSTD_compressBound(srcSize);
+        const size_t decompressSize = srcSize;
+        char* const decompressBuffer = (char*)malloc(decompressSize);
         size_t compressedSize;
-        size_t dSize;
 
-        ZSTD_CCtx* cctx = ZSTD_createCCtx();
-        ZSTD_Sequence* seqs = (ZSTD_Sequence*)malloc(srcSize * sizeof(ZSTD_Sequence));
-        size_t seqsSize;
+        ZSTD_CCtx* const cctx = ZSTD_createCCtx();
+        ZSTD_Sequence* const seqs = (ZSTD_Sequence*)malloc(srcSize * sizeof(ZSTD_Sequence));
+        size_t nbSeqs;
 
         if (seqs == NULL) goto _output_error;
         assert(cctx != NULL);
@@ -3063,36 +3495,37 @@ static int basicUnitTests(U32 const seed, double compressibility)
         /* Populate src with random data */
         RDG_genBuffer(CNBuffer, srcSize, compressibility, 0., seed);
 
-        /* Test with block delimiters roundtrip */
-        seqsSize = ZSTD_generateSequences(cctx, seqs, srcSize, src, srcSize);
+        /* Roundtrip Test with block delimiters generated by ZSTD_generateSequences() */
+        nbSeqs = ZSTD_generateSequences(cctx, seqs, srcSize, src, srcSize);
         ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
         ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_explicitBlockDelimiters);
-        compressedSize = ZSTD_compressSequences(cctx, dst, dstSize, seqs, seqsSize, src, srcSize);
+        compressedSize = ZSTD_compressSequences(cctx, dst, dstCapacity, seqs, nbSeqs, src, srcSize);
         if (ZSTD_isError(compressedSize)) {
             DISPLAY("Error in sequence compression with block delims\n");
             goto _output_error;
         }
-        dSize = ZSTD_decompress(decompressBuffer, decompressSize, dst, compressedSize);
-        if (ZSTD_isError(dSize)) {
-            DISPLAY("Error in sequence compression roundtrip with block delims\n");
-            goto _output_error;
-        }
+        {   size_t const dSize = ZSTD_decompress(decompressBuffer, decompressSize, dst, compressedSize);
+            if (ZSTD_isError(dSize)) {
+                DISPLAY("Error in sequence compression roundtrip with block delims\n");
+                goto _output_error;
+        }   }
         assert(!memcmp(decompressBuffer, src, srcSize));
 
-        /* Test with no block delimiters roundtrip */
-        seqsSize = ZSTD_mergeBlockDelimiters(seqs, seqsSize);
-        ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
-        ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_noBlockDelimiters);
-        compressedSize = ZSTD_compressSequences(cctx, dst, dstSize, seqs, seqsSize, src, srcSize);
+        /* Roundtrip Test with no block delimiters  */
+        {   size_t const nbSeqsAfterMerge = ZSTD_mergeBlockDelimiters(seqs, nbSeqs);
+            ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+            ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_noBlockDelimiters);
+            compressedSize = ZSTD_compressSequences(cctx, dst, dstCapacity, seqs, nbSeqsAfterMerge, src, srcSize);
+        }
         if (ZSTD_isError(compressedSize)) {
             DISPLAY("Error in sequence compression with no block delims\n");
             goto _output_error;
         }
-        dSize = ZSTD_decompress(decompressBuffer, decompressSize, dst, compressedSize);
-        if (ZSTD_isError(dSize)) {
-            DISPLAY("Error in sequence compression roundtrip with no block delims\n");
-            goto _output_error;
-        }
+        {   size_t const dSize = ZSTD_decompress(decompressBuffer, decompressSize, dst, compressedSize);
+            if (ZSTD_isError(dSize)) {
+                DISPLAY("Error in sequence compression roundtrip with no block delims\n");
+                goto _output_error;
+        }   }
         assert(!memcmp(decompressBuffer, src, srcSize));
 
         ZSTD_freeCCtx(cctx);
@@ -3156,7 +3589,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
         size_t const bound = ZSTD_compressBound(_3BYTESTESTLENGTH);
         size_t nbSeq = 1;
         while (nbSeq <= maxNbSeq) {
-          CHECK(ZSTD_compressCCtx(cctx, compressedBuffer, bound, CNBuffer, nbSeq * 3, 19));
+          CHECK_Z(ZSTD_compressCCtx(cctx, compressedBuffer, bound, CNBuffer, nbSeq * 3, 19));
           /* Check every sequence for the first 100, then skip more rapidly. */
           if (nbSeq < 100) {
             ++nbSeq;
@@ -3185,7 +3618,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
         size_t const bound = ZSTD_compressBound(CNBuffSize);
         size_t size = 1;
         while (size <= CNBuffSize) {
-          CHECK(ZSTD_compressCCtx(cctx, compressedBuffer, bound, CNBuffer, size, 3));
+          CHECK_Z(ZSTD_compressCCtx(cctx, compressedBuffer, bound, CNBuffer, size, 3));
           /* Check every size for the first 100, then skip more rapidly. */
           if (size < 100) {
             ++size;
@@ -3213,15 +3646,14 @@ static int basicUnitTests(U32 const seed, double compressibility)
         {   U32 u;
             for (u = 0; u < CNBuffSize; ++u) {
               ((BYTE*)CNBuffer)[u] = 255 - ((BYTE*)CNBuffer)[u];
-            }
-        }
+        }   }
         {   /* Compress the data */
             size_t const inputSize = 500;
             size_t const outputSize = ZSTD_compressBound(inputSize);
             void* const outputBuffer = malloc(outputSize);
             ZSTD_CCtx* const cctx = ZSTD_createCCtx();
             if (!outputBuffer || !cctx) goto _output_error;
-            CHECK(ZSTD_compress_usingDict(cctx, outputBuffer, outputSize, CNBuffer, inputSize, dictBuffer, dictSize, 1));
+            CHECK_Z(ZSTD_compress_usingDict(cctx, outputBuffer, outputSize, CNBuffer, inputSize, dictBuffer, dictSize, 1));
             free(outputBuffer);
             ZSTD_freeCCtx(cctx);
         }
@@ -3289,6 +3721,91 @@ static int basicUnitTests(U32 const seed, double compressibility)
         FSE_normalizeCount(norm, tableLog, count, nbSeq, maxSymbolValue, /* useLowProbCount */ 1);
     }
     DISPLAYLEVEL(3, "OK \n");
+
+    DISPLAYLEVEL(3, "test%3i : testing FSE_writeNCount() PR#2779: ", testNb++);
+    {
+        size_t const outBufSize = 9;
+        short const count[11] = {1, 0, 1, 0, 1, 0, 1, 0, 1, 9, 18};
+        unsigned const tableLog = 5;
+        unsigned const maxSymbolValue = 10;
+        BYTE* outBuf = (BYTE*)malloc(outBufSize*sizeof(BYTE));
+
+        /* Ensure that this write doesn't write out of bounds, and that
+         * FSE_writeNCount_generic() is *not* called with writeIsSafe == 1.
+         */
+        FSE_writeNCount(outBuf, outBufSize, count, maxSymbolValue, tableLog);
+        free(outBuf);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+    DISPLAYLEVEL(3, "test%3i : testing bitwise intrinsics PR#3045: ", testNb++);
+    {
+        U32 seed_copy = seed; /* need non-const seed to avoid compiler warning for FUZ_rand(&seed) */
+        U32 rand32 = FUZ_rand(&seed_copy);
+        U64 rand64 = ((U64)FUZ_rand(&seed_copy) << 32) | FUZ_rand(&seed_copy);
+        U32 lowbit_only_32 = 1;
+        U64 lowbit_only_64 = 1;
+        U32 highbit_only_32 = (U32)1 << 31;
+        U64 highbit_only_64 = (U64)1 << 63;
+        U32 i;
+        if (rand32 == 0) rand32 = 1; /* CLZ and CTZ are undefined on 0 */
+        if (rand64 == 0) rand64 = 1; /* CLZ and CTZ are undefined on 0 */
+
+        /* Test ZSTD_countTrailingZeros32 */
+        CHECK_EQ(ZSTD_countTrailingZeros32(lowbit_only_32), 0u);
+        CHECK_EQ(ZSTD_countTrailingZeros32(highbit_only_32), 31u);
+        CHECK_EQ(ZSTD_countTrailingZeros32(rand32), ZSTD_countTrailingZeros32_fallback(rand32));
+
+        /* Test ZSTD_countLeadingZeros32 */
+        CHECK_EQ(ZSTD_countLeadingZeros32(lowbit_only_32), 31u);
+        CHECK_EQ(ZSTD_countLeadingZeros32(highbit_only_32), 0u);
+        CHECK_EQ(ZSTD_countLeadingZeros32(rand32), ZSTD_countLeadingZeros32_fallback(rand32));
+
+        /* Test ZSTD_countTrailingZeros64 */
+        CHECK_EQ(ZSTD_countTrailingZeros64(lowbit_only_64), 0u);
+        CHECK_EQ(ZSTD_countTrailingZeros64(highbit_only_64), 63u);
+
+        /* Test ZSTD_countLeadingZeros64 */
+        CHECK_EQ(ZSTD_countLeadingZeros64(lowbit_only_64), 63u);
+        CHECK_EQ(ZSTD_countLeadingZeros64(highbit_only_64), 0u);
+
+        /* Test ZSTD_highbit32 */
+        CHECK_EQ(ZSTD_highbit32(lowbit_only_32), 0u);
+        CHECK_EQ(ZSTD_highbit32(highbit_only_32), 31u);
+
+        /* Test ZSTD_NbCommonBytes */
+        if (MEM_isLittleEndian()) {
+            if (MEM_64bits()) {
+                CHECK_EQ(ZSTD_NbCommonBytes(lowbit_only_32), 0u);
+                CHECK_EQ(ZSTD_NbCommonBytes(highbit_only_32), 3u);
+            } else {
+                CHECK_EQ(ZSTD_NbCommonBytes(lowbit_only_32), 0u);
+                CHECK_EQ(ZSTD_NbCommonBytes(highbit_only_32), 3u);
+            }
+        } else {
+            if (MEM_64bits()) {
+                CHECK_EQ(ZSTD_NbCommonBytes(lowbit_only_32), 7u);
+                CHECK_EQ(ZSTD_NbCommonBytes(highbit_only_32), 4u);
+            } else {
+                CHECK_EQ(ZSTD_NbCommonBytes(lowbit_only_32), 3u);
+                CHECK_EQ(ZSTD_NbCommonBytes(highbit_only_32), 0u);
+            }
+       }
+
+        /* Test MEM_ intrinsics */
+        CHECK_EQ(MEM_swap32(rand32), MEM_swap32_fallback(rand32));
+        CHECK_EQ(MEM_swap64(rand64), MEM_swap64_fallback(rand64));
+
+        /* Test fallbacks vs intrinsics on a range of small integers */
+        for (i=1; i <= 1000; i++) {
+            CHECK_EQ(MEM_swap32(i), MEM_swap32_fallback(i));
+            CHECK_EQ(MEM_swap64((U64)i), MEM_swap64_fallback((U64)i));
+            CHECK_EQ(ZSTD_countTrailingZeros32(i), ZSTD_countTrailingZeros32_fallback(i));
+            CHECK_EQ(ZSTD_countLeadingZeros32(i), ZSTD_countLeadingZeros32_fallback(i));
+        }
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
 #ifdef ZSTD_MULTITHREAD
     DISPLAYLEVEL(3, "test%3i : passing wrong full dict should fail on compressStream2 refPrefix ", testNb++);
     {   ZSTD_CCtx* cctx = ZSTD_createCCtx();
@@ -3337,7 +3854,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
 
         /* Enable MT, LDM, and use refPrefix() for a small dict */
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_nbWorkers, 2));
-        CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, 1));
+        CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, ZSTD_ps_enable));
         CHECK_Z(ZSTD_CCtx_refPrefix(cctx, dict, dictSize));
 
         CHECK_Z(ZSTD_compress2(cctx, dst, dstSize, src, srcSize));
@@ -3407,11 +3924,7 @@ static int basicUnitTests(U32 const seed, double compressibility)
                         DISPLAYLEVEL(3, "error! l: %d dict: %zu srcSize: %zu cctx size cpar: %zu, cctx size level: %zu\n",
                                      level, dictSize, srcSize, cctxSizeUsingCParams, cctxSizeUsingLevel);
                         goto _output_error;
-                    }
-                }
-            }
-        }
-    }
+    }   }   }   }   }
     DISPLAYLEVEL(3, "OK \n");
 
     DISPLAYLEVEL(3, "test%3i : thread pool API tests : \n", testNb++)
@@ -3477,7 +3990,7 @@ static int longUnitTests(U32 const seed, double compressibility)
             while (approxIndex <= (maxIndex / 4) * 3) {
                 CHECK_Z(ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_flush));
                 approxIndex += in.pos;
-                CHECK(in.pos == in.size);
+                CHECK_Z(in.pos == in.size);
                 in.pos = 0;
                 out.pos = 0;
             }
@@ -3503,7 +4016,7 @@ static int longUnitTests(U32 const seed, double compressibility)
             while (approxIndex <= maxIndex) {
                 CHECK_Z(ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_flush));
                 approxIndex += in.pos;
-                CHECK(in.pos == in.size);
+                CHECK_Z(in.pos == in.size);
                 in.pos = 0;
                 out.pos = 0;
             }
@@ -3527,8 +4040,7 @@ static int longUnitTests(U32 const seed, double compressibility)
     DISPLAYLEVEL(3, "OK \n");
 
     DISPLAYLEVEL(3, "longtest%3i : testing ldm no regressions in size for opt parser : ", testNb++);
-    {
-        size_t cSizeLdm;
+    {   size_t cSizeLdm;
         size_t cSizeNoLdm;
         ZSTD_CCtx* const cctx = ZSTD_createCCtx();
 
@@ -3536,7 +4048,7 @@ static int longUnitTests(U32 const seed, double compressibility)
 
         /* Enable checksum to verify round trip. */
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1));
-        CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, 1));
+        CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, ZSTD_ps_enable));
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 19));
 
         /* Round trip once with ldm. */
@@ -3546,7 +4058,7 @@ static int longUnitTests(U32 const seed, double compressibility)
 
         ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1));
-        CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, 0));
+        CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_enableLongDistanceMatching, ZSTD_ps_disable));
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_compressionLevel, 19));
 
         /* Round trip once without ldm. */
@@ -3587,7 +4099,7 @@ static int longUnitTests(U32 const seed, double compressibility)
         RDG_genBuffer(dict, dictSize, 0.5, 0.5, seed);
         RDG_genBuffer(CNBuffer, CNBuffSize, 0.6, 0.6, seed);
 
-        CHECK(cctx_params != NULL);
+        CHECK_Z(cctx_params != NULL);
 
         for (dictSize = CNBuffSize; dictSize; dictSize = dictSize >> 3) {
             DISPLAYLEVEL(3, "\n    Testing with dictSize %u ", (U32)dictSize);
@@ -3606,7 +4118,7 @@ static int longUnitTests(U32 const seed, double compressibility)
                     CHECK(cdict != NULL);
 
                     CHECK_Z(ZSTD_CCtx_refCDict(cctx, cdict));
-                    CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_forceAttachDict, attachPref));
+                    CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_forceAttachDict, (int)attachPref));
 
                     cSize = ZSTD_compress2(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize);
                     CHECK_Z(cSize);
@@ -3630,11 +4142,6 @@ static int longUnitTests(U32 const seed, double compressibility)
     free(compressedBuffer);
     free(decodedBuffer);
     return testResult;
-
-_output_error:
-    testResult = 1;
-    DISPLAY("Error detected in Unit tests ! \n");
-    goto _end;
 }
 
 
@@ -3887,9 +4394,9 @@ static int fuzzerTests(U32 seed, unsigned nbTests, unsigned startTest, U32 const
         DISPLAYLEVEL(5, "fuzzer t%u: Bufferless streaming compression test \n", testNb);
         {   U32 const testLog = FUZ_rand(&lseed) % maxSrcLog;
             U32 const dictLog = FUZ_rand(&lseed) % maxSrcLog;
-            int const cLevel = (FUZ_rand(&lseed) %
-                                (ZSTD_maxCLevel() -
-                                 (MAX(testLog, dictLog) / cLevelLimiter))) +
+            int const cLevel = (int)(FUZ_rand(&lseed) %
+                                ((U32)ZSTD_maxCLevel() -
+                                 (MAX(testLog, dictLog) / (U32)cLevelLimiter))) +
                                1;
             maxTestSize = FUZ_rLogLength(&lseed, testLog);
             if (maxTestSize >= dstBufferSize) maxTestSize = dstBufferSize-1;
@@ -3985,7 +4492,7 @@ static int fuzzerTests(U32 seed, unsigned nbTests, unsigned startTest, U32 const
     free(cBuffer);
     free(dstBuffer);
     free(mirrorBuffer);
-    return result;
+    return (int)result;
 
 _output_error:
     result = 1;
@@ -4022,7 +4529,7 @@ static unsigned readU32FromChar(const char** stringPtr)
 {
     unsigned result = 0;
     while ((**stringPtr >='0') && (**stringPtr <='9'))
-        result *= 10, result += **stringPtr - '0', (*stringPtr)++ ;
+        result *= 10, result += (unsigned)(**stringPtr - '0'), (*stringPtr)++ ;
     if ((**stringPtr=='K') || (**stringPtr=='M')) {
         result <<= 10;
         if (**stringPtr=='M') result <<= 10;
@@ -4164,7 +4671,7 @@ int main(int argc, const char** argv)
         }
     }
     if (!result)
-        result = fuzzerTests(seed, nbTests, testNb, maxDuration, ((double)proba) / 100, bigTests);
+        result = fuzzerTests(seed, (unsigned)nbTests, (unsigned)testNb, maxDuration, ((double)proba) / 100, bigTests);
     if (mainPause) {
         int unused;
         DISPLAY("Press Enter \n");
diff --git a/priv/zstd/tests/golden-compression/large-literal-and-match-lengths b/priv/zstd/tests/golden-compression/large-literal-and-match-lengths
new file mode 100644
index 0000000000000000000000000000000000000000..fb63c32c50f44563ccb9dc3544216cab833ad6c8
GIT binary patch
literal 199998
zcmV(rK<>Y!`4pKCN*rK&oW#Ftnh@o5k3It2HaNU?7wVH)A4?fMPqhGL$-?v#tK&yNMiW*sa>lim+Qn*G9-aIQ_+YlA_eJi3zD|d}12#ZOp_ttS6|fBaGJt
z@$5_Lf4TYP3?d-Xr|}Wj-W@lEK2#(|J4ffP1330+0}kSw!`x7!KNd0jOSfz4qxZf+
zm5+Uttaw@rA7ynHaIyFBJkI{;`JxX+2&ldubu}%6=c_3
zGsVwtGEL}&;`_24hva;SJ3Ogce|+wKU$gFT!N|sA7pW8~
z!#mV&8AfybdQQ$xio`)GWexb1ebUH-e1mz7KJ%|Fc-e-h(ug~~9!;sYf=X_O2x{g(
zhThts60NHxwCdRWicB(g4Q31G?bp-<1Wm|y!?pQjgAB8XMUM`7*!J}-B5ZC|G}%6s
z4$hlEhqJ;GiG_7vHaO?FxEbs~6?)YiA2xbP$#XymOX>@Tf?yEzMcR9o{el}nw-B#V
z&`_2TGU3@rY*iD9_?g+TsCf
zkuGP($BO*Llykr%P6e7$vP#a7&0;mW-ju)5tY^Lo^V_~=>N$-9W)(jN2F^2+cK?W5
z3o%6)$#8_H5h;|9YutW3f&fRQx9*B`{*q}M#`9GzyTjao6D51U>NX^4u@cur{RhGMMB_WIwZlWx-EW{X7C^gcmE=h$}N;0_cv?D+sOE=ariZ{4*VGunI`2$qXqNVbU^W@a
z0Ju%rd&r-sO6S>G%F?-B5P1WL*`f2YHN+Q74NF26)X5R%A||IiN|O-7T46~=
zA!1?!uq
zqYU+0lWZ6!(C-Z*IN|(|vw>
zTiECKiL=Ua3oi>2($ko&41X>i^FvO~l}ivdcufhk9Gkt!)RjY}a{|hYHFVYA|gx|CNWn!2M;fHWRAL
zW~MmxJ#^BQvp-oxAm0;^p{{st%-MB=`qh3+?mGH-FF
z<^2VI2nqaesV+QNHZ^f>vcbW`g_*PM6-=V=v0||Qr*LdjMYaws
zDCN`6OGD!#dlzJWrBpxbmGp{!i1%2W&vT>hlkJml-qLnxes$MzjA0*+IEG=ibY)z1
zRK%U89EwQYBq{tA3Ygfp13SU$;Z8tkqCneMh@I2=tR1R*Nd3uj9W#ZQZp>+AUo4`=
zy)}qT(%RoKVOVxt2OP%c9v-nI!bhdH_{|y%Amfa6U&pP*zuF^Nz}>@od8zZFh|~Tz
zQgMt;mjEc+Uw7iv4K9@1XFG3u16rAwYaKXYg6hPY06bJXKN_)yP!ayjjNS
zfsSQR$)<7>^}uu9isq=Ip_2(t`TbjAOnYrG$b1Kc(Y%=4?896kx(Y
zQL#&6`!;#q;rQ#~S;yDU
z7SAnG{ey8Gfn88g6;)3^zW}PBB^tF~j9`r848l|io*7Rt(Q(-SNk|K&FpD|%;J!pl
zeOkHEdZdRr@8w7xyd&i(^#)%&+<$6aqE1(Tn6;5hl*5bN*TvxPKCCd~cW)7d)(d{L
zqSPfbC!6RY^i3^%@e?u9saN7sTu0`UviFsO!l}92x~B>F4CgtClbwnd>SzE+;R*Nc
z#)JW>Ae;HAdx*$yX*7N!J;9wON}9moEfCM(|6W-M_iNDDB^*24qr$58-YVeU1A<*I!fl*uI$-Lk{_N<4}R(Dr&;<{7a>5
zOli_t8~Z7xK8RsGk}Af!Z#W^#c8M)z4Hcq$kDwtX=RTxf^HxJ-nTyTGYaHTS63U-@
zyVE9wpCiUQFK=*prQz0Zu>Jm1ivssZ$-x|36&>pjv|tMDiTxhtgk7Hrf5GqLs>Y_%
zalPKd96BF@PKcKMz-c9KseePZ97GtT9h($Du#e^9nZ|a0KGhw+2iikVe6i?==#Uwxe+$rPVidGf$d%
zMprV(24<#QU7*bG8RF)iEwQ=?C2;|qTbpwmV^R|OBM)?goZMzfBQ~QZJ&MwqLhen%
zTEmSL7nVl|4-S)76;aYH*)F-~mE3t}g}u#=wjOTDEjmC_k7GFdh^}_{^iLwAIYhU|34LOz`|d%5W+hhyWddX+?BHgJ||yBV@!Ib
z%fl#NzW)gGTb#;c>a)d05Z$%N?%pudghWxdEZi!eRmMAI7?V5BHTA~BdYvXXrz`|i
zHm&gCar2XM|Am~mPUzU7>ynK(9^`Q)d+Q(hytZUNyS-+~k1ml^&}Qqg`>$?3N~Lt#
z!!=%xwSmsQrXajy%F!uf|5
z7#=QzQAJWyfUa3aTj1==ixHKdvc<^&wC_~5wjUg_Yv%*0Srs0d
zbF;1FTynLD`cf?)Dag**ZDJpkI}C+a0$DH}WMT=HFNyp1p2OBkA4dj_K-;<%(QJT+
zbha_ajh1}aC-D38_Z_;V^ygu$@-|~jkd;J@efl_~IIcGZjVf^V)a^D`DuN<3zil!P
zs6Ldgb&3Yxuuh-|19d;j|15HbFvFL6{HyTlu9AadOeaoaQL{|Z2{Q>p&3uU
zya#Z7{|@OauwK8UrV?5uLRm5Z43cl|dbZ@)T7gd7KSquy*2~{$&a-x%b1*~zVzXVa
z0ch_bul@ilhL^AJV0lqL^OezR%9a$Af3d<6m2*dl72TVl#f^Z62s?V8Rdl7G83*Ta
z#pHoiTCxbg1eyW|_q@7KK8g*wE-Ovf2^Z^{+tf|Y7>_y8`;NL?5!@7k?jGnza^xN+
z)Z_=3_$1N?+{`b(Jboi@d9qhh(fF&O(?e!Xj7*Ca(h{f-zVA5GAl*ya7i|!`Ce+$w
zs%rKuWPK_N5->B=xe5vEDu*tpfgc;ALPO`K4`P?tiU~K%|ih*q89LSvv$ph&g!4
zlEGaoaE<3T3s=J{p|e69`&em+`73KL32d-M1~arT(H4dt#CYck9((-a7|S&vJuK}%
z%2~Yj({Nka%9z{3atWpY;WW)pn6rHB@Nj$HH<)yGjskiWyh|;YwB_DXj;~K!D3GqU
z@9-bsiGlqN7a1y?U+3GdLbDUYyH5)iF=o8K-|<3h)A$4om@mEw5Pj^H0j>t7)!I`d
zXNm#sqeyuUm#{F=B5te{Uo*T`rvQrwE7%x0*6&!&ZR?gAWGTwWED~
z{{5F9Km*d{6FeqB?q6;H>CZcyRGmCeRVDbUHZen|4ozZC2EB$zFRDNTzF!|>VY1vr
zsAIZU)VMA18|GC3uAdIEBga*Hx5m`O%uKkjWhQd^(I4gb|G#E)|1yMB(*=ytnG7Js
zQK$z8C)j;GBSH0pWWbcxi+Yudgdc2^W5Gjm3x|A=MA2m`0pwg$|B9A&VbMrFOgRU5
z-d~~n?oGpq`Bo`pg+Z;|BW&|a;79(vfINx@up;++5Iy{+@R-r<4c3mh+@{#>C1=St
znAQ?4_53HizV14gr95mXhs~KnJpx`eYwc%HYir?3@v~P;XO(%SxO@*fNna4Al^KW<
ziVlu~pbOArbLStQ{Cks^KbwyH{~VkC6RvqYeRl`oFXq
zCQDC3tEsL=pbYml)(?=@nOOmhQbM;B^f4^-#=gg(2RYoe0%F3UeyD5FXm~uoFDdPi
zp2cAbj~k}+Khv58-D5=5P_fdlttHSOpvNr-!7ZGr@DuE-RmN5~rPXE@l|2ufosa~*
zFhmTG<^qQY`_XBXVWU`PU{C`09P44UPwm&#_TAg8zQD`L8K~!%6AY~prLjAGdmo@n(uYAJbbY*M(*EZD(PSGWF~3Kyd<(B#R+QX<{N`A2ulk9
zqP|OeGn@5|G~g
z+{JuKGRqcq#2XSxK6#ZlQ^N=?1IQfL{pG1qP*pt
zxZP8n?mB=5Tjz(lD0S|ma{XR@H~rr)HmzL2dR+(l$&Wm~+$l3&<9iwmBcC3WA5@z>
z9{=IaY>hAe)hK@#CY{PHHSA(S!v)?SA7TKP+
zINTEr&qmaIWW=a@<5)P;2?^hDL0%XKo7hb0iq9_^2KJk#ykHv-d1L%K3^7MA$L*ps
z*3}OV*H^*&ip04;uR!(N)
z#XWc;YDF-`gaQy){q^)sNl
zs3s?#T#eXfWt9FEH8ju17jqU6LSF4FYLpX;mZe*eAaampvMc#!gsLqp}4y
zqO`CW5*{v276%j~#l{|)Dz-|$IR?{yCMAN4QB!1STLa>4+e!k*%Pc^_<
zcqJ|&E@<#}w;G&?Z-;FO-?BKi6V~$QPJ(n1nSq`P)F|7jTZ+LXo!HIAQnblfvfbKr
zAf*6yoX5TfP|S*cbIbaF*&yJv3U9tozM!PS){=t`nT2-+;ae5HcMIR;tXx2D7Ydqr
zGDlvTJHQ)1iG1*7c8^^UI-}33c9C{b9_9OjLQtAO1j#5%jai~lr#?FuG(RzQa3Q;h
zYU+TI9&42lm0qsmNlR&C`GLa3O=zi4UF`HzCkLvYgOD+h)z3$z7~;R-GofFO_u`N|
zwjkqpV5&-Flw%LGQ;@f?U>o!3FD$DTc)P+zw1eq&NkX*&KiKX7z@%=$P`~G{5-$Yh
z%^^a4+ajR*ZPEfdJCnSR&UZvzW67vg4%6W1b0f9g2tEOvNI)9_q;~Rf*Po|{RGNdx
zuG*<)bK@m_6{75pG+jk*p0U(VP@G1)aI-ifm^eA!s{y8YC;FNIlLGEV7*mP}N-{5t
za21$0T3ioAysN~Fg`JAgB@b0f4{_>uH1LdK&qC9-sP?@1-wa2QISS_ajmFa~bat*9
zND46?j22qLUbHq@fof+l6n`GSDX=NW+At-I(e9K^%RP{_8-2XlG?0y!2
z)?m_Yca|5qK6A`af@ZaP>Bw0D>ROUj%;yuTb|wt@4Ppb6TXcGqqOpkvY07RN8DR0hs7pFavb4DS}6yI$2+IV5}e)z0#ioi(V-$+bZt&Bzj{g3U9J~yB{Zvj
z91YoajWRuzrU17=FAy4pH
zWE8r<1(L!cK)h
zf~w0zF21_s+pdDSmJHL(sZ>&?mXCWg28W?@KWfZ
zw!EUZ@ZNfR%6jrQA00~_nUOpPW128E*wn&cut$La*bP%Vg3*Wpfc4sh#x-sqvs2Kt
z`A91Gx<-M&N>>2yo9b6@M|vshA~C}BjDVF=^i;aVtXUtj3l7^St_43rTOw*L@qcvT
z{|eBx@xFS1s{9^$9#I`1ErW&=yCNW3MCW98O`FN$ycY*P2Ks?-MosGpwj?pSzu_Hf
zAY>2Vor<(1l`)}cJI&strV&m_)x14rw5!x1MBFfkhsGU^&$_KnAi#fvfx+YZQ8o^9
z-(fri{;qE$`Df!|MT<0lFjr-LqFT*&$Vhz@x5!C)LQ-TI)sypts5YmrK4?SX$1dn3
zL8g!M$ggq}h*Nl1(dYyT$-YDQUiDkx_8-X8WXtp6>{{Vdt6Fxc~TD
zaAuldt;Kiwa!RFRqhz`w!aNqoWk>*D@uw-})D#@UWB%tlnWvYJc!L_}(g*qtOwl79
zj43zTH+Q+S81ISDM*Cz0KeI@ssE+|`RTmTo32M#ap~y!>Lm4Y1-xj|UDL^TKG~{T)
zo9(|_=h?hxrxPKJjI5s3zbs>jFi8D53fFbt_qhan9vtn@Uy3xh51Lq#2@?m42UrlL
zJa5|sI4j&>YY5h~06&f=?ORMGZz8Sv8n~H?lmB`>oZO&6EJZeyXUU~Yl%H!`beNHI&Wk@4Iq)tLxF|k@)9VB~VwG^*tL{3I(kBaB5V>3Emhx9}4l1`1HX*c$&UpmpENV
zwaAo7bs-7@xpfKgYlX3ssXymW+(hk8=n=cBR&NmSQV7af3ReT2t!WU;xZ
z0jLtlEdcsPzi)RLcV9^v@xeag0C29QzDq)lW^l2*hEbKx3>+iR`^HV{Ci}v+^&adg
zGL6mpm4muOog?a?-7om4F;qXv(|S*z~KGD)#PYWXNmgGIXh<=
zObxvI+dcV_Q*Xl-O|;n86Z3+xSxsq$*(@k(cl`>$Vr^#(ik*t=$D64}*O+!V7wYC(Xh#7z!|5D^2RQX%&RoJCp-|Cb{>ysiqmo>WFy%
zZ>`-4uA$9_BiH3q2z0uN2&AFshaJhMlSXu~eVXknUr(|WzfdEEti$+pd
zURbM)v@{&#Q%b~!$>z;~Nk{rWqbr5ic)ecphRAZ
zT_6S081=Kk)WX!Tg&)@brz7eTMi!CDC+n?S72>=g9ZhzQ(u=+LfPJoabOSE^@Ss$7&$ku1@tZ
zFVOI-+`w%iT7u^FvGm{^^xIW7H-E!|tjOx^aS$UK1}Y3p=4w0j-idxJO9kjdx6QTS
zp63`EDw~uGGg()(sgXWk@~CT=X?;%E^mEK4kfFvgzeX8#T5k4Kjj^tcs{@5=4jy+Y
z4!7TsEI|o?d~7W2w?N@0ptdzLAI!+-2{62G5Hz}x!v8EE)fPDl)tFDS8bFcDdvAuG
zw`TC(K-L#T{oR~eH8I*sm8$%GsFaoX^#BoZ;r2tL4Ur{-93VN_UpIhj1DlB1dWE#f
zXJJIQ|9Lq{#}wopP0;bYlZ$IlG4DnH%|6Xf_ntoB#za4urS!GOgYZ)Qy!WnKCIvx;
zRY*2NQv5kh;JI=u@UIIy@DMV*ep%^9MB-~UA542rijx2arP>y67
zJnD?bR(F3sn5Gf9E0-No7s9hPHMS`P{=SMUJLYnQ;zLRYa5
zDqb!_22%eeko!6rLEM|@umF8G7LYQJshCc7Ur7m5I4*qgBa6^x9S6K0wlHBsnxSYY
zNNd(Bh{=-+wguV>E5^0x*hKmdl3pT_%ynENQ8C+61){kIB&FPJ&%GL3|EGQbYOD2(
zRiuCNSpjYdR7(cL-z+z2
z?5Sie{QBihs87tH^=8D(<}&#IK6vOI)K%Mgzpudop4eA=7`JspjiZPzr6Ztcp_p@g
z>h0q7Y(H0OC5JPsTeTrKBqmox4X{8_TeU&3iXX7|>Rc8dboXczo27yR$Ts1-$rI4~
zZ=XZ#X{@$3;T=Gyq
z#cMqGo|aTVGEvZF?iGEY1NV>^=9=7wgzv-n?fY*)Y~hifgou!5IhA7!>&CX8j7
z0B`tD!F>1@>oqVV*`=bjmQl%L{S@$k{R}7@m;#qx^;n3b)Toj7+hfgcOpnJYF%t}+
z_9C&9Ql0}oTg8GA>dRCc!nadrik_<+%^8sdmr>ILf4exstWi>TL@47Z{8X;IqmN{m
z2t=frNs8%1A!8^NZfy`xJ0PL*DW)n5z*qb}Ky8$Fz|VKMSFm|&&{0iu8`Nd?o%btw
z;-N+yftQj9S_h?2PC)OONygM6o-$-s^8`BK;$#{!^38Dba+<{eIWUOMl6G)W4uB@D
zDE8}|y4O5f@rU+w?<#JTE}i-${~ui@m;zui!0j6@4?x?3O1CguV8C>nsf0er`6R<40Tm~QzNBwh;l%)O8tad+W5!pD?_uB%iTcvD1a~0*;ev)@~-IH+$^3~f@1r*
zXj}(b0V;V?AQJ1G-y|C6AMpP5n2xJSy+f6HW*>(UC}w);
zO95_7PnfB>PK8Y(b(D{
z+=GowBgAwK$H*WE5}Yxn)CI}Opzw`tSthj>Aa>12F^M;~Gt2Ot={XhVW5b&P_>FQrv)&*#l3k2_c
zW$1y((2^jO@d
z*;#dXEvyXy`_5GXZ9nJ5Fbg8_yh^Vc?G-o)NcAdGq9VG+d2mZ&daWE^Xn&wOrx&?f
zwGu*AJ3_+kW&pTW!CyK7Jw5(P=#ux+Q}_qSrtO(j>pkp|OPQOL9tV((L|2G2h?wSh
zms;+-67FJma*S)R@sDo5l*{gKM_p*kq`uix!ngsBi5T1~x*knzXxirb1k^xqhgQv9Q;%^1H>Qrh&e
zgy`hwL3+*E9$d7|g^}w=r+|Eonus)0z6oN?I1K(V<59!I5lnoQ-^|v?pQOyK?uIMA
z95}9`wngg_jrB=vunX~#c`TG&LgdW;hUdW*)=G^>x?r6FzI2VuR4my6jut(c!?E1i
zKZbJyquJidVxyNu%>m)E)IMexj4N6$Z|L$rbjywUcxOZwvAYyX^S8?+!Rt-(y!+TO
zz&7~qYd9qT9Q?X0dx>xr;=CFFzd5FyAuJ6~eE$_nq)o$yY48hz=PGs+V-GvB%m1|}6gO5PUN6LT1&pl%=
z7RB&+R6|?i3rDwv9!^_ZET#TTvAL~cUGMp6!=n9kWE3$*{P!No`6yLUcw}q8+3PCZ
z@O!IPM$!&0AUt!VHspzZLpEIf+X8iG_V_iuThkp!w-Nq8j4=*NIz-72VQqmOxcA=N
z05xG3YW3Qc1{3ncZ`_IN_EWZk2EFep_xy`%6W7caVVZ7YOdr`}!je1}2iWjxLgWS`
z#;Yhqu;C4B^Cuf<Q~;!5$Zn
z%wC6^D}PooNjnuIeB|-f$Oqzsld+f{ji3RvBj?G9hxxE8iekg^pmd=i`m4qQ^G`I1Bu##b}9aIRR1&8
z#q>|b5iY-R=gU>WIp#-1%Ed1hj~`&9;i8EcI$PtRnZlcFStxa??_Bg8a2jj5JN*Ls
z@;w*@A0$p;hcC^F4_#>i1AeH-gxMgHrqu7%#fM5>s08-WF=!nFT+x&vGt0;z
z{;XVb4iBzw@j{FmvyY!_;8qw6^G5^IIK3T4&SZKg+KY&=hV^lYi-aeKPgV$8jUV29
zW{4V!*vjHIXJ1Ow`q16Svon{P8oh?N`Z}vwvwBueuh$W4b+6`zQ;#M8yo~xY_$pUx;e&m+2rY?j#s7LakSxpERYJO{p-xuJXGUw
zf%inmX-SVchd&oa5&k(*^Gl7~qJf6dvkdpl*p&0o(C!X_)u{pz+P-Kh{0XE8hcgF)}@tYRPI&Csgw{yjXS*uy%k=0y0a
z;4QE287UEaBFgfGJF~qtKjb@N7V*-5X&!z|Au-?#GR>i3YD`0(nM#qzp-|4
zVSdzy-B$k{F#-_hjSk`yiC(x>uOId}wROrfac!_cOtxz&MvfdVi0j!CwWU7Y?-016
zU&4{*6J@iTO+5Kh7uZ1vjtUZG92GjDLpVEll!*xQ5pxAs(C%cbgH(n#mT)1ZR(w3p
ziv)R3tkfu;P?_zaA@`MaP{_pM`R~orI}8lkASpT_UT1y;b)XmW-qEx?kyxFE$ik|N
zqSmoD2VRFQ!D8b(zebsg#KS`!tJOx9c8yWl=I{uu$YYPM1p9C@<=E&H!I_OyT{Bw{
zFwMizbfV_wj${Mp%XO_H7@WxXP#39+)9BX2H^+%9^8PCMVl4h+Xo}?kqbU{|UqCB4
zB|wnUTaKaRRP*j$P#MHB7kx7<@L{>qpcIrD1dL<5r`VsabBnxPXrPdlLbC`O@&bm&
z1x!)F_cpKYYq8n3ML;K$8~%5%8@NR`2ID%BI%R5DSEAsRL}!(KMoW{t-lB-rzqInS
zrhrR*k=peC4eb&*Rj7`!h5))p>=X_hFsAL#!=#q&CA7&y&Tx(~sQPeUsdLJJg5@`R
zm_7AqvWCY%BFvB#NaOZg7
z*|#!S(t1Z3aH}!zLQ^6!8CC&#Og`sMbw`xQ`mB!a*C-3;GMk7_^x%`xXLL}1#7*%3
z3jxdSE<<7aX
zXiNvV1{hFMa=|Z93gppNHT4@;j`)J(H?aIgn(OCC{mTV={&m(gX_q^(x09C_E(2tY
zlt$@%>JbJWB!JHsLRt0v9$9WacPkteKd{2f@5q6
zapt*H
znGMdZ+j?&Rt|f}PQJoak$B=*fdguEZ5x;|H`4#~f&+TgZmmBdenz|V=9g9CE2sWAJ
zGUjQEW;j4WiN%dJM3D9|%2t$x$Z*S|BLbQjh6ZQLL^>Xx#k6XRQKj~M4*wtkxJ43L
zWkw_Gi@iW|)SDx`ff6yhtk+9+P
zUR4t}%0cb3E6fR-7$lRSwqpTlq-ne0elk9OmX5i@csldZoe>}oCEj?ktFINw&kK(^
z-Q^=dqotpo5ui5FQD-j6ktCvqF0z$npdty$nmigaOdJ|zWOu+YlOWqqb-qAAFy_m<
zAMkE9uNTIo=He}`e8-)?x|BEJD-z_I<;C|*M72nW7rvIeXxEAvj_eCVJ<*#~0@M~K
zmK7j*o85UYc8Coc=Mrg9S1JWHCO$p_i&Wd987dvuaTj+IQ++~|?O<35bKn|cwzvaT
znxCVsUQa)}vO|d!|-BU3N8kJ4ZUmyr71<10N#Rp
z7Sy1}M8^!yYj9<2lQYZ6!+=ej3_?%ybzP53h8nuEJlJ{zjU(!L&n*sRCR?(`kW=q5
z5b_-B-P*+yttg)?oEG4bo|m5Ei&q9Z$18r@$>`t-7Ic!Uf_gn7O=$Xmg3!kTgq9y*
z?+^Imn!fsSz!c5R_fBBTIFFwuFv@9tx*uhb(X_Slz~iQ#QcS>TJ(GB
zQu^mg_{L+HR6{z_QoJQ;w+iR`Updqv&@z?v*wlnc{AhGcP;5}Jdqp`9
zSOsfxMPfB6f->Hu1z)Fd)YiXG3f%QDK7-jXo}#QASMb#xo*`qCv^5x&`al?OCru~S
z;tFg%C(^&esuq~~P3|A$N-z*-#?Wn|{?>%M&5rk54^)Xr0pARikKljCG8#@$cAs6C
z1yyAkMQXXLabnqZX=w8KOVZIO9kDAf2iE!WHKt%^L4rnbUyoA*9b>)1Mp;<`=pq=D
zM6`+0>jsEVA`dO7v8|*sOlUf@D*XreBT+XWhe-r2brP;b(EPPcPj~rnML1~k)ecDA
zeseJ0ZFO)b4z!SI1wD=8kwzV!m4UvuqFZrj+{@xBDY#WRoCJw&2TBj=FCKj+61jVF
zrVOeflUgrz9eSKalq86zZ_$Jj^oyDdWUXWrSM%H51^9HQeqzg;YExOeCa>X6B;>2G
zRvZ?)Za^F&=woQu6EQ*=D;VG`he8Y4w4mtn6R-PpUHUN~)p-Ufz^~->=ccZ`2{suU
zFYPw*ddd^sLJTR=%fk$LUW%nniIjCxqw;qc0(?@ryC@^wsXhB{Vzw7+O^9+z@Ak>Q
zk%^`A?g*LC{D`K^_tyySTi*%5i|S9X{*!5E`0}gWxJR_Jw?h`@MT6quN8B?uwsI{8
zZFVAo%>=uusI{U==7%5?lDEctahO$~VzI&b^J{Rsv6%3UgXdP;<+TolSNDp-%dPGu
z_onio3$5NrF8_=Zuqq$Te{$nVTJUH{d1y2IKE5>cacQ;)(KKW=z+nYFOtbEqZ}Zc>
z0d;$2NhLzWZRE(e@J4r+!^8|&lIo+vrIS};iGtWm4{R;;cw?icJ;YsPQUmY@FYRe|
zHEflu79$d9*f($tT)LtdhaE_ac-ZE&ES$90~STSeyfgP;l=gvv|
z4LA$A9{v}0CL-)72>MqGuxF!1J=P~pPP0hWam&Y|Y!h_?9>RD|p(oXy*_9UQCEZC~
zdbS9^SPGqnIUqR;^?lPFdt}s?kYH68kbQ{Z-Ch-0O&#=Pf|d}?XqoV%(V9)sq70p@
zEj40#^$D^NdoOMVZJ+mWbh}#1#qUXx-K^DS#COY;Aw9^Fz;k_A+EA`eDf1FG{&oyl
zX!-@o>CBBaGo(MPZh)h%`*9!uU4#CIT*?RD&0RF0rUV#Etg$)2z_GXd;B-S$V;kM8
z+q62XB+>r&?Muw0RBfGI<(;n~*LJf##!0M`J8C-FEo*5}AUtgN*RHI4#R@)Fv*TjtR6g4?7(!w8G}6iNa!T9kI3vG9qT{CN
zycJoySW+3-t5NC));~&l5!OVwGk7hHwh~3I_}*o?-96A#FNcRN6eQ&L`20~HIc`Cz
z(xEm4CFc7ek9v+&+ZXbSta!yl-zwS<0;S%(U+7gKuDUHvFh>0Ju~siV!6eDff^0qu
zi0q$^X?4nCUh_QW3^ek83+VegV`P%k8o{`!2_HbH7Zi~j
zG$;`@{cfC+Xbr;Ny~*t0n!~SM|2tQg0$|C|<-^dD<3jbrtn;JYMh&o%V2z3L+A099
z0Q%B|!Il&`k**=@KCvZg(WXjWEh9u4g!HEcSpS^joSQi4sGp<~XjNC$yLIw)fTIl(
z`}9fd<_V2FjG&=rDU$ecmFJ7L8_T_v*C_BsY8D+ZUA9_3iw_?R)Qugd?~VtK8=$i<
zD4TYd7^z5ta1erpAYk_ICZ%Qs&aFTgL`M%DC5EZ*86D2@^U8270=U<0tiZ84E4W$u
z@C=|;NvOMyu(4N|C|^pgS8Y?~NC2v&Yf=|zIF(it
zgA7^P@g(i59eL~u7}yxn6vq8!Sfrk}wAG*e7~*D#RIKo|c2bpKpLGne`MmggVwghn
zPzVG)y(PAaeie-w)?U*~m>fl=diQr%&TRqt9aT=8zz4q#4vHfY<}>!&ihb)Ruz54D)3KT
zU}o2IYOXyJT-!C}xu1m>R{Ry^+MYrHa%)ZjXyr7IN<8tu@}WGC7}lWfKNgB~n1Rv<
zYsGvwhdM~nBPtAyT1qwym<(4sHXim2Rz#@_za^uKBc8^*Ip9r^r$h0cciQ2%TO(q$
ziGeaa&=M|@hD;~7LMyHTm@sK&HWT+-D^F}W)6&MDualUG$gJ$fbLh}zYFjFsnehp;?
z`2a&h$}=G_Seq$`rShfVsv*U9DeH$)0$xTv7!hk^eTPjRBMk>kr$@
zg%%ON+xyR4rf`iW
zP5a+?gzJ_GSJ1?f65?HrrggqvaE++16>>r`8GvnU1tm&0w$$WAKt
z_#dw0{KQ{#I}B8x|8HGD5G=U-C>%zDi_%Ks9^z0^ID
z2MjCKGxz-WkggKD(;fe8wRQ27>Xy9*yqbTXzw}7dh!>3xtNp12$_O(R#g8H>z
zGCL(%nN_hQHmii0-}dne$dBp~3GhQ#Jyj+E4gASwa3ooxXR8@asnxuz*^E2G_L!tJ
zWagXuw1Sf5)A`T|z&eC|QZ##`>4k`slo@gKzruLp)WU3y7(W~M*`zQ2aU?g*l`>P$8+=({5lbT}X!ox>YtVtc%H
z_d1$ys6ZV0wd(kg>y)dF)jSE)vgh6$1#~3}w;etUBor}#(yMvFJ++~%xBKKpQTLmo
z7z-N3Jum#(s;a(s8?#GLoJX^06@r#0W_RFF#+=qW)#c33RR*)o+>e)hPTHX)05Q}4
z2IhWYiYJAtWwOt)DJG{#mbH!_FX&!JQyulLl*jGswzv=SfYm-eoO}B27|Il!#OCVD
z>1dLTiZX#i6;e(Sskb7>69%cy@GH%>ga;0(9h5uG-Td=iGQkPIXjg{tfBNVnAK6j3
zY6w`&KcO65-}MxN#E%ua%*ThA5WQ*xglHd_omQ??!6Z*UxAvGwIZTZv
zt(YTkWaEof_X@hs&T7&DdrSC)XZMHoh{A%qs%3EmY`t6&Q3OHw{GXSf_Z>xI5};t?
z{t1)=YFi=FUS?98x*hJNmQ{v`15*PiFumXDl1aYS38_a^ieMmxKtUmW^F5!Z1Zc{Y
zNRj`42zj7|*f=U~fFZ|^3$ltawYs!=Dxu3wm$)+)jFn2_Y|03Zp@t-k)fGI)+TBAk
zAcx3Kq?4VkYV?*8a={;Qqoo);;ps24Uq=YXQf9B^aXJznEx2%pb$(od75u}nMEZdh
zV1~AjUOp5$=?rI1l!5KX2nGiMsPU)d`58RbkobH`k`J7!h-JWVX8C7I!K1z_u<4W&l{6=VF<8#-k;lJS-9nmAi$$c(bU?=A#|=N2c4aD1u&F
zIseg532>JVM|c6k2bOYNt%a+r#_g_m*p{7x1SaT3HLVAutQ|oLcMX}ww`_&_2
z?Pq1MlqZLufViqo*fK7>1{#bzeUX$mbZiHL`L|?f=~1B1x+QifNA!lS>Q%}SMI3?+
zvw}=Wh&dt&!c7BsDpdA-Br*(opiG`D;4##6Yk}JmrsLF#2io0sWzgpBKlp%T2uifb
z{D$sYPF-v(22;U)f}WRkE&H}0fBAq*2~jhVC>HHX$semHGX#I%V@kK{q1Q~5Jm2ZC
zG^QxGd~nz1G?8Da6hfiW9<~@vP16FMs44ko&dcy8O+o94;5dn93sugQ40Z0imu_`P
zQUJnA0gN$}Sn3)Pn>=>xl<;gmzp<_MeLiAEJqZN0zxcdl{*JO_EmV>R%*wc1)Nrlo
zO53{0*Ju26v4W=4B5C!GsAJiU<_Lbb|7KA;i?llW>)@F!lr4XeVuz|;2V}ZJ^LiI@
zu)_jm3*Q7-an^dIBCd$SNO!k6MOKQnjL}N?PcXR^0SUW(@P5^TOf5c8p4^*PX}RdC
zfEw#lyR0BlsvzNyNWwqcaEP<`+>5bVI{=LGsw4yJ?{J0Hj5e@h;K(InggPMvNPw?=
zbx(t3Dbz{ex07Edos
zt@(Yb&87E=%XdmHW&#XNoA2yAI{EIu-D@s!ikTi##Fq>Vhlm-`#+rF69)i%%AA@Vb
z-=Tui>Sz;s0wPr(fX~l^!3Vh*abQ4ZkZ*2l_zlt0g*;if+kuPDHZ-M01T{BQgvaZD
zSVeJAl0SGPuOi}yiJqoGK<^IU7=3+|I<4Wt)v`NhQ6pjMqejAvnk@AetAEvas_O6i
zNCBT&gP22p=7{JU>EZxKZy-+RQII{kGlf;k`SEinMnievJvu{YQ;HqVIrSFIk1)oR
zN-Z#*6!%%|wgFgE!tb}lv5HO{Tl1|XmhcLFt-e2I3Wl)ueK*d0FUjYJnbG;IsQ>P*
zWe9>hpswPhK})W-^z|h*2%a1d-6*oL2mJdjy7`w+M3NN$R4#r26PBXmpMN|(acl^!pZt{z
z_VmftW~7XqR{BS6;)jF1h}36F8%fM{HT^}6kY2ciqQ&Q=lT$wlzgYoi3sO3Atca+Jr!_%b~XWF<<
zp{m7$yJa5VCAcLh;A2dBupMF8+^D^?8-agj>=KIh1HAZZQl<71?c@!WpwlE^@{#D7
zz%;>;y3X_#=ziJx4*Rvo0VqY&Tcy~R^Xw^X4$A1IKH75Q$ho8X5X`cr|3Pv9?+)JU
zE@!Laed!^Bdb1sNC^BB5eQ8t$Ma*QV-h&LsE^F%mzZLw+yDcdjj<8(>hl$QK}VqM`SHQ0>7Vy-64~6
z!WT(DS7q*Co=-Bc7s8lPKg7aL+FQIj)Cj^w!>gq6>7z_Uu(y+d^5JW&+JE{ge1Qxp
z0@i!1k4;bSl0Zg}Q}?Z~7P
zs;Yocd73^*DiTmuUWnWTZZU(oA^iDuXEA@X=&_mEZ)qZx)2FoB>4z7SW>3mQ
zx^=8Cf)sskfx2ko*@2WlU1}Z^dd{IXjaMwmG>-AuD%&(KXE1p)>7$2*xqk!+$O=NZ
zmOD@?R48?68S9~9RIem58jj2<4X3GlijYZsvh0Zlj^%Fso?dXi9^9;#F9TE7LguLt
z$S>5UDzqIwm?X>+1AEc1wqd>>VQFWkXp+Qem?~|Qs`GuQ$kjL!8qE>s&+R={V=`t0
zLh&ZSXAE@Hk}J*oR0B?Mmtu0nir+1-8sykwb(jxk8SW*8gjyITH>2@Mg5OdRX9aZP
ztCO+)p&N%p(&iXdU&XOm<;>8wT!?60#vd=@?k=<#T6dHsw-k7H-ssXA&bex-$2zKV
zWC?MY!k?P({*V*HnT-Q-+Z@xcoq?s`3hciDM8Q~Hm$bw_28NWAL;ihMOgRW{-n`K8
z-v8HbmYbK>X0p*Q=bY@q36#CEt0Y8;6E-U}N_?(7r94#QE?fh5=3ecj*DI~e)Q(iz
z2$wz&GfM^p&5`=o2`2%yL_{M&nJz%St8j@KtN
z$M*)S;>dy2iKBsQ#uU{ANo*^a<8IZ01m-{>L7+z&
zt7|WUtCUnw!taFRJX3{)f+vMGVwL-&U$EJ^CJJCPCXX><@;xi_edXpRtq3UpT)blw&?RYVTVgk$<#B9MUIjLP)Qk`bSo9!
zj=q;aY-pbqdA`OsQzS|>wu~{g_Is5O5BdPwogh6E6gwv3t*qmAwXNcwQAd8fmgX4W
zD_RaS(>s&fUYfraLFFpQZ^jJB{e7fzlw4%^yZa0IOj;A@L|G4Ys@J)~gyb}9?kaE{
z6by_&Atfb@KN&g(LinOo!Dq>CS;mlS(<}NbD>1g*Oc&@U;V`EF>8s5MgK7k=0|fw{
z4pwU_ZP~zjfU#I{kNYsF5u45*Uoq*igx}!E3Z)-jN)as(1wB!~l3l8Z8NFc|4&Bgd
z3vcuYX6*X194%K_OaHmaP_VWljB7Y+LaSv72Y1^|Uu@|zD)49Eoull5SD{Oi8*@TL
znLn$?_fszN>!LKk`@Cq1$YGH8JHUAoIC1oc&|da+0g{Mhf&uRFhom2i8#P#au^C2*
z$x2pFK7$O!O^AvzU{+Sf(hQQedM+Wn*PI(=86!>=?#T!c82>(n7&}X~^I%Z$bb`9F
zynPOYDiD39pS&8T8hG1G>%$&-_g;ia#U3FYza}ZG%0|(~ReiV+ckuV637pY%EO)eb+xenEIq-
zg^YOa3=Z70A0AHI2RcT#f4@82xPS5K$17)GA&8RQ6KTel#Y^IWXjyR3Ag{jW!EZcG
zi}sk@Ez}7&P0xa8k>PvEr4LypN+CvhSiD4T#`V#}S5XwamnF!d0nrD3KW2K}>qVU^
z{WV|CbZaM+s6{ee6w1$$1RN`JYn>BR*sz{BrR3;a0SvYG>$AQGd<~7L9Vz&M=Mj;v
zFz{lGKX5ff>UHY`&hw?eheB|Un@n-Z{o^)?t-<9A%R5_Ga*QNo!v(o>a8gX}|`GV5kRI~D4n(^MTALH4piXzwv^;jYk|ZSJ}r3A-h(r$vvM
zhi<0_l!fM~O>V}Lx3Ox~l4fZhkCk0xJN*=*%s~d>k`-x7f6ba52GQR_`RLiMZgcRO
zrA~;=+e*G?Qq9h4NLTSi`DO%|y?d6UZ^0Pia>~NgRLq6ja0)ZL)UdmyMcPJo0?nhK
zxEPFD0HCMrjmow;^fW$E!XIT3b16c>a`_U#T8T@G6jgF69*6xltW8&v7M~GBZqRH8
z-~jqfCl2cfNm$m`_YJpo)TwUkO6)Xa#7RVZcV#Bx_zUZUw`w;kcVSE0HEQJ&`luv#
z1H-}p3mv9kVRl@sDUM!`w7M1AOJIiU3I#g!Kxa@>cd%7H<~bIP1u$doQsWLrh&0@4e
z1DkwXh?hz~j+$P{JPkT@`I%7iFQJ_hT<=Do9&jscgJ;8e+r?-OP63kec7w3IUzY;I
zKF3m;ANyvydzsvZqBGQMi$h!zwG0DH=!j&=vDi=n1{+IDP**9z`ZY1!w58%9V$;n;
zpjt^H4rXQ|iYFP;3OOIow><6$q5pYT3LLOy1<~Hba4b!9kIa;|AjN
zd4B@X8_j->!ZhoZ;p~8U_1PhJ)Y~Tu+Qx${;cSHnCZZv9az
zo4Ybs2~jweiynLT)c@#vqFw$BtJdf?NHk$Nn_FpPKaXw3IVz@iEx78XB~AYUWNaR_
zOE7)Xj(hi(iQgr?I86bQ7f`F6UKRvwvIm!6e>*R+OzLrwhZj40hdwFC>uCnYA$dd$
ztQL4}yjWv0t^LW+TjR8++zzUag5-6>$A>EEZVllrCIzBejcO)23+^d9sqFzG5|OG#
zRB`5D4t0ORAH}k2i8TX-sR?Jmjnq45P$b}R$Se1Haon3D6R|5@D^|_2>ze_FQmZi)
z#~k@45AZ`R98LNE3MNm`2gszE6mFrT<5wc7M36U(1diSD_Mg_P)oPb-!!eMn>|AI!q=3p`ixYCXA&a~VSz
zf2^uX$!3viq-=C&};`hSAcVblgLCBVi>RBg!^Iv&X#8N|{np)69@
z-5fHHOL>!**r;P$M4#R+qvxh8JyTgSo6e-Uy`lcLY1p+*G;vAQx+;EAm+r+$>!Jn#
zn-WC4G6Mf?x%VqxXV5K6u^twDL{?QBYU8Q)&&%`xRk&?%nM7FklwsB~Wsr4b;YI7^
zF8s5`cH+5uj&Huf%5j+6ih8CKLTOeV6c{jRn4-Frp|02JYrLgbTI`2Hy
zP75vim%B;6_*F30Db&+@c5!1CjUFw3wE&CkpVM#Gr%8=N?2py@%PKh
z1_;b`4DO7{;0oT4oMKf4OVHKGhJ#%-(6QAwNB4(OXftyKa0tE?zD0l^cXnAq+}`BE
z8*RSAx&iRH-qignRBBc=?W1NC32gcK7wF!b*@g$TbGfbr`WJc|DW|s_NQP{H5~eq%
zRIJqjORJcECY{5AR0PE$KAWa0X=o}!W&lUL;*yM(R!YAc>5Gg%H!LoKSdTLG#Kt{+
zKzu;zb+{cgRJMKl)$k=qmR4`wh*@fbSZJh21pbpIVEC|evS?$fv11%AM$d1n0+p%A
zX>YFW^p_8hmX7ZkDX3=dgxuK6@Bk8SH%#2IX#-k_DpFAj!LQ(<*U)9+Yi8XGe`0Nn
z=+A!(F3aU?_|B?vYzXIOxtEOP_TQ;|ylG5@UytB}u`IrssbQ9!`W(g82o~(O$ccOFut8JPlDatiIftLg9vJp
z8UhZc#s-gh?GYCCsxD9cp3?EKkU+^8l({8HKNA+(^j3rqlSxL>G34f1R{75Y+388z
zGP7QSd7k+oHv|*6`e_b_kN(<|#tLTr81$Wa>$-6~2+BaeEl6o5CDdq{nM2Jc3ACn(
zbRgzm!08$HJ56mp3}DaaU%OpGBUJ#rHbZ=5hb{p-jEXy@+8uKhh?bDAbm7{f7RCK*
z+$c_5!?;4-PGH>gZt3vGPm?96&mMtn!4t-_;d!ud*(8%q(iMsN#tGL}y4AHYc_~6(
zwP2E5osc9~ra*)hSsCz74J0}Si8?g&9cHcK9tzQKB{-34zCY5Zqza1Dm_K2A>7{=F
zSn_fJ*f~Li3bQbASxv8Sx)xdx`(QG1!q5hfcc!f7=RU3wQFtpu9toAA>`6bUBd9iN
zJn7f&(KBOcY`sShU|Crn>Hhv{0&Ne(1Ocah$=mcmDwgHNElAZlA1r=!oB3^`@-AluJA0N|TBH
z&+{}`{j?-`z*aQ`e*|X-r$bR*LIW>op_Fht7eH@%4v7(=L-~?xJ9d=vT2A!pWquWM
zYIV{rDMG2g&ida(s_I=lvedIQ|Nq=+$(Y3U;sf=!!!k#pI+MLjbc;ruj4a8e{90
zXGLR<^!ZZ?QfflphR5hYmy{|=_{jn${dzUtCeeXf~A-!3MpC#?#x)SHA
zZZy|NsE!KV$ghBi`YenI)V-r8ymH(1ztC22he?UAl-R-;9%6k^nF7ts3M-|F@C2!-
z;5%y!i`bi0E-^Vzqjrx8NhlNJK!aoIbKX8pD>39LpWz{DuN3!srL
zQ@DWI$T?L|%g`T>0_ZZ&6-8TQo~2_~tc85JPlG%OhO&sdNm4i&`xMOl3WRG|L)4Ys
z8m~)R{T$iIMjwaTrQ_H|%iVmj1Sia9g1i39qPGkb{#nxSzNODzQ1!yOiaTeSXfjQ8
zS(w&#qQH|*O&bR?PSZ@=*VD3f$AzHpF6!lDC|YGsMLmRTw7JCW@^obOlEOu>89P5Z;DQ*|-bX7jBVhgukDm4*^HJYEJI>|=t_nBIiSiaaFS?+(6n!I0jPssm8
zib$zc!q(+|ToOVb~B;d`zs&G^ET{bM_^djo_^W~Zsf
z&aUxncpv|W-$H*aTr$a{+NMzcYS7Q)b8S&kgqTelMGpZL&1kkVH{xUSlBTI+gdUep
zEO)?mxQOxspd<>xd!kLE>iH*?tsc-TgW`DEOKx>%lJLKpZp5LFOrza`*G~Hk->Z6I
z|E=cr*i&8)R%MP<)#yT^0AF>`C78X^hAD$$RdDGk1~~yX=g;p9CrzyHJ8MABCC!?F
zKsot*`4~)-1`(8CzLyQ`eu9Wb$LC`~Xm-2GXBo*O|1G6i4k#N6|Fqz_%tQC~yNqAu
z3OSB`=Dp#@QiTfF?uT0!u66jkUPgze&2cJQ(7V7>YjZ_^BPMNpy^=!{M78z_0yaKQC&8DX_h_vhXJ;+Jgt??x2!2u;8{z2&?Iaj5v+JrAbsnY
zqh)v@Q!%Y#tb_y2!|C~+G_;LnLK^q)E=)~-*HyHxxC?S;(KS)#DstPC?`&~Jk-NbX
z(Qg|IB8;T7)oxo{xBhc!<;t38LS0(0R
z$=mjRNCJoAj55b=Jp)I0z>Y)&>Fyl;*++|8S~vIAMA$yJ-xSyZjzSR%@Fh<&a8!(F
zb+b4=Qku|4S>O=%qxFmx8mE$t-k((G=QN_WlLTtBtoINa%LNkA4LtZK$w?QDMB%c3
zAeU4h*E9G2S%Uxp`wCVx-j!Yw8F!OCg<%HlUA3_u?yD*(^_D}30ZdF7@~UuxPX!wh
z+S=)v!c9oy-YXbVE@Pzg;1iPDG~*(=L%tM^#EjDI7$gd$zQ{z6+vosY`D(+Ka~~l8
zAqmy%yPBXPQM|pQ!zRS`PMvx#3kOuANK3E-V`BV6!4%#U8`Vt-jtMN7>)>iGKbp$l8CU^ROcf_R}()mB;Y9EUw}o@%opl3temRe1~=(w^H!swp#^V!UTkOpBgE_*y
zXG3WrJ~EN~K0M|P2~b$fc2Qe==zta#T2Z*|Jua~;Sx+64x-q0qociJ-Y!pWX9(Tgl
zbf!51lUCvrTo7vyt~%zK4;Lg$KdgW{R=NIr(ilH%ANeC%iVmo`I>tf}eML3U
z*a|aM{emi4W7XgID)=8w!m|5!=|m?>xO#WkET_g2C>~6QHyp54>a%hHl(LEOBUlah
z;WN@~1{2bfVJCO8ro`lq)X0Jwd`V_hg!*2D}4ELmaO;1uOx##X?cX8wxiSq^E-j(Bg*Pm8(s!;xeN7T
zTE!}alaESbEkN5WTSn|-ij@N5LytGdW!ud%>RYGwW4W4_iXaERPzmPk=552@QKp=I
zP45|ZJzk(B#6Uju>}alSjSGo(Oi|Y}|2kmA)P@245uNCnsML=HoyJcf>dPGR!G+*<
zcx>VInAkh+B35#&c=%+dWh$=}OX-l7T1Dt=`6tW|AeAV0_rQ-?yqwm6gDKl(hNfOC
z{}cjg%TjnV_z^v)5F6Uf{UkirFsEwVvgJ5DQaMT>Jn0^h2eoefN95W
zKn>T^7|T=~Dr7(cS2JfmLc6HCtR+GK0bs^B_Q{dZrmBv&;i=-lDik?J&+vC+ACusx
z_Q|gN+_Uv$0IuT)FEeGy@cR7t(FhfqB5AqDQJ;>YvQxXxM~P$I5aky?#$};X(ck<0
ze5R8ys^>cC%-cu-+RT3igX%6|qFU#~Gcq#g=Y#E>W4rQEHH^vWPBBlnB(7Az9;t=nCgA8n
z4M2PfWM-a|ad|#5+^sqF@SOf%LksbdM3Zv&#>1JZ`{DW2am#-!8#PHn=nx#0tc^y7
zBId1F-sR(;%Ch+LFQx$yzOHo#vTZ;JcRXqZehWsE)HTPnEHq`gydPqASS;M*1exR-
zh5)Chc%4+k67HrAT`#3wS~xD6IHp2*KkkXEbQ+AVO=igv$4Or;Qv?6qG&``t6Q;@w
zYuT~Oe@O(ZSo&8sAQ*TXZS^X$@Z;Tkp1^M)VF=>J@zj&?8`FpBncB`_zT-erfEej+
z+Zww-|1O{XrDj}no6-FPtj|TtliM((*LBbyG@D7y?dQW5vw|`LP2Ku`fr%Ih$^b00
zAb%5xaEw2AzQfT)al4G>Lu`otUejH1xg~M@*>CS5MO}8x-&cpYz4SeSRV;iy5BR2q
z(w+rnRp*dCCA97F!Nq!52|LfKT;?s>(M16RAoN1s*v**Lw*c}+8yeh|T;r=71C^$)
z3KN)+#jZE|FK03yw0x8`QGjRxv4eDNN8C;SoY-ZVcKZmlw@{}mF~^qIittELd_%_f
z*lKSpylcsrcP=|yB;5j?XOVN8ZsMMA*ENv{MPwJNWp>{aOHO!;*XARE{IYiU*qsf<$qoaAXg^uXc
z`5dW=OPSMPy}l(XA!Q~$7xJ&crzA!>NpFFUHM)rWVD_n4g?^X(X+ewla{Mj1q@)K8$ag!I|=j`gY^y2Z~{vkre`@UPfa8OQH
zoi0K>K6ngLKd`aRcD?g=0&PovBH?wFQ!cXPGNj_%9`p7FZvGuuCR6K?4*nXuJ~ew7
zq7bROg_DMDBp-_6Pug|`x-H^#8QDLX1G@I(^?deZj+T{|{{m+g(Yq_`aLzeyovyc&
zAt=i}0a=;*65l&mRRR>ddsB)mL`QB!|D)=E*B}Qy1_HG=yTm?P^Tdh^6xFqe3jfk=
zaLm2mG?^rHg>Su
zSZXk6`Z8Unt<v|NJOOF!))D7Rnq}34<&T_
zMMrAWpd}~Djr$1tS4ucN7tH8T#W?lI8@Rj`Hm#;j8y(*GLW?jcb|GrM&@)FIQis)6
z`H^{x7xrkm=N=zEtd;tC#8?z{WOgTkv;Oc%mh5UQK+=^XVAa?NQ#ppc4?dx@pD+(3ovQJO=?4be5_dszBTJ3CM)$71xJCZ27sk^l5
zy<`r9{Sk7Hh7CLP5ct@!X1J^%z!)e|97S9bQH0T@{^!hpJUry}s`wYeG+pT*eDB`2p|ErMI%CVJ4~
zbXNOMC^AAqhV5jxYwclJew^?Ty@HciN#q$rXtqb6LQ0L8=uYPhen*KI;Bm|e79brU
zS~=ASdrWpC3;cSc+2C8G^HtPKRe#jllA}7dBzM(t?Mgwxv7xVUQpEA*9f
z6|9xlRpS<_2&UcQ6Z^~)%uYJOOzX+UcGox?c(db*9_3ZW9WyZLbXB36MJI+cPtZyj
z(titm^;#wSpw9(A&fn*Tj)qH+pcA%1bXyWX=pZ%tA&g!LMQ6t;ze4S}bn&yEP={ZU
zeNJ(=q$|tYgg_?XT(~x>6??5gy`P6?kak8t8Q_==&m=5-6P*edOVq_Kh@3KuD?U^8
z0`#sen{^pTc3$->Ogw6PjASk7eu+5;nM4PPM3F)t=jRL{hJl9y1p?mhx42jZ0gHG%
zK^Uff{V7|>R$9F(+H~&t-rQnu(;)u-jyH`CY;`NaVWsF2$E&0D>`o3;7og`PZN**|
zqyc$+hLj3rW;sOz>wu4Xm5e2KF@)qA5lajQpL2E*f7Fq?zIJJiMltcZeh99A0yG;>
z7N*f%x|l=YOtTNNQKIZ(gQy+Gb6i>emEBVk_pMR*0p)sLRO2dt8wurdSi=?(w)4FV
zMoE}+z>4$nE%J;$Fanjr0g!948nKywNF?+Vo~qnfOu5>H3}?TzmXxIQc6SPQ72Br~
z=z9`UuPn_)%N@jB!lehB2SB%l5=;E|O1W0*r>b!#klDfI`_sl1l#Lf!a;xA?`@gms
zeX5OHUdhr-#ktiV{`bv{gr*5t?RU42{VvOSW30(WkH3yM14s=}Z&2BA0*n)J*koH_
zm4&Gf>810zW11hINn|-ZD(;6RN3oPeZ7H^0^2Xo4gl$ra7fU}BQo?)AbW)IKYsuU-
zLrOO~@yQBuX;iH_2>Y*qZe=M(5=DpxS{USC4_AbKcmxl1R8Yv{k#Xpr)tXXZjQ!6&
z9t#J%Kh*8Fnd0Rcf4?J4^5*H
zP^%e&c8%tcB{UiOW9%`Tau(ySZlj*=U^K&xDRw^1iB%>}^bhFXU{R`=S$t?pTw}em
zH&a(rFWvtY-|4my?7@0&QA`ffz$DLrqMTYEy?T8|T*B}u8XzLBCsy+>SoHZ?#9X6p
zc%%I2U8R-0(S|R66W$p9a7*cMT7Ukgz
z;d_%u=Ob2tR7ShVJ51Q>g`OTmf*M9=VyR~TN~s^D%@p#1TfoVzf{Dj=zLO+k0(lo4
z9hQEs0l7foizP{Ym&cQ6YhP#;2ymEI9a)aTn;R_^=ynMApr7en5;nqRQ};7gUir~y
zAbN-m_Ob_S|87*RVaF$+
z`A`!;R~UNfz3I6~sc!`(CP^+ji!pI#h<#u9bJ0x;&L%vZ;V^n@7CN1zY>cj^8JtE|
zyDH{fds6;KkF;YxC9VCFMKfmgDVzPxWw?yhzb$3OxUBHXZW)Q@XY>rM*`^QkATcaU
z%_YwQl`zirV*nG=jEfM#(*-@Aim?!no-M5aG$T`W52MFI`-aT$jS%%?tTp1AOfnW2
zQfA&GcKk)lDWp>p2@)~EkT8hnt^z~OD0go}7Dt^LO$!l~NKmCdN%O;ZuKaEiQGXZO
z&@!q8`XaifdZG|t%}TYbnCzo+;QV_e|I&T5>}Ra+44p=n9}NYaV?Z6!RKt+SH`1Q1
zY1W1_FU$DCEC#*H29k;=q{!^3Fp0qbCYuRNksHZ21QJzYky9ohDiD+lVP$c*}LOx(f@0
zTeLSkq+8#jgZaSS+7mdDRC9BN{;LCGWnKJs1j*sCzCefFd
z>u6e8*;rTjd^`x&4b5gcoxw5Uz`VXT%qs0Au_!fB&_yNV$YZLR2yc~d6B`G0bQm)b!Yq&quRWb{Sz~zR$t2o@8ps@=nyH)|S
za*bP!1HtvWtL;Y^GFoFqS&|}{Z)0pJDvy9MWAU2h(gaicjS69dIwL&zJ;-PEaVb-}
zev+F;JR$-9fZ^3HpRIHJ$N;vjQKMMY3liPHf&X<2mH3VWTptgY#ty$0*HwutRX~PQ
zLZ^gCJYmk$L8-+HaJ!hoP1T$TR5U9yx`dvV%@iBZ0;K)GzggfE3F+9q3=;%t7V!GK
z!0_winw!W+pk;0C%gagN!YYX}AmdczZ8$R}b;?G{O|Q(UT+T8e3P|xQVtPEuYwXDI
z*dH{N$Mgv%xGXz{8OlT_*XnN+YqngY%4N?ko_4JHX&y
z)G4Iu1v1N2_T0wZtP5e6-_eTCjZmJKflsaTzkxjO^+*}YQub8^Bg_>6g=AA`{ADSa
ze}ghXyD5j}P~#a+_CQiB?5yuALQYzJD)ICQVnDFJaszAV+{O00DqQ2xtUiRakjZ8m
z*nlf>PY5o)t~-L9BzM2uupd?-;8n%#u;!4pW)O#c9QcMr&5nVo2tu2t68w}7op&dE
zYOV+#S($PCAr~**d$33Jca(h=-{tFb-pD4BCVYr3-KJuRQSI?`UI)ijRBbN}b@KMG
z3=<=^67Os;f}JOdj*Pr1>Eti9c~g%h7Dhmy-g`a2sFFN(F
zL1swBih`fhJ(|CYhW)~D?OE%>@0k!?s0C5F-`*c85Q7t{QVidM4`>Z&0%{N2#4|~N
zU9m^A;O)MMEFz*!*g3%>uvB<>07g&O(`{+3}
z-9S;d^)Io+O$7~8(@X=384`AgokMp)Bu@-{YrX0u+BN=6cUkQeCwriul9P4{d;Q@R
z=wy3_!6_uwsm-a$`25})x6DcJ+eO>iYzI2X_^&p{x2xaQ!c8p_hwu0u>FHsvN3f=$
zN=PH*Kq#EmzjYw#sU?#?;9k@*y#~2>gLs6_5sERgg?pFp{-JV7Sy_F=f&011HWH*^mvQfE?(=mj$`~(q}g6OHJK~a8yO)PzOLbVA)P`
zbn=?$WsCIGqYYd*H=&Z|vxTo|yO^jaakd$UVH3}uSxI5w4U~Xvn7HiL3mBcI3f7T|
zymZ^U;%)woWLeIC!`)LPpzvf+4%y|<4)4jN6Lg)d$+{;RR;Qb)1=d^YS&?|(>=VmY
zSS)G-)<4zY3QLD39$RHaZTfMLL4Da$r7DWJ6+UmS$A4_R(Gbm
z>r~kc5ztF7BYUEampHqU1%xFss4R~@up^^a*)QAk>s&VujyvJ<)xviz4zy!#VYyq+
zI9R~SC-KD|tBW|5oF`63s&7SDMr^?+W8CZHOL~P)y8-^hUxNVDg%Z@cl9YAc@4>nq
ztcd^$JKFh9Kt*XzjxhwK>mUOqyPfguLohBMyM&3=@)
zgb&@E0ZnAd`ZXw-Z?}Xp6OM0nvw(Hn6WXIUGp|U?sDsa=bM@Y$Fzo@rV^?6zZ>hwF
zBeK6cwr~IF$Oh8jiE~l27P(+tPI`rDkIgH>^uz(CA5bP&I$i5;n{yVgS`ELbI1Iep
zsTeZ0A#2EVC?ulf6z&7JJE@q#JYKZP`DWGHE6^^HVh@tC_wz{$uA*n<)yW
zMRj(MyI{YFvUltp!8c8r0B9p+(}iX-l*^A2o&{6aG?i)e4U|IZ)iOZ%O;>D$f9SmU
zdhExMxNwUFqVnaGf%q=Yh&A?`8)9YPjwhUQW{}zS0<3CzR@Vc$a{mqi_|o5jZLQ9%
z?m%Rj8P7X!LCXHK+{^SqWn+IdIjI4qC4(uGOm5sA+mTK;Yl>ut67!H9`?m5yD59%g
z^7NOK!vN=yNb1KXtaAytumsjNX%kfEjIn&a6Wg$KyA-H+(8L=21BpLiPb)F@z%cfv{p$K0`$xyq<;W;W?O6$
zt4LQJf66Qs*5-=m?PxBW<%{xU|LH#GPns(N{(&UlS@RJx@(-}y^tJvCy^K5bcB8f$
z8-n^}?7-VnHJbI+IHt=W4Zu2kNjjGZ`_L=84OuYkS^NtojN&$bb=j?$AaY+eUVkn>
zAi7wc1P6!}^Eby`xJ$cfw&+)oY#nx|a#{KJ?MurA4FvZORohO)Q;U
z!F(%o9+8}i{*Q;e-DRbkP$5l#9IXrPw?IsRB3{xbf&z&X=?s~0$fQhd*KWLoQbG0J1J&z{5ceY
zI=j9gy$xJ(j}a`|EpQNhm{TBShgU+
z^GJK8u9%%{e!P`c8;3aTeD|70bb(cbC#kEUiNz_|T+II?sXZtjzZNvT9SnM!fU_)TH^VXv|rB
zp4m96kk1?ko3`w&awf#Xl>CNctNBa^|Ja5qKt+Zy6<^iXgqBG8+)Oxl?2SOGPj`p>
zqf5R>N~EJ=Lw_{c3cC~5Gu$D@Phhbp)F#EOxQ+@Q)YQaj
zz_%rI&=cIn@_im+g~_A;a>NXK#fU388G8&WLHf|Wy`7OD%Ht1kL9%De#)e9(ElcQt
z0{4+GWB9Fz7DldoDEuZd5{L1~yN|SlZ?Ssy8_Z+a*~KK-YmZ1!`#sDUsJV1n=d`y$
zr9V73xlr}sOSY|7-h*yKX;KGN;nqLVUcDgMOMVn$Q%Gnkb?z`mpT@nrp;->lyjTks>J9ODGc(UNv{f>?tFLEGtLQ#--Mj_j-A}mYw4D1gaPmv
zL-Z1q2&fAxZri)`6q{c_>&VCHwq`jRyi)3t#tXWmgPN}%rUFWCblbUpX4BjMW0Dp7
z9YWO1MWd!X3XaA};G?0nCQ+d}GIunihwS007*e)Y3;dA}vjbGv4Orq|s`W_Fl?g64
zJgs(jQY%93d#lGt>+ZcdFGb=eS7W3g?wMB=>zO;&1u(Gt-wTY1RQnD?&)$ed%!qg>
z@|Jn3VGg~?Ln3K~KWq*YM2gLD%__sOHi2jcog)18@zA89hz^&!UcNo8)ZSQ1(A@yQEW7
zFNA=lu7ee3y_}a#()o~WOni#9J{|m#u2&t(S8y#LvTkiEDi7VG+A>r>pbYHgMLy!y
zp;h#^z7Ha8f6Mr6>6uSt)4UMiR!-O%I+?ShD#owHG0oYK0cVq;#%P8Sho&?8MY)Oh
z5su2aHC?K%d=4xO4o_@;Fn*DeS|Gc2Mg%i7tm$Lq{GVc8!yNO2dx1I6cIQZzVW6GT
zE?U2w?o_PjLg^ZrNieFJ?&S`?W-X9&GJG-u+R?gBm|2@Y%~U?r-L}}29)NNwEcMYl
zj+#lg@*^m12EjL87ts)FpUh%X0RQ^9L)|AKCn>wB1Tc|B9epfScWt|}qnw^7ve!H@
zbFuAvf$-j0>9qh>?E@CLq1q^J?}1bWY80H!I6vl+y=&7q>9*)wnohIWs)t=y(yQli
z#st~<&Q`+JHg329_@Ju{0YE$DM4Hj8bk?+G?cAFJH)FBg3FVJq;fvYeWgr7$
zY&~d6DT}+3b(}K%?A&mm7Zi`Ic;iY^yi2@)lmRg>J~B8wW)@9lpA7XZeyU|iqvclY9oG?-oC^#1O_3IjtrZxxTO>Ma_Xu~0)^BK*4O`zCLXUtA@M
z)Og7F&ILxD2>8Rcq_r?ux)`MC)Icj7^zv
zcu}EFIO(?Ixm8S2i!?CwV1lUg#drNO0hM(+Kq@F_>#aQF=~SN$6n?F&2;W^-hRC1b
zc%NHTV?+P$MO!>MT|Hd5s?Iql7sW$wy#Q4?uPUzO*#6e!2hz<$zc`s7
z35?}_2*`T$9Y+}z6A+*DPUs$v!QiI2$!A0}Z
zR0c%zL{IDAe=vAxL~t(G)12>DPG^y1LRdW_sc7${@~e~)e}(hJF$H;FXqIG?hGAh$
zkomlwI^3f>3+o}W_r7pc1;%9mailVM4teoxm2%t$v?@b^2lqBZGv0u`;}%3LhcK32
zj062I+D#_h-?KdV@T$ZmGU$B>MMPxuHiE?v3L){`TR#xBLh
zBGTFYD_t2h!hOF<3sk5S`0QL7E5B!EC(^~c%-Z0%F$-fEmf>?S<^>vQM{aw)CvK5t
zEewPxc(_jUCPcF)FW9BN-j~$4?kn9L_Sw|8HM37*g3ST9%46b{T$^1Im|Q~mzI}$x
zHQ8J+N+ed|5%Bd{sf?-M#o^1Gifre?K`8FhYhp7Yt8%luE=F1yyKGtZeoip*
zN0$8|j4hp)=*4x-qiro(gERckFGrm-uL+<0XW3x#iQR*-cF-TegRdRyFuh>StI>rR
zzFrqArAe%(N?BXg{L<;&v!GodbvroRjaR`jTt8F{ksm6ZE){(VX&DCgKs&dHsywyt
zby+Sf!k)?KIsdCq>tNdbI6&rCh!QVKX)(_BE-Y}hGkcMuldtW1LMA`ETbeI9c?k#f
z4nh^h8%9ierlf;Cqy9}?`H7m9I=n%|PgY0voAZf3@ceszx0l+)y^s{%X2Iz1Os2q8
zJIX=LH_osx*|ILd^|B_9u3F3k!^GAHj69;P(NFl*pLt5ZLxvB&Gr>$=FQb<6J
z@t_uelOr@IYd}FE=q1|CuZfklX^Tf?`gJDEYwW~h5^cj1Xrkc#9@$@PUWJ!fU+|t4
z*w{P49^zbnMqNzZr&UI~uow)HxDGY6CZT=PQYqxUGGHker-K5Q!(*nmjz35fA}4C;
zcBoR*4?M~~Ba2y$L`Vdf$@+%4(Pm`zmUtYC
zG6_X!DZksHXzOFLL?-$!_BqU(ySH
z8~gV_G!)4z2kW0bnEnfDq8ky6=A`EUb
z_1vZUo$Eg&#Li7$9z-^^)k~d|qRTwA8uH}xOy_~`kRu2S(BlMvGV-eMyyoZ@7baFp
zcVL49>lh&~fG3ijSsanA!NAGy&|XgW2bMCE5zQ4==XZg9DPd9e6y`gzd&Hmv*Lx*&0(-?HKyrP!Em+=R?&zZ>alebFUp
zra*Cn+*@`4pEMV4&4(T{IOZ23LECSw!JA;6{7mdT(4GnV=0`C83*LG{
zpv;m>`DhkE7}4!iv*7DfiNu1|ed?Bp$%>Gdcb^%aHZN7cWN7-q+@@kL-5jJMohi#)
zow5%5R9r|l(Z3BlCE7Hwx9AZoslqtsmoXWu57t@Eh@|s2joX7h+3-c$N4E2;w|%GP
zksA4;1d5!7URtt1UBsp)sxL4CMj=@Q+uoI3^bvFlK`<~pj;0qtmHXIxC;L#78Wlo7
zl*(N04EB~cej~F=(nV+MbCav2WLG3ZzX+Y*vzz_Y6dA5Wx@XiEo`=_oZ~i4$@;bh+
zL=qjsm(*~WgE$D>uJa5O?x*dTuRk~1X^)NdE06Q6F7tTz
zjywL|d6-TBFd1ClK;?}E`ASu|)E@nx!JP1I0{_k`2kzE8VX`ol*gvC$Iv-yoxQ0;E
zIiQ1^$k++5YqPnLoEODE_)PO)0yB1Q{~a+65>H@%wieVk!a&anmrYEl6gXUFs*HaN
z{k&8oT6ck83-0Y^5tX&CyhW&KFrN#
zq{*$|$|l|7tn$G6X-%7i&8w*2IFkm$CR7LhU!hfChg>|q&<<)8HC}Mdl?nE>2E)u7
z6;X5BKs!6XV)WKc-?-{ALWI0RI7RLr0*nuNwM}O82S)?KypW3G6ha(ThJuj&8{Xop
zJW($E69mpBvrvQIsPzvP@Au~?0Pp14p7=*RBjS}gg$LD3Wh`Rid=0@4ax|YL*D?*JdslZ9Y4X*%kb2Rc7wp%y
zSe};mGtTJ1?2#-;j?zQT2aHhH_EjnTD+YGs)yx4{G8p~~nfM5gbW&*jqb=4#u|g!$
z9=J{-Flwt+eayCl05N>|!ts`pHlwM3xboPv{gsSuPiwbc)=a^E&UNKI$h2QOnsH)F
z;5nhoV_-k>Hjh}x9|HgS?Ar(q@cm2f)oockNfL<6rc`z*n$_p+#n6RKoc5tajh=$PJ{w;lo2C^oC
zx196+qwUp5A9l%0%2axPfXr$h%+zjQUWgz%b`XTYt^m#8Fu*_y#UD^TEPFaEa_5B_c1F6mfiLR+(Y_neV?dTiYdPySxD)+du
z2rqb~G1C3>z=G|N`KFkVdQ!y1e$5>r1UA$W>@$v@(=G@K0wxuAw^Rzz6)Ipi;8?YUy4o!hFsb+nMr6#X*R+#N8FJFJj2cTL$9jp
zGsmQYOv)~)<)#U#WtD!o{1EcZWZ?R8-0PLQB=k?YU1ET2cBUx=7yILIJKFIo*Zun7Q=lHw5~+`Rfhw)t-^OQ_bEhCylGHV*%GRyCczc{s(;=5k;p${b
z-nu6$L~Uq-lt8LIyxNo5LDY>)lH48lXZaB=)}OYez8*R8Zx}=puEAobt)K?B*0i7k
zH>9B1Xps&&q6wy(|7!X>(d+>ay7(_t6N@?ZvW2o<3&u{AHKC-gn08D|^VAVIj3v>C
zb1e|han%&>?UY$Ne7>RO6M>uKCt$nimYmrdvp4*F$e&v-?4`46IB4zw%3b@c7~VRc
zd0Exhg@cB9XO#o_R$6n4ghNxJ@T}1(JkLLHtaGLRB2rxkzx$r;=_HzxOKk
zbj!#U-k%|h8=cefUaM75`J-Nb4Q>m&!8vZLC;o0Gx&ee^4`ZK)`pqDGfgvmmtB4u)
zNr@&?B9zTkR@}F3ommE4R~fQbiz45ifs7DnT}o9i&gzM!aH&7M;#L%C2=qvPgk7Kt
z?k*>$q;tk@ja1~L4;CxnwvA7+7K^zt?H5oF;pG7yLckAd*>L_j
z+CK-rU&8ZIs#NOT^3a|SmlHOUE*wA2RZtQPT4UfQzh
zY(N<*64zTVKzBxZ^}=nXJL&iDYnCY1=DyyDS_JFg5r!(jWc?qPX+LR^g>U3;Sm7Q4
zALQQqd``LGW4C>}Ar8dMBQkOH(Qa6P!l<7~HcOfJPl=KgN$JGAsi&Rno3S?OMo_tq
z`@Z1~-D+kk;^_2+A41qU9=V;k`~MUVmAXZHKBS}yt^TO3QOs1{8hADW%`x2EM}RU0
zBcxd4`pe=NEFdw0(u!&waT~R8fs)2H#mzY`8{!8B?mD;mXmwdT%*KGMDUs_cyjpj$C=;${ngg+fJ{43V
z>IUPLaNn1&otihM6lI<1r2;K2l@gsm0zH6xKRW6bD>iJw%*mZiBAT}>TK0_i2$BES
zfgmH6vb-Wd_U~jzxm0rxL}5CEf$Mkp{zB*{&Hh)cng{EUlDrh^;vC@N@5=Az8l#|i^(T03ouiY
z>2mo+-c(TlyM$}KZSNC~{wgu)56l0BSJHd#nf>LL>gQ<<9`Y{<81SHzePMnas;eAP
z&tlS;JDaEqi89UtA>89VI$DTuhNrR!38cUR<{rWYu6t#i#Q|MWyHJ9uY%6WPuSD;j
z%&%9Cv*KjjUkSz40#E~Gx*r;dw8FJlQWy9h*V~*l6&?syfu?AXux!I|phlmW(PgQ=v*+ng(QD?I7mk3M10`8uH45~U!d+(3
zxqyhz4`Ry8dB#7R@;&p!0^)|5L8<)o$-UD8-hAzEf*95~V3ht+MW_I+*V$GfeHDy&
z*8Dw9W9Wq;$BH3$Jt$}br?D(v|2>(5}vyVd?d+qKW35n*@76hfx*
z*u}FjgNOMan7_MZ?7)T!Y)~>bhV%+AsrZ{5vszHZq<-p^jzY
z4Ny^WFqsS}d)dSL&i0u)_KGU+>}b|x;Q7M3th!j=~xz;8iIU;&DU(JpUQ&!2i6OySIM2R8DCy!=&EY
ziY+-|01AjFC7rS>8JK5YYXn>0GNr7FRN_JEqhYjQ9Tt+2d3
z!dF$bUkS55CnplI%BBja6Vqre1cgPIY6qfz(dqX>Bt+>jv$GXag{LJQo?P*I8WvJP
z;JFDyu;a27xJ2QxFT?=nIX}m@$(MP<-P&q#ySZq
zb{+Ll#C%ZDr$f)c#!+yHQA-m#f$TZ~qpb*Ck^R=2)4j*C&
zwun}mnm{iK;C0%a52VWHC!w?y8c7}TvF@Bx2Io;q#%zxItjhiHJ=%L1Pii_Q{|LXE
zBK6^~+*QsUy$QC@v?zaTkfS#9Rpl!FvEN^Uk%HJ?8PZ?D)&&Dnqysv^;v7Vg_hQSAg5RsZ&gQmBQ
zfe@TEQ;gRVBcUaHvC#rsmz-3F5uUkaUbelWnx}dVd+Y9kzIDe3L#3;knN~<5n90X`gF>VsSQrRW3Lzr2X
zks{Ip395-oC;{u?_au3x382*{G|Vqo0=m@^9<;T6K_yvy8_KD^jI~lWEbfkdv4+&P
z%s}Z0q15j_3mL^rV`VenybTgDW)6iY^4|~Wmg5vHPv!x#j5mCVj68I7HiGT?<=;@tmh$&i=4b)3&ssdb4_BKOQt?#L$+VDn#K+Mg(M5b?sI~aZ-qr$LKHuCL^sA`=yO#HXO8nu*52RwS-BcGm&R3R
z$;6P5rdg7R@_d<8D|9Uwx{g6S5Y&b$)l9LPQUR?3Z4SC306$~{Eh|B;=&zwww4AH4Yc(^GlNF*M
zH;IU?pQ_O0jhpV7^!{~zb>!j^0HH)`XS7NU4BVg!4JTC15jo|Hs}R|KXS7Whx*V=L
zK!hc}n_UZD(rG&5MX+S$^;FfPPDnTX&~rk^{IF|Zq6SJUlo7}&V7;&x9(xw@hPSp6
zv;k--QNmUxSRI<`nKOFAoajs7yW}4&=_-@ZICK9OP-8&o?7^l`~cJ{XIYoSe#;
zqtL9FGL6H)@7GiTwKgEKYjWT1P2M}~W%3WxoE46z4qn7^1Er@iqHp>-BBXIgbe5Tn
zK_&IH;jq)$d_ta_NwseqF0w~xw0SQegAUw#I$oAt~VkJzuwR_p0!5=_AQ
z!gm{Z1b2T{4-r<3@LJa&??cnP7mylJ>N_v~k5BFWs(Im-g?$7%vYX28y)qxzh~CR0
zfvw<&S7-(1?|0P3hJ8lP6(h8E%HNJ&_=6BucHY!a>k20{4B`*}fN;Kpkdss7F|+L>
zs<0*lWNoD14MAIOC3E85Hbrj9dPj$sI$X$2yY->r7v%o&Yy~>@7c4{)d^;X3Q<$}R
zsmUAt=2|S(a$>^|aOT8A7Q7B>WSdHBuB5?!M_x!vi`Fukth4zVs+rmWEz{egt!JAP
z6Q;Q2wTtr;9OqGA2TVPt-(KZR#(SK-gix6=HYsF|?wUw}Zl}))DwF+&&@SyMsMB#b
zWmwk$ls!=tOZ!S4jJ;4At^L##_&^JKOvOnph~KIIAhw;Dy(^I5)#@YYoN*Mtr5!Ew
z+W7^lFNl~S(W7&A;$~IK)#FydFNM@>ZHw{VQ)V^!X6R9XI`DvP19&fu(gGwfGaob(
zKoMf0B2kZQa@FxyEBSB6_;X3kjt2s|4SbY4scz{5N6r}p^)#h9Pf1MrJQKF1BN0xl
z?l%~dOSWdp443^xA~8;fk6dJM5e=YA)i35+bkj7ry!CEzQa8C5;5Uz|@I#?0KuUBE
zEB;<)%u$OMnu0GtR$|2=q@Cqq>NY-a_&-X(xO$N?+10fplMVc)Z{|HnpV8T&fzT&9ko)
z1YN`F-ag;d--~y{(WCJ`XZXLa!E#ogDOKtr8GvKeA+EXn#}*TdEfG&+<+|ua1U=>z
z$tL-$we^PX|NWs`L3{(Y=xs9{(02nQ6p>*FanpBUezGuiUgc!=LYrv_ABGm-G1w)y
zS*Ww!Lv{ox$P`k73G{Er*bzY*lXWNm{$nI^lS3?(M(Ys8>yNSC)Cgw~BC8d`uJl@w
zp_qwnBTD=v?xE<}F-zJ@s0@Othv)zz0o+gUrZhbJlxvK*+a`F8#;9FWXGOV%>yP$3dMIsR?xBCoLI%At-_^A@iZ{iN+7=?Gy@V}pv0Af3is}7y;Vab!t^T07$
z{5mHgY+Y{Zk>tIP86PrSSBNM)Y#nP7-ENyOuk1#+u6NyuD7c(R$Wikv@gpRDk`t(^
zuJ}d#3P7NT$8rc*(e`I+Xag8oXkUGy0KV1^SG6l@)26R=vh7*$!$(1@4$>io4VE>i
zOqRj{rEuLb^|oC5B_*9oi*o|=ryAwppsxJs%hxgJhtYJK5{O*il%m~Y?>t1`{UmA;
zW?So;;yApeQtJT+PM%Ut!QVA(LfuBl0iFmTKg`f0w+VW;(<1EdEY_I+#C`Iw@gSCB
zQ&;wmb`=bry@whi5l2)Yr-hisP^E5OmM(1fCW&BIWMSNaF;iRO#(>a^A}ImB!e+Oz
zi-yOzldGuYj_U%$GK^pUaTMBNg?97z0ZGlgPCyz*C8`Ij#U;Ydw3j;GL`0%q0Mh>E
z{LF-FdII>9VWx{r&WUg7PpZ{SVddh!`$Z<Yy8R~pD8Csny9n({8aAV*yi83dSbgjXN2dY
zE}OwWgcD_8?~%C5&wlpPYo<_fD2Zd3o~;X~6u6izKK8Tk74Sz*!Q#oc`w#5+M!hP+
zj=6{}r0%g=Ldw{4S%uh0uY|pUHjH7(z1^)3h`j=lj!0lgZ0EX-Nu+kP(R;epLO6m43&b`rxb>fA_F@}nLX252??I)AX;{j|2?
zTN_og6Xlx!z~qs7;HOS?w#wmA(81mqepgygA+N>GU(-1WP)Hzix3)F8$RF#GdMG-nE9KoV7?J4&Mc7GDv*KkrLOb4|ovb9^&
zftoq}L{0z+g_Gl3!@_`Oc;4B~0X^v$ZEDti2Vey%0fG@=cHnGaO&&yJ>_PuUm2%2j
zsR<)pGG#?FQyVS3TqM$nYN19LzP%%ET^~aH!FWZES8LwPGgV3rQv$=GYJU3Z99cQ}
zM@Y9D3M(kUH#7dsN+%`JVli#^@ll|TB4+AjI0GUzzfUs!TDbP33tSi~lkWqfmz0e^
z*z7^$nO;!c(AV+Z-^-=_rv&SJ<7D5JFmr_3fblP5%AfyD|F~|D8%hIUGm87YXAkp6
z+)~8`hqHLo#i~zN0BIzGvR|pfo9(T7^FGIbM4~Ib#}u8~#Z`E9m1WGGy%)EwQCFU)
z$lgw7$!F0Z;TF>ez_j{Dw8wf~^S>}!$4`@_dIDFbQp$Q#xEW?pvhg0ieX19UlolKH
zKn^>B?U3c;v2q{?ch=iWt7uyKIjKoTu#Z#8^n&Q~?M*v!rQ)&bq@UX_l?qdY3H?(G
z6Q;M{irmNxOTdDtA)LM(fm8Aregd9)1g-Ag^raeI6-L4BsS9MJ{~Ogu*rziXa;C;d
zOgref3c3Qr9gY9%OMaw-HiRPnBN1v5KfD_Ym#*WuRLwS0y|Tn)kugxHoV{T~*p6qY
zq>pxEc{W*Z%Kw>SZ{))fsbUvjgzkvnM<;#~f>ST)aln8lkH1OSr7dvKM;+!sd(!>7
zUnbW*k#^eQA@%5k<8(96nn)mnD0doJc$U0mx;E6z$~ZLf%}|^7lz|-R45%5SrZ5
z6@P;*1gbuU7T(%N5PGN4IP|9Q&K(&Rx)$U)VNRYrQiVMU+H5kPrUBDz~On$2lF{o%s0Mm@cd>A*lXqa>;QRkkwWc}$fli*oPr;3lr@Q8C!nF1e-A`M-=boIy4{4)n8jqA8G3*#>b48}+S*<5@gA
zZjO$orz!bN4Rr*_b{CYOKNeVaHV(^H7T!_L{Pl*bxZ|+vP&K!wW+hG#R+Gc}*e&#)
zM7mX@Mg$hVt7tTaLSu+6Im4NaWzmj
zL2+GY_sjiLEC=j74ag9@xp#hQp>U*^-7m)oLwNry!RL5Ts7I(j#3r8qf;$&<-#i_y
zb4*88_Rx`opA)eci)#<`ztR$IJJ{$yGP*~y3;n{LWM8;6{_9wbJG4ZU0g;HEeKF}t
zQ6L`vjEP(q*k7NH?R?9>44D|j`9?uxPY&wMYak#9{gT8tBEZ$0Fpf#p5ej7
zV)&f)bXeo!+2Yt=*`TNhHrwswO4euRJYDfHWPg77u%DyBHga$V6G2fZXR9jKxq?z|
z6{vfOH&>0`TiFfVxxaYDYW=1`sEv(sxE?ZhMRRC4shiF2v<@_guu1dLW!mywH+S^|
za200}(2TMkQq>rdE;52sidz#ClvaDuFN?g`){A7myH2!J$^t)C_l{3Roo!+ytiMkb
zXZa@-zGKC7J;$mUr&!~)9lFU{463d~K2$aVaTYyUI&B!aAF`)k=EM6pn0Z~0_QuK6
zCvkgEd69#^=0|=FqBAMGy_umAS*U?=9CH)WFjd)I<{XV)0HiGjpbiv~f|F72)ePx#
z1)l4Kzc~38=4FGHGJJ~1XbR)w!MPcXsvA>&QCVDvQ?Th;K9__(nzhlGl&(@-E1U@y
zdWTJzI&yM^sCE}Iw_{`)m*dlM`0-N=2y+HR+rsuO=`H8hYZn)+exE1sKnM7D!X1&0
zV2HgpBP?+w^vU8?4zp#GLBq!uPf#0*(n*fcZ=J2|j@p02*^DF%p?d
zB4-79k~<0$8_axIRlj;oYq@HhfX%Mbf+2j3H`DN2fgRXPjL$D2Z*Zxamf
zr&9e0(LEMeZDVx}oXM8KvXhvI9i{v^P%=iyANy06^ZOw2Jl9^CcSb^~)3RdiKJ)k{
zZq!W_Y!-<3@^H#ykwI$FFnyhxs$FMAkIrpFjFXe;(_T6EGRJ-a6f2$!<`8qyXNj-ZIYDB^F|cE_5f=a70ISKjxa~O|)A*71H?``MT1UV!
z4)~}Aikh#ygIbD;+ejSKq4IX%mk5X_u{`|u%1}1Ccy9B8`!L`?cMGfMUM5;h?RW#Q
zbT>(?!3FdN%{w
zh6n&DVzfYLy#*#NctW+29EZ^z$E&UgX-tH>qU1OXUjk1T$ph(KW%p~yw=vwxS
z1HXu;>xEb%Z685;wAFCL?NN|X@#j7Crg3^(U%RTq*&YZ@zuHaa8|L(5fG|;VL3vC5
zAtLb~S%`>>y!(_{V3CLS_`OFFcy#FelYbNpNPnra@coZGSZp8BV>7uN11;%0tmc0+
ze1GOuP+U{l{q(1r`zol#(m{ZxnHnxzZ6
zO2}mIL;_ET8fDjL2qL)8@VP%NVmkVn!n$96&Z?JTB^G3oa0O^JW$HH@@gs{8UC?u>
z5LbuysG|41R&Zk8<3!BH^NW6YGm0i!a9-G2EasCbRiLRmA@E2fXQ_KG_DX%gZ`I=N
z+;h%L(<#;e+%ho=Lzlf4E51y2s^HtBNN``_kXHbTb|jY$BIBkg)gnv7{YZ$Z3(&$M)ceN^Sq-lV!F5HfrYlni?mA{hO4s71
z)#>lrHB1bHCds-}H5&G@JaF)87{&eVRk8Sj#YPsQ0+MN{VRFk0WM*8hr5l8}Q2U&5
z`Q8+{`-OhkQKp_^xq78l!i3|SbVieLuPDdrMBehWSfD?d!M&K}N`Fe*7$)TUEiSSp
zI|$G8vp1fzH#nazCSnZtewu}!$LhjK?ndtjp#U57?lFySBYH3
zhVa^^WGmQtRH5*GyU~}3UGpbjF2ho_$G$zJr*_uUodKLM_BW$jG{RU7jzO*_ec5b{
zRWy`{S#1}?fCFH+Pl~Vldm>}EZ@~O`8}$DJ|I68l^f!1bA)a?qg?nw~XQ5>ORMdHP
zK4@MwiuDZ3yMLYM2qWKr!n*wq*G=eJKF5FhVKJW-F{x*~+Rq0jUxEk77D^8$o}r-m
zL>5h2@U!xR{t|-Q7-~j(bXIivX9=N#MtN0~(yD$Ljjc^_v`Yha40zJmjip51kV9v>
zX!&(ow-#pWfZND@t_w=l+(4FfRpmvdD~Qp+#fSU?fSvklz*V7PL#8nRbv>Z_1(67&
zpGu*n6-vgZ%81L;IFtq$!3EHNzHMwchiQ(sLJ-9xk$JET0;A?(BV%&N2X4`-Uh-m7
zocUGH%}Q{Q#%CXO?d+q^6`4RIAv~f=LLp6U
z1WegmC?A*%h%j{o_1iC$D#ffQRvpRoex;1v>s^c~;_a$C(cy04L@bba6DQXrUdWxJ
zWoaG{+`r!73W2va5#@;DOq@OC)uk>(
zx!`iQ*)I))v35Zv)`2^tS69TG09u;vxDz>JLFtl6=KSX%lrDBV8h>Ad^cK;IPAtRR
zck!F+M#mib!^^U#_tw3`{^~LZT{#prpBNNHAA0|a7J)&VSQRVlT0KK3!hh1gAK@u
zE6~3!n!Aj`XLH6Gydw=TJSjr7Ut^nUC3P@-7Q)pL9`72B3e35a{syOvf03{^%V3_?
zX&J0A?^xhfdxvU#dw1LhL&=9HifQYRno%nMl=<^JkOAK+Ye+Y8^;^O(omqJ;^9sBl
zl}~lK?+tXzb`Ehf0PUez0Cn6
zodDV`J?0Yx71oUxZ8OZzq0_}x#0_G&ROdv$!2o0ZKq9Dj3rtCBCEEo&MAwmT
z_!RB*i~uVt?DX4oWg8K$2M#gu4yQ;mGVk#4`Avu7E@M52+b@pZR%;oj
zYxRvxc!QvZX};d7G@O!G@x)bT?X23ro~V7C4~HB&7X7UH?x2)b(u`{MV^CWueQMzy&!-$=f}dG9Ze
z|JBTe*Wrl^k?ZL~;}Qhm9W9@@EJu}fV;i`H@xvRk}B{L<=ncRb(pKjlKgi5&R@1l
zOs6w*a%;`>Y>sFrte2-4XvM8%n51DVqNa7NoTszcTN671b`|zRSp|+ORntdO(Bx~x
zd38Dj6B6U29+`EI`17@Ggo6;8NAVo*CG&Tl1HVWNV_+kSfxI{1hv`6%y7nsS$A*fv
z3(C!ya9d)mu1vUQSCz=A!OH|J5QbYHei0EX96(9;YaX1JM_@d=ft6V7!M>y18EU*y
zP73Gr;`x&H&nGZ9PVJI`yr_;Ky0x{!$a}#Ict6q;=26S4k
zFb?InY!|a%GoNqyWkSoZ8XPhq%46sE89>3p<*%Wn?vQrI$}|_uXHAg2%CvYCC2RTk
zpaoM4QZFj_ZUgEO>f`4QT#3DfBn;s3QWws!-xETsFv|GFuML;$r}8=v!{COrU$;*N
zQHBTc@*J$3UFIHwW7U3KRvZ8Mz>VKyA{?CL3|Wj}2tEgP4@8shtG
z0jDjKZw*_Uk}2c)wLc@iUGf_w0>$qsoQEknamHjNWMWC#!nXo-@Kc9ykmYt72rU-H
z*mHDB8X7Ldr?7nK*gEeGm%s?f7
z`lFeo*z!cOs;x%XD|$WDhj_R>tJDN)mc?P8!Fb^K1(-
z^Sphw5FQPW<=mc>SLEPrxV+5dulfGUkrfvUgwGWG`kk&ky&r-H3||@Thzrq?u9+gMlh3c4vIfZ6Q~A)`mQWT^&^spP6Q+kI~bT0CP>b_Hv3}*L2x+
zn5-(&`aHRN`Mek2U^0gsTBVI#X`afXowBON_-h+0SAe@N4~C=5LyJe!SWTL^duF(0
zk~fuQ9Wq{1)<9dO!{~oHQEn}yYte1j(+i)-OZD07&&xE4P8xO@8F1KwBWG^ACJXe}
z#aBo(xNp86faKuPi_k5NL+k7S5U{KGx}1&nKy}%|0B0QvoiHFEs>LV8=G<
zA`E6jUrNYZhB=85{KS5m1IJeUlj~m@X2SuRT*$HM2xw4=YR!C3z^>r=gWy&)cD#dJ
zehWM8bqs+%>zNG3za~RbP6bI*ly}LP#)0N5I+TpvLCc9`7jSWV7a^y%;!OiE`U=3U
z$Lhv7IHk9thEB>9A5tNCkD`aBNoP$qV9&fHnCxi%vwqyg`ElK5RBCeM(wt*g$j^?4
zy4x|?^@v%w(HLbAqfgQaeRx-A&OvlQbmU~)Dq#7!iW!FlB7@Sb1DF(T33A*?GwN*e
z(|d0(=x7Kl$`r<~9gCNGQq8IqQm}s<1Dgh^6$(Gsd=6tIfw-AKVnTY(RH#%aea5Jb
z><1amRAaF?B&?Z~d&)1*^@m?hkI8Ovn=rKZ7`$XZIVrn>zL<=mqV?d8lfBpuZDUO0
zKX2uhJ1{XQ#rF>}KBNSG0!GPizm&UBV2vrsD{+$9f{D5iViuWw9(hp=lfJ99f0jdz
z2xuCTc1n}|I-QQ)BAU3zwu2E$DfwRc$wfC^KJs`ou`QfiGSgW-Cdw})xaZan8wUb{
zfk}OwW%DcZv$y@8!Js+RkL~lSaY03te#ls8Q*irRMpHdld1lXTolLeK7Ima
zx@?(qtzBIj!(%xGru^jPPC@(qr`__SK|zvB90RN8gW@uZF6Y`RSg_uZuSc+Piz^Iz
z+hYhZg`|nO#=Yv~4SQ3V#>(u(6A~Be%nya8&|B}bSCyFoU4Y(%i?JUvgcZKSl~9U2
zm%Qs0ssp)=fELvkH*N
zr3^1Q6lHKaq||Za8N>0|8`JszeuFP7MbvTW^f~}0iRF^vGY5T(`{6WX^rY3pf;vuD
zjDgSjsj3G^1(WmMf?Cx+M`-~*42(Z^Hq!#zd%?_}8bx?bkt##qOxCR7kQvV>pn_D{
zs-A|qBq61Y9Htz#ocXL@nmgO+k;$2@0041+HR(I`c_12df?ypP(T6>zBv;maJEls7UWv_hu@S<0rq(@7TTCJW=bqIg*?kF9lqG-
zJR+?fChUw9pKl}F#*Udpn8psxX?a1(n~!hMH;pCvxLP%|;1l1BVo^2dv$qqWwm?J1
z685sUL(hCvuk|mF#0D&I9H0ZD1GPuW8VQe^_yjYf1JA`gEv-352Q>fZ4%$2
zZOlD8wi*21oUq&fsItk5vxj{Bm$2kVQagKrbD4lR?bwW`YC%xTS#EnwUc4aCuZ+%GPSd&{4FgbV$GzUzs`bG|9(+3umv_g4X?O;~7A%A0Kbx^TScc2Gs9t
zq9Lg5?dlFXw!zOPg0C(^-H+E5JL;L>#q*m8r*2V!Pz}~>6!b)FG2o(2Ie6S>S
z#`k82Cj1uuQLx(1+y3Dr6QheDfMU7`0Puyo_+@5R4{bYZ)@1TWLZ+0~?99TWX_pKR
zQh(eRIlNjvb;3gM_=8s-Oh){s*X=jJ2qQfR>8ERH1{2c%aUXqVpam*S|Bl~mHD-w#
zI4AwH`Fk&>umjG92}!EFd)rtSIHr3i+qXOdpSdF5&K+1J^Ui9?s%u(xrjKfsR%FujR6PIKm<|CbB2xax_yFJ>t!V0LdXU()EhV*
zrM8P9P}dzHsdgDG@-i!`bLC}shZf0BQq-YV;t^0`7tuCJg-sfRM$QsiCPn-~P6Y9C
zOLt+)l%E!?kt!6IIwEdVF%db9zpA%Q>9IqVb?^9e`Bhs68IsPsbJ;Fu{zbTzrYdZv
zVA}=(e$TP6_br>;KdPSGFLZB)D_{ELjFh=ro0GTEI{L7~2)Yk|P?S4E?
zJHQ0c8KlURV7j9F2I6t=c6+f@kzTMV$;bv7>^)u!Q``$ivKegx6~EA?qfapITza5A
z(-D#u4RFwO>iCz-2-5&LK*ql|ET(z;KPfgJZ+(7xjcO5VWg=z$C#%RuQ1*4Q0!w&6
zI-%%K@+jN`b(?P0(g
z&MgM8PeVl&uu^G&^Sm5FOX%rg=#EMBnIq;32>@don+l5XOU#ss%cguyye|#^VXI2|
zb~mWq1v-Lvfi4|-YWz*2-n{X-XF%U)AKrT10NP2fS^(BK8&qe#*&5SoKf}yqYu-Q&
zI9RWUjr;f8fXjlbiBeytWm_Y{K&G}`lzu?}Xr9F;D;!($AC@(wLW0otp&;ts!&rO+
z%+=+iLClv~#z6$=R6)4+vL&>1`%2n!BSp$z9b9xn*a0_ZOE@<#G6UKIf;e02VzB4;
zH%c`gAYwEPy3cA=OZ7>M9G4Uwl(gle~g0?sPe5i}O}^C%{qC>q{V7=vd1EJ|TsNW&Nd6nhgG~g!vm5TuV!D#Lb~)y?P+8
zV&SFW2sEd#fDVX&JXGB4R&LfyP?&RvfsY|>ro*tIY2U@c5218Lt!Oql4DFHkH2uSC
z`tJ?A=!R36wK0n8tdWe#8Pt2oKIl(bu);kqt+K+ReGfmi|4cZYf7{`BQTX7D&&fJY
zv0c?0y&{rmGl+y>;}y3M^VZT9YSy$`AC-VX`34i%a@CT>9CeMsZrSu0So@{p~aQ;Irv(cu_Yv9mZEaX3N3
zH+K_Yza&I4iUBNh@7|mxlhXbMpwtHaZ(@E=DMkUDX`+|k1EKuhn1l9{j6v4*Sv#9A
za%{cchz84-f+pSQ88Q>#;!MW`acapz2?SdX`Ww|{6HE&cd4r-R-R+e0=>E)#IE}pC
zqO#s_1!R(_xZd~DJkbTngU)j393|cRkPK
z&HR{$9v8wzgqDVtKnM4N#=X9(L#pjk3KY7+aXiF+6zjgKE*NBK9ENiF$rdqF?l)02
z##P8gBhN^TYh1$ZszLaSn$UU@z9m{J2GC?~2dY>B0ppTIFv%|TUyx6XHG9GEN+PT`
zOXoTVxUu2zNph#YuKm273J|F@-@kSO(8G*bRjodQXcRI0FuyMV7^d80a1*wRQ@)45
zA>3di(20HO8lU_*shkdfc|RM6$OR9y6+-1N7GodSNW{S*&
zJ~9ziZ`(hVpXkSBR5d_6bmWQ*v^itlaDaU2TiE=bh~gdVi#B5jCMJSxVzaF6(2g2*
zfpPD;$|nES^RzpNgDpD$HwcMmE;7)8N*o8*(nH{@tCCcy;QfJc)N%`udxCWl(wDT~
zr|jsMg71cH$FG3RBdyyMn#*tt=sb2TH#%ou*s&Yptq`9@)^{#}nhkthd%eLXjtRIi
zM-P*;<;pciBSC^B#1~(VriUoh%~y?>su(w%nQ(Yb`E0nJt045!n2k&Tct0CAPv-(P`TgJ+@@ZAAd-
zTqV%Mdu38+VixIV=rxNo4{V~WU`SaLv6RBg_5X_}>*U`{l+mrQu6~kGhUE0^z4J5)
z9#kfP?D@{jmmQw6$WlD}{9(@M$d>`HezrFIvkRwPI5UOq=x3P=^vqAbT5KgEZWxvD
zxzL+?Q=TmPUWCd_jTIbKqqHD1#h2__k3U#RUjQDs?4Bx57wU5cM77bAhvOnZ%L7tkZui^{^V_7Bi~~&_
zoEk$3R#en@{VVxHkV}(m!clr2l+ZE40Ga2722nUD5`v2}^+OcNUa#tZpp5yNgA*`1
zE5wgIoMK$R0|ah`lBvH5Hfn6AD*UI-zM3)G0u-L38=b5
zq>|R^l|3?siix&t*L?_9VVQ!!4(7;v_%FYZjR(r)q%ovcGyX$&WCIEa1<1qd>_)_p
z+J13SvTnEd4Cq4I{qG{ZASu0oY#(_xh}VPjpksdJqw+cRvI@F@)6h32IwxkF`OCpa%=mFzs$_
zbW@7f6my|*@UuFiNjKo;%q6@{|G80_V@lqPOOu^OmoY{s{9%!nIbEBEbT%M{d?qa(
zN}F}YZ>;5_*y15;1T(aE7N<$2L92?j?TdwxP8_n)P;MuP61q_7S5(T`h9Ig;ZKp^m
z&igP9)ckN^W4a9FHiXY5sx1p0rqV4Obm~pN7=jvkU@cOC351d*Gf#Do{QzHD$9h3_
zlJywwf=6k$p|yq!%1qU(RSBjt2difRE?pc9OR>Q7Z!E0McoNN+`p+VvjedqfwfkoC
z6oS7P(UM^Vm@WP6jgok>?v?Drd)vo{796&frkqmTcElRj()(MjQqW|6xHxdLJZ11<
zEF%=oX*HS*iM(B6l}4FeA_CM6WD+79$HmsMgl{}&609Olxx8s}JNY03GUw0z;aJ`m
zf6~PEb<(DS=OAA&oGkRO(fpaP>mE3MB_*AahWt5BWL8$|A8mm7L#)!|FRc)B(s0_b
zW)xO+C{nMvx5)k{G?u&8faX43_w
zikW;WqR!v-Xyc$l*YJXFf>wcPvV&*{T}<8yIHnw*2UQ<2%Y(GZ=-p_1ajx9Th*Fo`GC&fcm
z7}s($AC6*G*HH7?`$gG|^fhb*EayOc^9|#YW;)k95g}e7%%q5EOoIc?U`+E%NdZ)o
z#dm5eGm()zZ8R=58P~RB&fTEE4Z2MSp$=0&|
zuq6GvSdrCn>fqr}!c!o649(rUPPBBr*b3vDH&OVlb@wul^70d(4Y
z6S}>EU5FSI$)bez&h2(E6sWX+^#I2dnvaMOaNjdgjpN#VVIEEE!2iNoU3!6>e-WTa
zYh4tbean$bOAX-XeF)(e%PRVP`guquPH(UO@;zhMAlz8w!@kzI<`R?Tb(i~ZM+e42
ziUtun>XSfVBjdU76$)5C^gkQ5nvtR^mr;bNtI3N%CN5L7<8sBNe3>rtZGvG_KD3LRzU9`N8krS!Y;g*u)99`!9=f{
zlu-h~4rPRi&{#wED6nisfs!jkep6w4gR1r*_>$5?6cxqNmaa$bH_sY&sMM@Jdb-1m
zr`7CJk~3BWtT=~A4I&On?)9SL)}!Yon=1vzUUKAq31a*#GX`SnFsBHH*DMm;;-V!1
z)Zl3LLwy721{zH4OurfFCP9sNhqe5Se|bO+6F!l4VTRe2)3y6^%SoCyzYy4&dh5{c
zp!QyZD2$^Fg$fWRBroEVVHblchT|LB
zv#e$}VJV9eld|x(UK@8OYhKM~CR@b;?s#rIEVLc~7BN16rVLzLHC$TxZa7RO;rsN4br2#CnfZzx6uRh}1Hht)xCN2(gq8nsP$i=-n+#@<0KeaAmAID|?sVVU>v)
z_yD`6G!po3=&YE?{gJUnf1unsRu(pMer6m&#k)S46LOVWAUseQKz>$OohDmlXY*<8qvibZ6->%2IR
zB9UA05asbkfz=}pce=E)}dhu0N50sy`WK^>}J4uK*)nJ0~n
z&-~huy-OMP$d~SY<)Ljvzz6VXt4X&}rr9?=B7Mqdm(Io__cUXM_uT78!+r4fG>)Sc
z!$t?OZxFng>Jp2oW;Nz_MdP@PabnQjcTJe<3&(d3#r7y6>*lGIp{#RploqHxJlD+(zlP$eLYm>#Lv<-EKCNLh0pKy+O
z)0Fwoohf`U!Gq;#mF0;mX$Ory!wLZnb_mo)d~5^18A9vsXudU~ZF@zGgTPGUnp9yB
zz1`>t6hpjDrcvc5h{LgboAp~>e~~exe@nOR)U2@r51BMQ)kLsU9vv*8wS{o`&XRyw
zM#hrEqLDOzBDIiyG*VvSu31IvN^ntVAPCc{hvK*}8yUOM+y>4H9+gP(8nlDNUmBKUjRg5;bYxrhLyY&5kfe_Qy9R+YAk#FI}HLb#rZ`WE55dSSqLsc7eiuM*TfNZ0EeqOaj%;
zRyOuiz-tzO@%7)RsjwmIm_`Rm%PpW|46LT@8tOYYO*-A{8`PJQd|k4?_(gh?r%zFp
zZ=mB%udUVc2KHTf`QEc+gdkCK!0q$s2~Ouk@zy_>H!`p&9%xXZ{jMzdgk>iCDmzj5
znJal3r{vrc;4Z%9ZiBf^ES-uWtQAKnt;*;(J!c~I*?uJr195c?fCcvVD1LyM3LD5D
zbnK|~M}QK?XzPEzlTrk_sR8OkI!ooS7kk3c?JfRQGms0$ojQ!ySR=0mTex_sKZ
zD-+df2)2W1IfwIXn48wbr@L_Ek{99_b>^UYG*hMBiffMfb&fdT?Ci}{H?^etTm%v`
z4(R@tZC{sisRhJ8eQ$q`ch7NO~q_+!h76I&}@m_Jez
zDaFWF`DU}V9aid~9Vv`I^boR;r0fGcg+s6Gr>z%6&En_QAimC2gZK>a=rXaHsVrXG
zw7VHAELZO`Pi#~80u2iYg{ET&#$p&_W;#{>YjrZFb2jAq99T0q4k4k-awEoFJfq@q
zH5~JXyHFcGjvy)gl|)kYJM~GW%SpL&YHi?7MDBsTqH=Ckj8A<)I{n(j;%!hoTW~0|
z7v9yg($%XL9C4peN6B;oa>;c|Ui*wF@6X=TKzpSsst&AhYReBNP{MdRYdt)6&gQoo
zl0kLsF+z}_nY2-|#T4-EI*ufDKE)~Vu@Xoli25!i)1m;NC~c@0yGsDL6L`rjOhp(O
zL16Q#$efdTx@F$m72tkMhNdS|2cp>!Azxkz_5;4|iH65r&itCGzQN;_p@t>G@+@tu
zf$qIUMMxX;LmQ62qRRmV`4;mx?pGT3p)dG#Pk0tG1(E&WnvQ$Hhb*%aT*183Gk8W!
zX8n(=G6;VEzFzWK1hqz}wFT&aXn(Xy%^u
zr2Ur{z7;F9Q1$HGTR&;YTY(nuEF&JAZcE-)^xj}jdpD=UwpcQjC`Emg(JW_?sL|t`
z|LA<#yYcN+vfp1S`Xc;dyA(>xjIdp2r;wQG$F
z-h`mKEZ17I?*rUaQQgK@iuQ3-06%SoKB_s5rx(Q;NzY@t&NEdtZ-DknDSQmq79C{a
zTm&=wTV+5d6o>R(rBFw~rY+D!uYTX!m9(k*ET6ZPczA3lhT0n}!2;7!hMq?V*QBuG^jl#*j&PzS&F*F+b9O9yr@**U#
zyAKg~r)&P@phhO0?{QGLD^!NU^)Cs98N%>j`#`6#9UW1((iZ}w7Kq~!hs7qb!TP9K
z2^jq_d;qEEW3v1EOU@Bx_H_a(8wtl;!7WYnJbOZDlLii_hupUepV*u`M@Ay8;ivdL
zO)~(^gIC*9ZS}UViczF_BzZUh4RcQJ-$~*PPR5uBt*W_SDLsX%W+rI*`&;HrA4S3QG^!?ZcBKPcXoy(#e{;vtZZpyAv?=ZFtV@9W0bj7Oo*>4so6o0G#@
zuFH_Wjk5`f$*9z5n4Wo1&_b>>2eN~rQm<@c9G87cqZ!x_QxPa)s`ZUisXb(qquA3c
z1o5Q_7bn0&e|&Qj8`lmJ^K5`Vr5ZRBVS4p6YAi{8i&h1p`^0VS#oQ24Mb51QSQAUxy3rytJXtK2~7xm;p
zD|02mCQRmh;`YcF-5dbI#kH<$N;4j@fQdRrM432LGdxlJ{06=DWk(OKh&}T6xWt5)
z!#B#8Hx$dOZWVjeDZREjrlL11Mu~~G`i9HkEVJOW1b!y8Ckvs2vvp<}dQW*Lk5?CRf~Tv1CFrA<1Vd3vz?LCgPKiHdVFT_mB!0E&p?vOm$}i^6}G
zG4qwZkObI%01sNjIpS>}?J^h$jy6$yb{0=gV3tsG^5wy}{l8i#9@A-BQ?R`yV|&NN
zno__h`@;iH53hOq5tE?A}6{
z8+FKTV5h!6^#sj&2EpukWFb#LG3-RDa(DvkNPXc~H>sy5^X%7JA?Ru-e^F
zq*HM9>qKT@B9uT!eEx;2MP?}YnJ!q#OyLPvZHdMyT6dfGhSY9~49J`CFUtv0c~-Jd
zQ3~>sU(zR}<_`Lt&~H4xpD0zNA0}OkBpX7N7Jy+Kb^-#K03*d5+J=e8BlSW)3J-_6
zt`8BV#!?fe=fo*_Tn+=O;v)Wi7lCBmrfw<>t?&AGU{XOsB|`$oRFg$ej9hY
zI{$HIIFo@6ld^Uul7h0ckaE%6*($7DclK%S@;1)GF`xfAnZOVA#LNK?aoGsIxVnE_
z>`6+G-zWpLVr@QhC%et{yPS!Ta~j&J+%Z%pV%5McXPLws+OeJ&MtTkl4O0SVi~@mB
z;h2x>A8Nw#Se=aS9u$2r8dgbS`>{ot_D&qNU03eQD)M7{mFPRw?daxqCU_8RoV_|I3`c$8--4PIMJu}fX6S+(M
zFF*os8m{}F+5BUWUJ@cT;U;cqe-&O}<ws&{z${0jQC
z^@GzARq{u4m6;5OP00ZEYeoqRSptT47*uhL?Me6YL^kXdjU63O$#vw|wpzRw9(}K!
z5pE_t@la>h{5TkDk3{vSO{7orz6Tpxhs(J%fOQf{!UP*w)@K>r?==1Kv{}-4Eir`Z
z9=kY%F|Q8qDO50<2*EADiIno{-KgsYng3HV=lt}WOk)OET^#4*JW5<8IBPUJ92pktRR$27cL0R6hC?ng6W$TG9PvtkO=zYO
zY#gi2>v0%iJo53|u!F^9Yz9Ho1x0+DyOFefYde&><^uma)yWuA@qhb&R$!PM$`BVI
zaZ8fZ75t#TD0j7qFYQHE?T4z4=;56Ds-gDQTakNgbSb{jeYS|ez5DLHr6B#%)Q&M*
z2hD$q%!1}#dQ75QIGvx+mZ})Hltav|JlSNEb#XJXJkdni4XVeE6ompyFvtuMI=L2fw$xA?5XyetLi
zL)O)Qh4O9&xJB8&y3P5#D??)`zV)ue@TGzJ)e8?gosiT1E)qa7AT>I8gHa&xcQpJvZXMU0dv{m6
zy~T6F@FyX=u~XKMv%pY9t;KDuB3dvCb||7Vm6|Lq$vW+DiVsE
z@aNs#k)=byRq(}7oe3f%?s1OAIHtY7(l(0zL535ppi8n{(2q`{4`?u5)FWn_PvQdhl4Yw7QW*7r*}8sgp4?t$
zF1MQc{`=m;dam#-WQ6KS9p#tj(Ve||)5t>ltJhCz@sEcnzJ;~|PZ7yFDVPWW8?c0E
z*NBNb^h~4FXU|6`B7Jl$Q(rMdrYH$Qmv+%OUAoYL4rzsIP!BH1FSUf?;LCXCHvaEbHtgDjz#eKdFQtN-NrPz@5PD>k
z5KsQ|ydUg@TZ@BP5&6yv=&~>ccOTAvfY11E=qOs0E}F-M-98>HeRr^2cO7#(AV8u)
z4smS&^sDQvj)edL5p3MReU8YjO%-NmJv=0)Nx>H2Lh(1V;^`71{TKP+a(YnWSTpN5
zFhD6Cs2JiY$(*O@PRFt(Tw-#7@J<&poT+h(1L?^|Dz$SJ?mp4}AQ|wX%bZvQ7$A=0
zGSGSIQ)El4<>7>6K45vYvEoL=PpY(+AJnl&Tq^nLk(3H`t`UY}GEDb-TdOq))u
z6>zXr!rla~&Rznx+2x^iBO1WZlW08ePE=3(Be5$!6*lW2I$s~XMYf=0h8qGmjRwA=
zEgc^27HlMHAkNt-&US&f2bL7~{rNJ%P^b~J-WNQ={E&Q%MW1l>OfexHG<6jfK(`#}
zQw~22bn$^kxKT#2TGd>o1KrU2^x=JHIT3zpF9D{$ByA_;)dvJk{W>Yg68ExlL6@Qy
z_g@)?^ro;?g=Ao|ffh6BDjQ=#W`4^ek$0M60MR~3}rUa%a
z1^Op3HvO_{jn%E*p@qka48$o9IG?GkLjslgBrGq>yPUuA+Tv0>lL@5lmXL~I&1YzB
zqXi5cs#Xm|q1+->eg@G`oq*_m%&WgIQQ%|6a>6U<)Wej})v_pvYvQ%@0@o$Q>cS)L|njNWo#*y0G!)I7?1Z+xkEIugxPL~LX!9S
z{i=tT3QDG$Eq|NhQYm}wPvtU;E5N~_4RrKWrvLvjusY>|M9|xG*K%XNp}pokB8IZq
z+&EhfN#AjyWxcfrK+gUKln$R=I375m!*+0j9oseO07cJ#Dyx6i8N~tn~KGpH*ky!Hb%sb?8t9J2@610
zx6K=naAWaJ{@|k`0O~A4zjiK-YkAk*)Yzp2
zwj9u?TwbLBTijnFLdNfAzzs4yJv}pbJG;**rRMP5b8ez&E9Gp`yOC2`^K-sS9Qp^;
zk{34AP3;tI9nlmu+i28pF?}D{U#jEV8=+X=W=w(>)N-Z*aGb{PBV}&kDkMNaiZ_du
zifHZOr5VdrwJ93Gb_XF)LmpL=%T=fHWWf2;Qu{f3yHcCN=o&q!NUAF4s(Q(P$>03m
zTZJ1_flqVVx>ds9h`@mVuG
zSwJ5^aw@G4y4p2i!Q5r57
z+j28bJPW@+gK;;0yiie@TSfm=A8RVh>V;W0&GB^3qjPYeHlO8FgH#0nQeNpOsn_Jp
z;x?j#xUYw(O8x|u7-+kQ_foTchm7~?^A(;b3p#~G-DEGYjWa`u3IBIfv^oeh9^8HI*vl|
zlfr^wL-|ROJq|I)DMLJx;L12#pkA=h=?e2HNi6)YM5p>LCNy}$S(L#ZLj->+Zdj<}
zUH)j;uuj>!(VR`wUfH2v+>|>U@4t>KD1*2ZDz541Sf35PtauL4Hn4lO#oryoCMr>Q
zz-Kydy*}6BEjU0p4aKAL;*E`BUhk^B@Zzq}IGh@!
zPK2{;K;80q2MzOgh76dSDPZsU`lTUs69rJO-uv?Q0LoUCnP($%aj!9n$z$?Nyc1i0
zTf0-fX8@R4^9w>&*%;!CE7y+BiSYh*K4tUerbvl>+bbSReZCjMHq<@|raB83m%N?h
zZ>A83CUTR+bPh2l+
zQgD|5L2Et}qXeE?1(hKQLYO;s#*}DT3O)*xNpU|*Cp->@HE=k0Gzk*g%?wVT4sByoY6Mt+Oz;YZDDL<9j7yMW8c}q@ae?yok
zcv&B4yV~hjpJ|(QpjB3T#`g!>o
zzxNK`EwFz|@1XZC0hl%8^2=I2+-eld!hza3gpp=29Ssq)bL@3Fa%{aami>Ne68Y-q
zB%;Q;!9A}L-@={0LG2yS^d7vRWW8>lHM~c>Sm&^}&2oOchhWUch0N-mYa{mR*ww#q
zE^zGt*-EI}KdC`tTqBWVKk!ex|DH)ureh(K>;f3hg%rn-=zF;Fo*1b5x4X5mAHt0T
z{gLU%4D-5EbS7|nfT1lw2B&SC={cjD{hBqnJH*_V(C
zr@YV%KpH>jze*H<6$W_|u0I1i$mk-^YqV{5Hr=6+TBU{xL(7s}ckVKBo1SLrtNJB(jVYyg
zig=e6JQQ~h`a{qJW>Am&id$}+7Dp`QBhxO|OoP%8NLzP}KzEVLjDBPkf>7kvaEj6v
zC@*9>W0)QZdh`O(cLYjU+W&F&34L4=o0mv*#uM9!kHpPi%gPj6;?J?#wNU>#ZAtyE
z_QJ@GB}gL;Hxu~R#>v#$u@420ZqLBUFWBp%0E^9IU%4J2-0)owOD{&m|pAX9$teQx?-XuzMgng4-?85$=x#Y+NcvGJr;YtJS7
zud#_4rv%JhsD_wBs}|TDG%O>c4!z<`zoT9p<&dW{6p)~Gwak)pfT;N5y8f}a^6MLg
z?0z(!gWCg>5P=dM9CdQsPA``45L3NY1^vj!@sjkQw#B)O9
zb?_QA{-z@Mf4Ga@2EVk$NSVZjx?@=QGwJqo%@ZoAL_M5#P}=U_sjD++ph8{mlYdRd
zOu_>mPb#~A=l8}ZcnNP--DF#r=Wpd=$E%%|ONr;n(uQiG{17-eR
z;g8##O9EFONIm7(8tfJqcEn;ibg=&qqr9;es((yu00=c1X?K_RA7yf{
zy7V%j1-gYUc&*}8%hk1+ZgkZZd$RiZg^A7Ncoq#OpL-32$^324@_}#ZDsTduD~sX7
z55l_BRfl^vf9eJqRSmjI1eC@Vg@g{mF${<;5}lq
zv_e{YgHb9|MEQ*J{M2hp@}%pTKh^|gcP(CXX3o#JLpizp9%nsbP|
z`_D7ziB-vbG|e&NSJoT~zy0~x-DUjP=wv9`d031wukd(?ai&scn;bog!7KP4`^|&U4k5**phmu^g>Ury=boyU@7L>p`_u5O7sTYl5FVK={^K#l5pwE
z{5UZM0J5|FV9}j@Vf?CY3^xP#B
zv|YJfS8z!PfY-Dq`;CFSZY4&}BADhCKGnB7NM3h!9wx#(Izs1nkkRv$J!D`6pN9M)
z5}SH*i?G9TjiGC=4E4mv6dY&K0dTBr#F?1A@JoL8$K+(nAUXBMFF?oZY%(JU%gTA3
zVZMSGR!ulagk9@}yun@Kq|04o^^b0VB{0OofkrvJ;^%sRww;&_3)v7HQ3A$1xDv8z
z^h2M+Rl1b36tB~~7eehDcLh4^j(En|KX-StW;}XkAahMPsgI($A$(yovj9tIuePO%
z-a2vIu)H!e#Nd|TP9Ha@ylHlU_YK02c)^qVM#lT$@FUoTSNF{383sVusUa3Z9)t8aKv
zb`jK0b~^mK$U~cDJ!GROd1He^salz`gW}D`VE8Ol*3I3#K)Rsmw`cgLyHnqgnvwrH
z|2~I3upSA<-ysI`D3eC_(4;xeMw&tc>iqzUIbgzx1JzFUuo6jp6|W;=MNOg~__)wAVxk
zrQVWsqj|$+4hh>YElQ3=>@pH@je@RR$n!@y+mF7L#TCCma?QkC_hbGHWJ=K#plj%q
zfGNX23~q6;P1(O71VD73gu2z&kE)&A3g0ACNq}Tu5j~;=g+q1
z1S%}~6uW~aheS6|XA5;_iz@0y%~3K7W>jA}$DqmPf-hmQOYg)Z4|D;a&Q8{Z9_(DD
z&0i}i+?}MQ9d&sR+4flEVV`%;esW|(l&)dtLvdChGMUmIKf6sv
z;x?i>*?L9JEXX<)Ycr=ZRNA?#M`pe?YV(bEBEb&%s_D>T^(OltumigaHiffmcN8;G
z$~3C{HL8$B77TR)US~E+6d<=7B6JZ;0c+QNk|Pt(LB%>v^X|akt)RD3kiMtqlgAYA
z9LX&v6W-#@z1Ds=5%P~p6s6Rn*Iw^25+W#I%iGPq74%(HzuDJ6346puNunIUK7%Ju
zY3mOf;RIvzSYkHrsUD-=?M`z?p=yK6OC;qve!fyb1~&0MFF&u@<8otDS0Gk3NHWGz
z(6CrI7FS%rqk*v{jz!q;6Y$q{tzcJ*G#yOT^+4Q=(m+1E6l-gScfa_T;cat8Xf)zc
zMrx?Re!@z#?|liAN%=n|47R;AFx#-K_+
z8A})fzgL)w3~%=hgxJ+e0`#>;jP6oTp?r?-nYguh8-Tgv=fh5rN$NeSD;>5~AyXJK
zs-*>MF3UR_rkS0Bs)oAr0WEzYta8K&Ff|9*T!LKsTTE&gV9b(3Fp{@KM~{u_1P)y`
zTVaz_j2nD54PiVDxHm80M%<391&C-gHNt9GNW`8__ik&OrLv|}ZcfZ@)OvDti}3z3
zkun1a9qiIhj_u4;MgtpjX_y8Gcr=3X1c;iGVYf?H>N>$H)g4e~GsBs(uhfma7{GpR
zTA?TB(?#qR!Z#b!E8a(?o)f
zda{udeB)v0s&}kq>~*V5>?=hy=HDZAB4twRmo}S^fjQy?)v(xJ^wP2
zsg+vx8an3C)_3jiF88h3eeygGdMj-<;~u{hL=AYl8vG$<8T2W8W1o1viwOvQff|*u
z2)sx`)XIl6|GlhX)fHzH7=CHN4Trk4D4`cwV?vO13H`^+Xq6-hHEW1d6+N`YQINWU
zo8oU!0-b1ahDttp&c7R+uEmvZu%TEtWI*`yd~sFeOvh7ueM)X~m}DYCdq?pZ_n%J;
zH;fzHQ>cK~SL{bHm-^nIfc-%tuu$v*AFJ;jQ`;U`k;O{)bCwVcd#-WL7o^T)`
zymmPen;e@Xf1TYD3V5hl@*-R^;3TmUH)#!f2g_jnQww8ccDB`AJMW!1AD#nICRyAU
zHk#Mf+jTeoQT=pJORRqJJi-TvdP+3O0?Uwgucf^-eR85Jcxa?Ae_!k?^m`)&MweE#
zlZ^;h&N^MD8WyopZN4UM>qEnz4@+nvivjar
z09clUckbJ>AMnB^vAPBT%uiVfLbJT4siP#1>lCfcgy^LeAHXO^tP4s*H+*SG%o7_Y
zp^J~ZtT-4V$Gu*1y`)z|Ex!Q{w^c>*zd9ir#G>(97CLQ$vbA@X8ya30o4Sf^oVbimcEnQUl?Z@+Mp(Ni``F@&(C3GAK=%c(7#_X
z@QM(x&9Ml#|A^Lp|Eti6nMjL`2iC@(>H0
z>UN4MwZPv#z-`j*_&B_o)EI*Q3W^b5mEeJ;wQP4n4vU?JR_r@88^3$G+;GddYDnYE
z@lhtThUb~it}_>KJOojIF30x98)zOh!~=pkuac?Nh+LrWoP|K|Iv(0)-{7$Kh%&5d
zxnIO9%;I2+LqS04Q6rEVI2c=Lvi*JhW(
zKCBS^9(+niRsjX>$TO|$w~Uh3n)@HXm1gy6COHQxE7W70;LHEHP_`|%8TJA_4~yYd
z${y1Lp!#S3H)!N2>NCKO%dtn|PaX80M}Fq?IfeKMwcl^+o6BIboaPf6uNF26G?N{
zzSxTY@wt#I&M9R2;G(rO|8Ev87$b3io{8W~@%4~JS`$Rr$B#HKYb)+v`P9wm*dLva
zYxI3yiyU^>Ei=73
z`=nFFE{4>cJw3dHz-JfkYJ(33I%;76znLJv?A_B>;9VC7@HE-3$+m6Vn5LSn$#zY)
zZQHhMvTfTn+3)jbycfT7zc0?`Vy|=7UVDAvM(nnKx2uGNj|I%FzsLCQIl(EMJ|73i
z+<*?kw@o8`JxZ~cFvE6N9*)$hnd?EsRXvCYL@Ua$D~@QJ-;Q1Vx+m}Z8g|f;|G6HF
zMxr5)>(ZxeswIQ;X~(bv0G7+ejyzWSza^920o?vOOKT34=YVIS1e2_tq&NF
zRF>Ay1RrVsi+S*ix#$zZn^@f$49n{Dc{tSAf1_mI!oF7F
zLMMjcHJPfBNVO~0NCLQ?6ctafAG1lrdMhXvbz8-?Mzw4~Ir2(1s`$gi>JrOF`IvfC
z9kr1@%9epf#G|+9o0eu@J=fq<%I(ilgq>!fzf2cWPEiY<95rKpOwOD5q0C;SQ!vWts{&p6FkL&~FGvc7y-%0_25mYKvZQozHDW`c-}V$!
zoShvE&)6KHV>i-$_OI;#YpuQyK}D6yeaZK~Sx~{IU4;8~JeGcSz}blpuX;tl^xE{*
zocFyqA{R0#G#
z>@eANv+?}NNJPUVzTF_P*kd!JgdildqBL(sfZtOK3j6EmL452w
z^iXWi>djvT2uIc1l#j-Ius9i}s!$E8J$$R+Hq8D4MZuX-zlTOUw8oi1v-ojOI{28?
zRU*@UY!0SLjQvl-{y37{=&X)TC_d}6Tz1Z{(h4W|FVz@TKIYUM6P*FNy*R}m9wJYK
z0i|ryHy@@NqN4vLUOiU0Gl3>^b)GVw+en?-;h3BmVn^LDbcgxl)PVuGEqPThJWYiW&`x+E_8mqc=C>7j64s5R{HgZDK8m6>wGmAkBLg6C#E%T*R
zOrVZKjPFcpzckFZ6#bY<F1UF*TiRT+(gI4eG{dstkWoM^msPg+ls{jos
z?-!Mz$Z`p|700k$TpvuD!kyO&b=L{8n)J2L3usR#PGp?0s%DOP{=}g?a}V5r&Erh4
z_>Q%L^SY?rw>I50gMjr3-Wg#zc0zN3w7=7wXBvDRME{&A-yA$HUD+bWC1mDo*m3+9
zBL3_U%=5E6FR0nDZni?{|9CKs5tAOrv&X5PVpnzXKnTV|W*ZG>a7y;JvldtDOuqlF
z0{Kjn)GfViwT2Xt?Pg))?)dKnH2Mhxul`K>q2}m?`r3{!ot>m2pAe0Rso?+yX}%RP
znX}-Fp`6-%LRi_L6}DWEU01(VpD8n-{vJ=ps;aI<{ceYFq)r06j=Iy=gMW=yzNCN`
zQJ@a%oWq9B{i3b@ky|KSNqXK4-anLOxhE1P8q@Y-nFvEBYM}uX%UjghoJsR$mU5p9
zX|VqqMf4(cwMz_DQ?qFEgq|1gWN?#qP4Lqd$sR#+;m3P#S#-aMg1w@5!S6u12eD-E
z1^Ny=<#_qaO62)n6;-*vqo;g4Uy+u~1cr$hU`dPN!rd!Y>*`&49uTHxb{XZQXp`kK
z38mdcaNamia5%B8XD`E9Rx-*Mx8M2fAT8t#v$Ff23ayJdcS_ql8nUB~uT;8fuAU6@
zjGKfH+e5@|F?aXqe+h9B74!GrIbvFWGlauxe&H4flu!#!bpD<3N9G3flFpNMSN5P1
zK~X1)jw@+{h#BOA?^{fjjH%k&9N6>^>f7hJvonjpOXLEou%#nL3GN`)aOP&);4uPGBbpP&maFR)GorkO_k
zrG!de7-7NTnv2>8sqEE@EOD%Ek#vqCrh>z%yBK0tK@3j`eILbG(c37Km|75y%G@T@qz$8YW1^ILf30zR{Y`~vs4|!b^|ff#wI&Iq)VA=r7)
zU(>sw+_d~Ji>f2qbrGRZxS~Lcp3DFv1T|az5v^wSpJpr&F`lpA#u9BJTiX@wO{UaZ!
z*m9fb1Xwng-?vcbAVl3hw=!3*TzExf6^PmaGIRC=q-ES!iQW6xEl7IQr_R2
zaAh^H@js^C9`A(SfZ_b%FxIyf(WY?;2gFx
zyC2Uf#Hir%5S|oqULs~7>3xaTONYb&@A{83iN$;iEO_zrd0Y#%<$FK3Y>TSXiz~
z`Bd+qXcc!%Hkmb|PE`)eB`^pH?1C>N`ZcWAm&Z9?oTIdRF#U_Byv(8%%TvS)(cL^-n#Ti@#@Om0Dh*9
zJP~}CX9z@|Isv>tH6vF?le>3wx&VWaq1IZ$qn&s)EhxJVK6FJyf_Co_MkCC
znA^sR97OrK6f!&Yj{NVBsKwm^+8mee)k(=E&1>WU9e=F|{P&owbA_T+;<7BYxxr{}
zJ6sh@!#U=Thzkc|U!ZsH%yAds=*nV{5@M~qY#{t`$OyBtfDZ^K~w!K|J;MLy=oOS>1
zUshpCLf(xO;dg?%gG92+XDyiT$UgjFD1Ibca8XvAvP~FkvJg_T{OiFh>E4w1L&=@7
z!JXIM<_&1cX4+t=V0pAo6Jd*rY_U=HOWbSoxrjq0Cv-F}=i+~nm2!61N(fE`8`Cq^;Eh2)HLJ_Q1H1-ncU-B|F~@fEzB;DnS5L5Ewm#<1|Dm2)6
z5b2{gXKR$wq49~gqx)J&2=FeGZFUmL0VOYP5mCmK$*sHrDlD5`7|zp|fnyKE&nHyh
zt?1BS%SO3!U8k%WwA6P|G(USW)pM*vBbS~W1UW`J8B^Mzv2IBoa$QGyD%~vfqYmIw
z?XqVmHKcD6yA*{s+M5_T)Eu?%iq2ISM^|}VD%a~py6uNl|EX8~>LJbN+JsQRbPLOv
z)8q@A#Oy(6)2Ogj;u;Kv4FQjBw>|-DsDWK*cY!DI@*BKh@KG0;;HFcS;He@lwT1F`
z1@VX8XyVfi$Zvf0N^|BNWVj!JCsIuA(&KKH)>G8iXbFI37-ctmk6|lU49R3AN
zoV`givc;NCMIRm5SjjsE@BSy6}D
z#Np)9#?HnY>)*rv%T`Z!Ghc+MwU%NHuDprrG4kKS*DYeDPF^N8DVYdRWwAv^m`=R0
z!4t?a+s1*@ba9-?6#G7v(Om%+A4*%a$dB4s*
zpDwwQW5;Ooxo48Gl&lU;viWXPjU>C%D0riYSzMZr@SnJJ=;a};L6+e|%DoxM9dAa8
zPh{`ma|NyMdMS5;8`a9+vdNrhl}P3el-Ytci0eoRhyHtWB|dIXe=gts-5@Eg#L;6?
zXunU-xZt~PLMKl2F7C~3+rVlN4(P@Wag6!AGHlOqVd^;O{p4*i7BUXm65W(kogE3z
z5Q+X~kXs;cslbp+ksIMibG#g|taFNuW_oo=y#Z6l*!QpO!(Pd
z9T5hHgjDZ6dw(B(9HBdJA#$c&i!_Kk{)hz*Ibd*yTIDny<9XMDHD2_0HME92^`&B!
z>$L{snL=8-LN2Q)Z=s1zUzZRu&%7Y4Jf}cro;1cWqo-o4Bl_0uBM*K|!2UNAi;ad7
zmjlvK{hR+*wuI${`yml5PD#(Wm&>+N%PcC9nl>XVtsTyWq$`)IsKW4g8dy4;9R#C#
z{_q}9h4DW((Xean>D61q8{m#mth9fS!#E2-tKD_9-Yy43F_4_9Z`fh$cUCxc_3(ewTuTYPy
zdA2xp+YfWp4KxwsprE;nq#8GiGLxh_uO4gnAllX7O(i4b6+y<((;*s)U|5p}9oWhu
z;u_7B-shi`4_+=brjJD}*kZbh)p@}T&yMHlp^(8YA7e(wLHY=1_7&U-xI9ha-zJGG
zbFM?QRKb}2@`D41R^)yBRx!7}yMW
zi+`ukh<0GM?reD>j~k**ncv;?bF&4*)vz&|JQ&LO(i+HFG+{`{;1SG^1OW~p00;mA
zfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@
z00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve
z0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe00Mvj
zAOHve0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe
z00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r
z03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>
z2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mA
zfB+x>2mk_r03ZMe00MvjAOHve0)PM@00;mAfB+x>2mk{AX99&U=xInYc>121e|s#`
zkq@0>1i&^pxH?_1&Jr{UWpVfwn?OuP`_sz^>C-dI!+(y>DBq>uULp}k1S6j}`BZpi
zZciPjiBbw_ct*~|Fp4V}mpKe0=NTU_-QxJY*0A0fWH7iiRDJGtgyEU5RIN>h*J_I|
zON{VyeGGcW=Pxvb!IW8g8RFhpPyFSoq%5r@ja>eAdm($~)An)*PeL?Ta*MLOh0hrz
zphhOaD|uWE&h=;xj<}aGu%VPMh{^mW+hTcE@YF4w6yu#(>81roXyS}z*I*`rP|}UW
zd3x@Z{*CRfS03B$oqIm7UkTF6(|>*DM7)lyp#Q08;2+xL#y@koai!Lgfn5rzaV5O%
zVs0F(##yAhN`I`8F_EBwJ>CNUJFxii0WDo@HEEWj$0x4T@m3d3rVW|4-dWCj2{Jh$p~lvl$J5-moEG<+5VR4u6KIFx2mPYv@UB&
z#BSf{SEvV9v?Pgg8S3rzBnqS&R<$_3*Kdgh(HZ1ZdfqQTD_#p>gp}c%$
zT0UM^e-3`SWAkp2MG=!h1jCEf=lF%3JFb4l^FwWL;6yCF>
z9>^FAHu#(LDXM+&ftdwj(XNEjvKoM$M9BU%H!0zL@mizrX`7uzR17{SZResfTB&S{Y9s-%LNYj?#
z;d6zjz;~8esQ1zG5>`YBu0*AnBg|5^IbA_J!C-O4o6GeLFNZim-CfwVZWp&|G|Do=XoMkF*dt~%SE
ztX=MgU;vV2jb;T^DN26g-=e9t!<@I43#>=zt@Ikwm=M{3R5SQ55z5+f9#x9R`=2w%
zaQY>$o_)!-j5q;yWpTWO!_uM52Mz>}tu(!ZT(ZLDHBjCK@DExER@f9X*9atByYC@U
z*8E3}kqaOKRx~O!tr~$RHe_44UkMw>d$Nbxu8yG>28H0-E5}b^4P*9jjBu!P^Ql$v
zJ`DJGq6$;{W67m%XgT_qwJl_qAv0(@3Z5qZNaF;936ugG1!ff~=3sTUwwS9%T(rpb
z<6<2l9@;Zi)=*7W%kGIGuw&R3OxCPlNg>_6A0{;%XyxOk#as{kPIE~Of|_DP+i0uD5nGJrr{DQ(8
zIwBybBkPweYJQ%?g(KRJbk-ZL8thpk)$H9I@Nh4?%MYFZ_#bI`%Ii&~{f5fM*g1I=Nw%(~Zbm|vvqBL6UfcDVqxD|d>
znxB^dqww;fZ6N1q%{1Z#6*$|Ox^YcjiKG)fQj|{_SB0+MFQi;2Yj{99w$Rg;Ox)q|
z5GQP`B4S7&FK*H~LJq;mI{Xx(cpGj`Oo|&;h1|}pTP!pkKBjvSA)sMQDmC6U-^=jf-7Q*;C|fc7L)bXYMb>a4B@^$l)~>M
zJ*0Yo)Z^RmvCzp^(C|rZijz$oh}c{n+_@I0WZHMNMG0Jj`_q`eK+Z~;aYB4APw#b$
z#ynT1v8VD&@+hJHDapUTtZlu$1m+q3m^&;XD;)FeH;B|O%+uviAZQs;#!H!J=!mU3
z+aD7CzAOrv^xj>18cOLJj!i
zX@=TuxroPI!sty9F!DWPa*4^01qyZVeux^UMNAG`;_dxH9n10Tn5PKLAsymnw6$|D
z+Fi1({rFUk0DF=e?ZeZ?z&}T7)Q7zc`ye<9)X9PI>AT)9A1a==@7fx`<014{2u=Rz
zH7B<%^AT;v6T>dV&p;QdjXB&)8*=$0u!!FSwI-_QRd*870hbHXB);Q+)$MM(PWU17
znlcvYpFMUI@#k4{%%PuRq{(QN8e0VP0y}D_xxk*VXpz&2RWaRNLW5-l(zN)OPVU4t
z&fZZreO)^=xjks6c8243BNfzFQY5E<%neg%QjP)&{_wdpkt_Lrt$`?XtYk@OB*cUt
z7|L_>s|;<&NgM8FL0yy4%>*{13<8Fcod={(McpZbYv6tP8e}fW!abzW{`KWXux4+C
z>|WZs`E46>t4&)L9Hb-7#y7J%D{{bcQ$lYs%r!%4t5Y+yi{m~(hi3|@bSaBwj}n0jM4-RA2G2`}
ze1MLb-f|TOKC>jhh}U6O-cg*@O(^1r45!DGyczi34RE6O|EB5d)nJSe@20lZmIXiham8H
zaUUu7`YcNv*n!1;Ob2dH(%Y}sfZWfE-?6Q3r=2-w
zbk;B^fxOu?a<#=YA1J2WLbpMaqdvrIf}OxS#r`lf;E_eaRYC`$#!+>%Yk!yE@D-HV
zs&KNBcODRYWT
zafr0vsgK%L`N|!{r@eDA9)XgSQ$IKg(s-@b+)Rj9Z+Q$}rinq2?$}Uu@XMho^i_qP
zka}cl)!L9SkrzHl15A?%-`Ef)nKA6`F?(Q1hENhRN@%?uW7f(y#HpEiI4ap|h4SX@
z?g_uW@c!7K@*jthY;2NPl=mxF0arEWw?=Qqtx?0_y
z!eUe{auH1y5cqi8E05-b6)iu74MB-$s-^Mm-%|}PS&mF%AchS(wV=PRiEBPVXoFK~
zuJzrO1@Dq(Eyj}36MgFXQIi}c!!R~8atEV)L!fy*a@3uQa&e%J?FSG1)V1oNdmz%m
zpq@J06$Qc(+8Ai%mtnBZ)7KZ|r?6wVXWydRf1zxk`&|*7NID!4P%Iokr96|(L2XPh
z_VE27wRNx8P?>$M$eR~6Rfv>}z?efVMe4rhnkU+nUBj|Q?x0v$GcLDyq%-!W;m?#H
zgW=z1%{v`q5=AMN-d$;3?04G^=8?}->QzdS6Mg=>=GY>(k=6XIroGbPk*LSj#1Iyq
z`Hxjqf~Miz+sAM8jYH;B!IsHzS%+0eS016lIM|hxn+8o@2=P#Ex6$}M9a_7nV@^1Y
z0ouF<5xlpb=AkdNmW(h~wc?Phjf}IH!32NZj$sHq-}kV`*@^h+E(OOlJ0Iumv{%Pc
z*9RP6ia~bSCKXc~ye@9+JhwPgom`^9JTN+C=~EgHw^X8Q6t$^htDCQH2zSE#J`k~R
zXfpMV*Q-Su&;~jb;V_v^JA1ZoMXctZA>dOPyJ3*MFOos4A&VE+R7uSu8{_>Tlj0k(
ze!V}Y$wCtbG!apNB;Z&IzF`=qLdCgol626*sgpXV2k~l(7J{L?x|ZEON|~~Hx@`z=
zicAebxD?fB2U{>3+OqlBlCkJ8WZsrQhr+a^v-}?T;RhLj1i9K&;O+hSkwPd0K6i-5
zM|L&6}x~#iyuF{#?$%ELwsnZ>Rv-6)eSAE(G4KM`&bFzads?)UL76N`LeH
zNFB5~2eFQHSlzQ_XyNaKTx+cfg~RJC;ool0
z;fDa@-o&K{kEDn|LaPL$K2Zm_5U*IVSrZ!2gI}s&;mIxrvr+?dUutdm3>mWEDj!%Bm%D
z;h)%-kYbn8G(jU(L|V}And$C_x`%V2xHp=>4Eyu`lv>XIcH`?|q`r*J9yEZB`Cfc4
zIG_G)(@1PVscx>OihPDJcbFckk3O^P2JNCkedq^g(V{7459bygQKjP|{EA+Ybh-e)!nh1}Ito1D!T$<*&>XObGsq=e)Zs#ih|?
z%82qkyt~Ws;~}`qHM8(uDnWAD#TjB#QaPTcH=zQ1KfKX3(pdD#a8pw%o{BnB{9roa
zlDPrUEfm()4g)>8hH2zq9QQ3sx^{cMON&lgsiFRx
zWHsZ_4D84lJegPI-=2k-h(ift*e!U;soIbR$fQe&_=&EPyla((CaK2n*Y53|tD>$j
ztU1!?Pa%ps8xi5OivN9TA2-)4yL-n+aY~O<(BP|3it2d1JxK^G1GS7ZO`q`=rq5YF
zDI(-Mh1qc)s%!B|NR?))q&@-*Oebm4gf1mLWCv<
z=AwTb*G55}AXCyruh54mmP5&2NvxVZ!iN@2yrVWyBBRCvSO)>=9=O+{fTUD7;IwvN@TstIMdjy*9q5Of{nD{3+4)o1->&cGM}_rhOPq^a%XzWaCT2kiED@gutg1U!oWZWg|JXP98*1
z=}^XqFHND`CTmuVlND!v=8DJoe9w9G?c;hJCUFr5+xv_}*irY6xAuQseOb@#kYcCd
z^MLo8WOnjHZJ@}B*au>8S*<1Gg~)RGp~4j
zA+?V~+Nhkb3LHLk|8Rb6Gjde@>hsWndL517>)xPd)!p~Rg(uA+NF-Ft;3N3jowABz
z{9L5+!J>k;Lr8jT+n`kbd`uYaRw>+fG;)8!BT(?lgRw5f}0?qqtN5JCeK6I}bdwB-8L^KI_sV65_H(
z-&^?KHo@8}^lR|SjIvi*4c;Qt47J?eq7TKXxfAmup5GCnb-Py6-h4k4O!o)*-AF9N
znFa&FkadK~9|^WC!xl8iw+JP7p3PwqL9!b&;jy?oPxC^rz?$b?`rgKk{cDB03m;yy
zs@ZzcH@y}?45|)LPWGh}8W=&wFqYxPdLIb&jjRtcQG5TzQN}41N-?WCt>mKF@sek8
zO!{aR&*5+CyjQ=3RzWn@t7y5O|C+sfEoD=dc8t_wRUuQe>tO?nMCCRbu__KvgwvRg
zoVAnaS5)o^T8g8zAbNedvVw@M
zk{&z^iD6Pgx)dxHW?`B%v+XZa*nPFMXRSV(QWMC)qz`ZB9G|$sX!W2j9WDNSkjtIp
zH18pb>4Ia#r^=d2q>SDX{7p`mCJLtl(W21l#U6GXwTxJvG(MNXJ011Fn%iDVL7Dw4
zYQ@wf@l)xCSj=~+;cqp=6GjTjm9+S%n@+uT2x3Q`;zARd55<@dv-Tp4BPzUe=f^s9
zcj}Cs&%?ur1JHsal3;)H*8CZsRCcWJ2MO?y*$63kZt8-O!vu6zA4bx?QFA7M>{PLg
zs}AfXF}A{Zslij=4h^<%VPgXWFX~^VEST3+76h
z%9zln=f9ykyG{3j(wEc=%O-Xs70{k*<)yUPZpom~A*|-P%RZhqrk#YMueSk`W*#x_xI`Ey<&)6Hlau0A_Atu1u3UdDlWmV(f+TeTE>oRb4GfkwqK3t7HitN3bZsZKa
z48Lz&T5vN%Z9}YKw(Gf?(U$%lEBHI1r1@pTE>NwPX@~cbWvxsOsx*%NtlwO7A}RuL
z$sF#6RGJ9?<2e56vr4qn0BhT=*qvM0zZ4&Um=@&vZC4wk+ZAs6u=1CX4Hj&=JBy@l
zI&bfH!7#5I6PFkrWS)ZQa+f$4WrD*ue-Wj0A&5~b*(l9?r4j*NEH*)AXFKAKP>VC)
zI0DNg4{RVks$vjbw8V;83-i?H{4c3JcEqM!C+`I?FL
z-So?4j1!HJ8dPIjYW2V0oiNf>V7hhmOEdzFE(K@$e>e#!R`2=@^Zcsv*wroH&v;Z9)rxW=aWt2Q75@XSbYF5~7wKFjBo%
zX6m?4?v0Uu5zVF}Zj)WNps187(P7uXMUu+Jy;T8P>`wWb4w3+NC628c4lTpN7-5Hz
z%Bl4WNvyNtZ)8wb_$)b+x(t&2nJe2(q~Wy4d~@kz=gT&{WE?l_(M#0OJe`wEHLesa
zGeIP#R=;i|ho+x}88f;MK@W8;_3#=KdF1R>qQW?TY}0q|&ea2-G@MJOB=rZjUoh2#vwbN;FLnA(G_9?wM57`Wx2Di?{p
zU$olgWh5P~h?@e@TZ*WdH1%6-pKfJI6sw*45_HY?B;;*
zI9u0@M}j2Qy))leHG7=ANB0@FrO?7Kfhs!0^2-^}Um4>#}9Lv?@MXcJBszs~Imq0HPX)5iXO!&xQ@
zE}qJzb2FZJkJ@_6h0PVI96D+*k-W&tEg4F^{2lS2Ix?cPUVi;@B&}G1h$L~zN#!wE
zRh5o&a%&Yw@ETDxlp6J$?E@h)P3}{*BawkU{^9uj#cx=|v&s`1PK>%)3sJ$kzvxLh
z+(c$liKeu!J2{ilx}zWq8edk$C5u~<@z&k{B2>oY@_zmd7Q}J+mR0^^zev>`*6p2K
z_~7|wXC*FXhZ`%zIPP|v4$rZIAUF7YTGGytP=isM+xFd^lhuhH8+F1KhKqR;k{Bj+fs6xRnr3*E#j&q
zvQ+*pDPX2v%4(fqf$eE?B%ZZ{&cr3B=-n=k{n$@&g5__McdiOb;EDx-xG_vRLM1H*
z^6TyJ?8kf`*4#YdHLm`ubxNr>b`X}SM~B(tk92d6H%-^jY^e+AJs#2vI@MS<u6*_t4wdd4*1c
z{^jFh4BhSf>($ANIRX7btYcilbc6?==ZxDk_uoX
zSAms;)%+3ZoyL+ck^A#7L@Uc9Jr1wdw$nZDJ9R|h{Lek>zB#8R@hiR+d;dP$)EkwP
zyynjQmYXdP&oPgGoP_wY_-S!`P)6xY>?;fX`n8gNAS(!}y#BMHpuP`lfkW&zgbk`Q
zjW1F(&+QU^Ju(^~*BNITwvryyXC%!1_mRFyD)(ybgO!?u^>B#ml-1t^MYKsQ=x=?T
zcE4a}nr^#2e9PYnJP4HW3F!lZ(K<+ow8V~$UF9=I_d2nl1R!7iY$fE+VVcR9+k1BL
zEr^VfcCx}7$!e)CRl6&9l?NH2Esn>1D|5Q}Xvg@g>@M@_d^DG>Gg1o-vszbSJd
zI&K^AL3~!*lD?Sj8;VD;`7mjic;{|>PSVeZIcy>YqIX+DeYk**H(>O6%#&7<65|$tdprmAiXb;j
ziXTbMxyI?*)_{E-HY4$73?bMcRa6=SYNd{a4h3T|xtBL0DJF_TOSX;KgGMbx?ZY5?r
zvpm;qDpyBQ4qe%b!<%lL?xBqEz}2LY9VzQ6Rcby9B_Cb)*SfL&Yi4f+s%H}^QsF`c
zRj(jno|n-|r>!#z#?H9sV-0d8jnoJJ?>}L`Rle9+s13nvVe=Bg-o`%P_X)YBcN?T|
z=}0z>BuYCI!-BOr!`xa1)g_b(9xHB$T`HWQbd>+FMZQsi9@B)?0MDv2N1k9PT)}5{
zNQ(t6MIE6Bd6np~b;WVjlg7F26W9Z>t1jx66^SyntLqF_PMUu-J_
zxS(c`^U=R}@LR~)Hp
z*9=Dvq4RB>|0Pb^LI1Gd-D+6+ECEBsPTof#-OX}Mai(!ci@
zDh$__AKcRhux?d6U7U%ZIAaPVM2)g?Yyz_@gBW5>B<=>6fCqbbr-d*M2J?fSOc#Azl2U
zTXvAuX-66ho8m%=i(0F+DZfIJ66=+270|8XP$}uCeJZ;#OYPbJt8tG|Vr(+*2O?n`
zA))e9_h&_9BOBg3=|32-Qn9y~KD-sJRm{eqj$e}+;sq&XkZ`WJFyPtm&5Mk$_9SvV
zRaX}E}MX479>t-G@_3!rpD(k
z?K;OfCrn*b=oJ>6+(c{4%UeE^9W^AXN5e;S*k-C57fX`jCr}*W&LNt@U=ZK96cA++
z+6F4#swaXc!B0oAW6t%r`p@way0OQd534gtdqITEdLFbtLA62CYMVVA{sF6=zTTvo
z#|=5$Ky7jc;#6XE@uJZy$I`xL4AP{X-yTX;mU*T|LC_-av$?1?i~xjLDVn#bqtNQG
zlU}BwQ+BN4`I+rczMK1~1n=NyVe(lgyURb3@5dug%iTk*@M}pqKQoW@IT;)pBZe)A
z)~gjBm>6enDmHqpiM9NX9_sFQ@b1^uIXHa={3}P!F71&?aUp5o<&G?PAGX5&(8+!~
z5p9`l+Q~k~#-+(fgk#mzXe^5p(EC?vnQHE>uyXG>K^B`g#M~o+Og^!xyVV!!(+T7VEo9)L&0y{gCQYYCna
zy6O?sIBI4f*7D`fEj^5WfGw|2kqtkoQ
zX05`DR~bUHy=m=QryV{$v)_M9L#Fm*Y_75vhWg6_{bU=s*!N4svB&7kNg(ClS{wbo
zTM|ueCK}xnjqDBwxk)Z%FWHgCdop|a{^9*<6m@GA`&Ze+nXwQlKxRLfn!5z1*I2(E
z%`mc+u=9&Xmq8R#<%>M_jR!|~BjcnN#G4Z{mL;YvRl!A18b+0i!RwYZV#X97s*|vp
z$v`wc?>Ci{4PWp{A1+u~(#cU6Qy7=U=4C%*Is_p!?(DjK7*{U}SQa_FHZ#!-;X5KGPu5yVSHj#=g}9>=tLysDe<2
zBGO#5yi`Z8v&+2)D?tql@(|Vv?IvPQG71ebgjykG?Iz*caKhTBvtO8mPEWtlGK&1c
zMmTmmM$u;8Y;!~}%qyEYcJYNu_De{HH7S&~<`ar+&{eY+Xo(h(t`|vZgV4*bbY29ej
zMlNQUni&l~_mQM7K_wepKzq<1+iKk8x{xL%ES0U_%|*WPK@R*hPB%Nb>P
z$L|X!k6BN&?t6I>cz!AI&{|MMN(w^PY0^Ur#MvyyKkguScK3{NS^iDh-8rNWf^aj1
zPQ|;}DI@w)R8c*iXSOf%Y2AiAI$hJ)Br;^Zl6?{CP^5zCGxuJ8XFw8K_m^p5($?>F
z$|wsI7a0OboO&hGf0m@2m(^h*b_W
z2Y+T|Gs5cxk;;8-#`)L;q+V3a;ZSBZ8yN`np{4SfB0-5v^&7uTn#YvB>nRS2Opsx=
zXhoe#{t}p7StH&Eh>|1y>x3{oLL@40HA!sw^ij@>Fgw{|K7SG7CLDEY98(8cF581N
z{k4C*Q?aZ3Mhn5UIS-b#=yNWRx3p6vM`)mkFs#^kFH^?;;lF`N6%lgTbIZd(6JRf`
z!B2)V!A`~L6bc`k+SV{&%MzSMrrEPh!K%w>yBQ}W&+eC;AYjMKN99sh?ri?VBn@fD
z>zDQUBGY`aBAh2a;VZa6eJ_bi6p9b0WcO2(=$|+{BPB&CX`5mlV+HJLRjP_3-3U}JsAYkwm7yNt(%DkB2
znGV~sWb&A5O3z&z8q4o|O$u&R
zhPln&ZgU>{oY1+ZK5N0_yT>Cw_)M`nm$8E*yjXD-OWn0v-kpdRG{nEXpGJ9xxO2P6
z=28pRJkUll=imy7tD4Sk^pyx8Z&T`^|CJ9i!I9i{%GBUqVsODoJ%m8o!@d0u<-5K}jBoq8Dzyy4{CgQW6DU84$8X;p^f2_uJU<
zD(?m5lh{2wH^b?&WyCvg9KB!h3J0^9-}Y*z3dv>ESNUB&I`9^!mau>`1m~dxajh
zPh>+qJYa&0Q@7nDYP#fY*?qwT2mj`HZtKfmkk>+|xoC5q3E#INqhNgPwHXBMn`CKn
z9}punCOPKwcbeab5p(dyT}@PbA5O9*kU&9Oo37|nuCA@hQ0Mx^d?5eB$rznk;{AzW
z^o2@Y+0EFpXZ4(B^TxB{hr}Er0}-ENA;&@3A5o59AM0SwW{=P8UE1^bk}W8oLXpgf
zvOHp=NCwt^_^nS{8zAfkSQZa!Nsws&{@QMYoj%RLSfVXYU>T&_7|Ic@
z8uiEXVL`3jScn`zk`7f+iPi2RSUwm*jASHrU92rnJaqUr{j)mF%*0*F@Oa^B>BF-9
zXL~vurI0Anq2EV<6*Hk5l0ob9gISsvgVgKbc5wu1FCd}rE~!IvWkp^nFpRX4XE#Q-ZLe~cJ&FEVi6q4oBg{b&4{Us_^|6xA9pm|C%vUn+*$w#d<=Fd4of
zs(841il#&PK|+x31)Y@JY}a?2hO~u9RWQNe+IIIkn%~p%lR?;!#DwyF1b8->iuV4B
z4XyViPO7Yg{N8s;eBzr-giZ>2BAaPEr2hjqK*+y0@tsBD+=0-lotjVoq|Be+=!<2G
z`^k-+f{ji3RvBj?G9hxxE8iekg^pmd=
zi`m4qQ^G`I1Bu##b}9aIRR1&8#q>|b5iY-R=gU>WIp#-1%Ed1hj~`&9;i8EcI$PtR
znZlcFStxa??_Bg8a2jj5JN*Ls@;w*@A0$p;hcC^F4_#>i1AeH-gxMgHrqu7%
z#fM5>s08-WF=!nFT+x&vGt0;z{;XVb4iBzw@j{FmvyY!_;8qw6^G5^IIK3T4&SZKg
z+KY&=hV^lYi-aeKPgV$8jUV29W{4V!*vjHIXJ1Ow`q16Svon{P8oh?N`Z}vwvwBueuh$
zW4b+6`zQ;#M8yo~xY_$pUx;e&m+2rY?
zj#s7LakSxpERYJO{p-xuJXGUwf%inmX-SVchd&oa5&k(*^Gl7~qJf6dvkdpl*
zp&0o(C!X_)u{pz+P-Kh{0XE8hcgF)}@
ztYRPI&Csgw{yjXS*uy%k=0y0a;4QE287UEaBFgfGJF~qtKjb@N7V*-5X&!z
z|Au-?#GR>i3YD`0(nM#qzp-|4VSdzy-B$k{F#-_hjSk`yiC(x>uOId}wROrfac!_c
zOtxz&MvfdVi0j!CwWU7Y?-016U&4{*6J@iTO+5Kh7uZ1vjtUZG92GjDLpVEll!*xQ
z5pxAs(C%cbgH(n#mT)1ZR(w3piv)R3tkfu;P?_zaA@`MaP{_pM`R~orI}8lkASpT_
zUT1y;b)XmW-qEx?kyxFE$ik|NqSmoD2VRFQ!D8b(zebsg#KS`!tJOx9c8yWl=I{uu
z$YYPM1p9C@<=E&H!I_OyT{Bw{FwMizbfV_wj${Mp%XO_H7@WxXP#39+)9BX2H^+%9
z^8PCMVl4h+Xo}?kqbU{|UqCB4B|wnUTaKaRRP*j$P#MHB7kx7<@L{>qpcIrD1dL<5
zr`VsabBnxPXrPdlLbC`O@&bm&1x!)F_cpKYYq8n3ML;K$8~%5%8@NR`2ID%BI%R5D
zSEAsRL}!(KMoW{t-lB-rzqInSrhrR*k=peC4eb&*Rj7`!h5))p>=X_hFsAL#!=#q&
zCA7&y&Tx(~sQPeUsdLJJg5@`Rm_7AqvWCY%BFvB#NaOZg7*|#!S(t1Z3aH}!zLQ^6!8CC&#Og`sMbw`xQ`mB!a
z*C-3;GMk7_^x%`xXLL}1#7*%33jxdSE<<7aXXiNvV1{hFMa=|Z93gppNHT4@;j`)J(H?aIgn(OCC
z{mTV={&m(gX_q^(x09C_E(2tYlt$@%>JbJWB!JH
zsLRt0v9$9WacPkteKd{2f@5q6apt*HnGMdZ+j?&Rt|f}PQJoak$B=*fdguEZ5x;|H`4#~f
z&+TgZmmBdenz|V=9g9CE2sWAJGUjQEW;j4WiN%dJM3D9|%2t$x$Z*S|BLbQjh6ZQL
zL^>Xx#k6XRQKj~M4*wtkxJ43LWkw_Gi@iW|)SDx`ff6yhtk+9+PUR4t}%0cb3E6fR-7$lRSwqpTlq-ne0elk9OmX5i@
zcsldZoe>}oCEj?ktFINw&kK(^-Q^=dqotpo5ui5FQD-j6ktCvqF0z$npdty$nmiga
zOdJ|zWOu+YlOWqqb-qAAFy_md!|-BU3N8kJ4ZU
zmyr71<10N#Rp7Sy1}M8^!yYj9<2lQYZ6!+=ej3_?%ybzP53h8nuE
zJlJ{zjU(!L&n*sRCR?(`kW=q55b_-B-P*+yttg)?oEG4bo|m5Ei&q9Z$18r@$>`t-
z7Ic!Uf_gn7O=$Xmg3!kTgq9y*?+^Imn!fsSz!c5R_fBBTIFFwuFv@9tx*uhb(X_Sl
zz~iQ#QcS>TJ(GBQu^mg_{L+HR6{z_QoJQ;w+iR`Updqv
z&@z?v*wlnc{AhGcP;5}Jdqp`9SOsfxMPfB6f->Hu1z)Fd)YiXG3f%QDK7-jXo}#QA
zSMb#xo*`qCv^5x&`al?OCru~S;tFg%C(^&esuq~~P3|A$N-z*-#?Wn|{?>%M&5rk5
z4^)Xr0pARikKljCG8#@$cAs6C1yyAkMQXXLabnqZX=w8KOVZIO9kDAf2iE!WHKt%^
zL4rnbUyoA*9b>)1Mp;<`=pq=DM6`+0>jsEVA`dO7v8|*sOlUf@D*XreBT+XWhe-r2
zbrP;b(EPPcPj~rnML1~k)ecDAeseJ0ZFO)b4z!SI1wD=8kwzV!m4UvuqFZrj+{@xB
zDY#WRoCJw&2TBj=FCKj+61jVFrVOeflUgrz9eSKalq86zZ_$Jj^oyDdWUXWrSM%H5
z1^9HQeqzg;YExOeCa>X6B;>2GRvZ?)Za^F&=woQu6EQ*=D;VG`he8Y4w4mtn6R-Pp
zUHUN~)p-Ufz^~->=ccZ`2{suUFYPw*ddd^sLJTR=%fk$LUW%nniIjCxqw;qc0(?@r
zyC@^wsXhB{Vzw7+O^9+z@Ak>Qk%^`A?g*LC{D`K^_tyySTi*%5i|S9X{*!5E`0}gW
zxJR_Jw?h`@MT6quN8B?uwsI{8ZFVAo%>=uusI{U==7%5?lDEctahO$~VzI&b^J{Rs
zv6%3UgXdP;<+TolSNDp-%dPGu_onio3$5NrF8_=Zuqq$Te{$nVTJUH{d1y2IKE5>c
zacQ;)(KKW=z+nYFOtbEqZ}Zc>0d;$2NhLzWZRE(e@J4r+!^8|&lIo+vrIS};iGtWm
z4{R;;cw?icJ;YsPQUmY@FYRe|HEflu79$d9*f($tT)LtdhaE_ac-ZE&ES$90~STSeyfgP;l=gvv|4LA$A9{v}0CL-)72>MqGuxF!1J=P~pPP0hWam&Y|
zY!h_?9>RD|p(oXy*_9UQCEZC~dbS9^SPGqnIUqR;^?lPFdt}s?kYH68kbQ{Z-Ch-0
zO&#=Pf|d}?XqoV%(V9)sq70p@Ej40#^$D^NdoOMVZJ+mWbh}#1#qUXx-K^DS#COY;
zAw9^Fz;k_A+EA`eDf1FG{&oylX!-@o>CBBaGo(MPZh)h%`*9!uU4#CIT*?RD&0RF0
zrUV#Etg$)2z_GXd;B-S$V;kM8+q62XB+>r&?Muw0RBfGI<(;n~*LJf##!0M`J8C-F
zEo*5}AUtgN*RHI4#R@)Fv*TjtR6g4?
z7(!w8G}6iNa!T9kI3vG9qT{CNycJoySW+3-t5NC));~&l5!OVwGk7hHwh~3I_}*o?
z-96A#FNcRN6eQ&L`20~HIc`Cz(xEm4CFc7ek9v+&+ZXbSta!yl-zwS<0;S%(U+7gK
zuDUHvFh>0Ju~siV!6eDff^0qui0q$^X?4nCUh_QW3^ek8
z3+VegV`P%k8o{`!2_HbH7Zi~jG$;`@{cfC+Xbr;Ny~*t0n!~SM|2tQg0$|C|<-^dD
z<3jbrtn;JYMh&o%V2z3L+A0990Q%B|!Il&`k**=@KCvZg(WXjWEh9u4g!HEcSpS^j
zoSQi4sGp<~XjNC$yLIw)fTIl(`}9fd<_V2FjG&=rDU$ecmFJ7L8_T_v*C_BsY8D+Z
zUA9_3iw_?R)Qugd?~VtK8=$iitgA7^P@g(i59eL~u7}yxn6vq8!Sfrk}wAG*e7~*D#
zRIKo|c2bpKpLGne`MmggVwghnPzVG)y(PAaeie-w)?U*~m>fl=diQr%&TRqt9aT=8
zzz4q
z#4vHfY<}>!&ihb)Ruz54D)3KTU}o2IYOXyJT-!C}xu1m>R{Ry^+MYrHa%)ZjXyr7I
zN<8tu@}WGC7}lWfKNgB~n1Rv7LMyHTm@sK&HWT+-D^F}W)6&MDu
zalUG$gJ$fbLh}zYFjFsnehp;?`2a&h$}=G_Seq$`rShfVsv*U9DeH
z$)0$xTv7!hk^eTPjRBMk>kr$@g%%ON+xyR4rf`iWP5a+?gzJ_GSJ1?f65?HrrggqvaE++16>>r`8GvnU1tm&0w$$WAKt_#dw0{KQ{#I}B8x|8HGD5G=U-C>%zDi_%Ks9^z0^ID2MjCKGxz-WkggKDA;BSMYDCyj{fT^Ct
zr&`8IZSq0(wM6)W`n6v&J0)3}Rk0*CtAv@~_VEhHkLnQ#@IzNURVDxp{K;l;Bw3iCfBl&g)^JPFgX=iVFzbR`P69X<;r6fuC(
zt9ijawV|xH`{YGY_nV>^3mU~eFZ|i6s=jv{vrAB%N3&=Zf|e&{ci>RQoYp(l<;>4j
z2D8oFkC%K-+My%>G1LAA=6+#{CxxnIvd^(8CZ|Z2wT>Sz=w3%t9rdo1$L;I3xDWDx
z)jmF)d;0De$`qW$=IYDoXp)VJGJ!)CQce-6w<5Ha9YtaipkU+v36ukBTOrb3W>TBF9qy%;RfdQIQv)b4z2E7QNxs$zsYg_b
zU?7D+K_PwfJ)frpXv&pHk^g@Pd7y>ZI4W*{A;*skvWhXay0m&Kq03E|xHA@vl}h4l
z$_S33h9r#D6+Fk<-9s`UhsaK(lbx+<^p+BG!5?v>r5HTn=`XWiM+nDKX0PROIuaf&
zxNwJceq4bS{KK$B`hgW-hPIDhJ`_6X3};T1f$hf#1_uGC@u%eZ89ddH_qtp}sz
zODAf(010=7TTyxoTA~t}G3&=W41lmEgANp-zhA+3L}kN}Y;lOe7eG2IFsKafq;&Jj
zmfiPL>ue=OvE91+)gxlxO0>xQhVEKUU2H1`Q^9_Mo|knk`?er|`G89aQ8SS!7VS#OAFC%b
z1b^RSO1JBw*G!Z=-|4V4rYN_3aM$KEkzc74LZQ+gwirxJ(*m8SDfwp3%kU>nLFDZgogf0K!TEj4_m0>KYN7Ja+7q@N7Q6v90!fK4L{Z2?Vvj_`GEP
zjGtfv^x6h;F&Fy
zEq{?>hpJu&WV%B0dKYrA!vbUr-vn53)_SBOu86`&cego3R*JQZ(MtGFFu4^03A=sp
ze$|3ZEk015+?!Wvx#+5Z8tYTLtRPXUAmNWl!av(^h_m?Ii?Lce0F3ggBm?X3aD~;3
zHn3vg$R%QgIw1r|fUkUYPlIJC)sU6dvtm0rj1t^QLKZ1&smzGE>pXhQ5lpkTntBT*
zd(hpAGjN3^l#FF)a0AAJqSZwwG-MUFQdg3Wk*6$Bp%&m+DJ>x!p8)TFopN`)&sJ;(
z|D!>1=);Z?9iBp%@WaJH;)s`)EwZ4HhAW%kW4W5fGT=X$u`c@kW1DcH!!5)%k2nYc6q$nI2NamkbPt
zh#Aqwnt3Z8g3!+&gKNRxp@P!tXcKw@B2^!N&(DLw2e}w=U_fS&Z*FV&4bjqtJXyHg
zfs4*IG^IrZH8)g*$LoJsMR8D)KX@dsBI1XMo~A)S?+)J>eSMTVt>MDevO8x{BVp>J
zM#7AmEcF(vf7N)Z>hJqV0iRiem_vT%i0B*X;s8f)AWr8|kUhCGg;mP=@pC6eLwVpm
zIzwkuiXF~5^%l&JFvgThEij!F_gU<=0a#PQ@3+LUicTC`^Q|P7@CtpczCUIPhOqX1
zH_m)7$>)ce(fO>X|L(122!cDHuHvIXORl!`^(8e3o*WO|D6+8!{QE7s`Ik>eB0*Ks
zss%E-s;_qOks<&U_C9bbQap=i&6&6lF+P~c9<@UTbO>=eFxC1YC|*?Rm4ARGD9Q95
zMe3)Ye>^>LYzVEN{FMv#^vTv{q>P+a`bTZzhl9O{)MrW?Nz8RM{Y8zCUbuv!#pk4x
zQ$Gp6SpkGmoGM62yYpgMqgU=md);L84r!rlQ{?_cb3NBm(an-#40yoWXkPqqvC9re
zi7Fnd4v^19SQm)m_h3PWoKge3z$>lq4IMJiG7jUs7am~67XZ@C-2q5r-Bex?Q`OSJ
zgBgTw%QBJ*kOCSf4RCFw?X{F+z6xbLQ)EtixSUlLt^*;c
zaW$uF|Ijr1F}V4{wmpUIofAg@7qa9M70zz#!F(FL;e&P46Or5Wgg!pxFsmyV@!Im9bwqqsJ*irfq!P~5{mW%y!dKT
zrS=l-k?5JgG{KR&&h!@Oe%bjB`?bdbC`Hp-rP!A9>?v#x%IKv&+H&K_
zxug0J%(A8bL2>}^4&Li7XT3WuOt45qt#0`%(BI#)1J
zsug%gWGz7gzn_KOA(L~$7fC-?W$s{}PcpCfdF*=s?dZRXy
zaZTi*aAb*Yc;5>QrNh};BjF@w1w{P}ffF@M{_tQ5fc7brV<
zIbp1QvrL0E4xLvbBXrEBA5@O1r;b~gF~Z6SAm}kX%vsltQVAit$icfO+Oomuv6!D&)uOu=Wj?5_yr>T32kV$;9?1=`BiT}B+L>6d(p7AVZI+>X=kQrlEi43Ds7ai^L?nu
z)i@Fw%@OC%?LAgwGG+uq@g~7%40O|yE6w{<15R+4VsgZa-z~2ie0ca$Z!
z6nJ;u=+YX_xoWA$I;wJH32~UhpPKOgkQ2k1jRSJq9Mi9zfu-OI?7sm-!B}0Fw8TCJ
zhLn>-{(V+VIS6jvywLF8|JQDoo0ryRve7T+ob190l)bX6Bt(f5HY+qre6BpDJXGT@
zTmyIJUhSmUE3M4bj#SzRmp%_OO9lkZk^0vOCjqrYL^KjPkpARY1Mj=q&Hc=w!gyU8
z!kYlsOyrsNaR(5Ojs3??DlwCq>9V_qA
z?SeMt+WPy7-RFb@z4m8)T_hM}EAP<`~~AS`IVQJCoa9n!gr71B`!J{xo6a9!G3l{{-{8m!r5|2O5iJk}
zJyF1tU8;x~y)!kgABz@h>9{`R#wK+43f5bE+M?voEv2sBTg0W$p{b_
z|2~BnJ4?0mU{LUMg1WN2eGY^w5Phbfyf8esE=HLZ4Z`Cz0hb~7Y5;)9;8ZziFcrQt
zDlDAsoJa}gc46i2m4UMQBDfofs5s|Ji@M4TVa5Y5gb?XGq^QFIsLU4|oOmWHm<2H${!Q~3eJ6l+C
zj3i{k23@V44R7ZyV+BBcg!YFc%Hkxoc_RoTeqzW@jBV{IfM_-NJj`gi;Y?0|_==mu
zd!u>N#btRD=fGd#KsqwVpgK4|)?9_eO~>t;ba73!eVR2>>Y_Od}}?=f%T
zuF#up?z$ccyCtrtMUR<>Zl?#7h32SDZpMy{S>0iK?dQH6=_R<
z&6*ts(ceP(=-IArbMTs_PKeFhO1@`O&CY5_SMf#pW(1hMdzPbb!5HFl%EHuC%!S)<
z3NyUau)C#2+D3K)&7+{W7>rr~pr`DO%CSk~9~4YzgFsc!2^>@;JUx9j0Glc3i9}j$V(nx)s_>V20}o1v>OVXHZjjuvI?h
zITnotFk}TailWfZu!68ecZEnze21CPThJ!ozY`;+p&#j{JML!^D@9|2T&kC-u%d*U
zQ3>B88IVn>GCgO_Vzfg8n|xb{mr6g5nqJ8~4LWrBnNadCp`8+3??#^sP8b{6|Dyo7|Tvpt#IdP}y-vin<O&pj4DJcWx
zpHJi@)7JFtk{~Y$<9=|
zDyDZWxay@PP5%L8Y#y~sFn!XFd-s-!-zB{`O#zb^P^+C@76fdv2bW)eJ1?T!{W
z7dv~0J}JlRX$Hn2c|;7X7I&1)^Gw
zY9=`g?kPK|?ExYZk*Y>iapqtSb$`Mi#j@Ix&eP5J){CQr}@$fTJRZlR;&Si%EFX7A3RmedSZ83?R+5+Vm(B%b!z;?zxVIl<0&{E0!&NNMtA<
z%);;sJXi5*J-Lf>8ABLb&ZN1$q5igM*tJbG
zaY@y>Dt=Ly?!`&#q6Pq)5=6W*0{?Bf_bXm!&@D@`9u|B=R#h8nyB`3oZKO@3xCwI84;*BigeSmj|Q(ef)Ot_@W;s|2xgIzVyvDG(6_lHqvGjj!S
z2)-4*MSvf7c3DE)-sHj?ZN9>~0r0ur)cq<{YF0Jvqh=HdZ29>Y=-!*zh6lBCxvm8I
z7kV2hr?(tPhHQWmrZ=WktknTatC)T!ox_4u1jQmgo2DvhXevTx07tyyl8lyCO1~TF
zi;O=vEG~jrk23Yd#yx#Nd_d}TxE(Z9wtf56@Fhr=R&U*iS!#n=XrxC3{*xwP_^@-b
zXk)6eV;n9<&u^;&m8r;SZ?5h1mk*DYj_(;MsAlej+}O+T01|FDOx&?)16qhGQc()Q
zui&BA&}HIlX59;aVr`7*&wmRp%jInN&Z=>22H1OZ>6?$$%;al!79pm
zJ#a5O2%MUbLPsthbFeg8FBG
zT^ZFXdBxvJ+D;r}q$jY5-S7%YR$Dbz3c9?M32su@%5aDi$I)LuP09tm%AS>{nS@;@
zik)`Va?qKL{jmU#Bv6H-Nt}Nrjz+kcKgY66XcHN2D!vaQdr*@{DP}Z2x$Z`
zzMU~og5BJSloF4F2x^iV0uH9e29J5|5f=8UE>HcQ(($m6K*<=Cxg|(H6BgR^R)i0e
zNk-B!I34V9)1YyIn#fRRFv;Lwsb1E&)4?iaVv+
z9di|kmXNP>;o71W#rF~x+lO?Fn9)WDZ6UMXQd9ZNVB$G|j
z6^Z)B3D;J-)wMBsDMDVgV3J&&kR(^8K!g=p8SqXGBsvC(IyCehX075L3ej&RIFV|;
zKhme93X0R1KVf_6rGEfe@^S#!IYEO8voLX4O|Nje7FrPdU@~&T&<2lprmW`YKCTc^
zcq>C536-MkNk6C~s5WXm>DTVjGh=9My+;pV1~4o?SqQZ)98$U(1{peBXSb*n$j+}V
zws=F%?@-m{#8bIXjb(CLWH$!Pz1Pd(5r@ifjGtIdDXWw79+#gk>dgm;w6gRq%x9xG
z<5r0C)4;d+q*RNBK0ylmkckx2P^Xoqy6%OvP^xEs=(q-PoWZqs{{aUzz46a(pR9W5
zh~HfGrmVn}OFQIBlZpP%^E6ofv?O`JRy72F1ZM}QLs4Er121TylyEy2KyP{ui4mbg
z`I2ipc9im3PW0+!eid?RbRmjt)U!1I|J-TGn8fzt1NFDVGDn~~
zlf6vj)(y|QdCIqlHW6fZ{3FU`+3jMeiZ_QI_bcneh%m$^OW3sM5#+jp@_g%nW6^yf
ztQO3;I9aMNzzc4D)|r28gr~9N5;y>QTB8wc4&&X-3q7jrOve
zlbcB7ZZWX23uE{W3#oJxUWP<;2;f9aj=2M7Q--pp{7*hq=gRFoDFr&8r)9cEl@mSC
zX=`PYdb=K*GkJbJL@rSaySEPJ-4(oDrifF^eo;S3@0^m*=q{RKi!Mj+L;(zSGZOci
zy!Q1o*}5-1;h+y1W9yb@MPrWT9fdoM4aP#kG!kZk7PH?}HKJL)uSJ~wu_z>amo{yE
z??L(@y;+E#CGFw566dOJG}lL{jtbq#uYid9EQ|@%y`v|*a@+O4&{lAVNr|tN*uoed
zVtr7V0?o_{E2W9>1gWUtJ8KM!*qc=@F*#48c8>{3D*X)F%@xsh+h;tJ76RvS**lG9
z{kAX(tVSfj#3;TCpph+8xPaQoIaN^0&>xQi=rYe0MO$Q^rDIpDg?zbBgFFd_vWU7#
zQaBm=6wLezglkws)Ro;DuS;9~9NEZ5ABWnd*n
zrO#ea^}@M|J7<|_GEH?^nAUcpz>`i*8wWB@(@fjf)3SBPg`n>)>g8i7T4hc}J%nqt
zxy0=9bY%9D!bPwdK$p>CGX(bPMZN?nZWrxzRX(j^3%6b>H4%t4nxCUO$wut=nOc=t
zzStjG?thb-ykB-t$p1x(NU2oAmayjcz!%qt!d%UhVxnBU>s~F>55qP~(;s)?d#)_{X1B6Rvr>VxyuJLSmAODEoLVqn>GRdRbrcnNB(9h#@ZBbE#m`xf*4*?a;
zXtpvp;$!oYrm16u9+yrmcffYIi1GrUBnrWMqD`Xe`6rdF9?&a;;&|ChZgppp@V}aF
z#G#K&quqknPWuest9oJot>*REQ(g~NWsX$U=t80ZUvsj@wBWhSL-+Q(j9=vnIgWnjz2U}Eg$md1hg%n}b@;npMu(=&aVlHTyTDUxb47n7
z>TWi*u%Didd%}GzTp2QEL!LPtAAEPpn@=~ZRGsZETe6dL-0tf*)a=vs)0_iQT{eAbmOcB20k*L`t(DBTtSMFC
zSxbD-By1rOtawx)ee0N`Wq2V|F|A^(gaggP>G_^Cw2fs#8u#xmOih2+RkW_S3vy@C
zHBsg&a@&*dY;i@AyTKCCZyO6DjHI*VNqSly55^u;QiL~zl%7}^seTaVwKR!#RGEsM
zz-eJoqQcNf(Vo;j(`qmsB6uGxz>kg8%{h3RX1Um0l7VcauGZVFv77wXq)Vt12n=
zmP3dEOiUQ^s&Il&1sf6C+Uc3XO-SS3D;QEPW2E!o6O!9B<0864z7&nbjMD8GBnqUy
z$V89Z=m1^$YQvUuA0YoB3DxVnnxG<4yuG8tCdBqmoq8?{2UMd-ORxiDV*EqF6y6jY
z)lCVG2`rfF;A$>En$61vy^=n1SpIXK2A4j5Daji?2J9AA?+Ue02#jdqQzAsWpDMkt
zE0^(F^?{(q)v1K2CO!N9GR~#Cxt=U4Ry%xMIb7Eb@RT#DB?s&f8qb@Oh_KgG=OaZ|
z6F)H|;3*q+T6QMm0rF0m_FPaTuG
zF{DnM`r;yN6h{Oecf!_mra1zWR^k&}5Ni*vI_i>Ydy~5s3%G68+PJ<{F6jBIsVcq`
z7LMgYQaTvin--S-KgQ#R%MA$<=I@}P*h!2Y(vMr83m%{k7bHtRtbjUJx&C|77(Z+u
z`6F724yd_0#zGK%MK#da3Nuyxf+|^K)!+Ck_#aKevio=GL?=tQdUx0?r^XT}9!!Tf
z9I#dDvvL5GvWf8{SPl2#Gtz7Z6Vj4lCwH=@#N>|D$buStNoG`p`d)jQBoBK
zk{@*_dV;ElLu4bSEQVNS1PlH$?=_S*8R=m2>GHSiWgJnKtoOyQB!fL^d4!<0qtpfS
zJAvmT%Ia4eUIufy3-w}J#VUl8k4j=KK-(-^M(ksXl>*{Jk2l9<+s!iSTc`G8xtf=X
zAP2rs3FhtQZNuMDrks6E?-_SJUZ5nzKtA;BXs&II3yF42QP(s7I$*@qh5`H$o#>gU
z)Q<$6#!n#X%N+8-h2VC0Y~l5o*gNhbR&uO(_++MKDz6kv>5!IMMd)n#C(I8Zl_+=j
zz>itHoYsJYDcfa+rd})m6as0>Qh*?b5$|5TiyX^iUob|~!^0SYi>G#_dI4@?j@#lX
zYG8iq5>ly!;ekJZX~%9r4cF5c%TydHWIzH}GiN?RyQsRXB|-oJV8%H1$&t{es*bne
zsp7yY6gfuE@ONV$li;WJ$*%m|v-M;EuHy$UGiAx}`uzCO2o;(lX}QNypN^ukQ@hSb
ziDTUmMmfSTIa+wGBW4qgYBGSyYf*r
zjLGRvI3#r|VB4M`V%e_ez}MHyySJb2!u|^IzaA2?{ie89%=8)IQ%I?#NqRm*(&+bi
z`^WSou2jGtsfFVv;OIdOKzs^hW}cIAc|I}RtvU7Zoc>=!3-OUelXCaQ!Q~qviS2arU4MXu5|~pZ9oWjJZc4g3r3UFHOI6p
zG-bKGA7XY`EZpM+ndBOV0H>#Tom9gT?xqc0FQr{tI4+tvrb2l??un~(8jP+@X2}u9
zNnb5f1OMGLJFvkMrpgOz*|E!iNd&7{`d2m}7>=ff7W
zf-(Y4-THrli5LjV04%d0e-nssj6Zn3!_h@?yNu>TY>56|(_L}7C2{=OZ|@;RU3Sdh
zSBJR0^gV%9EPOr>_@;%@o&{x9=a4=nwC(Z1#d=r?JI|_I<}KRMMF9gK^g`a)&6w4<
z0P;o~8r+p!6PS?2t~dKHXEGkNe3Ug&fM@}+gLG|2+)e+S*kze^`v|nR
zP^T*~$ClTM@JLa7L&o>mYHusNYsr{*E<0N!-2$Cwk#n1F;+}8UHIWEKWEZPtcHa}^
z$#Imr=UwpA30!qm-UFt2R)_rVfeaOjtt%{9uaK$4FX!Bb8}J61a@LA9%%k*Y4bKfZ
zMAP%5$(N9$qj|N3j_A|*9I1*+nbTmsz9lLlWhOr3qkdDcP0&Pv&}>wNhgkMWn^J0;
zwWplo5r>+sKhOMWvj-)8sknuvblh{2?=uKy4rr+`By7uGseD-9HmX(+P
z0%sP{yDRK)&N*(KuD6pRD9b(pS(*D1-#b`U0u;M@Q;IA^M{Y#_qw0UxAO}4L0<|~0
z#6DW{#EJ_P)wPHU|I%%6%)Q?|5KT>
z{_sea>}o7P>yE>wH4cRMf->^lid|YDcJSxL=sd&i>vfJR3Ki$FPfz&lp#jSGKyeIO
z?QCJy>%e0>k|?{WyR_=PWDbM<5ps`)4LkG@_}H;#xU3++7$``p9K5xigE~Zkn}9=C
zH;dObuCO3&j&U!77BmI)U>RHr<)#{B+{G>C^d)VU)RuvBRF^ojxcYvlO^$cvyQnU@
z_ocJENl?oRbS*=r7|h|>-A;%MAMcaM9sTywLSQXGH*YYxj!z~=r`!X3B@V=d-F;hw
z-$-ZP@p@83!Q{nHyYt*tdF*qihn_`TU#v;;TgCEMcJ`qG7+EpGfVy?3BJk?9xf&9m
z#n{dzC#PO5f?MDwdeGu@R{KvVGD1Rz?PRxW?O|AcobVC7f|FQD&eD;*Ekz^v*U^$PI0%SE6dx2KqlZ^xHhU4d#ypepND6Vc1AxL;Ft~1BrJRr
zoeCFA)Wt4{oHC0mK2!Aq^sX(Nbs0!@UiB+XJZgK4WG(1^i8%+ELaTbnf`x++uIjApZW2H;oQ#bt}PPrRWmJ
ztE2VoP7YKTpywoQ#a`9G4Z*62(EtuG#gJArqNxxm_y)9vk$USqU>UWs2#?0Tv`5=-BS|xtx@;^<$7LJ
z<0^j}3FUHF!xj;?^Sum4NtkoMiu3U;@{B$(0+qr6kZZ9Tv6+8JB=i%Ws@zyix!Q#c
zXTP+Tl%(`_cM5kE+ouuedlFKwEX_sB9mHM2r3ae_K(~bwOZ@jrxmN0@s&OWe*}>)e
z)5aB)jTc&StKd!hzqT2Ds*PJ-$4xz!*3_sxuirU_W>cejuIF3WjitjR`?zm7Nq
zNDWYLP}y(-j1zI#WLsgCg{cqerSrLCnjfD@WH~%4?uR8uv6Mw^DYjkm#^1k$ZBmLC
zOFtA+!h6niQjlkB$=o$VN;f+3$qI65RIND(`>%j*Whq7yMTiDk8023MSA>3e1P^so
zP{`wvap<1ano?kl{m(rf3kSPD)a|#K;`CN1@17r^dRh|_fP;|PoArlxyl`$0O`{P|s~LiJjpmRgG#UD1>@k~i7UQpOqn_?yG{cQ4c0SFCRVGgK
z59r=tQL30(d}vBsW4*FBQ&&?j-TxKe>9!H
zAR?|OR`V}d^!ZxET%&Gyqx~=U+cSmMYT==_;l
zA?HnH?ep9iQf`?R<>3n9dy_}!BUXS^M!U#6OxWs$o*qMj8b)Vgsb>F5sUM`x6!L*v
zz{#wFiN|)nlO$pSc^4cVmVT}Qxj^BIB}sjk$CGGlUuYEwaF|vdS&qV+8!Z*+b_n;N
zpXpo@Ho|37_cK;r`O#+}dWa47vIlGbb(3nK;YmHd*CUnWM|Kz2x38Wb^B!6c;Mn&S
z&N{--`(&*Lo9o(x1g*#kf?`Tm`SRcIs~#ApTaQW+kh<)anAk6q)BU~8=erTCCw;j#
zfd?Q+ZDC4_vd<$#g7+SjR&mv@O=JHK&F6fTkN_&DBkZzq7Rz>Yn$iDP((b?d`+KfJ
zE`{$LJMM-@Y0x)2Witr`SmN{YMjb*4gc&R7PDm;FG544;xqY^{HisU)^qEmsfFnT_
z-SYQCTUi>`QtJj`$0wlqP!m8`7<%cw>A6U$Zv`bLNiI5zF>z*yeP8%<(M=1^COn+s
zFnVhiI-R6!jIO2`oJLl=D&|~!QvOGev|~Oct^JclGiLQEoBhpYxQx}mEoH^HtnkWi
z8Hwg+^bD=prVsNVF)T~XCC>ttFwXU3029-Uix9!n1wEdMu@H}*Ev*1FBU5z`qsKz~
zhRpDd5cOlMHR75~G8Pz8X5J%q{6)(tq*D?J5;4J$Fo@@_0z=LycW*=%N1Yl?3lWt_
zP^CUe^TT(p{B9Cae;3-&GO7joBD$t}q7YxrO0}$*?4xqv{Cgz-(tWh-XRPlGoko@)
z4F#QJKpoOl!;r`~(w?np)`l}NUI7!bJfO!|3z$2-vjZSYe{KLhK*GPb(;^F44OO%h
zL6MXNIV94%tixNPGef$jW}`sENAW;jG&mP}5)J5tNUMgaJr*)akzlUvO541<)TzuD
zQJ9D88)P$#vQC#|po&|bt)%py#hF^bt=X04rAA)o-Vdj;skT#|E6zy!b2vGE=GTVf
z{Et{SuWd?$JlpPuV)tD@uF}iD8gZVw+f%eL9t%akXMDclSY5#JZIU$DCEC#*H29k;=q{!^3Fp0qbCYuRNksHZ21QJ
zzYky9ohDiD+lVP$c*}LOx(f@0TeLSkq+8#jgZaSS+7mdDRC9BN{;LCGWnKJs1j*sCzCefFd>u6e8*;rTjd^`x&4b5gcoxw5Uz`VXT%qs0Au_!fB&_yNV$YZLR2yc~d6B`G0bQm)b!Yq&qu
zRWb{Sz~zR$t2o@8ps@=nyH)|Sa*bP!1HtvWtL;Y^GFoFqS&|}{Z)0pJDvy9MWAU2h
z(gaicjS69dIwL&zJ;-PEaVb-}ev+F;JR$-9fZ^3HpRIHJ$N;vjQKMMY3liPHf&X<2
zmH3VWTptgY#ty$0*HwutRX~PQLZ^gCJYmk$L8-+HaJ!hoP1T$TR5U9yx`dvV%@iBZ
z0;K)GzggfE3F+9q3=;%t7V!GK!0_winw!W+pk;0C%gagN!YYX}AmdczZ8$R}b;?G{
zO|Q(UT+T8e3P|xQVtPEuYwXDI*dH{N$Mgv%xGXz{8OlT_*XnN+YqngY%4N?ko_4JH
zX&y)G4Iu1v1N2_T0wZtP5e6-_eTCjZmJKflsaTzkxjO
z^+*}YQub8^Bg_>6g=AA`{ADSae}ghXyD5j}P~#a+_CQiB?5yuALQYzJD)ICQVnDFJ
zaszAV+{O00DqQ2xtUiRakjZ8m*nlf>PY5o)t~-L9BzM2uupd?-;8n%#u;!4pW)O#c
z9QcMr&5nVo2tu2t68w}7op&dEYOV+#S($PCAr~**d$33Jca(h=-{tFb-pD4BCVYr3
z-KJuRQSI?`UI)ijRBbN}b@KMG3=<=^67Os;f}JOdj*Pr1>Eti9c~g%h7Dhmy-g`a2
zsFFN(FL1swBih`fhJ(|CYhW)~D?OE%>@0k!?s0C5F-`*c8
z5Q7t{QVidM4`>Z&0%{N2#4|~NU9m^A;O)MMEFz*!*g3%>uvB<>07g&O(`{+3}-9S;d^)Io+O$7~8(@X=384`AgokMp)Bu@-{YrX0u
z+BN=6cUkQeCwriul9P4{d;Q@R=wy3_!6_uwsm-a$`25})x6DcJ+eO>iYzI2X_^&p{
zx2xaQ!c8p_hwu0u>FHsvN3f=$N=PH*Kq#EmzjYw#sU?#?;9k@*y#~2>gLs6_5sERg
zg?pFp{-JV7Sy_F=f&011HWH*^mvQfE?(=mj$`~
z(q}g6OHJK~a8yO)PzOLbVA)P`bn=?$WsCIGqYYd*H=&Z|vxTo|yO^jaakd$UVH3}u
zSxI5w4U~Xvn7HiL3mBcI3f7T|ymZ^U;%)woWLeIC!`)LPpzvf+4%y|<4)4jN6Lg)d
z$+{;RR;Qb)1=d^YS&?|(>=VmYSS)G-)<4zY3QLD39$RHaZTfMLL4Da$r7DWJ6+UmS$A4_R(Gbm>r~kc5ztF7BYUEampHqU1%xFss4R~@up^^a*)QAk
z>s&VujyvJ<)xviz4zy!#VYyq+I9R~SC-KD|tBW|5oF`63s&7SDMr^?+W8CZHOL~P)
zy8-^hUxNVDg%Z@cl9YAc@4>nqtcd^$JKFh9Kt*Xzjx
zhwK>mUOqyPfguLohBMyM&3=@)gb&@E0ZnAd`ZXw-Z?}Xp6OM0nvw(Hn6WXIUGp|U?
zsDsa=bM@Y$Fzo@rV^?6zZ>hwFBeK6cwr~IF$Oh8jiE~l27P(+tPI`rDkIgH>^uz(C
zA5bP&I$i5;n{yVgS`ELbI1IepsTeZ0A#2EVC?ulf6z&7JJE@q#JYKZP`DWGHE6^^H
zVh@tC_wz{$uA*n<)yWMRj(MyI{YFvUltp!8c8r0B9p+(}iX-l*^A2o&{6a
zG?i)e4U|IZ)iOZ%O;>D$f9SmUdhExMxNwUFqVnaGf%q=Yh&A?`8)9YPjwhUQW{}zS
z0<3CzR@Vc$a{mqi_|o5jZLQ9%?m%Rj8P7X!LCXHK+{^SqWn+IdIjI4qC4(uGOm5sA
z+mTK;Yl>ut67!H9`?m5yD59%g^7NOK!vN=yNb1KXtaAytumsjNX%kfEjIn&a6Wg$K
zyA-H+(8L=21BpLiPb)F
z@z%cfv{p$K0`$xyq<;W;W?O6$t4LQJf66Qs*5-=m?PxBW<%{xU|LH#GPns(N{(&Ul
zS@RJx@(-}y^tJvCy^K5bcB8f$8-n^}?7-VnHJbI+IHt=W4Zu2kNjjGZ`_L=84OuYk
zS^NtojN&$bb=j?$AaY+eUVkn>Ai7wc1P6!}^Eby`xJ$cfw&+)oY#nx|a#{KJ?MurA
z4FvZORohO)Q;U!F(%o9+8}i{*Q;e-DRbkP$5l#9IXrPw?IsRB3{xb
zf&z&X=?s~
z0$fQhd*KWLoQbG0J1J&z{5ceYI=j9gy$xJ(j}a`|EpQNhm{TBShgU+^GJK8u9%%{e!P`c8;3aTeD|70bb(cbC#kEUiNz_|
zT+II?sX
zZtjzZNvT9SnM!fU_)TH^VXv|rBp4m96kk1?ko3`w&awf#Xl>CNctNBa^|Ja5qKt+Zy
z6<^iXgqBG8+)Oxl?2SOGPj`p>qf5R>N~EJ=Lw_{c3cC~5
zGu$D@Phhbp)F#EOxQ+@Q)YQajz_%rI&=cIn@_im+g~_A;a>NXK#fU388G8&WLHf|W
zy`7OD%Ht1kL9%De#)e9(ElcQt0{4+GWB9Fz7DldoDEuZd5{L1~yN|SlZ?Ssy8_Z+a
z*~KK-YmZ1!`#sDUsJV1n=d`y$r9V73xlr}sOSY|7-h*yKX;KGN;nqLVUcDgMOMVn$
zQ%Gnkb?z`mpT@nrp;->lyjTks>J9ODGc(UNv{f>?tFLE
zGtLQ#--Mj_j-A}mYw4D1gaPmvL-Z1q2&fAxZri)`6q{c_>&VCHwq`jRyi)3t#tXWm
zgPN}%rUFWCblbUpX4BjMW0Dp79YWO1MWd!X3XaA};G?0nCQ+d}GIunihwS007*e)Y
z3;dA}vjbGv4Orq|s`W_Fl?g64Jgs(jQY%93d#lGt>+ZcdFGb=eS7W3g?wMB=>zO;&
z1u(Gt-wTY1RQnD?&)$ed%!qg>@|Jn3VGg~?Ln3K~KWq*YM2gLD%__sOHi2jcog)18
z@zA89hz^&!UcNo8)ZSQ1(A@yQEW7FNA=lu7ee3y_}a#()o~WOni#9J{|m#u2&t(S8y#L
zvTkiEDi7VG+A>r>pbYHgMLy!yp;h#^z7Ha8f6Mr6>6uSt)4UMiR!-O%I+?ShD#owH
zG0oYK0cVq;#%P8Sho&?8MY)Oh5su2aHC?K%d=4xO4o_@;Fn*DeS|Gc2Mg%i7tm$Lq
z{GVc8!yNO2dx1I6cIQZzVW6GTE?U2w?o_PjLg^ZrNieFJ?&S`?W-X9&GJG-u+R?gB
zm|2@Y%~U?r-L}}29)NNwEcMYlj+#lg@*^m12EjL87ts)FpUh%X0RQ^9L)|AKCn>wB
z1Tc|B9epfScWt|}qnw^7ve!H@bFuAvf$-j0>9qh>?E@CLq1q^J?}1bWY80H!I6vl+
zy=&7q>9*)wnohIWs)t=y(yQli#st~<&Q`+JHg329_@Ju{0YE$DM4Hj8bk?+G?cAFJ
zH)FBg3FVJq;fvYeWgr7$Y&~d6DT}+3b(}K%?A&mm7Zi`Ic;iY^yi2@)lmRg>
zJ~B8wW)@9lpA7XZeyU|iqvclY9oG?-oC^#1O_3IjtrZxxTO
z>Ma_Xu~0)^BK*4O`zCLXUtA@M)Og7F&IL
zxD2>8Rcq_r?ux)`MC)Icj7^zvcu}EFIO(?Ixm8S2i!?CwV1lUg#drNO0hM(+Kq@F_
z>#aQF=~SN$6n?F&2;W^-hRC1bc%NHTV?+P$MO!>MT|Hd5s?Iql7sW$wy#Q4?uPUzO
z*#6e!2hz<`jOBg^$a?f0M;R3p5TEo;=pK&4mSex8Ij~Xga60x8N)3b*
zTBH566$Q~!gU6!5Mf1{B21N5jPwU@*FnDN0a4y%=obOjoXOUz=SUn=CXz!%*tCSIc
zh4aKQ1$kd+mSmHLVPQ;=`MjMv+@m`S>mjoDzHn3p#$^9-q%wF8dGTzOa@+>ADno$>
z_clW_-hjR17DOzEFqU471N|@BO(xvmvpo6ms>CHS=zRucL&~hIIMX>pLkTm3@`uC(Lnmlj=+m
zDx5|7cvzh`DA(#5;X+Tge`3u76U
z;d3zN1sZ8bZhO5aZjogz41_3nxK8sXM6)I@*rmSSm(;lKE8QLT+0?f+vrl4z%>lQ{
zW8#%un_Uu^TtfK1eTL07*<3J6Bv#@P@by`#jH%$o;me$gZ0Ev3DDKj0VlyGDa|p1b#dXc2Z7o`ZGyKmlN1ZdT37`CD*z8ruN~_!y67Jntyf}(u?xVCeMRxIfP
zo-z31`6PG0+u4Pqn`>cQN(H8e>(yW9lS>6azwxEBpj{w!J2>2pSHUq{KU55nA1a+L
z6@3Y583y)1JGY3cJhkt2SuQNXp2_Gr|Eo{yVA}mSK;~D75-&<=G0yfbEO50mdy%4(
zukCt5CO^AdnlCwd2?z8JLKVduMofC9q=P-9{!Lu@iJFx>ILW8r5JVC*`7Xh4Ly8+lmE@8i|3C
zNCcS4`i8jCW@Po2cpQr|2}Nirzv1l6IBr~s3*3}SCi*)NjkCLv3e|<0mN|TFM8Mm`
z-YR=)*Bn&%TCBaZEC}*ou1PmJ*SPIz8OK~QeX)Lub>46%KX&X{ZYRgtd8K#5`fCt}
z(3^CEr+y6!hI|f~y{rsPrCBwP_x#O-qgewy>GX)U$?nH1-T;zozU^#ur))H(1bgsc
zFH$q&2p{zlYM$I^GgWEBs9C4QlyX^<1)N;OR=H45q
zrvk+
zOTp+ZUiPsXnbX1|3~n>^+@<=R>pvvK&P`t)L^ie6OP!OV%RICi^5pYO=Yj8#BM1x7
z;{<>*@~ZH>=I9m|CRR#!V1one7$Gl!Cz7369FeWTz{&5>UQYK1mNJtO%@tPXcY%E=
zVNv$vCym!FwoZ+=Bq^3IeoRADHYGhm*u0?CVipaOrH$-XIuOHW&QVo1P+asxr~z~}
z{%58@97gbG-_V@OOO4{3fsSosL4XR{_X*@~GKw`mSuAQ36X>Z9JnXwwk)DsV{7;!4zSVyLBnG-?yACfke6uqQ=|Poap*!G
z_k=F89n%o=1QO-XAIj1dG~gl_+B}91+NCtaaJ*ZLcu*RbXhxs+kEvQ<#r`K~yV5cr
zI~N6xig^_~7Lf;!S6D<~y-_#GnJ$dnI%Nd&MI_
za(%chSl(go=%XcZzn7l&dvXL(@z@=D|Fn7(E-Loww@I=tIOtfWjjP(rE()ug7)m=*
zoXwPgijbFgpBbJuFIB)~
zX!^n2reZJM9Hb(hDa%})vJU%HTu3(2zYRMj+BC4Y=n*Ta!Z_xaF&V25)>+Por1Lh7
z+k-yY@I~53w)3mEeW&J;8u_9GikyaCTCzZ0#HJ>yFE9c|Az1|5-j!YS5p)VcFfcrh
zrWZh!``CLY`%se_6+%Fi%3SUY_Let(BeP12#<&6dTN>#Yj9{r!eobYV||IR80?$$eD
zvM`m{Kcj>?A73Q6hEUTvpo5#p*a@#|v$>I+7sWsLO!Hs@Gj?wO9Wf0OPhfwx7SuPw
zK+g%6O-!g1I9z6`jDHLLyi_AvcY$9E?(Jp~m9?+DMW|`x{LSqroqFvFL1(V4$$F4obSq)$0m%|S<
z#4zL=@q3uaB~We80fL((RB?|(OI77~cq$YpHm!X=7WneaUaOyWP5bF#1_m1G-6_m_
zS9cU?^3}PJdeNR2?ANteo|gAB&gj7Gkt|4#(nHM$j8ND1RVn-{26p4s%mG+382$^H
z_y~`5QfU37E!IM@LL|~2xK1K4YO7U!%(jC7F?{*L@s^S{qp5$m^4PTfm5gmqYqwt3
zOu>H6b>%(Cv|l@#abip0IibsAU_bIUk66ba0{{B#+XxQu{Y&oEZCN}?5{TlQcUR_z
zmQ=QGo1fGNXU|5hW~Q@c!m`Y{BX`l*Vucf-&Y*{7_%MEpt*lRZFf9ej3objDu%JkG
zaH&U#hS1&qEq#avvL=GJob&yo?bS#hcF9Z1RC<4a%xWIY)NWs1h#)$45QM?50L|bq
zz(5PdA5c9kdpay~u^?6rXeb_=4`8W>LmEStk$9PFliv!<`{44cDX7;H+Yfm;Gx8sW
zU6;3}DP=p+5~xrGPTGTsj9Kt)0N$a$v4L9pA)%;@B{+O6No$-e$=$3f0k>Pz>hj8(EO-KfDokM!L9x
zFY5o%z8leFK$b^qIq!Mjn)wFBBM4CjTW35!^D`}_l->t$Y@|c}8H!3zQpFFrjKX*p
z*NnL}eiDo{jlp6n_qehMFLicSWGT-r04NoYxF
zHo`JT+>r!4!_E0aud3=Z$E1Qx$}XwprU|KKm43MV5c17r;QDgh>y^7C^iR26Vt{OR
zrYQs%b}OA(mg}cvVlsiA8&2q?|JzS>U7fUk0NwC+VLva{rccjpdQi^sgHVrDy`k$#%Gpuryy36)HSHe
z)~&pFdz?_yA(u?y>SRgYx+f|`ZD@j&K&m~w+LPHq)QwA$+#UC4`4KJFpSGpG9y#%E
z7(^1T!D6Sapa!YWh3T>;VtD_%Bowi#heOg|c1?#!i$q
zp`@;uc1%q3)Dbw0CDDj;EfCLf)fDgTlvz7`zMYUHhyU-a4OoS=HBtgNAu$l>_-!T62noLsO#gtkEev&p&XibEYjt_3)3@
z%9rS^k!z98ML-B(svIe4Cz
zp*6ac+OFut4G+e0w~!0HtrQEX9U5}n8BVY*>1USBrE(mZTGQg5dY(5q^f7A?0$(ob
zSsE<4NN~-kl5P{f_bT>u%g7bppCOAIozwANt5s0>qh5XuZVS7?Ic}>b{%$6^0fb`@
zW1olm%^-Y%AuJ55h#B@ti6&Dbl+9FD+_!C=Sq5BJ8M0T4BHx~Yj1XvDN>wk;>WQUr
zsXx5pRupLn^hkb$U7!l?E+?j>bH;9sROF)%7AxSklu>J@Wy}UatDS1^H&D(Ni@7oF
z7f=u3kKL@^F!t+t8RO<<>0%G$tzmMGTdzTSvh1nb`shAO~h
z{U4WUKWUMLZ{%)R;T{1WBPLLr=9GZu{P;OP`Qr#zTpksYGx|p==6mjLfAPTxt+NC{}d0ExnW`xxn_Dt^c`AKbx`7wOIQsaEOI%;%{eX`;s*xqI=A^~by+*i#(=CTk?ShF
zT6eK16Ru~P1F?iY6;vYX2IG}*-ZI_efHHf+Jn
z$(>Cinzt-k_Kf%lk^k3$AS0ILNSBQ=6N3yZBUVm@zYm5#TaT_ll#vwWmhpw)a~lX*
zSpk@tTn;@a$|3J3z_g=5qj$NS#6L?
zVjk(f8`(*V$tNxgFjJH1a`{ExR8auCgloNR?-P#xDlzE~%m0N}(tGZi{pFYH=V=Wd
z@-GP(@Su}@VSXH{s~l0!V$zs9o2UwjGR^`a+~YkuT8MClr?LnMq`(5^9>N8#du5!(
z0bNnMP=cv!D{a27MDL!=uUCz;;$+-k3B}a{Py=PU9~y|X!nIdY7x*66+nh8N9tc)~
zL26iJ`B54g7RX^pse?pt?G)2g*0J!)Vg$hJa0n8xY{PM&MxU9{WvRZi=jl(;Yv!02
zj)0m2C0SuL3iOo1U1rg_fQZizV#>>T#y^|#J@do@;)a<)sr>ZGz0(5TeC=<77}hvo
zl>Smhr~s|k*;XNa6^wZ0M43!+LX@E;hEUSn)rVsQ>EsYTq6N!o@>Uai@fDL^VKCw}
z&RKAHdzK~3t^Q3m%|a0k_DpwSe}UU!ELawUYj~9BCGrh~`D)I$nkqtKCNMLfqp&@@
zIVi#?pwsMOv630cu=n+jtfE_e3HXzUDqx=~lSf%)f8&=m2?uAqiu$5Svjcn}wvhlr
z#GADKm}Cx21o*vjFzfi
zB_J*y3Pj~7@=90B6i$q|+yGcCjcAub*wO
zBiz&%vzfI@e-Rhs!G_)2oPLVk)d3{nvy6$W5(W)+ehZVst_Ks!FBv&1?D+BP&tS*9
z)&4=-wa=juVRy$ALZ{Zd7_EJZ@ZE5h{30^~GL=}ngrSBBPQ=*+<-LpvB4btQeH&(u!Wt*wRVftWaYO(-{~7GS
z|Gb*Jw|=@*PH8a1q~6+!EjeKT3Wz5qow6$#m}g#V1Y6%SrL2ln;hJ7CxMd%ETZjtU
zK#-sIfSSW=axf3Au)ICOS5>uN39~*YClayBrV6MN(`YUPh_6@o(ALa6fIx}}($QSi
zZ)f4d3LUMclDt&2z#;+=M7TWhqYqM0maHNXcep=-^w+FMHsD@r2cmw_>Gwh;MCmZI
zvlUW>rzIVpT=9Au7E(dtw9qT>dLr>7Vn50lEA3HoSbG6$!O1WcN=Es=_6N*fdD&kR
z+8$cNf^(D1OtR%UqYfCDw;-N;eN3o&v&Z|SH{p|Na7fza;=?Pv!`V_7@|tzHB5w(N6yW(lc
zH01oC5XTGiwVG89A7Ti$h*p}KKrag5b=sW|q{`xV@
z%Kh*?+ItvJYC0zW2)~;m_2IAFRn8r~3AWF)D1U2^qc<~9Z}<3ubISln*m0@_5S@A5
zU?2_>dL8bEngjM42hwGf+2-`;tkCF+9c}GphI`O9VeScE|5a%w4l2Fk=LIF45PGdd
zm-P5OQLl=nh$xJJizH3OM|2^8ge^(%o~ubI@UXG`nU2@XO7U>tX6rWazxuX$FUBVi
zkk5x&Z+@c7p&RmGP8~-DS!96eHD~X+ByV`2ZMVoQZ)|(Y-!Lst8C5aQ#x5GXuYy^z
zo}CPa!3Tv9k(yhBrnilO5S%qrjMow)p(T8=(E?kSoK%Jpp1EaU2U%U1xP@w0tyWlT
znKUW3Bg%dS!MU-(x8?Nb>$?4E=KHj_zp+^KV}49u=UjHq%Ot2%ep?MtXcKfW;=v(w
zn>+nS^uqKw0@}CXP%~ry+D-jPFdVZmwE*@lo*JOKpt8f;3-M$)e~@>?D8toCs~^T3
z1ZYe#ZW6yz*(I?hsBTu*_W1W#wafsgHd-zvPONGBNL_j#Kw&(@=1k0|EoBjc;_e!=r5_3rQNalk
z_!0DpfC)Cd*`zZEy{gN>-D4CGxr#kN(>?@GMaH6)Xfyx`T!Ufk4>BLD=OK)XoV;HP
z#tlewO=6o%raeBD_7qS_8%bn^_(f4Moe53>tOQDjfFI3uqIF6s)7UNGms{5d$8PHkLN>#A=G6qk{llJJt-xXyf{SCqtV@vzN09UqOADzP{&jwJ=+RMn$SNH_h^b3(`buxnqU21+ZG
z5y&ZEy|5P^dlvGBx3&?q0ca`Y73eofRT{WdveH*S9Pa0Fu1_Lnn&c1Uqn_nPg)lBP
z94>=-2VJrV=_@kvhoXVM_(5#p;jl;n2*Hi(uHXyQVa^LMu-aG7N@(jWiu(PdYI3tfRgREkIF+>7M@aHeg?6d
z^~@HJ*ssi1>*;3_Ou+iWcN=&FcYjt75mt=wTGt=%L({w$kQ!0yJ1_o^PwoAxdEu6Y
zeFQqPo67FJG9TE8-pe9^t>A}OXa(l)chts)eMZg|BeZtP-;Q4RgAi7B-qcU)3MVuS
z;t&6TaK3|(lT+j|v+X0QuqFd!ZKU4~L0fJmbK>4MMQ+J@M~9a>T*ytk^`YSxT@EJPA~J02}ln6-JS$s7IVS}fIaV#5z`=EOr5ybfw)n@Vf0q``hiUPw!e)-su_
zv-ujTnc4v@)7zr0XPXlfrnuy_i}Mp4=TTn=Og*OGUgb>2dz`(5P?<3{DP)fBnn;0e
zr_Tv0ll_L!F6}9(({VRtSl0lQJy8@(`$`>*y-*sh{nQlrKnr?I#YruQ->Lr~ww;*0
zE0EyT>Lci!aTLF$9WC_Q`30&kh?pVKqjPrRW>w17<5s~hh16_qi}Bu5W;OX{=uv<=
z@PKUtcrT680wgdqA2bp`5n`btQIBkL)$vy=`ESPfb4ksP2Liece3Uz>Zs`L@&KU&t
zG^IIDNlf}Y6Sk!z5l*b`HyD#kwr0ue8Lh>_dCwwzXU(zmtbYFCMyx=@G
zwWQ2kt)2+nS=!O0fMaJd$7{lB!kXwnK9H+zcfR#^G<>8Ys{!sVhJ+j*GNK?ZiWZ*o
z?6L7;zx5Rb-|e5xv#%2bUBl_#KHt>ei+97(qwzjx_`j~fa#o)yRq7!bfMeAmuDSfj
z788pt5l>^~y68m&J?0h3Ci$zi^@i^M{h?byd;_-VZ8IIvcLO98kzojN(|2KhvM_aC
z>k!53
zkFnm=2xkx?s};ho^jeXjn2Bv8O8g}5q3GE$OWI4Q41%hM=l~)C+)wbPG(7v1YmB(t
zCU}j;s9jTMMY)CK;QQ@`uRB;74g^Oanv5>j#ZqIm0XMRiHBgp#F#E@S-<+s)R7H~&
zjGF?=qK7L+AyV&wA|hE`wszq_WRE=HuMLf@VSS84sDGcqQbH+~pjJ63_;)2UBZkTa
zszqZ-uFCu1m_oX4zsTLi^N0z%pl%06A`vOK`wUe&W1D06sS?a@;tu5)g?G{Lzn_x;
zVmpqj4xR8}$&=0Vz%g3A2M85h$uX49cvQZZksT#>_)h*cio97
zxSUAHQS&SDBP4#36R4`L_(l8*K%j@matK(__GfEo0~lFoUwxqfzSa&`wJU1VrmuCf
z?OE`{M?tF&(jkQnmNlqMmcjw0aNRNWwp{xqC7nu(a{}|H8s*@iuKelC*D>ga(R7;<
zh+N;4qTOQeJVf99Bx(_6TkD$QIJ~7&>j4K&o>EQ0-!*JP-A2d(o(Ld6%+MpZ33|8F
zBJA!g)|mgqee$pIAeLfNSN4u}6%3udhZ-UgM^qrEg_y-qrEXr9E^PNEiC|Y`VcdZ+
zQ(NN3fY6H~DFMF1X1B46hR3**tEl9T>jJ|vj9>q86xv~hcJucENzJ@YKpICSst2pZ
zCBn|Mmpa}=M50~*(*EZB%!F%t0{D|*ri);XCd=f_=YFtmeAmKaZJmb+@i?&v)x`)F
z2+8Y(?ghWLeGo4o%NUDGxVB4h(lq*OKm4wN%xKLG#(CTqM%W$?EYGYQW{Y1HsPPVh
z%^r$@wJ}usu%B4wVE~AM#tb>*D@1L%ZfFPY%}@oEt5e|J!JzAatc0ie!O8(`ENGNc
z+ZTjxSC$}_iE$Z4yFA<9m&V3U);Iu7ks}1Pq!A)YxBFqGySv6!jU8CeBv_ziu34;~
z|6v|_%72TM5tl`3I{K_GpDJMpnsI&I`RPNr`=HIw_V!JoxB5A66xy(+?vxri;K?y*`z%Gh&Rh1f{1guQ__jA6;W-K`IZy#kTsCAA|J
zQ9s$_es;t)akXCl|AykIIoM@zKL*RW=rx6-^$J+}EV*AHtK7b*Rmw|T--?QY`h^~c
zzZsV=q>+^`Y!*C*xUad{u9b}Pp9`Cf^)HB>h8@!NEPBvW#sB@KLL;42UpzvF=NW|2
z%D@%{>E7JV7dLmlPFlVkeS@pn=UO5Gy&8Nh%vt5zelF+Xfz3P=ZCbl_62mU)+(=aN
zqaGIqXf^#hf3V&Cw6@|~8&$It<(mJ%V!H
zdC-Ui<`C1vOmkW!9rG^*{xlfUmVQG2M`2N84+>@=_&`*VurhS!s{(jfP^+jwc!krY
zHBaqT!SB5oAAZIl9%-)&tm<}=cBa%!hmLY-r3CoJ?R*2YSw)RU`n%2_s!HWkoVm8!fzCB+`g#p+*?Ky(4a2A42@Wctws^Yu?N=
zRZ0z00>hzde){PgSvmPfNVgjbD=5G>GycpUwpQJ{_@X6j@(10prQPcr;k
zxb~w9To@{o?*pQjl#M^w>_Ow1UQpf8*YVxo%ccFN1nYa_WZ#uAbA;M}@h@Y_pZ`t&
zxNeXeN&{dsiu=815A#OcQpE;`vv||Rs!vw{X(WQOU#Y^I?X7zAKF5GWqAR_}6rI||
zRd{ukWz3zu7q_iZSDvTH-cDx8XVDzj@gBZ?suzir78~_I4m*MEkmcjCav%tI*4s;~Xj=L?sYyn#k5kF?g6Q+@O*?X>
z;<4(apW82$3R8p${Zk7QrnldU+{g<{z=EhDoW2}^Q}P&o0-kyVt?u6Rr5aroM#1f=
z3uL7K8`VeHr!yFGrp8B1JLtFyx&p%;jsNRQex!pogd+YU5o!@Xyc-LbuH(2=%{EfK
zvczPOF;J+SyxwQ!nXpzyKpNbabeo4+k8r;lu<`pC!izmq
z`V{lu2!PBKP57t4e
zvd&Jsnjw|>N4io!a7&``J_wEw;mkE;L7#IJ?d0Y{
z-dxo3_epcz=0fBUn%vSAe}gRqsy>Dm-r7eHdZ*Di^rrC69T^t77UVf$PM$ncg*^${
zY%-taF;Uv7vUQmwXUL@;lXYq9PT5Yt;d#yn^Ep$@H@RU1;i0oUf?d^bzJfNVH`gW+-cio{^@gjsch_1Qx!lXf%dGV~8y|%QCwcz%EFRVI3)l!&JH$T27`_
zrR?!a@H{i)=_1vN%6iWI1LtQC)-(kaZ5FD4@wWBQrG-hCxFvqzadw}s>G_b;+MZsn
zoz)V-SPx$~pNU{`HBdG|ab0Kk%l%U<2kblz$Pm1_cYbQ2aHN;rFUJT&c>gQG=Xg-4
zN2ou
z>sX9Cv_zBvk%*mrG3iNBARhjViCh=hU!TTxC8HeT{B*26rw+L1Tzilf)+&Wf)q=i9
z7H(N(w5P1&ey4>fs6{3+CvVfZx}{Lf3sC$VwSB>6V*$v#ezj(!J{iZ>vjg4}+9x``Db7(lJo6YXD
z4m60cN%PTV+VWgCcl85s6=xC9jIth5)fka3GJ;czTN4wMR(sJei@ey@i)6mLPPA0Y
z0zXyvj!#9MZDJ&>zfTlr`6m>!}~Xwd0mk9#>vwsaeGgBk%PbHM}7^WGby{hnV}I`sDW`Da}&}qRoPwU9F1N8
zq%8)Z4iu4slTq*04C!+Pp6i6aIQbUlWrLP7e2T|t3ghF!xfzVA8&iH!SzL!xu<2Sp
zmxMl=wb7WAu2NhpoCy|shfSC|a&m;Ib{8?XV`LhaEO8|C$>LQGvt^S(!^aj+P#cQUNsiEOoyA%O+o$>Q
zaq7fU0VrMmbP$4(IO}O|o?Ps!?B=8RnGm=pnR=Wy=qDN6A#2muQ$mBnrFBcejn%E;
zI>7642lgQZ<%K#`Jc>$olo0<&FkUgte*WSejZ2Vfb1RMPuE}~%K1dR5VlpeIjzUGR
zzXUUG(AV1ObyEc~5}8UOX9aqaI|>sU%zRi?zj{q;xoVq$&92gdA$*ND)9_n?9oS5a
z&o3e5eEqM~9>9@r6AbUCQvC?gJr-DPV|5Lj$(F&glbDDdrTjTiGDgTB`%{tI4;x
z?KvIO_>uQFwd#{vN5C--_^1VnnyNF39l@^;{t2#6=KJpA{{P&T@FZu5fs
zFyKFT3#;c|CR$AGcmuF>H%n+|O{-ms$1MtJ^{>eBC49wHb2W$-#QZhfhw7(P%|=9e
zTl7}E7#6kDMZb%BHv`&+2mmQ!v_NRR1tuOVCSEn+j@(X!v2XG_{LRdzwB+o)ffmkyg2n)-9+0Sii$1#
z8PX+mqqt!_u5NcByWKjL`l(!FioCmIq3=fJvcA3g#lm_MDF`@Nf1}c)@#4~8Q{PL_
zh6~!AQDBSxREOc3r3<=B$Yk$C0#An;W!Gp3BDl}+xj!vpI{KNyx?g_Is+VCU7G#oe
z1!y&8>NgwlBa0GU&~vE}SBLkgqW8U4aAMx$M9jwXi+*`CiY8icUf5bJ=94K^ps70{
z@JJ+Qse3N=N`1g@)#C2lbIwcCDb@enGBFB6m%SD%zD#zi;M=1}a9`n&R{)B3C#lzv
zUSVGB$84^mDlI7V~CEC%!2#P4#|+p*ltK<5&71=NRVm%!o3HCzS;3c?k_&
zH@6>?>$O(UYhcr(RQLv9$E@v&xcwu^X~13IfW%&up>Gxlb|f;R247z=!fb5F?S#Obml2$+}ZD8uqa~aPVpv#r^G7vG{_;Mi!z1l4+=6
za?1;3W?ZhN8-%z}`I2*1(k4@xbcj>KR4~HozCRLLOnl0
z2mnKk06Rd$zvwtt7?lFySBYH3hVa^^WGmQtRH5*GyU~}3UGpbjF2ho_$G$zJr*_uU
zodKLM_BW$jG{RU7jzO*_ec5b{RWy`{S#1}?fCFH+Pl~Vldm>}EZ@~O`8}$DJ|I68l
z^f!1bA)a?qg?nw~XQ5>ORMdHPK4@MwiuDZ3yMLYM2qWKr!n*wq*G=eJKF5FhVKJW-
zF{x*~+Rq0jUxEk77D^8$o}r-mL>5h2@U!xR{t|-Q7-~j(bXIivX9=N#MtN0~(yD$L
zjjc^_v`Yha40zJmjip51kV9v>X!&(ow-#pWfZND@t_w=l+(4FfRpmvdD~Qp+#fSU?
zfSvklz*V7PL#8nRbv>Z_1(67&pGu*n6-vgZ%81L;IFtq$!3EHNzHMwchiQ(sLJ-9x
zk$JET0;A?(BV%&N2X4`-Uh-m7ocUGH%}Q{Q#%CXO?d+q^6`4RIAv~f=LLp6U1WegmC?A*%h%j{o_1iC$D#ffQRvpRoex;1v>s^c~
z;_a$C(cy04L@bba6DQXrUdWxJWoaG{+`r!73W2va5#@;DOq@OC)uk>(x!`iQ*)I))v35Zv)`2^tS69TG09u;vxDz>JLFtl6
z=KSX%lrDBV8h>Ad^cK;IPAtRRck!F+M#mib!^^U#_tw3`{^~LZT{#prpBNNHAA0|a
z7J)&VSQRVlT0KK3!hh1gAK@uE6~3!n!Aj`XLH6Gydw=TJSjr7Ut^nUC3P@-7Q)pL
z9`72B3e35a{syOvf03{^%V3_?X&J0A?^xhfdxvU#dw1LhL&=9HifQYRno%nMl=<^J
zkOAK+Ye+Y8^;^O(omqJ;^9sBll}~lK?+tXzb`Ehf0PUez0Cn6odDV`J?0Yx71oUxZ8OZzq0_}x#0_G&ROdv$!2o0Z
zKq9Dj3rtCBCEEo&MAwmT_!RB*i~uVt?DX4oWg8K$2M#gu4yQ;m
zGVk#4`Avu7E@M52+b@pZR%;ojYxRvxc!QvZX};d7G@O!G@x)bT?X23ro~V7C4~HB&7X7UH?x2)b(u
z`{MV^CWueQMzy&!-$=f}dG9Ze|JBTe*Wrl^k?ZL~;}Qhm9W9@@EJu}fV;i`H@xvR
zk}B{L<=ncRb(pKjlKgi5&R@1lOs6w*a%;`>Y>sFrte2-4XvM8%n51DVqNa7NoTszc
zTN671b`|zRSp|+ORntdO(Bx~xd38Dj6B6U29+`EI`17@Ggo6;8NAVo*CG&Tl1HVWN
zV_+kSfxI{1hv`6%y7nsS$A*fv3(C!ya9d)mu1vUQSCz=A!OH|J5QbYHei0EX96(9;
zYaX1JM_@d=ft6V7!M>y18EU*yP73Gr;`x&H&nGZ9PVJI`y
zr_;Ky0x{!$a}#Ict6q;=26S4kFb?InY!|a%GoNqyWkSoZ8XPhq%46sE89>3p<*%Wn
z?vQrI$}|_uXHAg2%CvYCC2RTkpaoM4QZFj_ZUgEO>f`4QT#3DfBn;s3QWws!-xETs
zFv|GFuML;$r}8=v!{COrU$;*NQHBTc@*J$3UFIHwW7U3KRvZ8Mz>
zVKyA{?CL3|Wj}2tEgP4@8shtG0jDjKZw*_Uk}2c)wLc@iUGf_w0>$qsoQEknamHjN
zWMWC#!nXo-@Kc9ykmYt72rU-H*mHDB8X7Ldr?7nK*gEeGm%s?f7`lFeo*z!cOs;x%XD|$WDhj_R>tJDN)mc?P8!Fb^K1(-^Sphw5FQPW<=mc>SLEPrxV+5dulfGUkrfvUgwGWG
z`kk&ky&r-H3||@Thzrq?u9+gMlh3c4vIfZ6Q~A)`mQWT^&^s
zpP6Q+kI~bT0CP>b_Hv3}*L2x+n5-(&`aHRN`Mek2U^0gsTBVI#X`afXowBON_-h+0
zSAe@N4~C=5LyJe!SWTL^duF(0k~fuQ9Wq{1)<9dO!{~oHQEn}yYte1j(+i)-OZD07
z&&xE4P8xO@8F1KwBWG^ACJXe}#aBo(xNp86faKuPi_k5NL+k7S5U{KGx}1&nKy}%|
z0B0QvoiHFEs>LV8=Gr=gWy&)cD#dJehWM8bqs+%>zNG3za~RbP6bI*ly}LP#)0N5I+Tpv
zLCc9`7jSWV7a^y%;!OiE`U=3U$Lhv7IHk9thEB>9A5tNCkD`aBNoP$qV9&fHnCxi%
zvwqyg`ElK5RBCeM(wt*g$j^?4y4x|?^@v%w(HLbAqfgQaeRx-A&OvlQbmU~)Dq#7!
ziW!FlB7@Sb1DF(T33A*?GwN*e(|d0(=x7Kl$`r<~9gCNGQq8IqQm}s<1Dgh^6$(Gs
zd=6tIfw-AKVnTY(RH#%aea5Jb><1amRAaF?B&?Z~d&)1*^@m?hkI8Ovn=rKZ7`$XZ
zIVrn>zL<=mqV?d8lfBpuZDUO0KX2uhJ1{XQ#rF>}KBNSG0!GPizm&UBV2vrsD{+$9
zf{D5iViuWw9(hp=lfJ99f0jdz2xuCTc1n}|I-QQ)BAU3zwu2E$DfwRc$wfC^KJs`o
zu`QfiGSgW-Cdw})xaZan8wUb{fk}OwW%DcZv$y@8!Js+RkL~lSaY03te#ls
z8Q*irRMpHdld1lXTolLeK7Imax@?(qtzBIj!(%xGru^jPPC@(qr`__SK|zvB90RN8
zgW@uZF6Y`RSg_uZuSc+Piz^Iz+hYhZg`|nO#=Yv~4SQ3V#>(u(6A~Be%nya8&|B}b
zSCyFoU4Y(%i?JUvgcZKSl~9U2m%Qs0ssp)=fELvkH*Nr3^1Q6lHKaq||Za8N>0|8`JszeuFP7MbvTW^f~}0
ziRF^vGY5T(`{6WX^rY3pf;vuDjDgSjsj3G^1(WmMf?Cx+M`-~*42(Z^Hq!#zd%?_}
z8bx?bkt##qOxCR7kQvV>pn_D{s-A|qBq61Y9Htz#ocXL@nmgO+k;$2
z@0041+HR(I`c_12df?ypP(T6>zBv;maJEls7UWv_hu@S<
z0rq(@7TTCJW=bqIg*?kF9lqG-JR+?fChUw9pKl}F#*Udpn8psxX?a1(n~!hMH;pCv
zxLP%|;1l1BVo^2dv$qqWwm?J1685sUL(hCvuk|mF#0D&I9H0ZD1GPuW8VQe
z^_yjYf1JA`gEv-352Q>fZ4%$2ZOlD8wi*21oUq&fsItk5vxj{Bm$2kVQagKrbD4lR
z?bwW`YC%xTS#EnwUc4aCuZ+%GPSd&{4FgbV$GzUzs`bG|9(+3umv_
zg4X?O;~7A%A0Kbx^TScc2Gs9tq9Lg5?dlFXw!zOPg0C(^-H+E5JL;L>#q*m8r
z*2V!Pz}~>6!b)FG2o(2Ie6S>S#`k82Cj1uuQLx(1+y3Dr6QheDfMU7`0Puyo_+@5R
z4{bYZ)@1TWLZ+0~?99TWX_pKRQh(eRIlNjvb;3gM_=8s-Oh){s*X=jJ2qQfR>8ERH
z1{2c%aUXqVpam*S|Bl~mHD-w#I4AwH`Fk&>umjG92}!EFd)rtSIHr3i+qXOdpSdF5
z&K+1J^Ui9?s%u(xrjKfsR%FujR6PIKm<|C
zbB2xax_yFJ>t!V0LdXU()EhV*rM8P9P}dzHsdgDG@-i!`bLC}shZf0BQq-YV;t^0`
z7tuCJg-sfRM$QsiCPn-~P6Y9COLt+)l%E!?kt!6IIwEdVF%db9zpA%Q>9IqVb?^9e
z`Bhs68IsPsbJ;Fu{zbTzrYdZvVA}=(e$TP6_br>;KdPSGFLZB)D_{ELj
zFh=ro0GTEI{L7~2)Yk|P?S4E?JHQ0c8KlURV7j9F2I6t=c6+f@kzTMV$;bv7>^)u!
zQ``$ivKegx6~EA?qfapITza5A(-D#u4RFwO>iCz-2-7w!rg{56DK;N(eSUh4Y7uK?
zB4zz2tH?)C_I0uXOL#y!q3BQYDBJ^en||%p?JzV^8ZG$SVVkTj!E>HBjyST0Am}Q3X1Sc
z%#?}CrhHAjFAe`;t4jKIH>lkOI)ZnBE**Mm{7s_Xyz#kbK;LH{-g?~t+DWfk0M%$HflK?LYjLAdv_CA4(=O4@QGMao|tTy#U&0XJt$
zI5#gc1KI+DI9uytu;=$TN;MuJVl)l9&uUfWN)6q>J(bWbh~j`U7euF~>JK8xVK|AU
zXbn>kX;0#f(Ql-=Xu{#Ocm`uA{E%M-(a?-o8gx7uh~PO5k*89GZc_4kx&K2i1rG##9Xf`+u?UDC1{ljbe?+v`@hEtfeF^cP~k&MY1)O*N2=ucU&!aXjnvcjT$4?ng4
zOgNo?+u?Xo_~4As$vRH4UDX@CB9dq`h=gF{6}J%eU+DNRdd-~I5+*t^{Xq=b@o13o
zI%+4nXuA3&3`N_B=1%+_?{dGZM1Qxn_LPmN8vjQ8Rdzth@PR%3S7D1~>A;tqm9Gg;
zuHa(4MhD0_NHVnjm~oae=?_Sc@;#TgeK!I!%$3>`Ta0E3=yNtT3aDxZcgE4kZpg^`
zIzF)#=xxAfg(Kp@y7%MDW>7GL6O2{eR81f3^{M;zy({|dWcXo4De-RxO`J9F0XW)m
zTg<0R<4D&O_FmRhaWyf-PSoKk2!+znvz(_^6#46_We7A>nujCj4Y8?p2Y8qd(fCxj
zabMt*V-?=Yn7FBO)Jm|29+}^5_U?1N52@6G85q9OveOqYRN(g1X~XJ8`WhKObZct
zgQ6zg?UeNB{>+LvjlABXvfgk7WRj@3-uKcx(FMqZ&T{A+CF#9wL>uB8ZP`2O!T_7N
zy$<2@w0!Bq@u{{FsLx7s5q^mWGu;2ls-;y}qhLs_jw=6uQE3Jj8w!>%OWk
z7-VT2hI0AI7BN%qH&HdlRmepn&q$1GT*B?DLHLZC(0UTSC0Z&5&}420s#pO5$fPCp&*!-S|;vMUY
zHe(1TCW35Yv#jmVjv97>aqqgyCjZs*v^$7{Ejs@<2#IGdGSGoa90%9ZL*T2cl2obS
z{ef`Qatn}qf^`wnm$cuf?C6++?}ly1uYk=Xt=klu%Ww4q3Aiyw50kUy$~8tKL4qX27hjI1hbYv|SB;pe7}?g6L;lPjijb~L
zUyf;HXmCwweRl62m;nB9R;vcEopjyOop#;PxrC?z%~m9EvMr8qs5}4+r_*|ojgT?`
zah2TPUx8eMXQCBtMF8nsCD6lrWm0Hj7U^f`HH$M3Y@)1SNLdrHl)}sP|BEN<-FZm+V@PKUhg$03NvPo+?il
z>T?D}wb7G@<03%J@FOu}0AKBr2y3waf^UIy>m-1*0nTVju5hQfUeKi@aSt5LGSA2A
zAc0J7_uLNi+oYC^15F*A8bb+IRMdF=EBQl^OOtEDQF}#dSY@d!ubTrOgr3(2*mZ`l1Mk78}eou=QE93pvvO)1Mv6@Q+qezn^iIG
zRO`*9Be`6ibGja0w9St*2#-Wgokdk}r(NZ@BOsB-d`=S}J%wu~*$gy{+qQYQkeZDb
zs@@%QOrO*FO58vw)|u@nOjbq^#-f6u8uEJZ@a@Wh4y7ZMMWWyc
zQgHDaqdAr7VI%tqsJcU>lGf^#Ju-%hiMDLleF#=znS#I$=E!^aFTaqD2g>B6F{D;A
z{zG?U0}2QQ$iwRFM#PZXesNKqp`N`_bThIUEoh}Z*5B>s4HExqh7ZPYINHrcYV6{`GxUH(s1
zn%^ig6St$F2Mf|L?QU&!Q;OCUbD?qYvpS+lH{j;XCA>}lxlx&8O5Ti1lbuGFF-9l+
zVUdP^2Gf*N^XEmDCA
zgpwsQPj!#|0AE_idO>!Q^%(AgM`^dAwT27IOx3GZ38pd!t7ic&T^tNcvB2|hEUe9V
z63v+U&my6XeuhD{`)2bLg1;Eil3@gxE&c3`l6bQ2mF&ZN+sB6%9JZCFoKoCVMq
z`&+G2&}4qNIB>H(W$
zLGqq;G+-A?v)O*Yx0aHEaYd=Rkb(4daq#I@dc9AzmTO
zq=;!ug9FZBO!G@g0aTO4cWNs$k&!%YG%ht6*S2HM-Jrk?!YJ1EHVGYs9bZ8Ag-5oa
zO;GHR73%1ddh&%iW8V1GaCV8B6;aIks4@X&%aYhw^p1R0kDNyx;^Z$SP)}HfrWN&M
zr@*7;-%!cva1lTsk=1eP;NeihQy_W_&E30Bv~<1L3geqMQTVNO_cD+2
z@)MsAAQ0>ID~vk-+w%&nESg;ZQaJwt(sbk=tBPb2ohKb}^^YAEHU+MGoiXO$Xa=ep
z)I%`M8rPT@D)LeuVyRJIEROog%``Yaiy3?Ru@ph+}w8Nj3>^W$U49b5w5D
zJ71YD%eQ6$E(P;dOPx6ry1jy3h!_;fqJ;L&?RGE}sI-6e0LK)XkBAU(-!oB-QG}_h$%{ZHHPo_3
zh-!+`U}wMLRkD2Y!hVX>pdK<-?(?>Z#mS)?pBWfGMky58O6~&XQBpix#Rf%FZt#9;
z|Ndb<^JPjjY(NUaVv^lmKXzScgA=XRrWWruT>UQ7ZW_3*u1<7VqPcdW2~e+iUaa2=
zPD(N$a(K6?_FjT3BZBQw319_|UU_|7l2P_6C{a8vcP=D$1=s9-e1_5k
zMRIFJ8UqMlSg(!pagsn(d~tdnHnf6JyBsaAl6s}^^6l6)KZG?M_VI)qgs8lk{7C^5pVOG&%j8tFAD4@&$qO;~UzutY$Z1DT@-5vhcQE8+RvbUd?AFTg3tHcy2r_v>pH!F+PB%
z3|w0^Tw3{VI7}tuu8q?!Ruc3_xs7wgdXH4U^*Ysv)H0i`q&_hSv6K&*azfnb-7QP<
zKmnd`Wvn_Ydzav0m5CYn0K29%68LWDteD9Ck+DU8pxilD7B+K!W*kAqyFQr{a+O*j
zJWv=wepXnWCR=4`UA3MCs)dgDJFvL>eHzXz-%-cM7>$F=n5(n>OAjwXmTQiCTZhWD
zQA-KqRnDd*VpI=A+)$sCGp6rC;E^|0*HidlCO8GK|E%j9mUc{#7fR^e<}gw0$_UgA
zSd<*R_-fx`w;05>XU&m|`zDBx>x2)g8A_AfUA!{h?P49ojTlMb6Q8IEqUkMs{UJuu
zQwE0FKCi}#MP$9}yf}{{kz4Q(0IK2k>aCNw-m^**85ReadH-&c-75
zG-HPM-0Mffeem`)j-wUBMhCHP5WJY`5{s#3HRg9kJa|s|axmXHzw6*x
zhE6?|`#Vi8lKY{+!SBBp!T)d&AocD+v4*KP@Is?ENXJ|kt26raMkG__2K_xG5~m(2
zY}|Cvga5rl&E9eNiVS@ziI0;EP?IKT_67Yp^&?%~BoFpVB3Z_mh*xWpIuQ$JNgI{f
z^(`4MMy)w1Ve+~ijpyxnV_Y0_Bvh}=A2dD(>f{B6jbR1_Hk%x0M}ZeH1d2t%uhHcb
z0WV)?Cf0JFtl74~;I4Wb;GaqeAB~7u>6=@}&$9Lqj{hCFe4<_Da8Byi7IIS&`}Y;G
zZFxS@FDbbYSoL;p8y{5OzEU!}JUK0OpgfsS1fpOd$2#t=)o^%sQs$)tDX`u$MpWqhO6
zM6gpH9W0==g>d=Kl7Lu7#*)LLku-lIwUB-^R{`IdF)Gzx(K-|zN{{G`UVo{ljeHYQ
z3%&36+jfInn6XYXP#8Ko^h)4sE3q|bze*|?fK2z0@Ea9eGo|SIa7l>ACFrK_ZA1&y
zE~QGhDB5u-v)yzKR;fhfu3Gc-vT(o}x{&;-46B!0T&swL)qlofja>@^%YprxfZZ9n
zC-EtChBo=`ywo=s!NA@A$T`9$!%PD}${iG5I`>xTfI+s*LJjxa{{$2TS03=N+zU7tR6b8Vz#6jlRRDyR!~
zfx>-8{XIZz=e*5K0@cn|Huh7%YZid<_1~zeup#T1Mh8mEEudoztfuW6>N_`0I^F9V
z)R&TcU9!LUMS7E`Pf?X`pyN%it<~}d_FZ}T-m_$cAW?I`?epjfPUl4N)<2jxGO#Ef
zXi%a3t}OV3WhVP7J5l(VD|s5H6-Owo%IG&eXCn34ekBb9
zadi!V1@`wSet?+@8^|AY?5OldfD*@O>wmtJQUtoG0qR3KOXaW^d&1D<9-`KCk3EK-dO-JN}o?Dzj(yIa#QY
zNCVzZxdT}i)|4&_9skl2DW@P*-RMtx=_IU1m{?!mI->=x(05)|*q6GRljf*^ktU+3
z3mjcjF&-o4L$9N{eA>G!6V+-6wu5Oohx2Tho7TjqyKv-^7vdOo=Ae2sQ>ERCYmWJK
zjyT}#?9Eg+wWRu71QIh2=>C>%Uzc*J1;jsnaDbtg=m#reDPILu5Q?!Hy@z}Z(yvZg
zl-Oj1LV1u^XqW}FAH*&MpPy`4$~WH2qzDW8ps{66E@Ug<
zhfYim7I$W_T;V0BJrAQDVLonyMy=Zz7&`7H59UVSDV<}+FgFIxK1zQHIdPZC5mNmY
zq3M$NW6OyXTPt3eKT;DZ#mHCrX0x>&R_dS~DU3h#5VDY@>;pW7L$BN
zzRpyG_zdvqGO?MdEMD8RyBRAiSMM@UY*Y9G4GRc`reg@kVi;p)I#vH`buy-NHstyo
zSTi>cA)(81BgS1kqvCNj9P@^|P#ZpuASwKnL{jxT^+~16Nx5@sZQxEs?t#6ca&A?O
zPklf-{o2IhZBRU0a454E-qo|x)vFd9ai35}$#ep8$#qL!`-~{>&)(BOd!;I>4yM>L3QjgLXe=Dv{ACf6!7gjjwE$H#VPTz5=bM6`Yt8Yq5z;M
zZKxN!O8~eNc*!kHMHm@DVDqWSoRfLFW!~Er;C@VorYBPeqS+83UtS6J1HSHwhR0pb
z{Fbn4
zSTdFFBK%^z6iUmEvic+4gbiAi>`o0k>%d(}
z`ogP)EV0Ezm@-e&5=awO1S<#h#6p7Q1^u
zUsl^g?~4H`(foi84jNbT`Z6d%Xm1K(c|Y>R=DEo1jGK%^;5+jjNxLBzptwc^Znh!2
zJLE$;4NA)cV~b)8Q9dcO}`rDYxKSK#Ic%+?HBKV;a^`Qouir!pXhP
zOFil_G#dIG;-ZuCA|$Z84-t5$YyRb+Mkbx_aZtD`REEO!FA0Si!th}GK&P-B9Z|Q^
z7XqUeh~p85#U`=A`lwk682vDO0IBC=viti>&Jktybpk3I3CCQ)Elu=1dqQZF1`el(
z+_wy$*ql2@Mk1}@r}#ZhGXTwlSKCr;^|r5yQKWe!c{l(Kb58ExN#YGo#+V4Ls<~e&
zJ%y@fKoi~tT0LdJf+*a`vk*pqiC&k2Tt=qm;%SiA?6=Hv$`&xel$fE)WfSzq|)
zR^j?j2Or4&NWB_V4Dy)`^m%RNQ+r~R(k?5T)fPZWPNB`|1oPV2q|9)oMW`B4gck~K
zOL&oYc7`Oygo4qhwBV6@{#t`p4d57px7r~tUP%R!+hFc4`k~&DlNN1@N@hF~?DlfzoB%aFf~vk8gGsMKhfo_SEvLasCivV)>huWVu*mwig38Q2d~
z5h!A+^^H@hJ!F%k*wZTn@udkDC%{90d~*{U*A5c%Y=Azc8aNYSdi6AFEJ=OjXtZx%
zQ^>E(!}?2-y&qC(Mmp#Z8&?Sq2Yq;ww+aW~D(#4&g0mm!1mE4bJ9Oyt&Rvb8Q3_2fY-b0xthOy+#z_Q)6A900<_wXSPQGaj*ki8@9^nK)E4JW>4o
z2EF!WM-Q!tJ@WUs#Dtf_H_Dec6w9k_6?@Yuy|y~0qBko>iHWxQhRfhAv*5G@ekQah
z3!#Ivb!HE@%MOkMzsZV>dOeH}09M;&Q9ELB06%VKrIC=*qWYv@OE8h&T~`$CK5Jz{
z>|@OyXAzxNBJVsvJ-qZ~G~ggb3bffAPQA$a8DM(?XXq+Ne89{=_DTg{I8n*Igd2&d|Lbt=8QUz
zvaz;aY)p@SD~MY9Q+ah!dbiRbKSyyItWB97aEvPxKka#?O*));da(RK%l}-7igPku
zB%#FsiiqQ~Khfli!he@B^Oe1j1lWE64_d=H;%y)8G8hPsHc@+a7EexKmQZu@<-xf9
zzgj0A(`i~$u)QT?d&k9^Qotzt!vjtauX+3B3nFZ{Xn1YDGJN~+NAHssX#MJ*Ke>Ks
zy8^M>^QFdC!EK1_-a?ogb;xdDr@lY+1kHK|!R&ctAx}Ut>_n>N2rxAgQFS!43oDFy
zP|)zY=9sY-df)%B+TBp3Q*ia`L}pe^d(2k@)OGv7D5jNduL!0A9Y*JuZsO1s`MFq~hNN_4}{k#o_L(G76f0MLc=+s?fA
z{h|9}71)TfwX*9Q{;x5bK}1+lCmMt@X{d43Y>${ug&V99tL444`jDQq_f{I
zEJ#bB#vBZ#Ck3U&HY!I>s_7LEQd-DXp#eA
z#7#4pAl0#)kfR!Y8+W@p|8ZtGlYtJCvUVntg0i!aa?#t_Dy&?0_G#|&HqOE^pZ__T
zzz_Ar%mEK^*$BS4x_?~kNlK63CQvzp<0)bHBn2+lpYQpkZos8}t6n!unR!NV&aT)^s;IPuqL=f8w&qaDf
zy&LGbRw0*N-K$3vE83h7({Sh(msIBClMv`e&P5+a@1F%PnHd#&up+)-Ji^jrVQkU*
zRG+Qg5fE%WGto2?xl8>oKmuCT?hd6<%QF&@{*J;v+Ra93+H;
z@(8H$N%!(ZHtZFR
z9UV}~b>!H#TD%t?eXpGnZYDhOP-oWsI2dY=MD?dlq)+p{2OC<4%egdwbrMO!1RGe^
zXBpk^H2v|kS<-kdF@)+KyEud~uMY1iR4|(e!7advl=AA`sOv14A<3sHz#D<)t5+Yn
zCtvWD@X3}S^FnC>gvygx17jofH&&&XT1Wt9IMUiaTsAd^6}fSgT-WQ20_yWMSPpPk+ghkJCwTS0{=VJ
z$rw`cfBSz{V3-`r5EmeEOOn$S{Gh)mceRNx?L}7YhpLX~;hg%aq4w5Wk$Y@(DZbEs
zwur#J`|iD^ApO$Rjxk#Y&3}r_g63X&Orl#jouAN_su;JFL(Hu_*<_P-aWkhBLg#t@3$_TdwTv0&)(Uq9)BeawP0w$D4uF(5Vl$p~CP48c(H(=Gi-0fO2#ip#S
zFS!XpZaD(C_^iXcECuL8*42N7@@@vWMcKc)&H215Lt`nv^{&M5rGfg@3lBP-kkkGy
z5&UI0Da)1bncJGcY9z
ziuIj@zF(3k<<%PzOuG{J_h=uKqId3m|AYT2W#|l*naB-OMW~-ebHUMnNB7vt#WoNn
zdavtuH2gbm9oL+DcUQZ;#dE^&Cn3ABQ`V2Oz)(c3#cix2S}+QBD55juX=xGkw-MmP
zl+%>!9dPiZDFVwX5{jJg=iS|rr9;70@WLkC@;o=;02414MT5A5g9%|vHK+j7E*?D+
zck8NWX*0i~?VFF}GljW`+#K0LaAkrb6+|6fVo)ppxj@NH{0|y(@nC;E<9=Tkpp`Mr
z3)8?K$IZeRb;Nc9LZgG1Ae+IA!%ADyHj4g1h7+xzOR`1J&Orz9i&qpUBeRM2SUok_bC<#KBcF{Opy3m0RX@zP~
z4=%_rwS?m3ST^Cd{kQT5W%HcEpD{-_D1!d)s$8^NK75H4l3V|<7bIg{++6#gcB$rB
z6a`d((?{HU+|-8#mj>+n?5X+tzjSadAAI^S&&-icXC|Z;*n#YCR
zJ{~N6cd%S{9dkP%K%zkoacuzftLvxS|6=r8WJS3(`!4}{`@i(*L
z=@KIS7y03GdQjq6GwV1oKq(!l7~(0(oTuqd$Fe0{Vse1+P8Twqsd0+~>B&YawR08j
zKGFUl8StUYoLB@HAdcfQ(0S@pWJ{~%;e=#9V0pB$;zqc7eACmK68>`7*&!s1dW?7d*oJkbI0q
zpK$d|F(Dl^brlsrw;btH4nGWZ@qtFTQAV*^)m)_m-O&2<;eBU05q@ef0j9nrZ71Z_
z2Lw(1Iw{B!_p)+9m!cN;Um1qIrpQ)@v0+snBEHBHuoWJqf
z;!-=438d|okcwcXjRt-d<+#*$e2GLKQfareAtG_Q%;A6#d!Yk<1!<5j~
zvM7ja;3eCCk2Pl#Vu^zIrI+We2ynbo5lF|Nk+tI^}^x
z(A#v^a$~-sz2-e4hO*e)I9m=$-*KU3y|o8G&i)3J4xe2(9yp=Hc5s3n+coI`MbCdK
ztAF!c3qD5tz9PY$dIC%p#JGf*d7FGff9cC
zXt}Nb)29{HlR0n^fB~_On=sWr_EmxGs1uvyPllcc@$Wq#MZfdt!~Nh0jONG(|3Fj>
zSCk_dHf)j!G=@!_avj`yFc{Xueq)1e(o$J&6YgR!8A#pxG%cEQJ2$$CG;0oykpbP4tyf=q%2%`2?sKob*A-1NR%07%|NO
zv9DxJAV?)^AE4V`BuaMwplG@1@6>C4Mcq>{Y_jV_eYjsL{@T{5s@LdMv8~HMzPz<8
zpyMA_2(^T`p&lSoRx=(>>vYFP|!
ziRWRcI@uog3HkYaBOh$ocrOHP%ZMTOPb}o%;dDq?4*rf!v9MGs-UZntA++QL>#_wjp4Kh4EJu`PZyU!`5=J4EeZlY)_
zcs`#F2NQk%l)
z8a=2;sw(EHddYvu-~8TNg&R}jfm=!mo$|Yd*mzQm@8H<6)-G7%zZT`FG#8_1n+dsc
z1p#5A@bPeaQ%Q~SSu;FYKp#MIDy>vY-Y7egc9vupdS0|AGgU~!Pl1Fjwwy~UJ^%ry
zmL?9){>-5cP(w#i8ZH>yax+dm3%@^uaW{UvP*Is%MgLSEYbwj?g;_Su@pR6kb8w(G
zpXF17R0RK0Ug;>Q*W}FNHll;LuZO5g{sfg6XuF8_QnP)Cc`dBez`B#8aPheWlUILx
zWdUe?qoTe{hW3;sCF11m7y(GilUo^=20=!4XVv|h=fNiq^xH-tUJ}D`&3_r?ViT*8
zYgQ$sE1#q!r+_);b=>h_+2t*1+jIR
zR>dxMvzpO2JDqSgRYnDr2Wl{JBS*BYyIN&bXK_oXtCef-Q-OnXK>qeLV-jX-a^jD!
zTO&Gwunj(jCp;%QjzaO1!h&H#`ALyI4l&0mLp+k;$~aq~Ua-*V3iByREc~xTr}{1?
zGUl))ZD1b-@SSg7M&{%F{+PT9KAoK4eS*`Z(Dlsg>nzm6*?gSZqbuIcDlpAEjO
zcn;7uuzR(|-yOszDp7dAXF6}aKG)$bI6ycJ#iR4$jg4Yn@2b4;;;zs*pHos!6<EP^4&ytzAf%R`$a?du}vKtJcvi8?K=Y5I|b^
z&p?Am5|NT@gmtt8)G(5vr?D@$gt?<4N9{N$qzx2qbTA9eSM7_lPBZhHC~Zm81jM3b
zjfvEu8=4?X?IcsnqR~jHQ^+t=({e>3c_kHi;d;2@w)N~Cq9nV+aPd`sy>K{}W&{KrPi%r5MZo2(MVkBF2kbfh
zL;K&4g8Dze{BAc6WU3+cz`DM=Gnd{Y@=Mx{d+iB!p*$IWUwCh3cHFj}y5*V+apoFW4vl=7KKBofhzoQ4w@e-~v+;o7q*KVHfDgLo!6i7h~C
zj~h=?3x{+O%0rE+b0h*xB4yV~hjpJ|(QpjB3T#`g!>ozxNK`EwFz|@1XZC0hl%8^2=I2+-eld!hza3gpp=2
z9Ssq)bL@3Fa%{aami>Ne68Y-qB%;Q;!9A}L-@={0LG2yS^d7vRWW8>lHM~c>Sm&^}
z&2oOchhWUch0N-mYa{mR*ww#qE^zGt*-EI}KdC`tTqBWVKk!ex|DH)ureh(K>;f3h
zg%rn-=zF;Fo*1b5x4X5mAHt0T{gLU%4D-5EbS7|nfT
z1lw2B&SC={cjD{hBqnJH*_V(Cr@YV%KpH>jze*H<6$W_|u0I1i$mk-^YqV{5Hr=6+
zTBU{xL(7s}ckVKBo1SLrtNJB(jVYygig=e6JQQ~h`a{qJW>Am&id$}+7Dp`QBhxO|OoP%8
zNLzP}KzEVLjDBPkf>7kvaEj6vC@*9>W0)QZdh`O(cLYjU+W&F&34L4=o0mv*#uM9!
zkHpPi%gPj6;?J?#wNU>#ZAtyE_QJ@GB}gL;Hxu~R#>v#$u@420ZqLBUFWBp%0E^9I
zU%4J2-0)owOD{&m|pAX9$teQx?-XuzMg
zng4-?85$=x#Y+NcvGJr;YtJS7ud#_4rv%JhsD_wBs}|TDG%O>c4!z<`zoT9p<&dW{
z6p)~Gwak)pfT;N5y8f}a^6MLg?0z(!gWCg>5P=dM9CdQsPA``45L3NY1^vj!@sjkQw#B)O9b?_QA{-z@Mf4Ga@2EVk$NSVZjx?@=QGwJqo%@ZoA
zL_M5#P}=U_sjD++ph8{mlYdRdOu_>mPb#~A=l8}ZcnNP--DF#r=
zWpd=$E%%|ONr;n(uQiG{17-eR;g8##O9EFONIm7(8tfJqcEn;ibg=&q
zqr9;es((yu00=c1X?K_RA7yf{y7V%j1-gYUc&*}8%hk1+ZgkZZd$RiZg^A7Ncoq#O
zpL-32$^324@_}#ZDsTduD~sX755l_BRfl^vf9eJ
zqRSmjI1eC@Vg@g{mF${<;5}lqv_e{YgHb9|MEQ*J{M2hp@}%pTKh^|gcP(CXX3o#J
zLpizp9%nsbP|`_D7ziB-vbG|e&NSJoT~zy0~x-DUjP=wv9`d031w
zukd(?ai&scn;bog!7KP4`^|&U4k5**phmu^g>Ury=boyU@7L>
zp`_u5O7sTYl5FVK={^K#l5pwE{5UZM0J5|FV9}j@Vf?CY3^xP#Bv|YJfS8z!PfY-Dq`;CFSZY4&}BADhCKGnB7NM3h!
z9wx#(Izs1nkkRv$J!D`6pN9M)5}SH*i?G9TjiGC=4E4mv6dY&K0dTBr#F?1A@JoL8
z$K+(nAUXBMFF?oZY%(JU%gTA3VZMSGR!ulagk9@}yun@Kq|04o^^b0VB{0OofkrvJ
z;^%sRww;&_3)v7HQ3A$1xDv8z^h2M+Rl1b36tB~~7eehDcLh4^j(En|KX-StW;}Xk
zAahMPsgI($A$(yovj9tIuePO%-a2vIu)H!e#Nd|TP9Ha@ylHlU_YK02c)^qVM#lT$
z@F
zUoTSNF{383sVusUa3Z9)t8aKvb`jK0b~^mK$U~cDJ!GROd1He^salz`gW}D`VE8Ol
z*3I3#K)Rsmw`cgLyHnqgnvwrH|2~I3upSA<-ysI`D3eC_(4;xeMw&tc>iqzUIbgzx1Jz
zFUuo6jp6|W;=MNOg~__)wAVxkrQVWsqj|$+4hh>YElQ3=>@pH@je@RR$n!@y+mF7L
z#TCCma?QkC_hbGHWJ=K#plj%qfGNX23~q6;P1(O71VD73gu2z
z&kE)&A3g0ACNq}Tu5j~;=g+q11S%}~6uW~aheS6|XA5;_iz@0y%~3K7W>jA}$DqmP
zf-hmQOYg)Z4|D;a&Q8{Z9_(DD&0i}i+?}MQ9d&sR+4flEVV`%;
zesW|(l&)dtLvdChGMUmIKf6sv;x?i>*?L9JEXX<)Ycr=ZRNA?#M`pe?YV(bEBEb&%
zs_D>T^(OltumigaHiffmcN8;G$~3C{HL8$B77TR)US~E+6d<=7B6JZ;0c+QNk|Pt(
zLB%>v^X|akt)RD3kiMtqlgAYA9LX&v6W-#@z1Ds=5%P~p6s6Rn*Iw^25+W#I%iGPq
z74%(HzuDJ6346puNunIUK7%JuY3mOf;RIvzSYkHrsUD-=?M`z?p=yK6OC;qve!fyb
z1~&0MFF&u@<8otDS0Gk3NHWGz(6CrI7FS%rqk*v{jz!q;6Y$q{tzcJ*G#yOT^+4Q=
z(m+1E6l-gScfa_T;cat8Xf)zcMrx?Re!@z#?|liAN%=n|47R;AFx#-K_+8A})fzgL)w3~%=hgxJ+e0`#>;jP6oTp?r?-nYguh
z8-Tgv=fh5rN$NeSD;>5~AyXJKs-*>MF3UR_rkS0Bs)oAr0WEzYta8K&Ff|9*T!LKs
zTTE&gV9b(3Fp{@KM~{u_1P)y`TVaz_j2nD54PiVDxHm80M%<391&C-gHNt9GNW`8_
z_ik&OrLv|}ZcfZ@)OvDti}3z3kun1a9qiIhj_u4;MgtpjX_y8Gcr=3X1c;iGVYf?H
z>N>$H)g4e~GsBs(uhfma7{GpRTA?TB(?#qR!Z#b!E8a(?o)fda{udeB)v0s&}kq>~*V5>?=hy=HDZ
zAB4twRmo}S^fjQy?)v(xJ^wP2sg+vx8an3C)_3jiF88h3eeygGdMj-<;~u{hL=AYl
z8vG$<8T2W8W1o1viwOvQff|*u2)sx`)XIl6|GlhX)fHzH7=CHN4Trk4D4`cwV?vO1
z3H`^+Xq6-hHEW1d6+N`YQINWUo8oU!0-b1ahDttp&c7R+uEmvZu%TEtWI*`yd~sFe
zOvh7ueM)X~m}DYCdq?pZ_n%J;H;fzHQ>cK~SL{bHm-^nIfc-%tuu$v*AFJ;jQ
z`;U`k;O{)bCwVcd#-WL7o^T)`ymmPen;e@Xf1TYD3V5hl@*-R^;3TmUH)#!f2g_jn
zQww8ccDB`AJMW!1AD#nICRyAUHk#Mf+jTeoQT=pJORRqJJi-TvdP+3O0?Uwgucf^-
zeR85Jcxa?Ae_!k?^m`)&MweE#lZ^;h&N^MD8WyopZN4UM>qEnz4@+nvivjar09clUckbJ>AMnB^vAPBT%uiVfLbJT4siP#1>lCfc
zgy^LeAHXO^tP4s*H+*SG%o7_Yp^J~ZtT-4V$Gu*1y`)z|Ex!Q{w^c>*zd9ir#G>(9
z7CLQ$vbA@X8ya30o4Sf^oVbimcEnQUl?Z@
z+Mp(Ni``F@&(C3GAK=%c(7#_X@QM(x&9Ml#|A^Lp|Eti6nMjL`2iC@(>H0>UN4MwZPv#z-`j*_&B_o)EI*Q3W^b5mEeJ;wQP4n
z4vU?JR_r@88^3$G+;GddYDnYE@lhtThUb~it}_>KJOojIF30x98)zOh!~=pkuac?N
zh+LrWoP|K|Iv(0)-{7$Kh%&5dxnIO9%;I2+LqS04Q6rEVI2c=Lvi*JhW(KCBS^9(+niRsjX>$TO|$w~Uh3n)@HXm1gy6COHQx
zE7W70;LHEHP_`|%8TJA_4~yYd${y1Lp!#S3H)!N2>NCKO%dtn|PaX80M}Fq?IfeK
zMwcl^+o6BIboaPf6uNF26G?N{zSxTY@wt#I&M9R2;G(rO|8Ev87$b3io{8W~@%4~J
zS`$Rr$B#HKYb)+v`P9wm*dLvaYxI3yiyU^>Ei=73`=nFFE{4>cJw3dHz-JfkYJ(33I%;76znLJv3&?Qa
zXu48|ipqo1+x(A!;&cm6>GJ=n(Rr62CA>c
zGy~3ATwfy~elserIFJdLD}MqyNS$}tK=(FTr#MdGeCFm07ji>K1Uw_6`DupvX+!r7
z`$Vzo2{PHw>xu`;davt%q`|mBq;xKZ+Dfg+j8yJ3j2;WWmr{EIR?(HmU)T|9vEHVz
zW3V)ZpG{D&Qys@4RYuoiq848$yReLZQEP-_A<6L+-D_w6zwZOiQM&kzAG~J)TxQgl
z=@O&&b6Jm8&CoN1;u4_%O=W#$T4-Rm`%C(yv6q<{C!WREqm_sX
z^qNft2xiQO)==#3$Zj5Bx%ZEzo}PvbFSd(aaLB&u^10*}+Ook7Qd1~-)_`PprK8<&
zi+SIY7htY~>gazvRja~D+RyINee`_}I1nQU95&Sx#<5mW!iaxvU?c8EiN?V;+l}Ef
zsi*$5g)!#jableTF&(_Kb;!zE3&)u*^2l=|oPv;UX-bj~dltA3lVT=98A^;@T{W`WqfJlTqp*l>Hs
zi!C07d0&Hhes>%78CdnRqfpGx1R|B%ePwbQoBPP&DQf2j{L`c^_ur599Z>%+d9h*+
zy>`VT&{CZfdj5ClMZVGM42=TimUrE>>-S?qS7nmAVU(tN-J7Q#hn%aZq1Zk>ds+t=
z>h%xHb8|?E?A}cKGtW=p{Z~UEjJAROx0LsB*lYzui)&}y#aB9ZA0Cz`vQkVTzdK}<
z$|fn0!nS`eG;U@(Pj1a6<}wJzkiXgR#7+crQ}LJmw4wr$N-(X6S195gY=*fO!lO?=
z(8+=Iv~Ux&(8Xtx%X=&gkJ;HEX5$z(FwGU?{vX1TBA||)ySxWHZR(&kH~*-sbPWGf
zk5fD`n4e``!z|*FPgZ(D_aB3(G|%w;W>-T)<-T+Ft9dX1mN>lWFY9hd>2Mik>S7p;
z@GicIf*P-mq8JpLu_H<>#1kircTtysC|V})^}H`zKAHc;K=;b6?Q*C;;1GnmvLR!1
z6I#b7ak*!Y6NM2t9%xGcP^KPL86h2YBXQQpJl2l(oG5c{jJA_Sj1LRQW4s)Ii*6IL
zavS)^-7?_dug0t!)VMw>0XzRFKSI~118g?OaFM|YnNIV3*eSi}Ayi54fB6Bwycvv<
zK~q_uX`;;KO){Jk3>n!9B~=0&NWkZdJTpg
z5H!h(E_9ZCaBW3Yyq5GEQU>@cmA*;WY}yZto_K0*d2{mU0gm?-9J1<3^{?jnDerJR
znl&P;q8}3=Ftg+d57W2~mN%pSE~+Vc%t29MxC*L2HC|x2U@)A6vh*BEY^|}kRrBB<
z$gxHku_?TOdOg}JO{b$f4nDCL>o#sIdF;8e_BlX`P)XBg41$O<*T0Mjjuvn1*B}*2
zjnuON7kxu@X)syv(I?|L55>ao5ku^V*}X&(Sy`ps=q)=O=*8VD+CTnx4{;4g)K>h!
zsgA-zPH|6tqiBRo^+c8f)GfOlQIbvVtPs=TQd3OC$>}^g|BP!f9%7X20+rQ@LB7~*
z+f}_^dIV-?aGRHR3w;^)3mK5O>iNf=Y)VlT@f$pF25L=Wou0z?q-~}*
z;HbHJv!0FT?oz$4?)PG$WVJ!%x`#yZ7UAM8TtGM=rai#(a~5p{Vhb0x`8hs>MpTBB
zb!eM@O7H>gEPE^AOXRE#PhE{HIF;NWF#!EM1&^&tW>ovq0?$?{@%Yl}>S~1T5ThuG
z*5)rpIfXN-MCP4vQ|rw%dB9NVCaMf}5r4$FkW0`Qlc1N)f)!*d*ZDRtvonqd;G@#|
zm&~930n~Q+3UnT3oMSzxA5eT1iH3?<)m!A4HLle%ku!LVEOSC;KgXEi6+~=>F87nc
z{iYX=-x19gS_IwZXw|KOuVDMBc+6BuVvgaDjZnZ`+BW~hm;|$}XU!;)&Y=&Edn>GO
z-Jq9riVafAS5gch5@bz4PiwVfEvvZrLa#;vg=5Lkhrk(7qH#7TZAv?n!OQap^i6QQ
zEQW3b$oa0Dgm6k8;PqhIU+gE{YY`2gyN-9&4WvPtI6B+htdR5u4Vdj^u`f0wo4+jVDOvxt&xIYlVkgeF``O^Avz!(473f0
z7zPP6uZ86l)E!wi|6)Ax$yLcgMdS|z(HmIxtrjESHKqirvdR;+uVV$9n^%Z+*hBmr
z?1j*=c2CY$wB;M=K1>z1X)@vCm`F^m4r25C@E!*qm;1QS581Fo&`#{X@ehyxt=c4B
zv2$1*mE+3Q9b>HQ)cUfZ8sGc73KNA)VD;U0-vPB>j#d78f=s8F(M~(
zO_~bQ^-m(*9Ky)`5*SR~6do6D?Hk?_(gZ_z{@$DLiB`u)1F(_=hW;&L2{&cxc3;+WxaJ(g!y<>(BempB>!t57K
zh8npymgd@0Vbs^Hm;P0|Cb&g&mSxdnDVR}-pdMj|guM)@AY8{^!J#+Hrk_yZ9#sEzzlLNw-o1k4;Il$e|qZJ=6ux&Cznz`d5e|h8u5zom$LYROo6uJcz0{$#)(Gt8_|)|p=ci*nWtDFlBE+KG<^7j5(u>jDjz~|s=J0Y{*xA2w7=0vQb^Q;UfJv~r
zx=ud&f^-8jx?|o^r6P9ON0ulnDD|F0>)lW2#jd9jF+S8y8#;52kDt7Bd#3<2&
za$Vb&kb2pxDF>U{pEySOO`PeGUB13uq6+!r-gQ;SY`&(pG1-xJSV>85A}zmEx>yY9
z-ser8GF!6-`u(^1YLxwTy-PpwXLHehZXvDEN!l}{5}90m)MQ^L7o_oVuzXtc?c>#V
zBpqF&4p#U%WH}@16Ll>p$)~}(K8vG&A#72ez1W8tt28k97sc>l!YMO$)b$KkGrp0i
zNar@>V1I2NS$rXaSf*pZQ;n^8HOA+77W89kXK^{`y2r?u=7WR+mkfgJOqE{O%!g(r
z=lTDvZ7358(*Q~4`*m#*sK(38`{wkp1_T4WW#2U-N5fEjIYL7(*v;Od!zdSSzzH|$
z`NQXW4*vHNfZmRXT-RfuOugwdn=2{vjVf5c7FC~Zh>X_vaz8f6yf2rz2p91p^`Lji
zzpQv_V2$Jpm~fuZCs#@El)X
zGH(tbVzN-R-d8srH%_zM0JYaol)m3=(O&65SO~S@)
zoTF{895OY)%}sTS;?G>T!2}A%ms-lxE|ljFCDr)*^{VfJQIiJVqqknEJz7M@o!OS>
zRhS?i@G8%
z9rD(y!d>({UL{a-yUDiedzEBJOI;b3Y2mf6M>TZG1GGYCk(qx%evz68*T!1KGRNg7
z;?PIneJ>{d5Piq%PFfpZm-9c|u2BV^Bz4hHBGTkhG=^6pu_GVGO{b%j_TSI`*Ym{O
zBB-nxUocL(^qQOOJnyv>A(VZlz@WL@7qdZwz8Z&fkN)mrx|?<|yX3(BJGW#&WQX2E
z-A7Z=$cCFjjs;<$qfKj0E}$l$i*qRKz<6;d+?3z^%0z%6o@9amL^ni-QTYnc7-RMz
zMa#6e;82}N?%37PC<_77)x5s^nBq?hA)J9hLW7mjg=s$QmAavRrVw^`TSSv`^VBbP
z$k!}SP!m-S2^k_IU;Liq;|$9UzM<5N(7IY9g*lSuG7Auc#o#GY>6#Vz;c7FILujrD
zS4}DHPpxN(e`-k7OstPi!_jI3>P>c&Z|)~zUs96GZy
zytZ;bjlX_7|fcKym@WQAZ2>T#<%lNvg;
ziUrLfte@|PT2lfsfue?l>lInU9Gb7DRZ(DJUwc@e)Z3=j^eeeefhKMCe!?(Qf$+-e
z1jF5f`;z=7SB`QqZM?VnC6UG}m($_hfp~8M8dq*(&E&?L|60SJ)nzV|n>s(Dj6dJx
A&Hw-a

literal 0
HcmV?d00001

diff --git a/priv/zstd/tests/gzip/Makefile b/priv/zstd/tests/gzip/Makefile
index a50350f..cca3109 100644
--- a/priv/zstd/tests/gzip/Makefile
+++ b/priv/zstd/tests/gzip/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/gzip/gzip-env.sh b/priv/zstd/tests/gzip/gzip-env.sh
index 120e52d..4570e81 100755
--- a/priv/zstd/tests/gzip/gzip-env.sh
+++ b/priv/zstd/tests/gzip/gzip-env.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/helin-segv.sh b/priv/zstd/tests/gzip/helin-segv.sh
index f182c80..b400c24 100644
--- a/priv/zstd/tests/gzip/helin-segv.sh
+++ b/priv/zstd/tests/gzip/helin-segv.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/help-version.sh b/priv/zstd/tests/gzip/help-version.sh
index ee0c19f..fcda1c3 100644
--- a/priv/zstd/tests/gzip/help-version.sh
+++ b/priv/zstd/tests/gzip/help-version.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 
 # Ensure that $SHELL is set to *some* value and exported.
 # This is required for dircolors, which would fail e.g., when
diff --git a/priv/zstd/tests/gzip/hufts.sh b/priv/zstd/tests/gzip/hufts.sh
index 9b9576c..49c3695 100644
--- a/priv/zstd/tests/gzip/hufts.sh
+++ b/priv/zstd/tests/gzip/hufts.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/init.sh b/priv/zstd/tests/gzip/init.sh
index 97e4e4b..ebd8410 100644
--- a/priv/zstd/tests/gzip/init.sh
+++ b/priv/zstd/tests/gzip/init.sh
@@ -13,7 +13,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 
 # Using this file in a test
 # =========================
diff --git a/priv/zstd/tests/gzip/keep.sh b/priv/zstd/tests/gzip/keep.sh
index ab9a218..f87b1a4 100644
--- a/priv/zstd/tests/gzip/keep.sh
+++ b/priv/zstd/tests/gzip/keep.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/list.sh b/priv/zstd/tests/gzip/list.sh
index 75912e1..e218d75 100644
--- a/priv/zstd/tests/gzip/list.sh
+++ b/priv/zstd/tests/gzip/list.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/memcpy-abuse.sh b/priv/zstd/tests/gzip/memcpy-abuse.sh
index 7d5c056..1478890 100644
--- a/priv/zstd/tests/gzip/memcpy-abuse.sh
+++ b/priv/zstd/tests/gzip/memcpy-abuse.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/mixed.sh b/priv/zstd/tests/gzip/mixed.sh
index 383a54f..b47f4a5 100644
--- a/priv/zstd/tests/gzip/mixed.sh
+++ b/priv/zstd/tests/gzip/mixed.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/null-suffix-clobber.sh b/priv/zstd/tests/gzip/null-suffix-clobber.sh
index 0efd0e3..5acfb32 100644
--- a/priv/zstd/tests/gzip/null-suffix-clobber.sh
+++ b/priv/zstd/tests/gzip/null-suffix-clobber.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/stdin.sh b/priv/zstd/tests/gzip/stdin.sh
index eef4cd8..d697ab8 100644
--- a/priv/zstd/tests/gzip/stdin.sh
+++ b/priv/zstd/tests/gzip/stdin.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/test-driver.sh b/priv/zstd/tests/gzip/test-driver.sh
index 649c084..0529cc8 100644
--- a/priv/zstd/tests/gzip/test-driver.sh
+++ b/priv/zstd/tests/gzip/test-driver.sh
@@ -16,7 +16,7 @@ scriptversion=2016-01-11.22; # UTC
 # GNU General Public License for more details.
 #
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 
 # As a special exception to the GNU General Public License, if you
 # distribute this file as part of a program that contains a
diff --git a/priv/zstd/tests/gzip/trailing-nul.sh b/priv/zstd/tests/gzip/trailing-nul.sh
index 7b15d5e..b33b98f 100644
--- a/priv/zstd/tests/gzip/trailing-nul.sh
+++ b/priv/zstd/tests/gzip/trailing-nul.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/unpack-invalid.sh b/priv/zstd/tests/gzip/unpack-invalid.sh
index fe8384d..ceda5ff 100644
--- a/priv/zstd/tests/gzip/unpack-invalid.sh
+++ b/priv/zstd/tests/gzip/unpack-invalid.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/z-suffix.sh b/priv/zstd/tests/gzip/z-suffix.sh
index a870a54..c0bf509 100644
--- a/priv/zstd/tests/gzip/z-suffix.sh
+++ b/priv/zstd/tests/gzip/z-suffix.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/zdiff.sh b/priv/zstd/tests/gzip/zdiff.sh
index d62a846..6e99b66 100644
--- a/priv/zstd/tests/gzip/zdiff.sh
+++ b/priv/zstd/tests/gzip/zdiff.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/zgrep-context.sh b/priv/zstd/tests/gzip/zgrep-context.sh
index c8648b7..d213426 100644
--- a/priv/zstd/tests/gzip/zgrep-context.sh
+++ b/priv/zstd/tests/gzip/zgrep-context.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/zgrep-f.sh b/priv/zstd/tests/gzip/zgrep-f.sh
index d0cf27f..1e73ed2 100644
--- a/priv/zstd/tests/gzip/zgrep-f.sh
+++ b/priv/zstd/tests/gzip/zgrep-f.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/zgrep-signal.sh b/priv/zstd/tests/gzip/zgrep-signal.sh
index a8c5388..dd8442c 100644
--- a/priv/zstd/tests/gzip/zgrep-signal.sh
+++ b/priv/zstd/tests/gzip/zgrep-signal.sh
@@ -15,7 +15,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/gzip/znew-k.sh b/priv/zstd/tests/gzip/znew-k.sh
index 6c239e2..5cf99ed 100644
--- a/priv/zstd/tests/gzip/znew-k.sh
+++ b/priv/zstd/tests/gzip/znew-k.sh
@@ -14,7 +14,7 @@
 # GNU General Public License for more details.
 
 # You should have received a copy of the GNU General Public License
-# along with this program.  If not, see .
+# along with this program.  If not, see .
 # limit so don't run it by default.
 
 . "${srcdir=.}/init.sh"; path_prepend_ .
diff --git a/priv/zstd/tests/invalidDictionaries.c b/priv/zstd/tests/invalidDictionaries.c
index b71f741..66caa9e 100644
--- a/priv/zstd/tests/invalidDictionaries.c
+++ b/priv/zstd/tests/invalidDictionaries.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/legacy.c b/priv/zstd/tests/legacy.c
index ac4938f..3be3864 100644
--- a/priv/zstd/tests/legacy.c
+++ b/priv/zstd/tests/legacy.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/libzstd_partial_builds.sh b/priv/zstd/tests/libzstd_builds.sh
similarity index 92%
rename from priv/zstd/tests/libzstd_partial_builds.sh
rename to priv/zstd/tests/libzstd_builds.sh
index 05dad8f..f9e1e76 100755
--- a/priv/zstd/tests/libzstd_partial_builds.sh
+++ b/priv/zstd/tests/libzstd_builds.sh
@@ -23,14 +23,19 @@ mustBeAbsent() {
 
 # default compilation : all features enabled - no zbuff
 $ECHO "testing default library compilation"
-CFLAGS= make -C $DIR/../lib libzstd.a > $INTOVOID
+CFLAGS= make -C $DIR/../lib libzstd libzstd.a > $INTOVOID
 nm $DIR/../lib/libzstd.a | $GREP "\.o" > tmplog
 isPresent "zstd_compress.o"
 isPresent "zstd_decompress.o"
 isPresent "zdict.o"
 isPresent "zstd_v07.o"
 mustBeAbsent "zbuff_compress.o"
-$RM $DIR/../lib/libzstd.a tmplog
+$RM tmplog
+
+# Check that the exec-stack bit isn't set
+readelf -lW $DIR/../lib/libzstd.so | $GREP "GNU_STACK" > tmplog
+mustBeAbsent "RWE"
+$RM $DIR/../lib/libzstd.a $DIR/../lib/libzstd.so* tmplog
 
 # compression disabled => also disable zdict
 $ECHO "testing with compression disabled"
diff --git a/priv/zstd/tests/longmatch.c b/priv/zstd/tests/longmatch.c
index a171c0e..547b261 100644
--- a/priv/zstd/tests/longmatch.c
+++ b/priv/zstd/tests/longmatch.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -36,27 +36,27 @@ compress(ZSTD_CStream *ctx, ZSTD_outBuffer out, const void *data, size_t size)
 int main(int argc, const char** argv)
 {
   ZSTD_CStream* ctx;
-  ZSTD_parameters params;
-  size_t rc;
-  unsigned windowLog;
+  unsigned windowLog = 18;
   (void)argc;
   (void)argv;
   /* Create stream */
-  ctx = ZSTD_createCStream();
+  ctx = ZSTD_createCCtx();
   if (!ctx) { return 1; }
   /* Set parameters */
-  memset(¶ms, 0, sizeof(params));
-  params.cParams.windowLog = 18;
-  params.cParams.chainLog = 13;
-  params.cParams.hashLog = 14;
-  params.cParams.searchLog = 1;
-  params.cParams.minMatch = 7;
-  params.cParams.targetLength = 16;
-  params.cParams.strategy = ZSTD_fast;
-  windowLog = params.cParams.windowLog;
-  /* Initialize stream */
-  rc = ZSTD_initCStream_advanced(ctx, NULL, 0, params, 0);
-  if (ZSTD_isError(rc)) { return 2; }
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_windowLog, windowLog)))
+    return 2;
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_chainLog, 13)))
+    return 2;
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_hashLog, 14)))
+    return 2;
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_searchLog, 1)))
+    return 2;
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_minMatch, 7)))
+    return 2;
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_targetLength, 16)))
+    return 2;
+  if (ZSTD_isError(ZSTD_CCtx_setParameter(ctx, ZSTD_c_strategy, ZSTD_fast)))
+    return 2;
   {
     U64 compressed = 0;
     const U64 toCompress = ((U64)1) << 33;
@@ -97,5 +97,6 @@ int main(int argc, const char** argv)
     free(srcBuffer);
     free(dstBuffer);
   }
+  ZSTD_freeCCtx(ctx);
   return 0;
 }
diff --git a/priv/zstd/tests/paramgrill.c b/priv/zstd/tests/paramgrill.c
index a0cfa58..8971c65 100644
--- a/priv/zstd/tests/paramgrill.c
+++ b/priv/zstd/tests/paramgrill.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -555,7 +555,7 @@ static int feasible(const BMK_benchResult_t results, const constraint_t target)
 }
 
 /* hill climbing value for part 1 */
-/* Scoring here is a linear reward for all set constraints normalized between 0 to 1
+/* Scoring here is a linear reward for all set constraints normalized between 0 and 1
  * (with 0 at 0 and 1 being fully fulfilling the constraint), summed with a logarithmic
  * bonus to exceeding the constraint value. We also give linear ratio for compression ratio.
  * The constant factors are experimental.
@@ -566,10 +566,10 @@ resultScore(const BMK_benchResult_t res, const size_t srcSize, const constraint_
     double cs = 0., ds = 0., rt, cm = 0.;
     const double r1 = 1, r2 = 0.1, rtr = 0.5;
     double ret;
-    if(target.cSpeed) { cs = res.cSpeed / (double)target.cSpeed; }
-    if(target.dSpeed) { ds = res.dSpeed / (double)target.dSpeed; }
-    if(target.cMem != (U32)-1) { cm = (double)target.cMem / res.cMem; }
-    rt = ((double)srcSize / res.cSize);
+    if(target.cSpeed) { cs = (double)res.cSpeed / (double)target.cSpeed; }
+    if(target.dSpeed) { ds = (double)res.dSpeed / (double)target.dSpeed; }
+    if(target.cMem != (U32)-1) { cm = (double)target.cMem / (double)res.cMem; }
+    rt = ((double)srcSize / (double)res.cSize);
 
     ret = (MIN(1, cs) + MIN(1, ds)  + MIN(1, cm))*r1 + rt * rtr +
          (MAX(0, log(cs))+ MAX(0, log(ds))+ MAX(0, log(cm))) * r2;
@@ -581,8 +581,8 @@ resultScore(const BMK_benchResult_t res, const size_t srcSize, const constraint_
 static double
 resultDistLvl(const BMK_benchResult_t result1, const BMK_benchResult_t lvlRes)
 {
-    double normalizedCSpeedGain1 = ((double)result1.cSpeed / lvlRes.cSpeed) - 1;
-    double normalizedRatioGain1 = ((double)lvlRes.cSize / result1.cSize) - 1;
+    double normalizedCSpeedGain1 = ((double)result1.cSpeed / (double)lvlRes.cSpeed) - 1;
+    double normalizedRatioGain1 = ((double)lvlRes.cSize / (double)result1.cSize) - 1;
     if(normalizedRatioGain1 < 0 || normalizedCSpeedGain1 < 0) {
         return 0.0;
     }
@@ -854,7 +854,7 @@ BMK_displayOneResult(FILE* f, winnerInfo_t res, const size_t srcSize)
     }
 
     {   double const ratio = res.result.cSize ?
-                            (double)srcSize / res.result.cSize : 0;
+                            (double)srcSize / (double)res.result.cSize : 0;
         double const cSpeedMBps = (double)res.result.cSpeed / MB_UNIT;
         double const dSpeedMBps = (double)res.result.dSpeed / MB_UNIT;
 
@@ -937,7 +937,7 @@ BMK_printWinnerOpt(FILE* f, const U32 cLevel, const BMK_benchResult_t result, co
         }
         fprintf(f, "================================\n");
         fprintf(f, "Level Bounds: R: > %.3f AND C: < %.1f MB/s \n\n",
-            (double)srcSize / g_lvltarget.cSize, (double)g_lvltarget.cSpeed / MB_UNIT);
+            (double)srcSize / (double)g_lvltarget.cSize, (double)g_lvltarget.cSpeed / MB_UNIT);
 
 
         fprintf(f, "Overall Winner: \n");
@@ -977,7 +977,7 @@ BMK_print_cLevelEntry(FILE* f, const int cLevel,
     }
     /* print comment */
     {   double const ratio = result.cSize ?
-                            (double)srcSize / result.cSize : 0;
+                            (double)srcSize / (double)result.cSize : 0;
         double const cSpeedMBps = (double)result.cSpeed / MB_UNIT;
         double const dSpeedMBps = (double)result.dSpeed / MB_UNIT;
 
@@ -1726,19 +1726,19 @@ static int allBench(BMK_benchResult_t* resultPtr,
 
     /* calculate uncertainty in compression / decompression runs */
     if (benchres.cSpeed) {
-        U64 const loopDurationC = (((U64)buf.srcSize * TIMELOOP_NANOSEC) / benchres.cSpeed);
+        double const loopDurationC = (double)(((U64)buf.srcSize * TIMELOOP_NANOSEC) / benchres.cSpeed);
         uncertaintyConstantC = ((loopDurationC + (double)(2 * g_clockGranularity))/loopDurationC);
     }
 
     if (benchres.dSpeed) {
-        U64 const loopDurationD = (((U64)buf.srcSize * TIMELOOP_NANOSEC) / benchres.dSpeed);
+        double const loopDurationD = (double)(((U64)buf.srcSize * TIMELOOP_NANOSEC) / benchres.dSpeed);
         uncertaintyConstantD = ((loopDurationD + (double)(2 * g_clockGranularity))/loopDurationD);
     }
 
     /* optimistic assumption of benchres */
     {   BMK_benchResult_t resultMax = benchres;
-        resultMax.cSpeed = (unsigned long long)(resultMax.cSpeed * uncertaintyConstantC * VARIANCE);
-        resultMax.dSpeed = (unsigned long long)(resultMax.dSpeed * uncertaintyConstantD * VARIANCE);
+        resultMax.cSpeed = (unsigned long long)((double)resultMax.cSpeed * uncertaintyConstantC * VARIANCE);
+        resultMax.dSpeed = (unsigned long long)((double)resultMax.dSpeed * uncertaintyConstantD * VARIANCE);
 
         /* disregard infeasible results in feas mode */
         /* disregard if resultMax < winner in infeas mode */
@@ -1850,8 +1850,8 @@ static int BMK_seed(winnerInfo_t* winners,
 
         if ((double)testResult.cSize <= ((double)winners[cLevel].result.cSize * (1. + (0.02 / cLevel))) ) {
             /* Validate solution is "good enough" */
-            double W_ratio = (double)buf.srcSize / testResult.cSize;
-            double O_ratio = (double)buf.srcSize / winners[cLevel].result.cSize;
+            double W_ratio = (double)buf.srcSize / (double)testResult.cSize;
+            double O_ratio = (double)buf.srcSize / (double)winners[cLevel].result.cSize;
             double W_ratioNote = log (W_ratio);
             double O_ratioNote = log (O_ratio);
             size_t W_DMemUsed = (1 << params.vals[wlog_ind]) + (16 KB);
@@ -1864,11 +1864,11 @@ static int BMK_seed(winnerInfo_t* winners,
             double W_CMemUsed_note = W_ratioNote * ( 50 + 13*cLevel) - log((double)W_CMemUsed);
             double O_CMemUsed_note = O_ratioNote * ( 50 + 13*cLevel) - log((double)O_CMemUsed);
 
-            double W_CSpeed_note = W_ratioNote * (double)( 30 + 10*cLevel) + log(testResult.cSpeed);
-            double O_CSpeed_note = O_ratioNote * (double)( 30 + 10*cLevel) + log(winners[cLevel].result.cSpeed);
+            double W_CSpeed_note = W_ratioNote * (double)( 30 + 10*cLevel) + log((double)testResult.cSpeed);
+            double O_CSpeed_note = O_ratioNote * (double)( 30 + 10*cLevel) + log((double)winners[cLevel].result.cSpeed);
 
-            double W_DSpeed_note = W_ratioNote * (double)( 20 + 2*cLevel) + log(testResult.dSpeed);
-            double O_DSpeed_note = O_ratioNote * (double)( 20 + 2*cLevel) + log(winners[cLevel].result.dSpeed);
+            double W_DSpeed_note = W_ratioNote * (double)( 20 + 2*cLevel) + log((double)testResult.dSpeed);
+            double O_DSpeed_note = O_ratioNote * (double)( 20 + 2*cLevel) + log((double)winners[cLevel].result.dSpeed);
 
             if (W_DMemUsed_note < O_DMemUsed_note) {
                 /* uses too much Decompression memory for too little benefit */
@@ -2652,7 +2652,7 @@ static int usage_advanced(void)
                                 (unsigned)g_timeLimit_s, (double)g_timeLimit_s / 3600);
     DISPLAY( " -v           : Prints Benchmarking output\n");
     DISPLAY( " -D           : Next argument dictionary file\n");
-    DISPLAY( " -s           : Seperate Files\n");
+    DISPLAY( " -s           : Separate Files\n");
     return 0;
 }
 
@@ -2707,7 +2707,7 @@ int main(int argc, const char** argv)
     const char* dictFileName = NULL;
     U32 main_pause = 0;
     int cLevelOpt = 0, cLevelRun = 0;
-    int seperateFiles = 0;
+    int separateFiles = 0;
     double compressibility = COMPRESSIBILITY_DEFAULT;
     U32 memoTableLog = PARAM_UNSET;
     constraint_t target = { 0, 0, (U32)-1 };
@@ -2895,7 +2895,7 @@ int main(int argc, const char** argv)
 
                 case 's':
                     argument++;
-                    seperateFiles = 1;
+                    separateFiles = 1;
                     break;
 
                 case 'q':
@@ -2940,7 +2940,7 @@ int main(int argc, const char** argv)
             result = benchSample(compressibility, cLevelRun);
         }
     } else {
-        if(seperateFiles) {
+        if(separateFiles) {
             for(i = 0; i < argc - filenamesStart; i++) {
                 if (g_optimizer) {
                     result = optimizeForSize(argv+filenamesStart + i, 1, dictFileName, target, paramTarget, cLevelOpt, cLevelRun, memoTableLog);
diff --git a/priv/zstd/tests/playTests.sh b/priv/zstd/tests/playTests.sh
index fa748c0..5f595f6 100755
--- a/priv/zstd/tests/playTests.sh
+++ b/priv/zstd/tests/playTests.sh
@@ -2,6 +2,10 @@
 
 set -e
 
+unset ZSTD_CLEVEL
+unset ZSTD_NBTHREADS
+
+
 die() {
     println "$@" 1>&2
     exit 1
@@ -88,7 +92,6 @@ SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
 PRGDIR="$SCRIPT_DIR/../programs"
 TESTDIR="$SCRIPT_DIR/../tests"
 UNAME=$(uname)
-ZSTDGREP="$PRGDIR/zstdgrep"
 
 detectedTerminal=false
 if [ -t 0 ] && [ -t 1 ]
@@ -124,6 +127,13 @@ case "$UNAME" in
     Darwin | FreeBSD | OpenBSD | NetBSD) MTIME="stat -f %m" ;;
 esac
 
+assertSameMTime() {
+    MT1=$($MTIME "$1")
+    MT2=$($MTIME "$2")
+    echo MTIME $MT1 $MT2
+    [ "$MT1" = "$MT2" ] || die "mtime on $1 doesn't match mtime on $2 ($MT1 != $MT2)"
+}
+
 GET_PERMS="stat -c %a"
 case "$UNAME" in
     Darwin | FreeBSD | OpenBSD | NetBSD) GET_PERMS="stat -f %Lp" ;;
@@ -159,10 +169,11 @@ if [ -z "${DATAGEN_BIN}" ]; then
   DATAGEN_BIN="$TESTDIR/datagen"
 fi
 
-ZSTD_BIN="$EXE_PREFIX$ZSTD_BIN"
+# Why was this line here ? Generates a strange ZSTD_BIN when EXE_PREFIX is non empty
+# ZSTD_BIN="$EXE_PREFIX$ZSTD_BIN"
 
 # assertions
-[ -n "$ZSTD_BIN" ] || die "zstd not found at $ZSTD_BIN! \n Please define ZSTD_BIN pointing to the zstd binary. You might also consider rebuilding zstd follwing the instructions in README.md"
+[ -n "$ZSTD_BIN" ] || die "zstd not found at $ZSTD_BIN! \n Please define ZSTD_BIN pointing to the zstd binary. You might also consider rebuilding zstd following the instructions in README.md"
 [ -n "$DATAGEN_BIN" ] || die "datagen not found at $DATAGEN_BIN! \n Please define DATAGEN_BIN pointing to the datagen binary. You might also consider rebuilding zstd tests following the instructions in README.md. "
 println "\nStarting playTests.sh isWindows=$isWindows EXE_PREFIX='$EXE_PREFIX' ZSTD_BIN='$ZSTD_BIN' DATAGEN_BIN='$DATAGEN_BIN'"
 
@@ -174,12 +185,19 @@ else
 fi
 
 
+zstd -vvV
 
 println "\n===>  simple tests "
 
 datagen > tmp
+zstd -h
+zstd -H
+zstd -V
 println "test : basic compression "
 zstd -f tmp                      # trivial compression case, creates tmp.zst
+zstd -f -z tmp
+zstd -f -k tmp
+zstd -f -C tmp
 println "test : basic decompression"
 zstd -df tmp.zst                 # trivial decompression case (overwrites tmp)
 println "test : too large compression level => auto-fix"
@@ -194,6 +212,7 @@ zstd -c --fast=0 tmp > $INTOVOID && die "--fast must not accept value 0"
 println "test : too large numeric argument"
 zstd --fast=9999999999 -f tmp  && die "should have refused numeric value"
 println "test : set compression level with environment variable ZSTD_CLEVEL"
+
 ZSTD_CLEVEL=12  zstd -f tmp # positive compression level
 ZSTD_CLEVEL=-12 zstd -f tmp # negative compression level
 ZSTD_CLEVEL=+12 zstd -f tmp # valid: verbose '+' sign
@@ -205,6 +224,11 @@ ZSTD_CLEVEL=3a7 zstd -f tmp # malformed env var, warn and revert to default sett
 ZSTD_CLEVEL=50000000000 zstd -f tmp # numeric value too large, warn and revert to default setting
 println "test : override ZSTD_CLEVEL with command line option"
 ZSTD_CLEVEL=12  zstd --fast=3 -f tmp # overridden by command line option
+
+# temporary envvar changes in the above tests would actually persist in macos /bin/sh
+unset ZSTD_CLEVEL
+
+
 println "test : compress to stdout"
 zstd tmp -c > tmpCompressed
 zstd tmp --stdout > tmpCompressed       # long command format
@@ -244,10 +268,13 @@ zstd -dc - < tmp.zst > $INTOVOID
 zstd -d    < tmp.zst > $INTOVOID   # implicit stdout when stdin is used
 zstd -d  - < tmp.zst > $INTOVOID
 println "test : impose memory limitation (must fail)"
-zstd -d -f tmp.zst -M2K -c > $INTOVOID && die "decompression needs more memory than allowed"
-zstd -d -f tmp.zst --memlimit=2K -c > $INTOVOID && die "decompression needs more memory than allowed"  # long command
-zstd -d -f tmp.zst --memory=2K -c > $INTOVOID && die "decompression needs more memory than allowed"  # long command
-zstd -d -f tmp.zst --memlimit-decompress=2K -c > $INTOVOID && die "decompression needs more memory than allowed"  # long command
+datagen -g500K > tmplimit
+zstd -f tmplimit
+zstd -d -f tmplimit.zst -M2K -c > $INTOVOID && die "decompression needs more memory than allowed"
+zstd -d -f tmplimit.zst --memlimit=2K -c > $INTOVOID && die "decompression needs more memory than allowed"  # long command
+zstd -d -f tmplimit.zst --memory=2K -c > $INTOVOID && die "decompression needs more memory than allowed"  # long command
+zstd -d -f tmplimit.zst --memlimit-decompress=2K -c > $INTOVOID && die "decompression needs more memory than allowed"  # long command
+rm -f tmplimit tmplimit.zst
 println "test : overwrite protection"
 zstd -q tmp && die "overwrite check failed!"
 println "test : force overwrite"
@@ -263,6 +290,10 @@ zstd -q -f tmpro
 println "test: --no-progress flag"
 zstd tmpro -c --no-progress | zstd -d -f -o "$INTOVOID" --no-progress
 zstd tmpro -cv --no-progress | zstd -dv -f -o "$INTOVOID" --no-progress
+println "test: --progress flag"
+zstd tmpro -c | zstd -d -f -o "$INTOVOID" --progress 2>&1 | grep -E "[A-Za-z0-9._ ]+: [0-9]+ bytes"
+zstd tmpro -c | zstd -d -f -q -o "$INTOVOID" --progress 2>&1 | grep -E "[A-Za-z0-9._ ]+: [0-9]+ bytes"
+zstd tmpro -c | zstd -d -f -v -o "$INTOVOID" 2>&1 | grep -E "[A-Za-z0-9._ ]+: [0-9]+ bytes"
 rm -f tmpro tmpro.zst
 println "test: overwrite input file (must fail)"
 zstd tmp -fo tmp && die "zstd compression overwrote the input file"
@@ -286,16 +317,12 @@ zstd -d -f tmp_corrupt.zst --no-check
 zstd -d -f tmp_corrupt.zst --check --no-check # final flag overrides
 zstd -d -f tmp.zst --no-check
 
-println "\n===> zstdgrep tests"
-ln -sf "$ZSTD_BIN" zstdcat
-rm -f tmp_grep
-echo "1234" > tmp_grep
-zstd -f tmp_grep
-lines=$(ZCAT=./zstdcat "$ZSTDGREP" 2>&1 "1234" tmp_grep tmp_grep.zst | wc -l)
-test 2 -eq $lines
-ZCAT=./zstdcat "$ZSTDGREP" 2>&1 "1234" tmp_grep_bad.zst && die "Should have failed"
-ZCAT=./zstdcat "$ZSTDGREP" 2>&1 "1234" tmp_grep_bad.zst | grep "No such file or directory" || true
-rm -f tmp_grep*
+if [ "$isWindows" = false ] && [ "$UNAME" != "AIX" ]; then
+  if [ -n "$(which readelf)" ]; then
+    println "test: check if binary has executable stack (#2963)"
+    readelf -lW "$ZSTD_BIN" | grep 'GNU_STACK .* RW ' || die "zstd binary has executable stack!"
+  fi
+fi
 
 println "\n===>  --exclude-compressed flag"
 rm -rf precompressedFilterTestDir
@@ -325,6 +352,7 @@ zstd --long --rm -r precompressedFilterTestDir
 # Files should get compressed again without the --exclude-compressed flag.
 test -f precompressedFilterTestDir/input.5.zst.zst
 test -f precompressedFilterTestDir/input.6.zst.zst
+rm -rf precompressedFilterTestDir
 println "Test completed"
 
 
@@ -338,8 +366,8 @@ zstd < tmpPrompt -o tmpPrompt.zst -f    # should successfully overwrite with -f
 zstd -q -d -f tmpPrompt.zst -o tmpPromptRegenerated
 $DIFF tmpPromptRegenerated tmpPrompt    # the first 'y' character should not be swallowed
 
-echo 'yes' | zstd tmpPrompt -o tmpPrompt.zst  # accept piped "y" input to force overwrite when using files
-echo 'yes' | zstd < tmpPrompt -o tmpPrompt.zst && die "should have aborted immediately and failed to overwrite"
+echo 'yes' | zstd tmpPrompt -v -o tmpPrompt.zst  # accept piped "y" input to force overwrite when using files
+echo 'yes' | zstd < tmpPrompt -v -o tmpPrompt.zst && die "should have aborted immediately and failed to overwrite"
 zstd tmpPrompt - < tmpPrompt -o tmpPromp.zst --rm && die "should have aborted immediately and failed to remove"
 
 println "Test completed"
@@ -349,18 +377,45 @@ println "\n===>  recursive mode test "
 # combination of -r with empty list of input file
 zstd -c -r < tmp > tmp.zst
 
+# combination of -r with empty folder
+mkdir -p tmpEmptyDir
+zstd -r tmpEmptyDir
+rm -rf tmpEmptyDir
+
 
 println "\n===>  file removal"
 zstd -f --rm tmp
 test ! -f tmp  # tmp should no longer be present
 zstd -f -d --rm tmp.zst
-test ! -f tmp.zst   # tmp.zst should no longer be present
+test ! -f tmp.zst  # tmp.zst should no longer be present
+println "test: --rm is disabled when output is stdout"
+test -f tmp
+zstd --rm tmp -c > $INTOVOID
+test -f tmp # tmp shall still be there
+zstd -f --rm tmp -c > $INTOVOID
+test -f tmp # tmp shall still be there
+zstd -f tmp -c > $INTOVOID --rm
+test -f tmp # tmp shall still be there
+println "test: --rm is disabled when multiple inputs are concatenated into a single output"
+cp tmp tmp2
+zstd --rm tmp tmp2 -c > $INTOVOID
+test -f tmp
+test -f tmp2
+rm -f tmp3.zst
+echo 'y' | zstd -v tmp tmp2 -o tmp3.zst --rm # prompt for confirmation
+test -f tmp
+test -f tmp2
+zstd -f tmp tmp2 -o tmp3.zst --rm # just warns, no prompt
+test -f tmp
+test -f tmp2
+zstd -q tmp tmp2 -o tmp3.zst --rm && die "should refuse to concatenate"
+
 println "test : should quietly not remove non-regular file"
 println hello > tmp
 zstd tmp -f -o "$DEVDEVICE" 2>tmplog > "$INTOVOID"
-grep -v "Refusing to remove non-regular file" tmplog
+grep "Refusing to remove non-regular file" tmplog && die
 rm -f tmplog
-zstd tmp -f -o "$INTOVOID" 2>&1 | grep -v "Refusing to remove non-regular file"
+zstd tmp -f -o "$INTOVOID" 2>&1 | grep "Refusing to remove non-regular file" && die
 println "test : --rm on stdin"
 println a | zstd --rm > $INTOVOID   # --rm should remain silent
 rm -f tmp
@@ -385,44 +440,46 @@ println "\n===>  decompression only tests "
 dd bs=1048576 count=1 if=/dev/zero of=tmp
 zstd -d -o tmp1 "$TESTDIR/golden-decompression/rle-first-block.zst"
 $DIFF -s tmp1 tmp
-rm -f tmp*
 
+touch tmp_empty
+zstd -d -o tmp2 "$TESTDIR/golden-decompression/empty-block.zst"
+$DIFF -s tmp2 tmp_empty
+rm -f tmp*
 
 println "\n===>  compress multiple files"
 println hello > tmp1
 println world > tmp2
 zstd tmp1 tmp2 -o "$INTOVOID" -f
 zstd tmp1 tmp2 -c | zstd -t
-zstd tmp1 tmp2 -o tmp.zst
+echo 'y' | zstd -v tmp1 tmp2 -o tmp.zst
 test ! -f tmp1.zst
 test ! -f tmp2.zst
 zstd tmp1 tmp2
 zstd -t tmp1.zst tmp2.zst
 zstd -dc tmp1.zst tmp2.zst
 zstd tmp1.zst tmp2.zst -o "$INTOVOID" -f
-zstd -d tmp1.zst tmp2.zst -o tmp
+echo 'y' | zstd -v -d tmp1.zst tmp2.zst -o tmp
 touch tmpexists
 zstd tmp1 tmp2 -f -o tmpexists
 zstd tmp1 tmp2 -q -o tmpexists && die "should have refused to overwrite"
 println gooder > tmp_rm1
 println boi > tmp_rm2
 println worldly > tmp_rm3
-echo 'y' | zstd tmp_rm1 tmp_rm2 -o tmp_rm3.zst --rm     # tests the warning prompt for --rm with multiple inputs into once source
-test ! -f tmp_rm1
-test ! -f tmp_rm2
+echo 'y' | zstd -v tmp_rm1 tmp_rm2 -v -o tmp_rm3.zst
+test -f tmp_rm1
+test -f tmp_rm2
 cp tmp_rm3.zst tmp_rm4.zst
-echo 'Y' | zstd -d tmp_rm3.zst tmp_rm4.zst -o tmp_rm_out --rm
-test ! -f tmp_rm3.zst
-test ! -f tmp_rm4.zst
-echo 'yes' | zstd tmp_rm_out tmp_rm3 -c --rm && die "compressing multiple files to stdout with --rm should fail unless -f is specified"
-echo 'yes' | zstd tmp_rm_out tmp_rm3 -c --rm -v && die "compressing multiple files to stdout with --rm should fail unless -f is specified"
+echo 'Y' | zstd -v -d tmp_rm3.zst tmp_rm4.zst -v -o tmp_rm_out --rm
+test -f tmp_rm3.zst
+test -f tmp_rm4.zst
 println gooder > tmpexists1
 zstd tmpexists1 tmpexists -c --rm -f > $INTOVOID
-
 # Bug: PR #972
 if [ "$?" -eq 139 ]; then
   die "should not have segfaulted"
 fi
+test -f tmpexists1
+test -f tmpexists
 println "\n===>  multiple files and shell completion "
 datagen -s1        > tmp1 2> $INTOVOID
 datagen -s2 -g100K > tmp2 2> $INTOVOID
@@ -526,19 +583,9 @@ if [ "$isWindows" = false ] ; then
     zstd -f -d tmp1.zst -o tmp1.out
     assertFilePermissions tmp1.out 400
 
-    rm -f tmp1.zst tmp1.out
-
     umask 0666
     chmod 0666 tmp1 tmp2
 
-    println "test : respect umask when copying permissions in file -> file compression "
-    zstd -f tmp1 -o tmp1.zst
-    assertFilePermissions tmp1.zst 0
-    println "test : respect umask when copying permissions in file -> file decompression "
-    chmod 0666 tmp1.zst
-    zstd -f -d tmp1.zst -o tmp1.out
-    assertFilePermissions tmp1.out 0
-
     rm -f tmp1.zst tmp1.out
 
     println "test : respect umask when compressing from stdin input "
@@ -583,6 +630,17 @@ if [ -n "$READFROMBLOCKDEVICE" ] ; then
     rm -f tmp.img tmp.img.zst tmp.img.copy
 fi
 
+println "\n===>  zstd created file timestamp tests"
+datagen > tmp
+touch -m -t 200001010000.00 tmp
+println "test : copy mtime in file -> file compression "
+zstd -f tmp -o tmp.zst
+assertSameMTime tmp tmp.zst
+println "test : copy mtime in file -> file decompression "
+zstd -f -d tmp.zst -o tmp.out
+assertSameMTime tmp.zst tmp.out
+rm -f tmp
+
 println "\n===>  compress multiple files into an output directory, --output-dir-flat"
 println henlo > tmp1
 mkdir tmpInputTestDir
@@ -689,9 +747,13 @@ test -f tmp2
 test -f tmp3
 test -f tmp4
 
+println "test : survive the list of files with too long filenames (--filelist=FILE)"
+datagen -g5M > tmp_badList
+zstd -qq -f --filelist=tmp_badList && die "should have failed : file name length is too long"  # printing very long text garbage on console will cause CI failure
+
 println "test : survive a list of files which is text garbage (--filelist=FILE)"
 datagen > tmp_badList
-zstd -f --filelist=tmp_badList && die "should have failed : list is text garbage"
+zstd -qq -f --filelist=tmp_badList && die "should have failed : list is text garbage"  # printing very long text garbage on console will cause CI failure
 
 println "test : survive a list of files which is binary garbage (--filelist=FILE)"
 datagen -P0 -g1M > tmp_badList
@@ -719,6 +781,7 @@ rm -rf tmp*
 println "test : show-default-cparams regular"
 datagen > tmp
 zstd --show-default-cparams -f tmp
+zstd --show-default-cparams -d tmp.zst && die "error: can't use --show-default-cparams in decompression mode"
 rm -rf tmp*
 
 println "test : show-default-cparams recursive"
@@ -729,6 +792,13 @@ datagen -g257000 > tmp_files/tmp3
 zstd --show-default-cparams -f -r tmp_files
 rm -rf tmp*
 
+println "test : show compression parameters in verbose mode"
+datagen > tmp
+zstd -vv tmp 2>&1 | \
+grep -q -E -- "--zstd=wlog=[[:digit:]]+,clog=[[:digit:]]+,hlog=[[:digit:]]+,\
+slog=[[:digit:]]+,mml=[[:digit:]]+,tlen=[[:digit:]]+,strat=[[:digit:]]+"
+rm -rf tmp*
+
 println "\n===>  Advanced compression parameters "
 println "Hello world!" | zstd --zstd=windowLog=21,      - -o tmp.zst && die "wrong parameters not detected!"
 println "Hello world!" | zstd --zstd=windowLo=21        - -o tmp.zst && die "wrong parameters not detected!"
@@ -749,6 +819,16 @@ println "Hello world 1!" | zstd -df
 println "Hello world 2!" | zstd -dcf
 println "Hello world 3!" > tmp1
 zstd -dcf tmp1
+println "" | zstd -df > tmp1
+println "" > tmp2
+$DIFF -q tmp1 tmp2
+println "1" | zstd -df > tmp1
+println "1" > tmp2
+$DIFF -q tmp1 tmp2
+println "12" | zstd -df > tmp1
+println "12" > tmp2
+$DIFF -q tmp1 tmp2
+rm -rf tmp*
 
 
 println "\n===>  frame concatenation "
@@ -757,6 +837,9 @@ println "world!" > world.tmp
 cat hello.tmp world.tmp > helloworld.tmp
 zstd -c hello.tmp > hello.zst
 zstd -c world.tmp > world.zst
+zstd -c hello.tmp world.tmp > helloworld.zst
+zstd -dc helloworld.zst > result.tmp
+$DIFF helloworld.tmp result.tmp
 cat hello.zst world.zst > helloworld.zst
 zstd -dc helloworld.zst > result.tmp
 cat result.tmp
@@ -787,7 +870,7 @@ rm -f ./*.tmp ./*.zstd
 println "frame concatenation tests completed"
 
 
-if [ "$isWindows" = false ] && [ "$UNAME" != 'SunOS' ] && [ "$UNAME" != "OpenBSD" ] ; then
+if [ "$isWindows" = false ] && [ "$UNAME" != 'SunOS' ] && [ "$UNAME" != "OpenBSD" ] && [ "$UNAME" != "AIX" ]; then
 println "\n**** flush write error test **** "
 
 println "println foo | zstd > /dev/full"
@@ -909,11 +992,23 @@ cat tmp | zstd -14 -f --size-hint=11050 | zstd -t  # slightly too high
 cat tmp | zstd -14 -f --size-hint=10950 | zstd -t  # slightly too low
 cat tmp | zstd -14 -f --size-hint=22000 | zstd -t  # considerably too high
 cat tmp | zstd -14 -f --size-hint=5500  | zstd -t  # considerably too low
+println "test : allows and interprets K,KB,KiB,M,MB and MiB suffix"
+cat tmp | zstd -14 -f --size-hint=11K | zstd -t
+cat tmp | zstd -14 -f --size-hint=11KB | zstd -t
+cat tmp | zstd -14 -f --size-hint=11KiB | zstd -t
+cat tmp | zstd -14 -f --size-hint=1M  | zstd -t
+cat tmp | zstd -14 -f --size-hint=1MB  | zstd -t
+cat tmp | zstd -14 -f --size-hint=1MiB  | zstd -t
 
 
 println "\n===>  dictionary tests "
-
-println "- test with raw dict (content only) "
+println "- Test high/low compressibility corpus training"
+datagen -g12M -P90 > tmpCorpusHighCompress
+datagen -g12M -P5 > tmpCorpusLowCompress
+zstd --train -B2K tmpCorpusHighCompress -o tmpDictHighCompress
+zstd --train -B2K tmpCorpusLowCompress -o tmpDictLowCompress
+rm -f tmpCorpusHighCompress tmpCorpusLowCompress tmpDictHighCompress tmpDictLowCompress
+println "- Test with raw dict (content only) "
 datagen > tmpDict
 datagen -g1M | $MD5SUM > tmp1
 datagen -g1M | zstd -D tmpDict | zstd -D tmpDict -dvq | $MD5SUM > tmp2
@@ -934,6 +1029,11 @@ println "- Dictionary compression with btlazy2 strategy"
 zstd -f tmp -D tmpDict --zstd=strategy=6
 zstd -d tmp.zst -D tmpDict -fo result
 $DIFF "$TESTFILE" result
+if [ -e /proc/self/fd/0 ]; then
+    println "- Test rejecting irregular dictionary file"
+    cat tmpDict | zstd -f tmp -D /proc/self/fd/0 && die "Piped dictionary should fail!"
+    cat tmpDict | zstd -d tmp.zst -D /proc/self/fd/0 -f && die "Piped dictionary should fail!"
+fi
 if [ -n "$hasMT" ]
 then
     println "- Test dictionary compression with multithreading "
@@ -982,17 +1082,20 @@ zstd -o tmpDict --train "$TESTDIR"/*.c "$PRGDIR"/*.c
 test -f tmpDict
 zstd --train "$TESTDIR"/*.c "$PRGDIR"/*.c
 test -f dictionary
-println "- Test dictionary training fails"
-echo "000000000000000000000000000000000" > tmpz
-zstd --train tmpz tmpz tmpz tmpz tmpz tmpz tmpz tmpz tmpz && die "Dictionary training should fail : source is all zeros"
 if [ -n "$hasMT" ]
 then
-  zstd --train -T0 tmpz tmpz tmpz tmpz tmpz tmpz tmpz tmpz tmpz && die "Dictionary training should fail : source is all zeros"
   println "- Create dictionary with multithreading enabled"
   zstd --train -T0 "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpDict
 fi
 rm -f tmp* dictionary
 
+println "- Test --memory for dictionary compression"
+datagen -g12M -P90 > tmpCorpusHighCompress
+zstd --train -B2K tmpCorpusHighCompress -o tmpDictHighCompress --memory=10K && die "Dictionary training should fail : --memory too low (10K)"
+zstd --train -B2K tmpCorpusHighCompress -o tmpDictHighCompress --memory=5MB 2> zstTrainWithMemLimitStdErr
+cat zstTrainWithMemLimitStdErr | grep "setting manual memory limit for dictionary training data at 5 MB"
+cat zstTrainWithMemLimitStdErr | grep "Training samples set too large (12 MB); training on 5 MB only..."
+rm zstTrainWithMemLimitStdErr
 
 println "\n===>  fastCover dictionary builder : advanced options "
 TESTFILE="$PRGDIR"/zstdcli.c
@@ -1014,6 +1117,7 @@ println "- Create dictionaries with shrink-dict flag enabled"
 zstd --train-fastcover=steps=1,shrink "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpShrinkDict
 zstd --train-fastcover=steps=1,shrink=1 "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpShrinkDict1
 zstd --train-fastcover=steps=1,shrink=5 "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpShrinkDict2
+zstd --train-fastcover=shrink=5,steps=1 "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpShrinkDict3
 println "- Create dictionary with size limit"
 zstd --train-fastcover=steps=1 "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpDict2 --maxdict=4K
 println "- Create dictionary using all samples for both training and testing"
@@ -1083,6 +1187,10 @@ zstd -t tmp3 && die "bad file not detected !"   # detects 0-sized files as bad
 println "test --rm and --test combined "
 zstd -t --rm tmp1.zst
 test -f tmp1.zst   # check file is still present
+cp tmp1.zst tmp2.zst
+zstd -t tmp1.zst tmp2.zst --rm
+test -f tmp1.zst   # check file is still present
+test -f tmp2.zst   # check file is still present
 split -b16384 tmp1.zst tmpSplit.
 zstd -t tmpSplit.* && die "bad file not detected !"
 datagen | zstd -c | zstd -t
@@ -1109,7 +1217,15 @@ zstd -rqi0b1e2 tmp1
 println "benchmark decompression only"
 zstd -f tmp1
 zstd -b -d -i0 tmp1.zst
+println "benchmark can fail - decompression on invalid data"
+zstd -b -d -i0 tmp1 && die "invalid .zst data => benchmark should have failed"
 
+GZIPMODE=1
+zstd --format=gzip -V || GZIPMODE=0
+if [ $GZIPMODE -eq 1 ]; then
+    println "benchmark mode is only compatible with zstd"
+    zstd --format=gzip -b tmp1 && die "-b should be incompatible with gzip format!"
+fi
 
 println "\n===>  zstd compatibility tests "
 
@@ -1285,7 +1401,7 @@ println "\n===>  tar extension tests "
 rm -f tmp tmp.tar tmp.tzst tmp.tgz tmp.txz tmp.tlz4 tmp1.zstd
 
 datagen > tmp
-tar cf tmp.tar tmp
+tar -cf tmp.tar tmp
 zstd tmp.tar -o tmp.tzst
 rm -f tmp.tar
 zstd -d tmp.tzst
@@ -1293,21 +1409,21 @@ zstd -d tmp.tzst
 rm -f tmp.tar tmp.tzst
 
 if [ $GZIPMODE -eq 1 ]; then
-    tar czf tmp.tgz tmp
+    tar -f - -c tmp | gzip > tmp.tgz
     zstd -d tmp.tgz
     [ -e tmp.tar ] || die ".tgz failed to decompress to .tar!"
     rm -f tmp.tar tmp.tgz
 fi
 
 if [ $LZMAMODE -eq 1 ]; then
-    tar c tmp | zstd --format=xz > tmp.txz
+    tar -f - -c tmp | zstd --format=xz > tmp.txz
     zstd -d tmp.txz
     [ -e tmp.tar ] || die ".txz failed to decompress to .tar!"
     rm -f tmp.tar tmp.txz
 fi
 
 if [ $LZ4MODE -eq 1 ]; then
-    tar c tmp | zstd --format=lz4 > tmp.tlz4
+    tar -f - -c tmp | zstd --format=lz4 > tmp.tlz4
     zstd -d tmp.tlz4
     [ -e tmp.tar ] || die ".tlz4 failed to decompress to .tar!"
     rm -f tmp.tar tmp.tlz4
@@ -1347,6 +1463,8 @@ if [ -n "$hasMT" ]
 then
     println "\n===>  zstdmt round-trip tests "
     roundTripTest -g4M "1 -T0"
+    roundTripTest -g4M "1 -T0 --auto-threads=physical"
+    roundTripTest -g4M "1 -T0 --auto-threads=logical"
     roundTripTest -g8M "3 -T2"
     roundTripTest -g8M "19 --long"
     roundTripTest -g8000K "2 --threads=2"
@@ -1366,6 +1484,8 @@ then
     ZSTD_NBTHREADS=50000000000 zstd -f mt_tmp # numeric value too large, warn and revert to default setting=
     ZSTD_NBTHREADS=2  zstd -f mt_tmp # correct usage
     ZSTD_NBTHREADS=1  zstd -f mt_tmp # correct usage: single thread
+    # temporary envvar changes in the above tests would actually persist in macos /bin/sh
+    unset ZSTD_NBTHREADS
     rm -f mt_tmp*
 
     println "\n===>  ovLog tests "
@@ -1441,7 +1561,7 @@ datagen -g0 > tmp5
 zstd tmp5
 zstd -l tmp5.zst
 zstd -l tmp5* && die "-l must fail on non-zstd file"
-zstd -lv tmp5.zst | grep "Decompressed Size: 0.00 KB (0 B)"  # check that 0 size is present in header
+zstd -lv tmp5.zst | grep "Decompressed Size: 0 B (0 B)"  # check that 0 size is present in header
 zstd -lv tmp5* && die "-l must fail on non-zstd file"
 
 println "\n===>  zstd --list/-l test with no content size field "
@@ -1500,6 +1620,44 @@ elif [ "$longCSize19wlog23" -gt "$optCSize19wlog23" ]; then
     exit 1
 fi
 
+println "\n===>  zstd asyncio tests "
+
+addFrame() {
+    datagen -g2M -s$2 >> tmp_uncompressed
+    datagen -g2M -s$2 | zstd -1 --format=$1 >> tmp_compressed.zst
+}
+
+addTwoFrames() {
+  addFrame $1 1
+  addFrame $1 2
+}
+
+testAsyncIO() {
+  roundTripTest -g2M "3 --asyncio --format=$1"
+  roundTripTest -g2M "3 --no-asyncio --format=$1"
+}
+
+rm -f tmp_compressed tmp_uncompressed
+testAsyncIO zstd
+addTwoFrames zstd
+if [ $GZIPMODE -eq 1 ]; then
+  testAsyncIO gzip
+  addTwoFrames gzip
+fi
+if [ $LZMAMODE -eq 1 ]; then
+  testAsyncIO lzma
+  addTwoFrames lzma
+fi
+if [ $LZ4MODE -eq 1 ]; then
+  testAsyncIO lz4
+  addTwoFrames lz4
+fi
+cat tmp_uncompressed | $MD5SUM > tmp2
+zstd -d tmp_compressed.zst --asyncio -c | $MD5SUM > tmp1
+$DIFF -q tmp1 tmp2
+rm tmp1
+zstd -d tmp_compressed.zst --no-asyncio -c | $MD5SUM > tmp1
+$DIFF -q tmp1 tmp2
 
 if [ "$1" != "--test-large-data" ]; then
     println "Skipping large data tests"
@@ -1515,8 +1673,10 @@ then
     println "\n===>   adaptive mode "
     roundTripTest -g270000000 " --adapt"
     roundTripTest -g27000000 " --adapt=min=1,max=4"
+    roundTripTest -g27000000 " --adapt=min=-2,max=-1"
     println "===>   test: --adapt must fail on incoherent bounds "
     datagen > tmp
+    zstd --adapt= tmp && die "invalid compression parameter"
     zstd -f -vv --adapt=min=10,max=9 tmp && die "--adapt must fail on incoherent bounds"
 
     println "\n===>   rsyncable mode "
@@ -1635,6 +1795,7 @@ println "- Create first dictionary"
 zstd --train-cover=k=46,d=8,split=80 "$TESTDIR"/*.c "$PRGDIR"/*.c -o tmpDict
 cp "$TESTFILE" tmp
 zstd -f tmp -D tmpDict
+zstd -f tmp -D tmpDict --patch-from=tmpDict && die "error: can't use -D and --patch-from=#at the same time"
 zstd -d tmp.zst -D tmpDict -fo result
 $DIFF "$TESTFILE" result
 zstd --train-cover=k=56,d=8 && die "Create dictionary without input file (should error)"
@@ -1645,6 +1806,7 @@ println "- Create dictionary using shrink-dict flag"
 zstd --train-cover=steps=256,shrink "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict
 zstd --train-cover=steps=256,shrink=1 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict1
 zstd --train-cover=steps=256,shrink=5 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict2
+zstd --train-cover=shrink=5,steps=256 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpShrinkDict3
 println "- Create dictionary with short dictID"
 zstd --train-cover=k=46,d=8,split=80 "$TESTDIR"/*.c "$PRGDIR"/*.c --dictID=1 -o tmpDict1
 cmp tmpDict tmpDict1 && die "dictionaries should have different ID !"
diff --git a/priv/zstd/tests/poolTests.c b/priv/zstd/tests/poolTests.c
index 08f31c0..9e62722 100644
--- a/priv/zstd/tests/poolTests.c
+++ b/priv/zstd/tests/poolTests.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/rateLimiter.py b/priv/zstd/tests/rateLimiter.py
index 2629372..3e65081 100755
--- a/priv/zstd/tests/rateLimiter.py
+++ b/priv/zstd/tests/rateLimiter.py
@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/.gitignore b/priv/zstd/tests/regression/.gitignore
index 1b2618f..3da209d 100644
--- a/priv/zstd/tests/regression/.gitignore
+++ b/priv/zstd/tests/regression/.gitignore
@@ -1,3 +1,4 @@
 # regression test artifacts
 data-cache
+cache
 test
diff --git a/priv/zstd/tests/regression/Makefile b/priv/zstd/tests/regression/Makefile
index d0d7bcf..ba8b430 100644
--- a/priv/zstd/tests/regression/Makefile
+++ b/priv/zstd/tests/regression/Makefile
@@ -1,5 +1,5 @@
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -15,7 +15,7 @@ CURL_LDFLAGS := $(shell curl-config --libs) -pthread
 
 PROGDIR := ../../programs
 LIBDIR := ../../lib
-ZSTD_CPPFLAGS := -I$(PROGDIR) -I$(LIBDIR) -I$(LIBDIR)/common
+ZSTD_CPPFLAGS := -I$(PROGDIR) -I$(LIBDIR) -I$(LIBDIR)/common -Wno-deprecated-declarations
 
 REGRESSION_CFLAGS = $(CFLAGS) $(CURL_CFLAGS)
 REGRESSION_CPPFLAGS = $(CPPFLAGS) $(ZSTD_CPPFLAGS)
@@ -46,6 +46,7 @@ result.o: result.c result.h
 test.o: test.c data.h config.h method.h
 	$(CC) $(REGRESSION_CFLAGS) $(REGRESSION_CPPFLAGS) $< -c -o $@
 
+.PHONY: libzstd.a
 libzstd.a:
 	$(MAKE) -C $(LIBDIR) libzstd.a-mt
 	cp $(LIBDIR)/libzstd.a .
diff --git a/priv/zstd/tests/regression/config.c b/priv/zstd/tests/regression/config.c
index 4c66dd1..1d86fff 100644
--- a/priv/zstd/tests/regression/config.c
+++ b/priv/zstd/tests/regression/config.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -183,7 +183,7 @@ static config_t no_pledged_src_size_with_dict = {
 };
 
 static param_value_t const ldm_param_values[] = {
-    {.param = ZSTD_c_enableLongDistanceMatching, .value = 1},
+    {.param = ZSTD_c_enableLongDistanceMatching, .value = ZSTD_ps_enable},
 };
 
 static config_t ldm = {
@@ -204,7 +204,7 @@ static config_t mt = {
 
 static param_value_t const mt_ldm_param_values[] = {
     {.param = ZSTD_c_nbWorkers, .value = 2},
-    {.param = ZSTD_c_enableLongDistanceMatching, .value = 1},
+    {.param = ZSTD_c_enableLongDistanceMatching, .value = ZSTD_ps_enable},
 };
 
 static config_t mt_ldm = {
@@ -215,7 +215,7 @@ static config_t mt_ldm = {
 
 static param_value_t mt_advanced_param_values[] = {
     {.param = ZSTD_c_nbWorkers, .value = 2},
-    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_lcm_uncompressed},
+    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_ps_disable},
 };
 
 static config_t mt_advanced = {
@@ -258,7 +258,7 @@ static config_t small_clog = {
 
 static param_value_t const uncompressed_literals_param_values[] = {
     {.param = ZSTD_c_compressionLevel, .value = 3},
-    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_lcm_uncompressed},
+    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_ps_disable},
 };
 
 static config_t uncompressed_literals = {
@@ -269,7 +269,7 @@ static config_t uncompressed_literals = {
 
 static param_value_t const uncompressed_literals_opt_param_values[] = {
     {.param = ZSTD_c_compressionLevel, .value = 19},
-    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_lcm_uncompressed},
+    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_ps_disable},
 };
 
 static config_t uncompressed_literals_opt = {
@@ -280,7 +280,7 @@ static config_t uncompressed_literals_opt = {
 
 static param_value_t const huffman_literals_param_values[] = {
     {.param = ZSTD_c_compressionLevel, .value = -1},
-    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_lcm_huffman},
+    {.param = ZSTD_c_literalCompressionMode, .value = ZSTD_ps_enable},
 };
 
 static config_t huffman_literals = {
diff --git a/priv/zstd/tests/regression/config.h b/priv/zstd/tests/regression/config.h
index dd88937..a4b542a 100644
--- a/priv/zstd/tests/regression/config.h
+++ b/priv/zstd/tests/regression/config.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/data.c b/priv/zstd/tests/regression/data.c
index 341b02d..43f085f 100644
--- a/priv/zstd/tests/regression/data.c
+++ b/priv/zstd/tests/regression/data.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/data.h b/priv/zstd/tests/regression/data.h
index e54e6a1..a4ee920 100644
--- a/priv/zstd/tests/regression/data.h
+++ b/priv/zstd/tests/regression/data.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/levels.h b/priv/zstd/tests/regression/levels.h
index 3b211f8..d15b120 100644
--- a/priv/zstd/tests/regression/levels.h
+++ b/priv/zstd/tests/regression/levels.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -35,21 +35,23 @@ LEVEL(1)
 LEVEL(3)
 LEVEL(4)
 /* ROW_LEVEL triggers the row hash (force enabled and disabled) with different
- * dictionary strategies, and 16/32 row entries based on the level/searchLog.
- * 1 == disabled, 2 == enabled.
+ * dictionary strategies, and 16/32/64 row entries based on the level/searchLog.
+ * 1 == enabled, 2 == disabled.
  */
 ROW_LEVEL(5, 1)
-ROW_LEVEL(5, 2)
+ROW_LEVEL(5, 2) /* 16-entry rows */
 LEVEL(5)
 LEVEL(6)
 ROW_LEVEL(7, 1)
-ROW_LEVEL(7, 2)
+ROW_LEVEL(7, 2) /* 16-entry rows */
 LEVEL(7)
 
 LEVEL(9)
 
+ROW_LEVEL(11, 1)
+ROW_LEVEL(11, 2) /* 32-entry rows */
 ROW_LEVEL(12, 1)
-ROW_LEVEL(12, 2)
+ROW_LEVEL(12, 2) /* 64-entry rows */
 LEVEL(13)
 
 LEVEL(16)
diff --git a/priv/zstd/tests/regression/method.c b/priv/zstd/tests/regression/method.c
index 55b1154..f84a15e 100644
--- a/priv/zstd/tests/regression/method.c
+++ b/priv/zstd/tests/regression/method.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/method.h b/priv/zstd/tests/regression/method.h
index 1a36a93..8efdd33 100644
--- a/priv/zstd/tests/regression/method.h
+++ b/priv/zstd/tests/regression/method.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/result.c b/priv/zstd/tests/regression/result.c
index 1f879c1..8ccb875 100644
--- a/priv/zstd/tests/regression/result.c
+++ b/priv/zstd/tests/regression/result.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/result.h b/priv/zstd/tests/regression/result.h
index 197fa90..8a761ea 100644
--- a/priv/zstd/tests/regression/result.h
+++ b/priv/zstd/tests/regression/result.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/regression/results.csv b/priv/zstd/tests/regression/results.csv
index b94d550..188eea9 100644
--- a/priv/zstd/tests/regression/results.csv
+++ b/priv/zstd/tests/regression/results.csv
@@ -1,74 +1,74 @@
 Data,                               Config,                             Method,                             Total compressed size
-silesia.tar,                        level -5,                           compress simple,                    6738593
-silesia.tar,                        level -3,                           compress simple,                    6446372
-silesia.tar,                        level -1,                           compress simple,                    6186042
-silesia.tar,                        level 0,                            compress simple,                    4861425
-silesia.tar,                        level 1,                            compress simple,                    5334885
-silesia.tar,                        level 3,                            compress simple,                    4861425
-silesia.tar,                        level 4,                            compress simple,                    4799630
-silesia.tar,                        level 5,                            compress simple,                    4719256
-silesia.tar,                        level 6,                            compress simple,                    4677721
-silesia.tar,                        level 7,                            compress simple,                    4613541
-silesia.tar,                        level 9,                            compress simple,                    4555426
-silesia.tar,                        level 13,                           compress simple,                    4491764
-silesia.tar,                        level 16,                           compress simple,                    4381332
-silesia.tar,                        level 19,                           compress simple,                    4281605
-silesia.tar,                        uncompressed literals,              compress simple,                    4861425
-silesia.tar,                        uncompressed literals optimal,      compress simple,                    4281605
-silesia.tar,                        huffman literals,                   compress simple,                    6186042
-github.tar,                         level -5,                           compress simple,                    46856
-github.tar,                         level -3,                           compress simple,                    43754
-github.tar,                         level -1,                           compress simple,                    42490
-github.tar,                         level 0,                            compress simple,                    38441
-github.tar,                         level 1,                            compress simple,                    39265
-github.tar,                         level 3,                            compress simple,                    38441
-github.tar,                         level 4,                            compress simple,                    38467
-github.tar,                         level 5,                            compress simple,                    39693
-github.tar,                         level 6,                            compress simple,                    39621
-github.tar,                         level 7,                            compress simple,                    39213
-github.tar,                         level 9,                            compress simple,                    36758
-github.tar,                         level 13,                           compress simple,                    35621
-github.tar,                         level 16,                           compress simple,                    40255
-github.tar,                         level 19,                           compress simple,                    32837
-github.tar,                         uncompressed literals,              compress simple,                    38441
-github.tar,                         uncompressed literals optimal,      compress simple,                    32837
-github.tar,                         huffman literals,                   compress simple,                    42490
-silesia,                            level -5,                           compress cctx,                      6737607
-silesia,                            level -3,                           compress cctx,                      6444677
-silesia,                            level -1,                           compress cctx,                      6178460
-silesia,                            level 0,                            compress cctx,                      4849552
-silesia,                            level 1,                            compress cctx,                      5313204
-silesia,                            level 3,                            compress cctx,                      4849552
-silesia,                            level 4,                            compress cctx,                      4786970
-silesia,                            level 5,                            compress cctx,                      4707794
-silesia,                            level 6,                            compress cctx,                      4666383
-silesia,                            level 7,                            compress cctx,                      4603381
-silesia,                            level 9,                            compress cctx,                      4546001
-silesia,                            level 13,                           compress cctx,                      4482135
-silesia,                            level 16,                           compress cctx,                      4377465
-silesia,                            level 19,                           compress cctx,                      4293330
-silesia,                            long distance mode,                 compress cctx,                      4849552
-silesia,                            multithreaded,                      compress cctx,                      4849552
-silesia,                            multithreaded long distance mode,   compress cctx,                      4849552
-silesia,                            small window log,                   compress cctx,                      7084179
-silesia,                            small hash log,                     compress cctx,                      6555021
-silesia,                            small chain log,                    compress cctx,                      4931148
-silesia,                            explicit params,                    compress cctx,                      4794479
-silesia,                            uncompressed literals,              compress cctx,                      4849552
-silesia,                            uncompressed literals optimal,      compress cctx,                      4293330
-silesia,                            huffman literals,                   compress cctx,                      6178460
-silesia,                            multithreaded with advanced params, compress cctx,                      4849552
-github,                             level -5,                           compress cctx,                      205285
-github,                             level -5 with dict,                 compress cctx,                      47294
-github,                             level -3,                           compress cctx,                      190643
-github,                             level -3 with dict,                 compress cctx,                      48047
-github,                             level -1,                           compress cctx,                      175568
-github,                             level -1 with dict,                 compress cctx,                      43527
-github,                             level 0,                            compress cctx,                      136335
+silesia.tar,                        level -5,                           compress simple,                    6861055
+silesia.tar,                        level -3,                           compress simple,                    6505483
+silesia.tar,                        level -1,                           compress simple,                    6179047
+silesia.tar,                        level 0,                            compress simple,                    4854086
+silesia.tar,                        level 1,                            compress simple,                    5327717
+silesia.tar,                        level 3,                            compress simple,                    4854086
+silesia.tar,                        level 4,                            compress simple,                    4791503
+silesia.tar,                        level 5,                            compress simple,                    4677740
+silesia.tar,                        level 6,                            compress simple,                    4613242
+silesia.tar,                        level 7,                            compress simple,                    4576661
+silesia.tar,                        level 9,                            compress simple,                    4552899
+silesia.tar,                        level 13,                           compress simple,                    4502956
+silesia.tar,                        level 16,                           compress simple,                    4360546
+silesia.tar,                        level 19,                           compress simple,                    4265911
+silesia.tar,                        uncompressed literals,              compress simple,                    4854086
+silesia.tar,                        uncompressed literals optimal,      compress simple,                    4265911
+silesia.tar,                        huffman literals,                   compress simple,                    6179047
+github.tar,                         level -5,                           compress simple,                    52115
+github.tar,                         level -3,                           compress simple,                    45678
+github.tar,                         level -1,                           compress simple,                    42560
+github.tar,                         level 0,                            compress simple,                    38831
+github.tar,                         level 1,                            compress simple,                    39200
+github.tar,                         level 3,                            compress simple,                    38831
+github.tar,                         level 4,                            compress simple,                    38893
+github.tar,                         level 5,                            compress simple,                    39651
+github.tar,                         level 6,                            compress simple,                    39282
+github.tar,                         level 7,                            compress simple,                    38110
+github.tar,                         level 9,                            compress simple,                    36760
+github.tar,                         level 13,                           compress simple,                    35501
+github.tar,                         level 16,                           compress simple,                    40466
+github.tar,                         level 19,                           compress simple,                    32276
+github.tar,                         uncompressed literals,              compress simple,                    38831
+github.tar,                         uncompressed literals optimal,      compress simple,                    32276
+github.tar,                         huffman literals,                   compress simple,                    42560
+silesia,                            level -5,                           compress cctx,                      6857372
+silesia,                            level -3,                           compress cctx,                      6503412
+silesia,                            level -1,                           compress cctx,                      6172202
+silesia,                            level 0,                            compress cctx,                      4842075
+silesia,                            level 1,                            compress cctx,                      5306632
+silesia,                            level 3,                            compress cctx,                      4842075
+silesia,                            level 4,                            compress cctx,                      4779186
+silesia,                            level 5,                            compress cctx,                      4666323
+silesia,                            level 6,                            compress cctx,                      4603066
+silesia,                            level 7,                            compress cctx,                      4566984
+silesia,                            level 9,                            compress cctx,                      4543018
+silesia,                            level 13,                           compress cctx,                      4493990
+silesia,                            level 16,                           compress cctx,                      4360041
+silesia,                            level 19,                           compress cctx,                      4296055
+silesia,                            long distance mode,                 compress cctx,                      4842075
+silesia,                            multithreaded,                      compress cctx,                      4842075
+silesia,                            multithreaded long distance mode,   compress cctx,                      4842075
+silesia,                            small window log,                   compress cctx,                      7082951
+silesia,                            small hash log,                     compress cctx,                      6526141
+silesia,                            small chain log,                    compress cctx,                      4912197
+silesia,                            explicit params,                    compress cctx,                      4794052
+silesia,                            uncompressed literals,              compress cctx,                      4842075
+silesia,                            uncompressed literals optimal,      compress cctx,                      4296055
+silesia,                            huffman literals,                   compress cctx,                      6172202
+silesia,                            multithreaded with advanced params, compress cctx,                      4842075
+github,                             level -5,                           compress cctx,                      204407
+github,                             level -5 with dict,                 compress cctx,                      52059
+github,                             level -3,                           compress cctx,                      193253
+github,                             level -3 with dict,                 compress cctx,                      46787
+github,                             level -1,                           compress cctx,                      175468
+github,                             level -1 with dict,                 compress cctx,                      43585
+github,                             level 0,                            compress cctx,                      136332
 github,                             level 0 with dict,                  compress cctx,                      41534
-github,                             level 1,                            compress cctx,                      142465
-github,                             level 1 with dict,                  compress cctx,                      42157
-github,                             level 3,                            compress cctx,                      136335
+github,                             level 1,                            compress cctx,                      142365
+github,                             level 1 with dict,                  compress cctx,                      42259
+github,                             level 3,                            compress cctx,                      136332
 github,                             level 3 with dict,                  compress cctx,                      41534
 github,                             level 4,                            compress cctx,                      136199
 github,                             level 4 with dict,                  compress cctx,                      41725
@@ -80,237 +80,241 @@ github,                             level 7,                            compress
 github,                             level 7 with dict,                  compress cctx,                      38755
 github,                             level 9,                            compress cctx,                      135122
 github,                             level 9 with dict,                  compress cctx,                      39398
-github,                             level 13,                           compress cctx,                      134064
+github,                             level 13,                           compress cctx,                      132878
 github,                             level 13 with dict,                 compress cctx,                      39948
-github,                             level 16,                           compress cctx,                      134064
+github,                             level 16,                           compress cctx,                      133209
 github,                             level 16 with dict,                 compress cctx,                      37568
-github,                             level 19,                           compress cctx,                      134064
+github,                             level 19,                           compress cctx,                      132879
 github,                             level 19 with dict,                 compress cctx,                      37567
-github,                             long distance mode,                 compress cctx,                      141102
-github,                             multithreaded,                      compress cctx,                      141102
-github,                             multithreaded long distance mode,   compress cctx,                      141102
-github,                             small window log,                   compress cctx,                      141102
+github,                             long distance mode,                 compress cctx,                      141069
+github,                             multithreaded,                      compress cctx,                      141069
+github,                             multithreaded long distance mode,   compress cctx,                      141069
+github,                             small window log,                   compress cctx,                      141069
 github,                             small hash log,                     compress cctx,                      138949
 github,                             small chain log,                    compress cctx,                      139242
 github,                             explicit params,                    compress cctx,                      140932
-github,                             uncompressed literals,              compress cctx,                      136335
-github,                             uncompressed literals optimal,      compress cctx,                      134064
-github,                             huffman literals,                   compress cctx,                      175568
-github,                             multithreaded with advanced params, compress cctx,                      141102
-silesia,                            level -5,                           zstdcli,                            6737655
-silesia,                            level -3,                           zstdcli,                            6444725
-silesia,                            level -1,                           zstdcli,                            6178508
-silesia,                            level 0,                            zstdcli,                            4849600
-silesia,                            level 1,                            zstdcli,                            5313252
-silesia,                            level 3,                            zstdcli,                            4849600
-silesia,                            level 4,                            zstdcli,                            4787018
-silesia,                            level 5,                            zstdcli,                            4707842
-silesia,                            level 6,                            zstdcli,                            4666431
-silesia,                            level 7,                            zstdcli,                            4603429
-silesia,                            level 9,                            zstdcli,                            4546049
-silesia,                            level 13,                           zstdcli,                            4482183
-silesia,                            level 16,                           zstdcli,                            4360299
-silesia,                            level 19,                           zstdcli,                            4283285
-silesia,                            long distance mode,                 zstdcli,                            4840806
-silesia,                            multithreaded,                      zstdcli,                            4849600
-silesia,                            multithreaded long distance mode,   zstdcli,                            4840806
-silesia,                            small window log,                   zstdcli,                            7095967
+github,                             uncompressed literals,              compress cctx,                      136332
+github,                             uncompressed literals optimal,      compress cctx,                      132879
+github,                             huffman literals,                   compress cctx,                      175468
+github,                             multithreaded with advanced params, compress cctx,                      141069
+silesia,                            level -5,                           zstdcli,                            6857420
+silesia,                            level -3,                           zstdcli,                            6503460
+silesia,                            level -1,                           zstdcli,                            6172250
+silesia,                            level 0,                            zstdcli,                            4842123
+silesia,                            level 1,                            zstdcli,                            5306680
+silesia,                            level 3,                            zstdcli,                            4842123
+silesia,                            level 4,                            zstdcli,                            4779234
+silesia,                            level 5,                            zstdcli,                            4666371
+silesia,                            level 6,                            zstdcli,                            4603114
+silesia,                            level 7,                            zstdcli,                            4567032
+silesia,                            level 9,                            zstdcli,                            4543066
+silesia,                            level 13,                           zstdcli,                            4494038
+silesia,                            level 16,                           zstdcli,                            4360089
+silesia,                            level 19,                           zstdcli,                            4296103
+silesia,                            long distance mode,                 zstdcli,                            4833785
+silesia,                            multithreaded,                      zstdcli,                            4842123
+silesia,                            multithreaded long distance mode,   zstdcli,                            4833785
+silesia,                            small window log,                   zstdcli,                            7095048
 silesia,                            small hash log,                     zstdcli,                            6526189
 silesia,                            small chain log,                    zstdcli,                            4912245
-silesia,                            explicit params,                    zstdcli,                            4795856
-silesia,                            uncompressed literals,              zstdcli,                            5128030
-silesia,                            uncompressed literals optimal,      zstdcli,                            4317944
-silesia,                            huffman literals,                   zstdcli,                            5326316
-silesia,                            multithreaded with advanced params, zstdcli,                            5128030
-silesia.tar,                        level -5,                           zstdcli,                            6738934
-silesia.tar,                        level -3,                           zstdcli,                            6448419
-silesia.tar,                        level -1,                           zstdcli,                            6186912
-silesia.tar,                        level 0,                            zstdcli,                            4861512
-silesia.tar,                        level 1,                            zstdcli,                            5336318
-silesia.tar,                        level 3,                            zstdcli,                            4861512
-silesia.tar,                        level 4,                            zstdcli,                            4800529
-silesia.tar,                        level 5,                            zstdcli,                            4720121
-silesia.tar,                        level 6,                            zstdcli,                            4678661
-silesia.tar,                        level 7,                            zstdcli,                            4614424
-silesia.tar,                        level 9,                            zstdcli,                            4556062
-silesia.tar,                        level 13,                           zstdcli,                            4491768
-silesia.tar,                        level 16,                           zstdcli,                            4356831
-silesia.tar,                        level 19,                           zstdcli,                            4264491
-silesia.tar,                        no source size,                     zstdcli,                            4861508
-silesia.tar,                        long distance mode,                 zstdcli,                            4853226
-silesia.tar,                        multithreaded,                      zstdcli,                            4861512
-silesia.tar,                        multithreaded long distance mode,   zstdcli,                            4853226
-silesia.tar,                        small window log,                   zstdcli,                            7101576
-silesia.tar,                        small hash log,                     zstdcli,                            6529290
+silesia,                            explicit params,                    zstdcli,                            4795432
+silesia,                            uncompressed literals,              zstdcli,                            5120614
+silesia,                            uncompressed literals optimal,      zstdcli,                            4319566
+silesia,                            huffman literals,                   zstdcli,                            5321417
+silesia,                            multithreaded with advanced params, zstdcli,                            5120614
+silesia.tar,                        level -5,                           zstdcli,                            6862049
+silesia.tar,                        level -3,                           zstdcli,                            6506509
+silesia.tar,                        level -1,                           zstdcli,                            6179789
+silesia.tar,                        level 0,                            zstdcli,                            4854164
+silesia.tar,                        level 1,                            zstdcli,                            5329010
+silesia.tar,                        level 3,                            zstdcli,                            4854164
+silesia.tar,                        level 4,                            zstdcli,                            4792352
+silesia.tar,                        level 5,                            zstdcli,                            4678682
+silesia.tar,                        level 6,                            zstdcli,                            4614125
+silesia.tar,                        level 7,                            zstdcli,                            4578719
+silesia.tar,                        level 9,                            zstdcli,                            4552903
+silesia.tar,                        level 13,                           zstdcli,                            4502960
+silesia.tar,                        level 16,                           zstdcli,                            4360550
+silesia.tar,                        level 19,                           zstdcli,                            4265915
+silesia.tar,                        no source size,                     zstdcli,                            4854160
+silesia.tar,                        long distance mode,                 zstdcli,                            4845745
+silesia.tar,                        multithreaded,                      zstdcli,                            4854164
+silesia.tar,                        multithreaded long distance mode,   zstdcli,                            4845745
+silesia.tar,                        small window log,                   zstdcli,                            7100701
+silesia.tar,                        small hash log,                     zstdcli,                            6529264
 silesia.tar,                        small chain log,                    zstdcli,                            4917022
-silesia.tar,                        explicit params,                    zstdcli,                            4821274
-silesia.tar,                        uncompressed literals,              zstdcli,                            5129559
-silesia.tar,                        uncompressed literals optimal,      zstdcli,                            4307457
-silesia.tar,                        huffman literals,                   zstdcli,                            5347610
-silesia.tar,                        multithreaded with advanced params, zstdcli,                            5129559
-github,                             level -5,                           zstdcli,                            207285
+silesia.tar,                        explicit params,                    zstdcli,                            4820713
+silesia.tar,                        uncompressed literals,              zstdcli,                            5122571
+silesia.tar,                        uncompressed literals optimal,      zstdcli,                            4310145
+silesia.tar,                        huffman literals,                   zstdcli,                            5342074
+silesia.tar,                        multithreaded with advanced params, zstdcli,                            5122571
+github,                             level -5,                           zstdcli,                            206407
 github,                             level -5 with dict,                 zstdcli,                            48718
-github,                             level -3,                           zstdcli,                            192643
+github,                             level -3,                           zstdcli,                            195253
 github,                             level -3 with dict,                 zstdcli,                            47395
-github,                             level -1,                           zstdcli,                            177568
+github,                             level -1,                           zstdcli,                            177468
 github,                             level -1 with dict,                 zstdcli,                            45170
-github,                             level 0,                            zstdcli,                            138335
+github,                             level 0,                            zstdcli,                            138332
 github,                             level 0 with dict,                  zstdcli,                            43148
-github,                             level 1,                            zstdcli,                            144465
+github,                             level 1,                            zstdcli,                            144365
 github,                             level 1 with dict,                  zstdcli,                            43682
-github,                             level 3,                            zstdcli,                            138335
+github,                             level 3,                            zstdcli,                            138332
 github,                             level 3 with dict,                  zstdcli,                            43148
 github,                             level 4,                            zstdcli,                            138199
 github,                             level 4 with dict,                  zstdcli,                            43251
 github,                             level 5,                            zstdcli,                            137121
 github,                             level 5 with dict,                  zstdcli,                            40728
 github,                             level 6,                            zstdcli,                            137122
-github,                             level 6 with dict,                  zstdcli,                            40630
+github,                             level 6 with dict,                  zstdcli,                            40636
 github,                             level 7,                            zstdcli,                            137122
-github,                             level 7 with dict,                  zstdcli,                            40747
+github,                             level 7 with dict,                  zstdcli,                            40745
 github,                             level 9,                            zstdcli,                            137122
-github,                             level 9 with dict,                  zstdcli,                            41338
-github,                             level 13,                           zstdcli,                            136064
-github,                             level 13 with dict,                 zstdcli,                            41743
-github,                             level 16,                           zstdcli,                            136064
+github,                             level 9 with dict,                  zstdcli,                            41393
+github,                             level 13,                           zstdcli,                            134878
+github,                             level 13 with dict,                 zstdcli,                            41900
+github,                             level 16,                           zstdcli,                            135209
 github,                             level 16 with dict,                 zstdcli,                            39577
-github,                             level 19,                           zstdcli,                            136064
+github,                             level 19,                           zstdcli,                            134879
 github,                             level 19 with dict,                 zstdcli,                            39576
-github,                             long distance mode,                 zstdcli,                            138335
-github,                             multithreaded,                      zstdcli,                            138335
-github,                             multithreaded long distance mode,   zstdcli,                            138335
-github,                             small window log,                   zstdcli,                            138335
+github,                             long distance mode,                 zstdcli,                            138332
+github,                             multithreaded,                      zstdcli,                            138332
+github,                             multithreaded long distance mode,   zstdcli,                            138332
+github,                             small window log,                   zstdcli,                            138332
 github,                             small hash log,                     zstdcli,                            137590
 github,                             small chain log,                    zstdcli,                            138341
 github,                             explicit params,                    zstdcli,                            136197
-github,                             uncompressed literals,              zstdcli,                            167915
-github,                             uncompressed literals optimal,      zstdcli,                            159227
-github,                             huffman literals,                   zstdcli,                            144465
-github,                             multithreaded with advanced params, zstdcli,                            167915
-github.tar,                         level -5,                           zstdcli,                            46860
-github.tar,                         level -5 with dict,                 zstdcli,                            44575
-github.tar,                         level -3,                           zstdcli,                            43758
-github.tar,                         level -3 with dict,                 zstdcli,                            41451
-github.tar,                         level -1,                           zstdcli,                            42494
-github.tar,                         level -1 with dict,                 zstdcli,                            41135
-github.tar,                         level 0,                            zstdcli,                            38445
+github,                             uncompressed literals,              zstdcli,                            167911
+github,                             uncompressed literals optimal,      zstdcli,                            154667
+github,                             huffman literals,                   zstdcli,                            144365
+github,                             multithreaded with advanced params, zstdcli,                            167911
+github.tar,                         level -5,                           zstdcli,                            52119
+github.tar,                         level -5 with dict,                 zstdcli,                            50978
+github.tar,                         level -3,                           zstdcli,                            45682
+github.tar,                         level -3 with dict,                 zstdcli,                            44660
+github.tar,                         level -1,                           zstdcli,                            42564
+github.tar,                         level -1 with dict,                 zstdcli,                            41155
+github.tar,                         level 0,                            zstdcli,                            38835
 github.tar,                         level 0 with dict,                  zstdcli,                            37999
-github.tar,                         level 1,                            zstdcli,                            39269
-github.tar,                         level 1 with dict,                  zstdcli,                            38284
-github.tar,                         level 3,                            zstdcli,                            38445
+github.tar,                         level 1,                            zstdcli,                            39204
+github.tar,                         level 1 with dict,                  zstdcli,                            38093
+github.tar,                         level 3,                            zstdcli,                            38835
 github.tar,                         level 3 with dict,                  zstdcli,                            37999
-github.tar,                         level 4,                            zstdcli,                            38471
+github.tar,                         level 4,                            zstdcli,                            38897
 github.tar,                         level 4 with dict,                  zstdcli,                            37952
-github.tar,                         level 5,                            zstdcli,                            39697
-github.tar,                         level 5 with dict,                  zstdcli,                            39032
-github.tar,                         level 6,                            zstdcli,                            39625
-github.tar,                         level 6 with dict,                  zstdcli,                            38614
-github.tar,                         level 7,                            zstdcli,                            39217
-github.tar,                         level 7 with dict,                  zstdcli,                            37871
-github.tar,                         level 9,                            zstdcli,                            36762
-github.tar,                         level 9 with dict,                  zstdcli,                            36641
-github.tar,                         level 13,                           zstdcli,                            35625
-github.tar,                         level 13 with dict,                 zstdcli,                            38730
-github.tar,                         level 16,                           zstdcli,                            40259
-github.tar,                         level 16 with dict,                 zstdcli,                            33643
-github.tar,                         level 19,                           zstdcli,                            32841
-github.tar,                         level 19 with dict,                 zstdcli,                            32899
-github.tar,                         no source size,                     zstdcli,                            38442
+github.tar,                         level 5,                            zstdcli,                            39655
+github.tar,                         level 5 with dict,                  zstdcli,                            39071
+github.tar,                         level 6,                            zstdcli,                            39286
+github.tar,                         level 6 with dict,                  zstdcli,                            38638
+github.tar,                         level 7,                            zstdcli,                            38114
+github.tar,                         level 7 with dict,                  zstdcli,                            37886
+github.tar,                         level 9,                            zstdcli,                            36764
+github.tar,                         level 9 with dict,                  zstdcli,                            36632
+github.tar,                         level 13,                           zstdcli,                            35505
+github.tar,                         level 13 with dict,                 zstdcli,                            37134
+github.tar,                         level 16,                           zstdcli,                            40470
+github.tar,                         level 16 with dict,                 zstdcli,                            33378
+github.tar,                         level 19,                           zstdcli,                            32280
+github.tar,                         level 19 with dict,                 zstdcli,                            32716
+github.tar,                         no source size,                     zstdcli,                            38832
 github.tar,                         no source size with dict,           zstdcli,                            38004
-github.tar,                         long distance mode,                 zstdcli,                            39730
-github.tar,                         multithreaded,                      zstdcli,                            38445
-github.tar,                         multithreaded long distance mode,   zstdcli,                            39730
+github.tar,                         long distance mode,                 zstdcli,                            40236
+github.tar,                         multithreaded,                      zstdcli,                            38835
+github.tar,                         multithreaded long distance mode,   zstdcli,                            40236
 github.tar,                         small window log,                   zstdcli,                            198544
 github.tar,                         small hash log,                     zstdcli,                            129874
 github.tar,                         small chain log,                    zstdcli,                            41673
-github.tar,                         explicit params,                    zstdcli,                            41227
-github.tar,                         uncompressed literals,              zstdcli,                            41126
-github.tar,                         uncompressed literals optimal,      zstdcli,                            35392
-github.tar,                         huffman literals,                   zstdcli,                            38781
-github.tar,                         multithreaded with advanced params, zstdcli,                            41126
-silesia,                            level -5,                           advanced one pass,                  6737607
-silesia,                            level -3,                           advanced one pass,                  6444677
-silesia,                            level -1,                           advanced one pass,                  6178460
-silesia,                            level 0,                            advanced one pass,                  4849552
-silesia,                            level 1,                            advanced one pass,                  5313204
-silesia,                            level 3,                            advanced one pass,                  4849552
-silesia,                            level 4,                            advanced one pass,                  4786970
-silesia,                            level 5 row 1,                      advanced one pass,                  4710236
-silesia,                            level 5 row 2,                      advanced one pass,                  4707794
-silesia,                            level 5,                            advanced one pass,                  4707794
-silesia,                            level 6,                            advanced one pass,                  4666383
-silesia,                            level 7 row 1,                      advanced one pass,                  4596296
-silesia,                            level 7 row 2,                      advanced one pass,                  4603381
-silesia,                            level 7,                            advanced one pass,                  4603381
-silesia,                            level 9,                            advanced one pass,                  4546001
-silesia,                            level 12 row 1,                     advanced one pass,                  4519288
-silesia,                            level 12 row 2,                     advanced one pass,                  4521397
-silesia,                            level 13,                           advanced one pass,                  4482135
-silesia,                            level 16,                           advanced one pass,                  4360251
-silesia,                            level 19,                           advanced one pass,                  4283237
-silesia,                            no source size,                     advanced one pass,                  4849552
-silesia,                            long distance mode,                 advanced one pass,                  4840738
-silesia,                            multithreaded,                      advanced one pass,                  4849552
-silesia,                            multithreaded long distance mode,   advanced one pass,                  4840758
-silesia,                            small window log,                   advanced one pass,                  7095919
+github.tar,                         explicit params,                    zstdcli,                            41385
+github.tar,                         uncompressed literals,              zstdcli,                            41529
+github.tar,                         uncompressed literals optimal,      zstdcli,                            35401
+github.tar,                         huffman literals,                   zstdcli,                            38857
+github.tar,                         multithreaded with advanced params, zstdcli,                            41529
+silesia,                            level -5,                           advanced one pass,                  6857372
+silesia,                            level -3,                           advanced one pass,                  6503412
+silesia,                            level -1,                           advanced one pass,                  6172202
+silesia,                            level 0,                            advanced one pass,                  4842075
+silesia,                            level 1,                            advanced one pass,                  5306632
+silesia,                            level 3,                            advanced one pass,                  4842075
+silesia,                            level 4,                            advanced one pass,                  4779186
+silesia,                            level 5 row 1,                      advanced one pass,                  4666323
+silesia,                            level 5 row 2,                      advanced one pass,                  4670136
+silesia,                            level 5,                            advanced one pass,                  4666323
+silesia,                            level 6,                            advanced one pass,                  4603066
+silesia,                            level 7 row 1,                      advanced one pass,                  4566984
+silesia,                            level 7 row 2,                      advanced one pass,                  4564868
+silesia,                            level 7,                            advanced one pass,                  4566984
+silesia,                            level 9,                            advanced one pass,                  4543018
+silesia,                            level 11 row 1,                     advanced one pass,                  4505046
+silesia,                            level 11 row 2,                     advanced one pass,                  4503116
+silesia,                            level 12 row 1,                     advanced one pass,                  4505046
+silesia,                            level 12 row 2,                     advanced one pass,                  4503116
+silesia,                            level 13,                           advanced one pass,                  4493990
+silesia,                            level 16,                           advanced one pass,                  4360041
+silesia,                            level 19,                           advanced one pass,                  4296055
+silesia,                            no source size,                     advanced one pass,                  4842075
+silesia,                            long distance mode,                 advanced one pass,                  4833710
+silesia,                            multithreaded,                      advanced one pass,                  4842075
+silesia,                            multithreaded long distance mode,   advanced one pass,                  4833737
+silesia,                            small window log,                   advanced one pass,                  7095000
 silesia,                            small hash log,                     advanced one pass,                  6526141
 silesia,                            small chain log,                    advanced one pass,                  4912197
-silesia,                            explicit params,                    advanced one pass,                  4795856
-silesia,                            uncompressed literals,              advanced one pass,                  5127982
-silesia,                            uncompressed literals optimal,      advanced one pass,                  4317896
-silesia,                            huffman literals,                   advanced one pass,                  5326268
-silesia,                            multithreaded with advanced params, advanced one pass,                  5127982
-silesia.tar,                        level -5,                           advanced one pass,                  6738593
-silesia.tar,                        level -3,                           advanced one pass,                  6446372
-silesia.tar,                        level -1,                           advanced one pass,                  6186042
-silesia.tar,                        level 0,                            advanced one pass,                  4861425
-silesia.tar,                        level 1,                            advanced one pass,                  5334885
-silesia.tar,                        level 3,                            advanced one pass,                  4861425
-silesia.tar,                        level 4,                            advanced one pass,                  4799630
-silesia.tar,                        level 5 row 1,                      advanced one pass,                  4722324
-silesia.tar,                        level 5 row 2,                      advanced one pass,                  4719256
-silesia.tar,                        level 5,                            advanced one pass,                  4719256
-silesia.tar,                        level 6,                            advanced one pass,                  4677721
-silesia.tar,                        level 7 row 1,                      advanced one pass,                  4606715
-silesia.tar,                        level 7 row 2,                      advanced one pass,                  4613541
-silesia.tar,                        level 7,                            advanced one pass,                  4613541
-silesia.tar,                        level 9,                            advanced one pass,                  4555426
-silesia.tar,                        level 12 row 1,                     advanced one pass,                  4529459
-silesia.tar,                        level 12 row 2,                     advanced one pass,                  4530256
-silesia.tar,                        level 13,                           advanced one pass,                  4491764
-silesia.tar,                        level 16,                           advanced one pass,                  4356827
-silesia.tar,                        level 19,                           advanced one pass,                  4264487
-silesia.tar,                        no source size,                     advanced one pass,                  4861425
-silesia.tar,                        long distance mode,                 advanced one pass,                  4847754
-silesia.tar,                        multithreaded,                      advanced one pass,                  4861508
-silesia.tar,                        multithreaded long distance mode,   advanced one pass,                  4853222
-silesia.tar,                        small window log,                   advanced one pass,                  7101530
-silesia.tar,                        small hash log,                     advanced one pass,                  6529232
+silesia,                            explicit params,                    advanced one pass,                  4795432
+silesia,                            uncompressed literals,              advanced one pass,                  5120566
+silesia,                            uncompressed literals optimal,      advanced one pass,                  4319518
+silesia,                            huffman literals,                   advanced one pass,                  5321369
+silesia,                            multithreaded with advanced params, advanced one pass,                  5120566
+silesia.tar,                        level -5,                           advanced one pass,                  6861055
+silesia.tar,                        level -3,                           advanced one pass,                  6505483
+silesia.tar,                        level -1,                           advanced one pass,                  6179047
+silesia.tar,                        level 0,                            advanced one pass,                  4854086
+silesia.tar,                        level 1,                            advanced one pass,                  5327717
+silesia.tar,                        level 3,                            advanced one pass,                  4854086
+silesia.tar,                        level 4,                            advanced one pass,                  4791503
+silesia.tar,                        level 5 row 1,                      advanced one pass,                  4677740
+silesia.tar,                        level 5 row 2,                      advanced one pass,                  4682161
+silesia.tar,                        level 5,                            advanced one pass,                  4677740
+silesia.tar,                        level 6,                            advanced one pass,                  4613242
+silesia.tar,                        level 7 row 1,                      advanced one pass,                  4576661
+silesia.tar,                        level 7 row 2,                      advanced one pass,                  4575393
+silesia.tar,                        level 7,                            advanced one pass,                  4576661
+silesia.tar,                        level 9,                            advanced one pass,                  4552899
+silesia.tar,                        level 11 row 1,                     advanced one pass,                  4514432
+silesia.tar,                        level 11 row 2,                     advanced one pass,                  4513604
+silesia.tar,                        level 12 row 1,                     advanced one pass,                  4514049
+silesia.tar,                        level 12 row 2,                     advanced one pass,                  4513797
+silesia.tar,                        level 13,                           advanced one pass,                  4502956
+silesia.tar,                        level 16,                           advanced one pass,                  4360546
+silesia.tar,                        level 19,                           advanced one pass,                  4265911
+silesia.tar,                        no source size,                     advanced one pass,                  4854086
+silesia.tar,                        long distance mode,                 advanced one pass,                  4840452
+silesia.tar,                        multithreaded,                      advanced one pass,                  4854160
+silesia.tar,                        multithreaded long distance mode,   advanced one pass,                  4845741
+silesia.tar,                        small window log,                   advanced one pass,                  7100655
+silesia.tar,                        small hash log,                     advanced one pass,                  6529206
 silesia.tar,                        small chain log,                    advanced one pass,                  4917041
-silesia.tar,                        explicit params,                    advanced one pass,                  4807380
-silesia.tar,                        uncompressed literals,              advanced one pass,                  5129458
-silesia.tar,                        uncompressed literals optimal,      advanced one pass,                  4307453
-silesia.tar,                        huffman literals,                   advanced one pass,                  5347335
-silesia.tar,                        multithreaded with advanced params, advanced one pass,                  5129555
-github,                             level -5,                           advanced one pass,                  205285
+silesia.tar,                        explicit params,                    advanced one pass,                  4806855
+silesia.tar,                        uncompressed literals,              advanced one pass,                  5122473
+silesia.tar,                        uncompressed literals optimal,      advanced one pass,                  4310141
+silesia.tar,                        huffman literals,                   advanced one pass,                  5341705
+silesia.tar,                        multithreaded with advanced params, advanced one pass,                  5122567
+github,                             level -5,                           advanced one pass,                  204407
 github,                             level -5 with dict,                 advanced one pass,                  46718
-github,                             level -3,                           advanced one pass,                  190643
+github,                             level -3,                           advanced one pass,                  193253
 github,                             level -3 with dict,                 advanced one pass,                  45395
-github,                             level -1,                           advanced one pass,                  175568
+github,                             level -1,                           advanced one pass,                  175468
 github,                             level -1 with dict,                 advanced one pass,                  43170
-github,                             level 0,                            advanced one pass,                  136335
+github,                             level 0,                            advanced one pass,                  136332
 github,                             level 0 with dict,                  advanced one pass,                  41148
 github,                             level 0 with dict dms,              advanced one pass,                  41148
 github,                             level 0 with dict dds,              advanced one pass,                  41148
 github,                             level 0 with dict copy,             advanced one pass,                  41124
 github,                             level 0 with dict load,             advanced one pass,                  42252
-github,                             level 1,                            advanced one pass,                  142465
+github,                             level 1,                            advanced one pass,                  142365
 github,                             level 1 with dict,                  advanced one pass,                  41682
 github,                             level 1 with dict dms,              advanced one pass,                  41682
 github,                             level 1 with dict dds,              advanced one pass,                  41682
-github,                             level 1 with dict copy,             advanced one pass,                  41674
-github,                             level 1 with dict load,             advanced one pass,                  43755
-github,                             level 3,                            advanced one pass,                  136335
+github,                             level 1 with dict copy,             advanced one pass,                  41698
+github,                             level 1 with dict load,             advanced one pass,                  43814
+github,                             level 3,                            advanced one pass,                  136332
 github,                             level 3 with dict,                  advanced one pass,                  41148
 github,                             level 3 with dict dms,              advanced one pass,                  41148
 github,                             level 3 with dict dds,              advanced one pass,                  41148
@@ -322,16 +326,16 @@ github,                             level 4 with dict dms,              advanced
 github,                             level 4 with dict dds,              advanced one pass,                  41251
 github,                             level 4 with dict copy,             advanced one pass,                  41216
 github,                             level 4 with dict load,             advanced one pass,                  41159
-github,                             level 5 row 1,                      advanced one pass,                  135121
-github,                             level 5 row 1 with dict dms,        advanced one pass,                  38938
-github,                             level 5 row 1 with dict dds,        advanced one pass,                  38732
-github,                             level 5 row 1 with dict copy,       advanced one pass,                  38934
-github,                             level 5 row 1 with dict load,       advanced one pass,                  40725
-github,                             level 5 row 2,                      advanced one pass,                  134584
-github,                             level 5 row 2 with dict dms,        advanced one pass,                  38758
-github,                             level 5 row 2 with dict dds,        advanced one pass,                  38728
-github,                             level 5 row 2 with dict copy,       advanced one pass,                  38759
-github,                             level 5 row 2 with dict load,       advanced one pass,                  41518
+github,                             level 5 row 1,                      advanced one pass,                  134584
+github,                             level 5 row 1 with dict dms,        advanced one pass,                  38758
+github,                             level 5 row 1 with dict dds,        advanced one pass,                  38728
+github,                             level 5 row 1 with dict copy,       advanced one pass,                  38759
+github,                             level 5 row 1 with dict load,       advanced one pass,                  41518
+github,                             level 5 row 2,                      advanced one pass,                  135121
+github,                             level 5 row 2 with dict dms,        advanced one pass,                  38938
+github,                             level 5 row 2 with dict dds,        advanced one pass,                  38732
+github,                             level 5 row 2 with dict copy,       advanced one pass,                  38934
+github,                             level 5 row 2 with dict load,       advanced one pass,                  40725
 github,                             level 5,                            advanced one pass,                  135121
 github,                             level 5 with dict,                  advanced one pass,                  38758
 github,                             level 5 with dict dms,              advanced one pass,                  38758
@@ -341,270 +345,294 @@ github,                             level 5 with dict load,             advanced
 github,                             level 6,                            advanced one pass,                  135122
 github,                             level 6 with dict,                  advanced one pass,                  38671
 github,                             level 6 with dict dms,              advanced one pass,                  38671
-github,                             level 6 with dict dds,              advanced one pass,                  38630
+github,                             level 6 with dict dds,              advanced one pass,                  38636
 github,                             level 6 with dict copy,             advanced one pass,                  38669
 github,                             level 6 with dict load,             advanced one pass,                  40695
-github,                             level 7 row 1,                      advanced one pass,                  135122
-github,                             level 7 row 1 with dict dms,        advanced one pass,                  38771
-github,                             level 7 row 1 with dict dds,        advanced one pass,                  38771
-github,                             level 7 row 1 with dict copy,       advanced one pass,                  38745
-github,                             level 7 row 1 with dict load,       advanced one pass,                  40695
-github,                             level 7 row 2,                      advanced one pass,                  134584
-github,                             level 7 row 2 with dict dms,        advanced one pass,                  38758
-github,                             level 7 row 2 with dict dds,        advanced one pass,                  38747
-github,                             level 7 row 2 with dict copy,       advanced one pass,                  38755
-github,                             level 7 row 2 with dict load,       advanced one pass,                  41030
+github,                             level 7 row 1,                      advanced one pass,                  134584
+github,                             level 7 row 1 with dict dms,        advanced one pass,                  38758
+github,                             level 7 row 1 with dict dds,        advanced one pass,                  38745
+github,                             level 7 row 1 with dict copy,       advanced one pass,                  38755
+github,                             level 7 row 1 with dict load,       advanced one pass,                  43154
+github,                             level 7 row 2,                      advanced one pass,                  135122
+github,                             level 7 row 2 with dict dms,        advanced one pass,                  38860
+github,                             level 7 row 2 with dict dds,        advanced one pass,                  38766
+github,                             level 7 row 2 with dict copy,       advanced one pass,                  38834
+github,                             level 7 row 2 with dict load,       advanced one pass,                  40695
 github,                             level 7,                            advanced one pass,                  135122
 github,                             level 7 with dict,                  advanced one pass,                  38758
 github,                             level 7 with dict dms,              advanced one pass,                  38758
-github,                             level 7 with dict dds,              advanced one pass,                  38747
+github,                             level 7 with dict dds,              advanced one pass,                  38745
 github,                             level 7 with dict copy,             advanced one pass,                  38755
 github,                             level 7 with dict load,             advanced one pass,                  40695
 github,                             level 9,                            advanced one pass,                  135122
 github,                             level 9 with dict,                  advanced one pass,                  39437
 github,                             level 9 with dict dms,              advanced one pass,                  39437
-github,                             level 9 with dict dds,              advanced one pass,                  39338
+github,                             level 9 with dict dds,              advanced one pass,                  39393
 github,                             level 9 with dict copy,             advanced one pass,                  39398
 github,                             level 9 with dict load,             advanced one pass,                  41710
-github,                             level 12 row 1,                     advanced one pass,                  134180
+github,                             level 11 row 1,                     advanced one pass,                  135367
+github,                             level 11 row 1 with dict dms,       advanced one pass,                  39671
+github,                             level 11 row 1 with dict dds,       advanced one pass,                  39671
+github,                             level 11 row 1 with dict copy,      advanced one pass,                  39651
+github,                             level 11 row 1 with dict load,      advanced one pass,                  41360
+github,                             level 11 row 2,                     advanced one pass,                  135367
+github,                             level 11 row 2 with dict dms,       advanced one pass,                  39671
+github,                             level 11 row 2 with dict dds,       advanced one pass,                  39671
+github,                             level 11 row 2 with dict copy,      advanced one pass,                  39651
+github,                             level 11 row 2 with dict load,      advanced one pass,                  41360
+github,                             level 12 row 1,                     advanced one pass,                  134402
 github,                             level 12 row 1 with dict dms,       advanced one pass,                  39677
 github,                             level 12 row 1 with dict dds,       advanced one pass,                  39677
 github,                             level 12 row 1 with dict copy,      advanced one pass,                  39677
 github,                             level 12 row 1 with dict load,      advanced one pass,                  41166
-github,                             level 12 row 2,                     advanced one pass,                  134180
+github,                             level 12 row 2,                     advanced one pass,                  134402
 github,                             level 12 row 2 with dict dms,       advanced one pass,                  39677
 github,                             level 12 row 2 with dict dds,       advanced one pass,                  39677
 github,                             level 12 row 2 with dict copy,      advanced one pass,                  39677
 github,                             level 12 row 2 with dict load,      advanced one pass,                  41166
-github,                             level 13,                           advanced one pass,                  134064
-github,                             level 13 with dict,                 advanced one pass,                  39743
-github,                             level 13 with dict dms,             advanced one pass,                  39743
-github,                             level 13 with dict dds,             advanced one pass,                  39743
+github,                             level 13,                           advanced one pass,                  132878
+github,                             level 13 with dict,                 advanced one pass,                  39900
+github,                             level 13 with dict dms,             advanced one pass,                  39900
+github,                             level 13 with dict dds,             advanced one pass,                  39900
 github,                             level 13 with dict copy,            advanced one pass,                  39948
-github,                             level 13 with dict load,            advanced one pass,                  42626
-github,                             level 16,                           advanced one pass,                  134064
+github,                             level 13 with dict load,            advanced one pass,                  42624
+github,                             level 16,                           advanced one pass,                  133209
 github,                             level 16 with dict,                 advanced one pass,                  37577
 github,                             level 16 with dict dms,             advanced one pass,                  37577
 github,                             level 16 with dict dds,             advanced one pass,                  37577
 github,                             level 16 with dict copy,            advanced one pass,                  37568
-github,                             level 16 with dict load,            advanced one pass,                  42340
-github,                             level 19,                           advanced one pass,                  134064
+github,                             level 16 with dict load,            advanced one pass,                  42338
+github,                             level 19,                           advanced one pass,                  132879
 github,                             level 19 with dict,                 advanced one pass,                  37576
 github,                             level 19 with dict dms,             advanced one pass,                  37576
 github,                             level 19 with dict dds,             advanced one pass,                  37576
 github,                             level 19 with dict copy,            advanced one pass,                  37567
 github,                             level 19 with dict load,            advanced one pass,                  39613
-github,                             no source size,                     advanced one pass,                  136335
+github,                             no source size,                     advanced one pass,                  136332
 github,                             no source size with dict,           advanced one pass,                  41148
-github,                             long distance mode,                 advanced one pass,                  136335
-github,                             multithreaded,                      advanced one pass,                  136335
-github,                             multithreaded long distance mode,   advanced one pass,                  136335
-github,                             small window log,                   advanced one pass,                  136335
+github,                             long distance mode,                 advanced one pass,                  136332
+github,                             multithreaded,                      advanced one pass,                  136332
+github,                             multithreaded long distance mode,   advanced one pass,                  136332
+github,                             small window log,                   advanced one pass,                  136332
 github,                             small hash log,                     advanced one pass,                  135590
 github,                             small chain log,                    advanced one pass,                  136341
 github,                             explicit params,                    advanced one pass,                  137727
-github,                             uncompressed literals,              advanced one pass,                  165915
-github,                             uncompressed literals optimal,      advanced one pass,                  157227
-github,                             huffman literals,                   advanced one pass,                  142465
-github,                             multithreaded with advanced params, advanced one pass,                  165915
-github.tar,                         level -5,                           advanced one pass,                  46856
-github.tar,                         level -5 with dict,                 advanced one pass,                  44571
-github.tar,                         level -3,                           advanced one pass,                  43754
-github.tar,                         level -3 with dict,                 advanced one pass,                  41447
-github.tar,                         level -1,                           advanced one pass,                  42490
-github.tar,                         level -1 with dict,                 advanced one pass,                  41131
-github.tar,                         level 0,                            advanced one pass,                  38441
+github,                             uncompressed literals,              advanced one pass,                  165911
+github,                             uncompressed literals optimal,      advanced one pass,                  152667
+github,                             huffman literals,                   advanced one pass,                  142365
+github,                             multithreaded with advanced params, advanced one pass,                  165911
+github.tar,                         level -5,                           advanced one pass,                  52115
+github.tar,                         level -5 with dict,                 advanced one pass,                  50974
+github.tar,                         level -3,                           advanced one pass,                  45678
+github.tar,                         level -3 with dict,                 advanced one pass,                  44656
+github.tar,                         level -1,                           advanced one pass,                  42560
+github.tar,                         level -1 with dict,                 advanced one pass,                  41151
+github.tar,                         level 0,                            advanced one pass,                  38831
 github.tar,                         level 0 with dict,                  advanced one pass,                  37995
 github.tar,                         level 0 with dict dms,              advanced one pass,                  38003
 github.tar,                         level 0 with dict dds,              advanced one pass,                  38003
 github.tar,                         level 0 with dict copy,             advanced one pass,                  37995
 github.tar,                         level 0 with dict load,             advanced one pass,                  37956
-github.tar,                         level 1,                            advanced one pass,                  39265
-github.tar,                         level 1 with dict,                  advanced one pass,                  38280
-github.tar,                         level 1 with dict dms,              advanced one pass,                  38290
-github.tar,                         level 1 with dict dds,              advanced one pass,                  38290
-github.tar,                         level 1 with dict copy,             advanced one pass,                  38280
-github.tar,                         level 1 with dict load,             advanced one pass,                  38729
-github.tar,                         level 3,                            advanced one pass,                  38441
+github.tar,                         level 1,                            advanced one pass,                  39200
+github.tar,                         level 1 with dict,                  advanced one pass,                  38089
+github.tar,                         level 1 with dict dms,              advanced one pass,                  38294
+github.tar,                         level 1 with dict dds,              advanced one pass,                  38294
+github.tar,                         level 1 with dict copy,             advanced one pass,                  38089
+github.tar,                         level 1 with dict load,             advanced one pass,                  38364
+github.tar,                         level 3,                            advanced one pass,                  38831
 github.tar,                         level 3 with dict,                  advanced one pass,                  37995
 github.tar,                         level 3 with dict dms,              advanced one pass,                  38003
 github.tar,                         level 3 with dict dds,              advanced one pass,                  38003
 github.tar,                         level 3 with dict copy,             advanced one pass,                  37995
 github.tar,                         level 3 with dict load,             advanced one pass,                  37956
-github.tar,                         level 4,                            advanced one pass,                  38467
+github.tar,                         level 4,                            advanced one pass,                  38893
 github.tar,                         level 4 with dict,                  advanced one pass,                  37948
 github.tar,                         level 4 with dict dms,              advanced one pass,                  37954
 github.tar,                         level 4 with dict dds,              advanced one pass,                  37954
 github.tar,                         level 4 with dict copy,             advanced one pass,                  37948
 github.tar,                         level 4 with dict load,             advanced one pass,                  37927
-github.tar,                         level 5 row 1,                      advanced one pass,                  39788
-github.tar,                         level 5 row 1 with dict dms,        advanced one pass,                  39365
-github.tar,                         level 5 row 1 with dict dds,        advanced one pass,                  39233
-github.tar,                         level 5 row 1 with dict copy,       advanced one pass,                  39715
-github.tar,                         level 5 row 1 with dict load,       advanced one pass,                  39209
-github.tar,                         level 5 row 2,                      advanced one pass,                  39693
-github.tar,                         level 5 row 2 with dict dms,        advanced one pass,                  39024
-github.tar,                         level 5 row 2 with dict dds,        advanced one pass,                  39028
-github.tar,                         level 5 row 2 with dict copy,       advanced one pass,                  39040
-github.tar,                         level 5 row 2 with dict load,       advanced one pass,                  39037
-github.tar,                         level 5,                            advanced one pass,                  39693
-github.tar,                         level 5 with dict,                  advanced one pass,                  39040
-github.tar,                         level 5 with dict dms,              advanced one pass,                  39024
-github.tar,                         level 5 with dict dds,              advanced one pass,                  39028
-github.tar,                         level 5 with dict copy,             advanced one pass,                  39040
-github.tar,                         level 5 with dict load,             advanced one pass,                  39037
-github.tar,                         level 6,                            advanced one pass,                  39621
-github.tar,                         level 6 with dict,                  advanced one pass,                  38622
-github.tar,                         level 6 with dict dms,              advanced one pass,                  38608
-github.tar,                         level 6 with dict dds,              advanced one pass,                  38610
-github.tar,                         level 6 with dict copy,             advanced one pass,                  38622
-github.tar,                         level 6 with dict load,             advanced one pass,                  38962
-github.tar,                         level 7 row 1,                      advanced one pass,                  39206
-github.tar,                         level 7 row 1 with dict dms,        advanced one pass,                  37954
-github.tar,                         level 7 row 1 with dict dds,        advanced one pass,                  37954
-github.tar,                         level 7 row 1 with dict copy,       advanced one pass,                  38071
-github.tar,                         level 7 row 1 with dict load,       advanced one pass,                  38584
-github.tar,                         level 7 row 2,                      advanced one pass,                  39213
-github.tar,                         level 7 row 2 with dict dms,        advanced one pass,                  37848
-github.tar,                         level 7 row 2 with dict dds,        advanced one pass,                  37867
-github.tar,                         level 7 row 2 with dict copy,       advanced one pass,                  37848
-github.tar,                         level 7 row 2 with dict load,       advanced one pass,                  38582
-github.tar,                         level 7,                            advanced one pass,                  39213
-github.tar,                         level 7 with dict,                  advanced one pass,                  37848
-github.tar,                         level 7 with dict dms,              advanced one pass,                  37848
-github.tar,                         level 7 with dict dds,              advanced one pass,                  37867
-github.tar,                         level 7 with dict copy,             advanced one pass,                  37848
-github.tar,                         level 7 with dict load,             advanced one pass,                  38582
-github.tar,                         level 9,                            advanced one pass,                  36758
-github.tar,                         level 9 with dict,                  advanced one pass,                  36457
-github.tar,                         level 9 with dict dms,              advanced one pass,                  36549
-github.tar,                         level 9 with dict dds,              advanced one pass,                  36637
-github.tar,                         level 9 with dict copy,             advanced one pass,                  36457
-github.tar,                         level 9 with dict load,             advanced one pass,                  36350
-github.tar,                         level 12 row 1,                     advanced one pass,                  36435
+github.tar,                         level 5 row 1,                      advanced one pass,                  39651
+github.tar,                         level 5 row 1 with dict dms,        advanced one pass,                  39059
+github.tar,                         level 5 row 1 with dict dds,        advanced one pass,                  39067
+github.tar,                         level 5 row 1 with dict copy,       advanced one pass,                  39082
+github.tar,                         level 5 row 1 with dict load,       advanced one pass,                  38999
+github.tar,                         level 5 row 2,                      advanced one pass,                  39701
+github.tar,                         level 5 row 2 with dict dms,        advanced one pass,                  39365
+github.tar,                         level 5 row 2 with dict dds,        advanced one pass,                  39233
+github.tar,                         level 5 row 2 with dict copy,       advanced one pass,                  39715
+github.tar,                         level 5 row 2 with dict load,       advanced one pass,                  39158
+github.tar,                         level 5,                            advanced one pass,                  39651
+github.tar,                         level 5 with dict,                  advanced one pass,                  39082
+github.tar,                         level 5 with dict dms,              advanced one pass,                  39059
+github.tar,                         level 5 with dict dds,              advanced one pass,                  39067
+github.tar,                         level 5 with dict copy,             advanced one pass,                  39082
+github.tar,                         level 5 with dict load,             advanced one pass,                  38999
+github.tar,                         level 6,                            advanced one pass,                  39282
+github.tar,                         level 6 with dict,                  advanced one pass,                  38656
+github.tar,                         level 6 with dict dms,              advanced one pass,                  38636
+github.tar,                         level 6 with dict dds,              advanced one pass,                  38634
+github.tar,                         level 6 with dict copy,             advanced one pass,                  38656
+github.tar,                         level 6 with dict load,             advanced one pass,                  38648
+github.tar,                         level 7 row 1,                      advanced one pass,                  38110
+github.tar,                         level 7 row 1 with dict dms,        advanced one pass,                  37858
+github.tar,                         level 7 row 1 with dict dds,        advanced one pass,                  37882
+github.tar,                         level 7 row 1 with dict copy,       advanced one pass,                  37865
+github.tar,                         level 7 row 1 with dict load,       advanced one pass,                  37436
+github.tar,                         level 7 row 2,                      advanced one pass,                  38077
+github.tar,                         level 7 row 2 with dict dms,        advanced one pass,                  38012
+github.tar,                         level 7 row 2 with dict dds,        advanced one pass,                  38014
+github.tar,                         level 7 row 2 with dict copy,       advanced one pass,                  38101
+github.tar,                         level 7 row 2 with dict load,       advanced one pass,                  37402
+github.tar,                         level 7,                            advanced one pass,                  38110
+github.tar,                         level 7 with dict,                  advanced one pass,                  37865
+github.tar,                         level 7 with dict dms,              advanced one pass,                  37858
+github.tar,                         level 7 with dict dds,              advanced one pass,                  37882
+github.tar,                         level 7 with dict copy,             advanced one pass,                  37865
+github.tar,                         level 7 with dict load,             advanced one pass,                  37436
+github.tar,                         level 9,                            advanced one pass,                  36760
+github.tar,                         level 9 with dict,                  advanced one pass,                  36484
+github.tar,                         level 9 with dict dms,              advanced one pass,                  36567
+github.tar,                         level 9 with dict dds,              advanced one pass,                  36628
+github.tar,                         level 9 with dict copy,             advanced one pass,                  36484
+github.tar,                         level 9 with dict load,             advanced one pass,                  36401
+github.tar,                         level 11 row 1,                     advanced one pass,                  36081
+github.tar,                         level 11 row 1 with dict dms,       advanced one pass,                  36963
+github.tar,                         level 11 row 1 with dict dds,       advanced one pass,                  36963
+github.tar,                         level 11 row 1 with dict copy,      advanced one pass,                  36557
+github.tar,                         level 11 row 1 with dict load,      advanced one pass,                  36434
+github.tar,                         level 11 row 2,                     advanced one pass,                  36110
+github.tar,                         level 11 row 2 with dict dms,       advanced one pass,                  36963
+github.tar,                         level 11 row 2 with dict dds,       advanced one pass,                  36963
+github.tar,                         level 11 row 2 with dict copy,      advanced one pass,                  36557
+github.tar,                         level 11 row 2 with dict load,      advanced one pass,                  36459
+github.tar,                         level 12 row 1,                     advanced one pass,                  36081
 github.tar,                         level 12 row 1 with dict dms,       advanced one pass,                  36986
 github.tar,                         level 12 row 1 with dict dds,       advanced one pass,                  36986
 github.tar,                         level 12 row 1 with dict copy,      advanced one pass,                  36609
-github.tar,                         level 12 row 1 with dict load,      advanced one pass,                  36419
-github.tar,                         level 12 row 2,                     advanced one pass,                  36435
+github.tar,                         level 12 row 1 with dict load,      advanced one pass,                  36434
+github.tar,                         level 12 row 2,                     advanced one pass,                  36110
 github.tar,                         level 12 row 2 with dict dms,       advanced one pass,                  36986
 github.tar,                         level 12 row 2 with dict dds,       advanced one pass,                  36986
 github.tar,                         level 12 row 2 with dict copy,      advanced one pass,                  36609
-github.tar,                         level 12 row 2 with dict load,      advanced one pass,                  36424
-github.tar,                         level 13,                           advanced one pass,                  35621
-github.tar,                         level 13 with dict,                 advanced one pass,                  38726
-github.tar,                         level 13 with dict dms,             advanced one pass,                  38903
-github.tar,                         level 13 with dict dds,             advanced one pass,                  38903
-github.tar,                         level 13 with dict copy,            advanced one pass,                  38726
-github.tar,                         level 13 with dict load,            advanced one pass,                  36372
-github.tar,                         level 16,                           advanced one pass,                  40255
-github.tar,                         level 16 with dict,                 advanced one pass,                  33639
-github.tar,                         level 16 with dict dms,             advanced one pass,                  33544
-github.tar,                         level 16 with dict dds,             advanced one pass,                  33544
-github.tar,                         level 16 with dict copy,            advanced one pass,                  33639
-github.tar,                         level 16 with dict load,            advanced one pass,                  39353
-github.tar,                         level 19,                           advanced one pass,                  32837
-github.tar,                         level 19 with dict,                 advanced one pass,                  32895
-github.tar,                         level 19 with dict dms,             advanced one pass,                  32672
-github.tar,                         level 19 with dict dds,             advanced one pass,                  32672
-github.tar,                         level 19 with dict copy,            advanced one pass,                  32895
-github.tar,                         level 19 with dict load,            advanced one pass,                  32676
-github.tar,                         no source size,                     advanced one pass,                  38441
+github.tar,                         level 12 row 2 with dict load,      advanced one pass,                  36459
+github.tar,                         level 13,                           advanced one pass,                  35501
+github.tar,                         level 13 with dict,                 advanced one pass,                  37130
+github.tar,                         level 13 with dict dms,             advanced one pass,                  37220
+github.tar,                         level 13 with dict dds,             advanced one pass,                  37220
+github.tar,                         level 13 with dict copy,            advanced one pass,                  37130
+github.tar,                         level 13 with dict load,            advanced one pass,                  36010
+github.tar,                         level 16,                           advanced one pass,                  40466
+github.tar,                         level 16 with dict,                 advanced one pass,                  33374
+github.tar,                         level 16 with dict dms,             advanced one pass,                  33206
+github.tar,                         level 16 with dict dds,             advanced one pass,                  33206
+github.tar,                         level 16 with dict copy,            advanced one pass,                  33374
+github.tar,                         level 16 with dict load,            advanced one pass,                  39081
+github.tar,                         level 19,                           advanced one pass,                  32276
+github.tar,                         level 19 with dict,                 advanced one pass,                  32712
+github.tar,                         level 19 with dict dms,             advanced one pass,                  32555
+github.tar,                         level 19 with dict dds,             advanced one pass,                  32555
+github.tar,                         level 19 with dict copy,            advanced one pass,                  32712
+github.tar,                         level 19 with dict load,            advanced one pass,                  32479
+github.tar,                         no source size,                     advanced one pass,                  38831
 github.tar,                         no source size with dict,           advanced one pass,                  37995
-github.tar,                         long distance mode,                 advanced one pass,                  39757
-github.tar,                         multithreaded,                      advanced one pass,                  38441
-github.tar,                         multithreaded long distance mode,   advanced one pass,                  39726
+github.tar,                         long distance mode,                 advanced one pass,                  40252
+github.tar,                         multithreaded,                      advanced one pass,                  38831
+github.tar,                         multithreaded long distance mode,   advanced one pass,                  40232
 github.tar,                         small window log,                   advanced one pass,                  198540
 github.tar,                         small hash log,                     advanced one pass,                  129870
 github.tar,                         small chain log,                    advanced one pass,                  41669
-github.tar,                         explicit params,                    advanced one pass,                  41227
-github.tar,                         uncompressed literals,              advanced one pass,                  41122
-github.tar,                         uncompressed literals optimal,      advanced one pass,                  35388
-github.tar,                         huffman literals,                   advanced one pass,                  38777
-github.tar,                         multithreaded with advanced params, advanced one pass,                  41122
-silesia,                            level -5,                           advanced one pass small out,        6737607
-silesia,                            level -3,                           advanced one pass small out,        6444677
-silesia,                            level -1,                           advanced one pass small out,        6178460
-silesia,                            level 0,                            advanced one pass small out,        4849552
-silesia,                            level 1,                            advanced one pass small out,        5313204
-silesia,                            level 3,                            advanced one pass small out,        4849552
-silesia,                            level 4,                            advanced one pass small out,        4786970
-silesia,                            level 5 row 1,                      advanced one pass small out,        4710236
-silesia,                            level 5 row 2,                      advanced one pass small out,        4707794
-silesia,                            level 5,                            advanced one pass small out,        4707794
-silesia,                            level 6,                            advanced one pass small out,        4666383
-silesia,                            level 7 row 1,                      advanced one pass small out,        4596296
-silesia,                            level 7 row 2,                      advanced one pass small out,        4603381
-silesia,                            level 7,                            advanced one pass small out,        4603381
-silesia,                            level 9,                            advanced one pass small out,        4546001
-silesia,                            level 12 row 1,                     advanced one pass small out,        4519288
-silesia,                            level 12 row 2,                     advanced one pass small out,        4521397
-silesia,                            level 13,                           advanced one pass small out,        4482135
-silesia,                            level 16,                           advanced one pass small out,        4360251
-silesia,                            level 19,                           advanced one pass small out,        4283237
-silesia,                            no source size,                     advanced one pass small out,        4849552
-silesia,                            long distance mode,                 advanced one pass small out,        4840738
-silesia,                            multithreaded,                      advanced one pass small out,        4849552
-silesia,                            multithreaded long distance mode,   advanced one pass small out,        4840758
-silesia,                            small window log,                   advanced one pass small out,        7095919
+github.tar,                         explicit params,                    advanced one pass,                  41385
+github.tar,                         uncompressed literals,              advanced one pass,                  41525
+github.tar,                         uncompressed literals optimal,      advanced one pass,                  35397
+github.tar,                         huffman literals,                   advanced one pass,                  38853
+github.tar,                         multithreaded with advanced params, advanced one pass,                  41525
+silesia,                            level -5,                           advanced one pass small out,        6857372
+silesia,                            level -3,                           advanced one pass small out,        6503412
+silesia,                            level -1,                           advanced one pass small out,        6172202
+silesia,                            level 0,                            advanced one pass small out,        4842075
+silesia,                            level 1,                            advanced one pass small out,        5306632
+silesia,                            level 3,                            advanced one pass small out,        4842075
+silesia,                            level 4,                            advanced one pass small out,        4779186
+silesia,                            level 5 row 1,                      advanced one pass small out,        4666323
+silesia,                            level 5 row 2,                      advanced one pass small out,        4670136
+silesia,                            level 5,                            advanced one pass small out,        4666323
+silesia,                            level 6,                            advanced one pass small out,        4603066
+silesia,                            level 7 row 1,                      advanced one pass small out,        4566984
+silesia,                            level 7 row 2,                      advanced one pass small out,        4564868
+silesia,                            level 7,                            advanced one pass small out,        4566984
+silesia,                            level 9,                            advanced one pass small out,        4543018
+silesia,                            level 11 row 1,                     advanced one pass small out,        4505046
+silesia,                            level 11 row 2,                     advanced one pass small out,        4503116
+silesia,                            level 12 row 1,                     advanced one pass small out,        4505046
+silesia,                            level 12 row 2,                     advanced one pass small out,        4503116
+silesia,                            level 13,                           advanced one pass small out,        4493990
+silesia,                            level 16,                           advanced one pass small out,        4360041
+silesia,                            level 19,                           advanced one pass small out,        4296055
+silesia,                            no source size,                     advanced one pass small out,        4842075
+silesia,                            long distance mode,                 advanced one pass small out,        4833710
+silesia,                            multithreaded,                      advanced one pass small out,        4842075
+silesia,                            multithreaded long distance mode,   advanced one pass small out,        4833737
+silesia,                            small window log,                   advanced one pass small out,        7095000
 silesia,                            small hash log,                     advanced one pass small out,        6526141
 silesia,                            small chain log,                    advanced one pass small out,        4912197
-silesia,                            explicit params,                    advanced one pass small out,        4795856
-silesia,                            uncompressed literals,              advanced one pass small out,        5127982
-silesia,                            uncompressed literals optimal,      advanced one pass small out,        4317896
-silesia,                            huffman literals,                   advanced one pass small out,        5326268
-silesia,                            multithreaded with advanced params, advanced one pass small out,        5127982
-silesia.tar,                        level -5,                           advanced one pass small out,        6738593
-silesia.tar,                        level -3,                           advanced one pass small out,        6446372
-silesia.tar,                        level -1,                           advanced one pass small out,        6186042
-silesia.tar,                        level 0,                            advanced one pass small out,        4861425
-silesia.tar,                        level 1,                            advanced one pass small out,        5334885
-silesia.tar,                        level 3,                            advanced one pass small out,        4861425
-silesia.tar,                        level 4,                            advanced one pass small out,        4799630
-silesia.tar,                        level 5 row 1,                      advanced one pass small out,        4722324
-silesia.tar,                        level 5 row 2,                      advanced one pass small out,        4719256
-silesia.tar,                        level 5,                            advanced one pass small out,        4719256
-silesia.tar,                        level 6,                            advanced one pass small out,        4677721
-silesia.tar,                        level 7 row 1,                      advanced one pass small out,        4606715
-silesia.tar,                        level 7 row 2,                      advanced one pass small out,        4613541
-silesia.tar,                        level 7,                            advanced one pass small out,        4613541
-silesia.tar,                        level 9,                            advanced one pass small out,        4555426
-silesia.tar,                        level 12 row 1,                     advanced one pass small out,        4529459
-silesia.tar,                        level 12 row 2,                     advanced one pass small out,        4530256
-silesia.tar,                        level 13,                           advanced one pass small out,        4491764
-silesia.tar,                        level 16,                           advanced one pass small out,        4356827
-silesia.tar,                        level 19,                           advanced one pass small out,        4264487
-silesia.tar,                        no source size,                     advanced one pass small out,        4861425
-silesia.tar,                        long distance mode,                 advanced one pass small out,        4847754
-silesia.tar,                        multithreaded,                      advanced one pass small out,        4861508
-silesia.tar,                        multithreaded long distance mode,   advanced one pass small out,        4853222
-silesia.tar,                        small window log,                   advanced one pass small out,        7101530
-silesia.tar,                        small hash log,                     advanced one pass small out,        6529232
+silesia,                            explicit params,                    advanced one pass small out,        4795432
+silesia,                            uncompressed literals,              advanced one pass small out,        5120566
+silesia,                            uncompressed literals optimal,      advanced one pass small out,        4319518
+silesia,                            huffman literals,                   advanced one pass small out,        5321369
+silesia,                            multithreaded with advanced params, advanced one pass small out,        5120566
+silesia.tar,                        level -5,                           advanced one pass small out,        6861055
+silesia.tar,                        level -3,                           advanced one pass small out,        6505483
+silesia.tar,                        level -1,                           advanced one pass small out,        6179047
+silesia.tar,                        level 0,                            advanced one pass small out,        4854086
+silesia.tar,                        level 1,                            advanced one pass small out,        5327717
+silesia.tar,                        level 3,                            advanced one pass small out,        4854086
+silesia.tar,                        level 4,                            advanced one pass small out,        4791503
+silesia.tar,                        level 5 row 1,                      advanced one pass small out,        4677740
+silesia.tar,                        level 5 row 2,                      advanced one pass small out,        4682161
+silesia.tar,                        level 5,                            advanced one pass small out,        4677740
+silesia.tar,                        level 6,                            advanced one pass small out,        4613242
+silesia.tar,                        level 7 row 1,                      advanced one pass small out,        4576661
+silesia.tar,                        level 7 row 2,                      advanced one pass small out,        4575393
+silesia.tar,                        level 7,                            advanced one pass small out,        4576661
+silesia.tar,                        level 9,                            advanced one pass small out,        4552899
+silesia.tar,                        level 11 row 1,                     advanced one pass small out,        4514432
+silesia.tar,                        level 11 row 2,                     advanced one pass small out,        4513604
+silesia.tar,                        level 12 row 1,                     advanced one pass small out,        4514049
+silesia.tar,                        level 12 row 2,                     advanced one pass small out,        4513797
+silesia.tar,                        level 13,                           advanced one pass small out,        4502956
+silesia.tar,                        level 16,                           advanced one pass small out,        4360546
+silesia.tar,                        level 19,                           advanced one pass small out,        4265911
+silesia.tar,                        no source size,                     advanced one pass small out,        4854086
+silesia.tar,                        long distance mode,                 advanced one pass small out,        4840452
+silesia.tar,                        multithreaded,                      advanced one pass small out,        4854160
+silesia.tar,                        multithreaded long distance mode,   advanced one pass small out,        4845741
+silesia.tar,                        small window log,                   advanced one pass small out,        7100655
+silesia.tar,                        small hash log,                     advanced one pass small out,        6529206
 silesia.tar,                        small chain log,                    advanced one pass small out,        4917041
-silesia.tar,                        explicit params,                    advanced one pass small out,        4807380
-silesia.tar,                        uncompressed literals,              advanced one pass small out,        5129458
-silesia.tar,                        uncompressed literals optimal,      advanced one pass small out,        4307453
-silesia.tar,                        huffman literals,                   advanced one pass small out,        5347335
-silesia.tar,                        multithreaded with advanced params, advanced one pass small out,        5129555
-github,                             level -5,                           advanced one pass small out,        205285
+silesia.tar,                        explicit params,                    advanced one pass small out,        4806855
+silesia.tar,                        uncompressed literals,              advanced one pass small out,        5122473
+silesia.tar,                        uncompressed literals optimal,      advanced one pass small out,        4310141
+silesia.tar,                        huffman literals,                   advanced one pass small out,        5341705
+silesia.tar,                        multithreaded with advanced params, advanced one pass small out,        5122567
+github,                             level -5,                           advanced one pass small out,        204407
 github,                             level -5 with dict,                 advanced one pass small out,        46718
-github,                             level -3,                           advanced one pass small out,        190643
+github,                             level -3,                           advanced one pass small out,        193253
 github,                             level -3 with dict,                 advanced one pass small out,        45395
-github,                             level -1,                           advanced one pass small out,        175568
+github,                             level -1,                           advanced one pass small out,        175468
 github,                             level -1 with dict,                 advanced one pass small out,        43170
-github,                             level 0,                            advanced one pass small out,        136335
+github,                             level 0,                            advanced one pass small out,        136332
 github,                             level 0 with dict,                  advanced one pass small out,        41148
 github,                             level 0 with dict dms,              advanced one pass small out,        41148
 github,                             level 0 with dict dds,              advanced one pass small out,        41148
 github,                             level 0 with dict copy,             advanced one pass small out,        41124
 github,                             level 0 with dict load,             advanced one pass small out,        42252
-github,                             level 1,                            advanced one pass small out,        142465
+github,                             level 1,                            advanced one pass small out,        142365
 github,                             level 1 with dict,                  advanced one pass small out,        41682
 github,                             level 1 with dict dms,              advanced one pass small out,        41682
 github,                             level 1 with dict dds,              advanced one pass small out,        41682
-github,                             level 1 with dict copy,             advanced one pass small out,        41674
-github,                             level 1 with dict load,             advanced one pass small out,        43755
-github,                             level 3,                            advanced one pass small out,        136335
+github,                             level 1 with dict copy,             advanced one pass small out,        41698
+github,                             level 1 with dict load,             advanced one pass small out,        43814
+github,                             level 3,                            advanced one pass small out,        136332
 github,                             level 3 with dict,                  advanced one pass small out,        41148
 github,                             level 3 with dict dms,              advanced one pass small out,        41148
 github,                             level 3 with dict dds,              advanced one pass small out,        41148
@@ -616,16 +644,16 @@ github,                             level 4 with dict dms,              advanced
 github,                             level 4 with dict dds,              advanced one pass small out,        41251
 github,                             level 4 with dict copy,             advanced one pass small out,        41216
 github,                             level 4 with dict load,             advanced one pass small out,        41159
-github,                             level 5 row 1,                      advanced one pass small out,        135121
-github,                             level 5 row 1 with dict dms,        advanced one pass small out,        38938
-github,                             level 5 row 1 with dict dds,        advanced one pass small out,        38732
-github,                             level 5 row 1 with dict copy,       advanced one pass small out,        38934
-github,                             level 5 row 1 with dict load,       advanced one pass small out,        40725
-github,                             level 5 row 2,                      advanced one pass small out,        134584
-github,                             level 5 row 2 with dict dms,        advanced one pass small out,        38758
-github,                             level 5 row 2 with dict dds,        advanced one pass small out,        38728
-github,                             level 5 row 2 with dict copy,       advanced one pass small out,        38759
-github,                             level 5 row 2 with dict load,       advanced one pass small out,        41518
+github,                             level 5 row 1,                      advanced one pass small out,        134584
+github,                             level 5 row 1 with dict dms,        advanced one pass small out,        38758
+github,                             level 5 row 1 with dict dds,        advanced one pass small out,        38728
+github,                             level 5 row 1 with dict copy,       advanced one pass small out,        38759
+github,                             level 5 row 1 with dict load,       advanced one pass small out,        41518
+github,                             level 5 row 2,                      advanced one pass small out,        135121
+github,                             level 5 row 2 with dict dms,        advanced one pass small out,        38938
+github,                             level 5 row 2 with dict dds,        advanced one pass small out,        38732
+github,                             level 5 row 2 with dict copy,       advanced one pass small out,        38934
+github,                             level 5 row 2 with dict load,       advanced one pass small out,        40725
 github,                             level 5,                            advanced one pass small out,        135121
 github,                             level 5 with dict,                  advanced one pass small out,        38758
 github,                             level 5 with dict dms,              advanced one pass small out,        38758
@@ -635,270 +663,294 @@ github,                             level 5 with dict load,             advanced
 github,                             level 6,                            advanced one pass small out,        135122
 github,                             level 6 with dict,                  advanced one pass small out,        38671
 github,                             level 6 with dict dms,              advanced one pass small out,        38671
-github,                             level 6 with dict dds,              advanced one pass small out,        38630
+github,                             level 6 with dict dds,              advanced one pass small out,        38636
 github,                             level 6 with dict copy,             advanced one pass small out,        38669
 github,                             level 6 with dict load,             advanced one pass small out,        40695
-github,                             level 7 row 1,                      advanced one pass small out,        135122
-github,                             level 7 row 1 with dict dms,        advanced one pass small out,        38771
-github,                             level 7 row 1 with dict dds,        advanced one pass small out,        38771
-github,                             level 7 row 1 with dict copy,       advanced one pass small out,        38745
-github,                             level 7 row 1 with dict load,       advanced one pass small out,        40695
-github,                             level 7 row 2,                      advanced one pass small out,        134584
-github,                             level 7 row 2 with dict dms,        advanced one pass small out,        38758
-github,                             level 7 row 2 with dict dds,        advanced one pass small out,        38747
-github,                             level 7 row 2 with dict copy,       advanced one pass small out,        38755
-github,                             level 7 row 2 with dict load,       advanced one pass small out,        41030
+github,                             level 7 row 1,                      advanced one pass small out,        134584
+github,                             level 7 row 1 with dict dms,        advanced one pass small out,        38758
+github,                             level 7 row 1 with dict dds,        advanced one pass small out,        38745
+github,                             level 7 row 1 with dict copy,       advanced one pass small out,        38755
+github,                             level 7 row 1 with dict load,       advanced one pass small out,        43154
+github,                             level 7 row 2,                      advanced one pass small out,        135122
+github,                             level 7 row 2 with dict dms,        advanced one pass small out,        38860
+github,                             level 7 row 2 with dict dds,        advanced one pass small out,        38766
+github,                             level 7 row 2 with dict copy,       advanced one pass small out,        38834
+github,                             level 7 row 2 with dict load,       advanced one pass small out,        40695
 github,                             level 7,                            advanced one pass small out,        135122
 github,                             level 7 with dict,                  advanced one pass small out,        38758
 github,                             level 7 with dict dms,              advanced one pass small out,        38758
-github,                             level 7 with dict dds,              advanced one pass small out,        38747
+github,                             level 7 with dict dds,              advanced one pass small out,        38745
 github,                             level 7 with dict copy,             advanced one pass small out,        38755
 github,                             level 7 with dict load,             advanced one pass small out,        40695
 github,                             level 9,                            advanced one pass small out,        135122
 github,                             level 9 with dict,                  advanced one pass small out,        39437
 github,                             level 9 with dict dms,              advanced one pass small out,        39437
-github,                             level 9 with dict dds,              advanced one pass small out,        39338
+github,                             level 9 with dict dds,              advanced one pass small out,        39393
 github,                             level 9 with dict copy,             advanced one pass small out,        39398
 github,                             level 9 with dict load,             advanced one pass small out,        41710
-github,                             level 12 row 1,                     advanced one pass small out,        134180
+github,                             level 11 row 1,                     advanced one pass small out,        135367
+github,                             level 11 row 1 with dict dms,       advanced one pass small out,        39671
+github,                             level 11 row 1 with dict dds,       advanced one pass small out,        39671
+github,                             level 11 row 1 with dict copy,      advanced one pass small out,        39651
+github,                             level 11 row 1 with dict load,      advanced one pass small out,        41360
+github,                             level 11 row 2,                     advanced one pass small out,        135367
+github,                             level 11 row 2 with dict dms,       advanced one pass small out,        39671
+github,                             level 11 row 2 with dict dds,       advanced one pass small out,        39671
+github,                             level 11 row 2 with dict copy,      advanced one pass small out,        39651
+github,                             level 11 row 2 with dict load,      advanced one pass small out,        41360
+github,                             level 12 row 1,                     advanced one pass small out,        134402
 github,                             level 12 row 1 with dict dms,       advanced one pass small out,        39677
 github,                             level 12 row 1 with dict dds,       advanced one pass small out,        39677
 github,                             level 12 row 1 with dict copy,      advanced one pass small out,        39677
 github,                             level 12 row 1 with dict load,      advanced one pass small out,        41166
-github,                             level 12 row 2,                     advanced one pass small out,        134180
+github,                             level 12 row 2,                     advanced one pass small out,        134402
 github,                             level 12 row 2 with dict dms,       advanced one pass small out,        39677
 github,                             level 12 row 2 with dict dds,       advanced one pass small out,        39677
 github,                             level 12 row 2 with dict copy,      advanced one pass small out,        39677
 github,                             level 12 row 2 with dict load,      advanced one pass small out,        41166
-github,                             level 13,                           advanced one pass small out,        134064
-github,                             level 13 with dict,                 advanced one pass small out,        39743
-github,                             level 13 with dict dms,             advanced one pass small out,        39743
-github,                             level 13 with dict dds,             advanced one pass small out,        39743
+github,                             level 13,                           advanced one pass small out,        132878
+github,                             level 13 with dict,                 advanced one pass small out,        39900
+github,                             level 13 with dict dms,             advanced one pass small out,        39900
+github,                             level 13 with dict dds,             advanced one pass small out,        39900
 github,                             level 13 with dict copy,            advanced one pass small out,        39948
-github,                             level 13 with dict load,            advanced one pass small out,        42626
-github,                             level 16,                           advanced one pass small out,        134064
+github,                             level 13 with dict load,            advanced one pass small out,        42624
+github,                             level 16,                           advanced one pass small out,        133209
 github,                             level 16 with dict,                 advanced one pass small out,        37577
 github,                             level 16 with dict dms,             advanced one pass small out,        37577
 github,                             level 16 with dict dds,             advanced one pass small out,        37577
 github,                             level 16 with dict copy,            advanced one pass small out,        37568
-github,                             level 16 with dict load,            advanced one pass small out,        42340
-github,                             level 19,                           advanced one pass small out,        134064
+github,                             level 16 with dict load,            advanced one pass small out,        42338
+github,                             level 19,                           advanced one pass small out,        132879
 github,                             level 19 with dict,                 advanced one pass small out,        37576
 github,                             level 19 with dict dms,             advanced one pass small out,        37576
 github,                             level 19 with dict dds,             advanced one pass small out,        37576
 github,                             level 19 with dict copy,            advanced one pass small out,        37567
 github,                             level 19 with dict load,            advanced one pass small out,        39613
-github,                             no source size,                     advanced one pass small out,        136335
+github,                             no source size,                     advanced one pass small out,        136332
 github,                             no source size with dict,           advanced one pass small out,        41148
-github,                             long distance mode,                 advanced one pass small out,        136335
-github,                             multithreaded,                      advanced one pass small out,        136335
-github,                             multithreaded long distance mode,   advanced one pass small out,        136335
-github,                             small window log,                   advanced one pass small out,        136335
+github,                             long distance mode,                 advanced one pass small out,        136332
+github,                             multithreaded,                      advanced one pass small out,        136332
+github,                             multithreaded long distance mode,   advanced one pass small out,        136332
+github,                             small window log,                   advanced one pass small out,        136332
 github,                             small hash log,                     advanced one pass small out,        135590
 github,                             small chain log,                    advanced one pass small out,        136341
 github,                             explicit params,                    advanced one pass small out,        137727
-github,                             uncompressed literals,              advanced one pass small out,        165915
-github,                             uncompressed literals optimal,      advanced one pass small out,        157227
-github,                             huffman literals,                   advanced one pass small out,        142465
-github,                             multithreaded with advanced params, advanced one pass small out,        165915
-github.tar,                         level -5,                           advanced one pass small out,        46856
-github.tar,                         level -5 with dict,                 advanced one pass small out,        44571
-github.tar,                         level -3,                           advanced one pass small out,        43754
-github.tar,                         level -3 with dict,                 advanced one pass small out,        41447
-github.tar,                         level -1,                           advanced one pass small out,        42490
-github.tar,                         level -1 with dict,                 advanced one pass small out,        41131
-github.tar,                         level 0,                            advanced one pass small out,        38441
+github,                             uncompressed literals,              advanced one pass small out,        165911
+github,                             uncompressed literals optimal,      advanced one pass small out,        152667
+github,                             huffman literals,                   advanced one pass small out,        142365
+github,                             multithreaded with advanced params, advanced one pass small out,        165911
+github.tar,                         level -5,                           advanced one pass small out,        52115
+github.tar,                         level -5 with dict,                 advanced one pass small out,        50974
+github.tar,                         level -3,                           advanced one pass small out,        45678
+github.tar,                         level -3 with dict,                 advanced one pass small out,        44656
+github.tar,                         level -1,                           advanced one pass small out,        42560
+github.tar,                         level -1 with dict,                 advanced one pass small out,        41151
+github.tar,                         level 0,                            advanced one pass small out,        38831
 github.tar,                         level 0 with dict,                  advanced one pass small out,        37995
 github.tar,                         level 0 with dict dms,              advanced one pass small out,        38003
 github.tar,                         level 0 with dict dds,              advanced one pass small out,        38003
 github.tar,                         level 0 with dict copy,             advanced one pass small out,        37995
 github.tar,                         level 0 with dict load,             advanced one pass small out,        37956
-github.tar,                         level 1,                            advanced one pass small out,        39265
-github.tar,                         level 1 with dict,                  advanced one pass small out,        38280
-github.tar,                         level 1 with dict dms,              advanced one pass small out,        38290
-github.tar,                         level 1 with dict dds,              advanced one pass small out,        38290
-github.tar,                         level 1 with dict copy,             advanced one pass small out,        38280
-github.tar,                         level 1 with dict load,             advanced one pass small out,        38729
-github.tar,                         level 3,                            advanced one pass small out,        38441
+github.tar,                         level 1,                            advanced one pass small out,        39200
+github.tar,                         level 1 with dict,                  advanced one pass small out,        38089
+github.tar,                         level 1 with dict dms,              advanced one pass small out,        38294
+github.tar,                         level 1 with dict dds,              advanced one pass small out,        38294
+github.tar,                         level 1 with dict copy,             advanced one pass small out,        38089
+github.tar,                         level 1 with dict load,             advanced one pass small out,        38364
+github.tar,                         level 3,                            advanced one pass small out,        38831
 github.tar,                         level 3 with dict,                  advanced one pass small out,        37995
 github.tar,                         level 3 with dict dms,              advanced one pass small out,        38003
 github.tar,                         level 3 with dict dds,              advanced one pass small out,        38003
 github.tar,                         level 3 with dict copy,             advanced one pass small out,        37995
 github.tar,                         level 3 with dict load,             advanced one pass small out,        37956
-github.tar,                         level 4,                            advanced one pass small out,        38467
+github.tar,                         level 4,                            advanced one pass small out,        38893
 github.tar,                         level 4 with dict,                  advanced one pass small out,        37948
 github.tar,                         level 4 with dict dms,              advanced one pass small out,        37954
 github.tar,                         level 4 with dict dds,              advanced one pass small out,        37954
 github.tar,                         level 4 with dict copy,             advanced one pass small out,        37948
 github.tar,                         level 4 with dict load,             advanced one pass small out,        37927
-github.tar,                         level 5 row 1,                      advanced one pass small out,        39788
-github.tar,                         level 5 row 1 with dict dms,        advanced one pass small out,        39365
-github.tar,                         level 5 row 1 with dict dds,        advanced one pass small out,        39233
-github.tar,                         level 5 row 1 with dict copy,       advanced one pass small out,        39715
-github.tar,                         level 5 row 1 with dict load,       advanced one pass small out,        39209
-github.tar,                         level 5 row 2,                      advanced one pass small out,        39693
-github.tar,                         level 5 row 2 with dict dms,        advanced one pass small out,        39024
-github.tar,                         level 5 row 2 with dict dds,        advanced one pass small out,        39028
-github.tar,                         level 5 row 2 with dict copy,       advanced one pass small out,        39040
-github.tar,                         level 5 row 2 with dict load,       advanced one pass small out,        39037
-github.tar,                         level 5,                            advanced one pass small out,        39693
-github.tar,                         level 5 with dict,                  advanced one pass small out,        39040
-github.tar,                         level 5 with dict dms,              advanced one pass small out,        39024
-github.tar,                         level 5 with dict dds,              advanced one pass small out,        39028
-github.tar,                         level 5 with dict copy,             advanced one pass small out,        39040
-github.tar,                         level 5 with dict load,             advanced one pass small out,        39037
-github.tar,                         level 6,                            advanced one pass small out,        39621
-github.tar,                         level 6 with dict,                  advanced one pass small out,        38622
-github.tar,                         level 6 with dict dms,              advanced one pass small out,        38608
-github.tar,                         level 6 with dict dds,              advanced one pass small out,        38610
-github.tar,                         level 6 with dict copy,             advanced one pass small out,        38622
-github.tar,                         level 6 with dict load,             advanced one pass small out,        38962
-github.tar,                         level 7 row 1,                      advanced one pass small out,        39206
-github.tar,                         level 7 row 1 with dict dms,        advanced one pass small out,        37954
-github.tar,                         level 7 row 1 with dict dds,        advanced one pass small out,        37954
-github.tar,                         level 7 row 1 with dict copy,       advanced one pass small out,        38071
-github.tar,                         level 7 row 1 with dict load,       advanced one pass small out,        38584
-github.tar,                         level 7 row 2,                      advanced one pass small out,        39213
-github.tar,                         level 7 row 2 with dict dms,        advanced one pass small out,        37848
-github.tar,                         level 7 row 2 with dict dds,        advanced one pass small out,        37867
-github.tar,                         level 7 row 2 with dict copy,       advanced one pass small out,        37848
-github.tar,                         level 7 row 2 with dict load,       advanced one pass small out,        38582
-github.tar,                         level 7,                            advanced one pass small out,        39213
-github.tar,                         level 7 with dict,                  advanced one pass small out,        37848
-github.tar,                         level 7 with dict dms,              advanced one pass small out,        37848
-github.tar,                         level 7 with dict dds,              advanced one pass small out,        37867
-github.tar,                         level 7 with dict copy,             advanced one pass small out,        37848
-github.tar,                         level 7 with dict load,             advanced one pass small out,        38582
-github.tar,                         level 9,                            advanced one pass small out,        36758
-github.tar,                         level 9 with dict,                  advanced one pass small out,        36457
-github.tar,                         level 9 with dict dms,              advanced one pass small out,        36549
-github.tar,                         level 9 with dict dds,              advanced one pass small out,        36637
-github.tar,                         level 9 with dict copy,             advanced one pass small out,        36457
-github.tar,                         level 9 with dict load,             advanced one pass small out,        36350
-github.tar,                         level 12 row 1,                     advanced one pass small out,        36435
+github.tar,                         level 5 row 1,                      advanced one pass small out,        39651
+github.tar,                         level 5 row 1 with dict dms,        advanced one pass small out,        39059
+github.tar,                         level 5 row 1 with dict dds,        advanced one pass small out,        39067
+github.tar,                         level 5 row 1 with dict copy,       advanced one pass small out,        39082
+github.tar,                         level 5 row 1 with dict load,       advanced one pass small out,        38999
+github.tar,                         level 5 row 2,                      advanced one pass small out,        39701
+github.tar,                         level 5 row 2 with dict dms,        advanced one pass small out,        39365
+github.tar,                         level 5 row 2 with dict dds,        advanced one pass small out,        39233
+github.tar,                         level 5 row 2 with dict copy,       advanced one pass small out,        39715
+github.tar,                         level 5 row 2 with dict load,       advanced one pass small out,        39158
+github.tar,                         level 5,                            advanced one pass small out,        39651
+github.tar,                         level 5 with dict,                  advanced one pass small out,        39082
+github.tar,                         level 5 with dict dms,              advanced one pass small out,        39059
+github.tar,                         level 5 with dict dds,              advanced one pass small out,        39067
+github.tar,                         level 5 with dict copy,             advanced one pass small out,        39082
+github.tar,                         level 5 with dict load,             advanced one pass small out,        38999
+github.tar,                         level 6,                            advanced one pass small out,        39282
+github.tar,                         level 6 with dict,                  advanced one pass small out,        38656
+github.tar,                         level 6 with dict dms,              advanced one pass small out,        38636
+github.tar,                         level 6 with dict dds,              advanced one pass small out,        38634
+github.tar,                         level 6 with dict copy,             advanced one pass small out,        38656
+github.tar,                         level 6 with dict load,             advanced one pass small out,        38648
+github.tar,                         level 7 row 1,                      advanced one pass small out,        38110
+github.tar,                         level 7 row 1 with dict dms,        advanced one pass small out,        37858
+github.tar,                         level 7 row 1 with dict dds,        advanced one pass small out,        37882
+github.tar,                         level 7 row 1 with dict copy,       advanced one pass small out,        37865
+github.tar,                         level 7 row 1 with dict load,       advanced one pass small out,        37436
+github.tar,                         level 7 row 2,                      advanced one pass small out,        38077
+github.tar,                         level 7 row 2 with dict dms,        advanced one pass small out,        38012
+github.tar,                         level 7 row 2 with dict dds,        advanced one pass small out,        38014
+github.tar,                         level 7 row 2 with dict copy,       advanced one pass small out,        38101
+github.tar,                         level 7 row 2 with dict load,       advanced one pass small out,        37402
+github.tar,                         level 7,                            advanced one pass small out,        38110
+github.tar,                         level 7 with dict,                  advanced one pass small out,        37865
+github.tar,                         level 7 with dict dms,              advanced one pass small out,        37858
+github.tar,                         level 7 with dict dds,              advanced one pass small out,        37882
+github.tar,                         level 7 with dict copy,             advanced one pass small out,        37865
+github.tar,                         level 7 with dict load,             advanced one pass small out,        37436
+github.tar,                         level 9,                            advanced one pass small out,        36760
+github.tar,                         level 9 with dict,                  advanced one pass small out,        36484
+github.tar,                         level 9 with dict dms,              advanced one pass small out,        36567
+github.tar,                         level 9 with dict dds,              advanced one pass small out,        36628
+github.tar,                         level 9 with dict copy,             advanced one pass small out,        36484
+github.tar,                         level 9 with dict load,             advanced one pass small out,        36401
+github.tar,                         level 11 row 1,                     advanced one pass small out,        36081
+github.tar,                         level 11 row 1 with dict dms,       advanced one pass small out,        36963
+github.tar,                         level 11 row 1 with dict dds,       advanced one pass small out,        36963
+github.tar,                         level 11 row 1 with dict copy,      advanced one pass small out,        36557
+github.tar,                         level 11 row 1 with dict load,      advanced one pass small out,        36434
+github.tar,                         level 11 row 2,                     advanced one pass small out,        36110
+github.tar,                         level 11 row 2 with dict dms,       advanced one pass small out,        36963
+github.tar,                         level 11 row 2 with dict dds,       advanced one pass small out,        36963
+github.tar,                         level 11 row 2 with dict copy,      advanced one pass small out,        36557
+github.tar,                         level 11 row 2 with dict load,      advanced one pass small out,        36459
+github.tar,                         level 12 row 1,                     advanced one pass small out,        36081
 github.tar,                         level 12 row 1 with dict dms,       advanced one pass small out,        36986
 github.tar,                         level 12 row 1 with dict dds,       advanced one pass small out,        36986
 github.tar,                         level 12 row 1 with dict copy,      advanced one pass small out,        36609
-github.tar,                         level 12 row 1 with dict load,      advanced one pass small out,        36419
-github.tar,                         level 12 row 2,                     advanced one pass small out,        36435
+github.tar,                         level 12 row 1 with dict load,      advanced one pass small out,        36434
+github.tar,                         level 12 row 2,                     advanced one pass small out,        36110
 github.tar,                         level 12 row 2 with dict dms,       advanced one pass small out,        36986
 github.tar,                         level 12 row 2 with dict dds,       advanced one pass small out,        36986
 github.tar,                         level 12 row 2 with dict copy,      advanced one pass small out,        36609
-github.tar,                         level 12 row 2 with dict load,      advanced one pass small out,        36424
-github.tar,                         level 13,                           advanced one pass small out,        35621
-github.tar,                         level 13 with dict,                 advanced one pass small out,        38726
-github.tar,                         level 13 with dict dms,             advanced one pass small out,        38903
-github.tar,                         level 13 with dict dds,             advanced one pass small out,        38903
-github.tar,                         level 13 with dict copy,            advanced one pass small out,        38726
-github.tar,                         level 13 with dict load,            advanced one pass small out,        36372
-github.tar,                         level 16,                           advanced one pass small out,        40255
-github.tar,                         level 16 with dict,                 advanced one pass small out,        33639
-github.tar,                         level 16 with dict dms,             advanced one pass small out,        33544
-github.tar,                         level 16 with dict dds,             advanced one pass small out,        33544
-github.tar,                         level 16 with dict copy,            advanced one pass small out,        33639
-github.tar,                         level 16 with dict load,            advanced one pass small out,        39353
-github.tar,                         level 19,                           advanced one pass small out,        32837
-github.tar,                         level 19 with dict,                 advanced one pass small out,        32895
-github.tar,                         level 19 with dict dms,             advanced one pass small out,        32672
-github.tar,                         level 19 with dict dds,             advanced one pass small out,        32672
-github.tar,                         level 19 with dict copy,            advanced one pass small out,        32895
-github.tar,                         level 19 with dict load,            advanced one pass small out,        32676
-github.tar,                         no source size,                     advanced one pass small out,        38441
+github.tar,                         level 12 row 2 with dict load,      advanced one pass small out,        36459
+github.tar,                         level 13,                           advanced one pass small out,        35501
+github.tar,                         level 13 with dict,                 advanced one pass small out,        37130
+github.tar,                         level 13 with dict dms,             advanced one pass small out,        37220
+github.tar,                         level 13 with dict dds,             advanced one pass small out,        37220
+github.tar,                         level 13 with dict copy,            advanced one pass small out,        37130
+github.tar,                         level 13 with dict load,            advanced one pass small out,        36010
+github.tar,                         level 16,                           advanced one pass small out,        40466
+github.tar,                         level 16 with dict,                 advanced one pass small out,        33374
+github.tar,                         level 16 with dict dms,             advanced one pass small out,        33206
+github.tar,                         level 16 with dict dds,             advanced one pass small out,        33206
+github.tar,                         level 16 with dict copy,            advanced one pass small out,        33374
+github.tar,                         level 16 with dict load,            advanced one pass small out,        39081
+github.tar,                         level 19,                           advanced one pass small out,        32276
+github.tar,                         level 19 with dict,                 advanced one pass small out,        32712
+github.tar,                         level 19 with dict dms,             advanced one pass small out,        32555
+github.tar,                         level 19 with dict dds,             advanced one pass small out,        32555
+github.tar,                         level 19 with dict copy,            advanced one pass small out,        32712
+github.tar,                         level 19 with dict load,            advanced one pass small out,        32479
+github.tar,                         no source size,                     advanced one pass small out,        38831
 github.tar,                         no source size with dict,           advanced one pass small out,        37995
-github.tar,                         long distance mode,                 advanced one pass small out,        39757
-github.tar,                         multithreaded,                      advanced one pass small out,        38441
-github.tar,                         multithreaded long distance mode,   advanced one pass small out,        39726
+github.tar,                         long distance mode,                 advanced one pass small out,        40252
+github.tar,                         multithreaded,                      advanced one pass small out,        38831
+github.tar,                         multithreaded long distance mode,   advanced one pass small out,        40232
 github.tar,                         small window log,                   advanced one pass small out,        198540
 github.tar,                         small hash log,                     advanced one pass small out,        129870
 github.tar,                         small chain log,                    advanced one pass small out,        41669
-github.tar,                         explicit params,                    advanced one pass small out,        41227
-github.tar,                         uncompressed literals,              advanced one pass small out,        41122
-github.tar,                         uncompressed literals optimal,      advanced one pass small out,        35388
-github.tar,                         huffman literals,                   advanced one pass small out,        38777
-github.tar,                         multithreaded with advanced params, advanced one pass small out,        41122
-silesia,                            level -5,                           advanced streaming,                 6882505
-silesia,                            level -3,                           advanced streaming,                 6568376
-silesia,                            level -1,                           advanced streaming,                 6183403
-silesia,                            level 0,                            advanced streaming,                 4849552
-silesia,                            level 1,                            advanced streaming,                 5314162
-silesia,                            level 3,                            advanced streaming,                 4849552
-silesia,                            level 4,                            advanced streaming,                 4786970
-silesia,                            level 5 row 1,                      advanced streaming,                 4710236
-silesia,                            level 5 row 2,                      advanced streaming,                 4707794
-silesia,                            level 5,                            advanced streaming,                 4707794
-silesia,                            level 6,                            advanced streaming,                 4666383
-silesia,                            level 7 row 1,                      advanced streaming,                 4596296
-silesia,                            level 7 row 2,                      advanced streaming,                 4603381
-silesia,                            level 7,                            advanced streaming,                 4603381
-silesia,                            level 9,                            advanced streaming,                 4546001
-silesia,                            level 12 row 1,                     advanced streaming,                 4519288
-silesia,                            level 12 row 2,                     advanced streaming,                 4521397
-silesia,                            level 13,                           advanced streaming,                 4482135
-silesia,                            level 16,                           advanced streaming,                 4360251
-silesia,                            level 19,                           advanced streaming,                 4283237
-silesia,                            no source size,                     advanced streaming,                 4849516
-silesia,                            long distance mode,                 advanced streaming,                 4840738
-silesia,                            multithreaded,                      advanced streaming,                 4849552
-silesia,                            multithreaded long distance mode,   advanced streaming,                 4840758
-silesia,                            small window log,                   advanced streaming,                 7112062
+github.tar,                         explicit params,                    advanced one pass small out,        41385
+github.tar,                         uncompressed literals,              advanced one pass small out,        41525
+github.tar,                         uncompressed literals optimal,      advanced one pass small out,        35397
+github.tar,                         huffman literals,                   advanced one pass small out,        38853
+github.tar,                         multithreaded with advanced params, advanced one pass small out,        41525
+silesia,                            level -5,                           advanced streaming,                 6854744
+silesia,                            level -3,                           advanced streaming,                 6503319
+silesia,                            level -1,                           advanced streaming,                 6172207
+silesia,                            level 0,                            advanced streaming,                 4842075
+silesia,                            level 1,                            advanced streaming,                 5306388
+silesia,                            level 3,                            advanced streaming,                 4842075
+silesia,                            level 4,                            advanced streaming,                 4779186
+silesia,                            level 5 row 1,                      advanced streaming,                 4666323
+silesia,                            level 5 row 2,                      advanced streaming,                 4670136
+silesia,                            level 5,                            advanced streaming,                 4666323
+silesia,                            level 6,                            advanced streaming,                 4603066
+silesia,                            level 7 row 1,                      advanced streaming,                 4566984
+silesia,                            level 7 row 2,                      advanced streaming,                 4564868
+silesia,                            level 7,                            advanced streaming,                 4566984
+silesia,                            level 9,                            advanced streaming,                 4543018
+silesia,                            level 11 row 1,                     advanced streaming,                 4505046
+silesia,                            level 11 row 2,                     advanced streaming,                 4503116
+silesia,                            level 12 row 1,                     advanced streaming,                 4505046
+silesia,                            level 12 row 2,                     advanced streaming,                 4503116
+silesia,                            level 13,                           advanced streaming,                 4493990
+silesia,                            level 16,                           advanced streaming,                 4360041
+silesia,                            level 19,                           advanced streaming,                 4296055
+silesia,                            no source size,                     advanced streaming,                 4842039
+silesia,                            long distance mode,                 advanced streaming,                 4833710
+silesia,                            multithreaded,                      advanced streaming,                 4842075
+silesia,                            multithreaded long distance mode,   advanced streaming,                 4833737
+silesia,                            small window log,                   advanced streaming,                 7111103
 silesia,                            small hash log,                     advanced streaming,                 6526141
 silesia,                            small chain log,                    advanced streaming,                 4912197
-silesia,                            explicit params,                    advanced streaming,                 4795887
-silesia,                            uncompressed literals,              advanced streaming,                 5127982
-silesia,                            uncompressed literals optimal,      advanced streaming,                 4317896
-silesia,                            huffman literals,                   advanced streaming,                 5331168
-silesia,                            multithreaded with advanced params, advanced streaming,                 5127982
-silesia.tar,                        level -5,                           advanced streaming,                 6982759
-silesia.tar,                        level -3,                           advanced streaming,                 6641283
-silesia.tar,                        level -1,                           advanced streaming,                 6190795
-silesia.tar,                        level 0,                            advanced streaming,                 4861427
-silesia.tar,                        level 1,                            advanced streaming,                 5336939
-silesia.tar,                        level 3,                            advanced streaming,                 4861427
-silesia.tar,                        level 4,                            advanced streaming,                 4799630
-silesia.tar,                        level 5 row 1,                      advanced streaming,                 4722329
-silesia.tar,                        level 5 row 2,                      advanced streaming,                 4719261
-silesia.tar,                        level 5,                            advanced streaming,                 4719261
-silesia.tar,                        level 6,                            advanced streaming,                 4677729
-silesia.tar,                        level 7 row 1,                      advanced streaming,                 4606715
-silesia.tar,                        level 7 row 2,                      advanced streaming,                 4613544
-silesia.tar,                        level 7,                            advanced streaming,                 4613544
-silesia.tar,                        level 9,                            advanced streaming,                 4555432
-silesia.tar,                        level 12 row 1,                     advanced streaming,                 4529459
-silesia.tar,                        level 12 row 2,                     advanced streaming,                 4530258
-silesia.tar,                        level 13,                           advanced streaming,                 4491765
-silesia.tar,                        level 16,                           advanced streaming,                 4356834
-silesia.tar,                        level 19,                           advanced streaming,                 4264392
-silesia.tar,                        no source size,                     advanced streaming,                 4861423
-silesia.tar,                        long distance mode,                 advanced streaming,                 4847754
-silesia.tar,                        multithreaded,                      advanced streaming,                 4861508
-silesia.tar,                        multithreaded long distance mode,   advanced streaming,                 4853222
-silesia.tar,                        small window log,                   advanced streaming,                 7118769
-silesia.tar,                        small hash log,                     advanced streaming,                 6529235
+silesia,                            explicit params,                    advanced streaming,                 4795452
+silesia,                            uncompressed literals,              advanced streaming,                 5120566
+silesia,                            uncompressed literals optimal,      advanced streaming,                 4319518
+silesia,                            huffman literals,                   advanced streaming,                 5321370
+silesia,                            multithreaded with advanced params, advanced streaming,                 5120566
+silesia.tar,                        level -5,                           advanced streaming,                 6856523
+silesia.tar,                        level -3,                           advanced streaming,                 6505954
+silesia.tar,                        level -1,                           advanced streaming,                 6179056
+silesia.tar,                        level 0,                            advanced streaming,                 4859271
+silesia.tar,                        level 1,                            advanced streaming,                 5327708
+silesia.tar,                        level 3,                            advanced streaming,                 4859271
+silesia.tar,                        level 4,                            advanced streaming,                 4797470
+silesia.tar,                        level 5 row 1,                      advanced streaming,                 4677748
+silesia.tar,                        level 5 row 2,                      advanced streaming,                 4682169
+silesia.tar,                        level 5,                            advanced streaming,                 4677748
+silesia.tar,                        level 6,                            advanced streaming,                 4613246
+silesia.tar,                        level 7 row 1,                      advanced streaming,                 4576664
+silesia.tar,                        level 7 row 2,                      advanced streaming,                 4575394
+silesia.tar,                        level 7,                            advanced streaming,                 4576664
+silesia.tar,                        level 9,                            advanced streaming,                 4552900
+silesia.tar,                        level 11 row 1,                     advanced streaming,                 4514433
+silesia.tar,                        level 11 row 2,                     advanced streaming,                 4513604
+silesia.tar,                        level 12 row 1,                     advanced streaming,                 4514049
+silesia.tar,                        level 12 row 2,                     advanced streaming,                 4513797
+silesia.tar,                        level 13,                           advanced streaming,                 4502956
+silesia.tar,                        level 16,                           advanced streaming,                 4360546
+silesia.tar,                        level 19,                           advanced streaming,                 4265911
+silesia.tar,                        no source size,                     advanced streaming,                 4859267
+silesia.tar,                        long distance mode,                 advanced streaming,                 4840452
+silesia.tar,                        multithreaded,                      advanced streaming,                 4854160
+silesia.tar,                        multithreaded long distance mode,   advanced streaming,                 4845741
+silesia.tar,                        small window log,                   advanced streaming,                 7117559
+silesia.tar,                        small hash log,                     advanced streaming,                 6529209
 silesia.tar,                        small chain log,                    advanced streaming,                 4917021
-silesia.tar,                        explicit params,                    advanced streaming,                 4807401
-silesia.tar,                        uncompressed literals,              advanced streaming,                 5129461
-silesia.tar,                        uncompressed literals optimal,      advanced streaming,                 4307400
-silesia.tar,                        huffman literals,                   advanced streaming,                 5352360
-silesia.tar,                        multithreaded with advanced params, advanced streaming,                 5129555
-github,                             level -5,                           advanced streaming,                 205285
+silesia.tar,                        explicit params,                    advanced streaming,                 4806873
+silesia.tar,                        uncompressed literals,              advanced streaming,                 5127423
+silesia.tar,                        uncompressed literals optimal,      advanced streaming,                 4310141
+silesia.tar,                        huffman literals,                   advanced streaming,                 5341712
+silesia.tar,                        multithreaded with advanced params, advanced streaming,                 5122567
+github,                             level -5,                           advanced streaming,                 204407
 github,                             level -5 with dict,                 advanced streaming,                 46718
-github,                             level -3,                           advanced streaming,                 190643
+github,                             level -3,                           advanced streaming,                 193253
 github,                             level -3 with dict,                 advanced streaming,                 45395
-github,                             level -1,                           advanced streaming,                 175568
+github,                             level -1,                           advanced streaming,                 175468
 github,                             level -1 with dict,                 advanced streaming,                 43170
-github,                             level 0,                            advanced streaming,                 136335
+github,                             level 0,                            advanced streaming,                 136332
 github,                             level 0 with dict,                  advanced streaming,                 41148
 github,                             level 0 with dict dms,              advanced streaming,                 41148
 github,                             level 0 with dict dds,              advanced streaming,                 41148
 github,                             level 0 with dict copy,             advanced streaming,                 41124
 github,                             level 0 with dict load,             advanced streaming,                 42252
-github,                             level 1,                            advanced streaming,                 142465
+github,                             level 1,                            advanced streaming,                 142365
 github,                             level 1 with dict,                  advanced streaming,                 41682
 github,                             level 1 with dict dms,              advanced streaming,                 41682
 github,                             level 1 with dict dds,              advanced streaming,                 41682
-github,                             level 1 with dict copy,             advanced streaming,                 41674
-github,                             level 1 with dict load,             advanced streaming,                 43755
-github,                             level 3,                            advanced streaming,                 136335
+github,                             level 1 with dict copy,             advanced streaming,                 41698
+github,                             level 1 with dict load,             advanced streaming,                 43814
+github,                             level 3,                            advanced streaming,                 136332
 github,                             level 3 with dict,                  advanced streaming,                 41148
 github,                             level 3 with dict dms,              advanced streaming,                 41148
 github,                             level 3 with dict dds,              advanced streaming,                 41148
@@ -910,16 +962,16 @@ github,                             level 4 with dict dms,              advanced
 github,                             level 4 with dict dds,              advanced streaming,                 41251
 github,                             level 4 with dict copy,             advanced streaming,                 41216
 github,                             level 4 with dict load,             advanced streaming,                 41159
-github,                             level 5 row 1,                      advanced streaming,                 135121
-github,                             level 5 row 1 with dict dms,        advanced streaming,                 38938
-github,                             level 5 row 1 with dict dds,        advanced streaming,                 38732
-github,                             level 5 row 1 with dict copy,       advanced streaming,                 38934
-github,                             level 5 row 1 with dict load,       advanced streaming,                 40725
-github,                             level 5 row 2,                      advanced streaming,                 134584
-github,                             level 5 row 2 with dict dms,        advanced streaming,                 38758
-github,                             level 5 row 2 with dict dds,        advanced streaming,                 38728
-github,                             level 5 row 2 with dict copy,       advanced streaming,                 38759
-github,                             level 5 row 2 with dict load,       advanced streaming,                 41518
+github,                             level 5 row 1,                      advanced streaming,                 134584
+github,                             level 5 row 1 with dict dms,        advanced streaming,                 38758
+github,                             level 5 row 1 with dict dds,        advanced streaming,                 38728
+github,                             level 5 row 1 with dict copy,       advanced streaming,                 38759
+github,                             level 5 row 1 with dict load,       advanced streaming,                 41518
+github,                             level 5 row 2,                      advanced streaming,                 135121
+github,                             level 5 row 2 with dict dms,        advanced streaming,                 38938
+github,                             level 5 row 2 with dict dds,        advanced streaming,                 38732
+github,                             level 5 row 2 with dict copy,       advanced streaming,                 38934
+github,                             level 5 row 2 with dict load,       advanced streaming,                 40725
 github,                             level 5,                            advanced streaming,                 135121
 github,                             level 5 with dict,                  advanced streaming,                 38758
 github,                             level 5 with dict dms,              advanced streaming,                 38758
@@ -929,234 +981,254 @@ github,                             level 5 with dict load,             advanced
 github,                             level 6,                            advanced streaming,                 135122
 github,                             level 6 with dict,                  advanced streaming,                 38671
 github,                             level 6 with dict dms,              advanced streaming,                 38671
-github,                             level 6 with dict dds,              advanced streaming,                 38630
+github,                             level 6 with dict dds,              advanced streaming,                 38636
 github,                             level 6 with dict copy,             advanced streaming,                 38669
 github,                             level 6 with dict load,             advanced streaming,                 40695
-github,                             level 7 row 1,                      advanced streaming,                 135122
-github,                             level 7 row 1 with dict dms,        advanced streaming,                 38771
-github,                             level 7 row 1 with dict dds,        advanced streaming,                 38771
-github,                             level 7 row 1 with dict copy,       advanced streaming,                 38745
-github,                             level 7 row 1 with dict load,       advanced streaming,                 40695
-github,                             level 7 row 2,                      advanced streaming,                 134584
-github,                             level 7 row 2 with dict dms,        advanced streaming,                 38758
-github,                             level 7 row 2 with dict dds,        advanced streaming,                 38747
-github,                             level 7 row 2 with dict copy,       advanced streaming,                 38755
-github,                             level 7 row 2 with dict load,       advanced streaming,                 41030
+github,                             level 7 row 1,                      advanced streaming,                 134584
+github,                             level 7 row 1 with dict dms,        advanced streaming,                 38758
+github,                             level 7 row 1 with dict dds,        advanced streaming,                 38745
+github,                             level 7 row 1 with dict copy,       advanced streaming,                 38755
+github,                             level 7 row 1 with dict load,       advanced streaming,                 43154
+github,                             level 7 row 2,                      advanced streaming,                 135122
+github,                             level 7 row 2 with dict dms,        advanced streaming,                 38860
+github,                             level 7 row 2 with dict dds,        advanced streaming,                 38766
+github,                             level 7 row 2 with dict copy,       advanced streaming,                 38834
+github,                             level 7 row 2 with dict load,       advanced streaming,                 40695
 github,                             level 7,                            advanced streaming,                 135122
 github,                             level 7 with dict,                  advanced streaming,                 38758
 github,                             level 7 with dict dms,              advanced streaming,                 38758
-github,                             level 7 with dict dds,              advanced streaming,                 38747
+github,                             level 7 with dict dds,              advanced streaming,                 38745
 github,                             level 7 with dict copy,             advanced streaming,                 38755
 github,                             level 7 with dict load,             advanced streaming,                 40695
 github,                             level 9,                            advanced streaming,                 135122
 github,                             level 9 with dict,                  advanced streaming,                 39437
 github,                             level 9 with dict dms,              advanced streaming,                 39437
-github,                             level 9 with dict dds,              advanced streaming,                 39338
+github,                             level 9 with dict dds,              advanced streaming,                 39393
 github,                             level 9 with dict copy,             advanced streaming,                 39398
 github,                             level 9 with dict load,             advanced streaming,                 41710
-github,                             level 12 row 1,                     advanced streaming,                 134180
+github,                             level 11 row 1,                     advanced streaming,                 135367
+github,                             level 11 row 1 with dict dms,       advanced streaming,                 39671
+github,                             level 11 row 1 with dict dds,       advanced streaming,                 39671
+github,                             level 11 row 1 with dict copy,      advanced streaming,                 39651
+github,                             level 11 row 1 with dict load,      advanced streaming,                 41360
+github,                             level 11 row 2,                     advanced streaming,                 135367
+github,                             level 11 row 2 with dict dms,       advanced streaming,                 39671
+github,                             level 11 row 2 with dict dds,       advanced streaming,                 39671
+github,                             level 11 row 2 with dict copy,      advanced streaming,                 39651
+github,                             level 11 row 2 with dict load,      advanced streaming,                 41360
+github,                             level 12 row 1,                     advanced streaming,                 134402
 github,                             level 12 row 1 with dict dms,       advanced streaming,                 39677
 github,                             level 12 row 1 with dict dds,       advanced streaming,                 39677
 github,                             level 12 row 1 with dict copy,      advanced streaming,                 39677
 github,                             level 12 row 1 with dict load,      advanced streaming,                 41166
-github,                             level 12 row 2,                     advanced streaming,                 134180
+github,                             level 12 row 2,                     advanced streaming,                 134402
 github,                             level 12 row 2 with dict dms,       advanced streaming,                 39677
 github,                             level 12 row 2 with dict dds,       advanced streaming,                 39677
 github,                             level 12 row 2 with dict copy,      advanced streaming,                 39677
 github,                             level 12 row 2 with dict load,      advanced streaming,                 41166
-github,                             level 13,                           advanced streaming,                 134064
-github,                             level 13 with dict,                 advanced streaming,                 39743
-github,                             level 13 with dict dms,             advanced streaming,                 39743
-github,                             level 13 with dict dds,             advanced streaming,                 39743
+github,                             level 13,                           advanced streaming,                 132878
+github,                             level 13 with dict,                 advanced streaming,                 39900
+github,                             level 13 with dict dms,             advanced streaming,                 39900
+github,                             level 13 with dict dds,             advanced streaming,                 39900
 github,                             level 13 with dict copy,            advanced streaming,                 39948
-github,                             level 13 with dict load,            advanced streaming,                 42626
-github,                             level 16,                           advanced streaming,                 134064
+github,                             level 13 with dict load,            advanced streaming,                 42624
+github,                             level 16,                           advanced streaming,                 133209
 github,                             level 16 with dict,                 advanced streaming,                 37577
 github,                             level 16 with dict dms,             advanced streaming,                 37577
 github,                             level 16 with dict dds,             advanced streaming,                 37577
 github,                             level 16 with dict copy,            advanced streaming,                 37568
-github,                             level 16 with dict load,            advanced streaming,                 42340
-github,                             level 19,                           advanced streaming,                 134064
+github,                             level 16 with dict load,            advanced streaming,                 42338
+github,                             level 19,                           advanced streaming,                 132879
 github,                             level 19 with dict,                 advanced streaming,                 37576
 github,                             level 19 with dict dms,             advanced streaming,                 37576
 github,                             level 19 with dict dds,             advanced streaming,                 37576
 github,                             level 19 with dict copy,            advanced streaming,                 37567
 github,                             level 19 with dict load,            advanced streaming,                 39613
-github,                             no source size,                     advanced streaming,                 136335
+github,                             no source size,                     advanced streaming,                 136332
 github,                             no source size with dict,           advanced streaming,                 41148
-github,                             long distance mode,                 advanced streaming,                 136335
-github,                             multithreaded,                      advanced streaming,                 136335
-github,                             multithreaded long distance mode,   advanced streaming,                 136335
-github,                             small window log,                   advanced streaming,                 136335
+github,                             long distance mode,                 advanced streaming,                 136332
+github,                             multithreaded,                      advanced streaming,                 136332
+github,                             multithreaded long distance mode,   advanced streaming,                 136332
+github,                             small window log,                   advanced streaming,                 136332
 github,                             small hash log,                     advanced streaming,                 135590
 github,                             small chain log,                    advanced streaming,                 136341
 github,                             explicit params,                    advanced streaming,                 137727
-github,                             uncompressed literals,              advanced streaming,                 165915
-github,                             uncompressed literals optimal,      advanced streaming,                 157227
-github,                             huffman literals,                   advanced streaming,                 142465
-github,                             multithreaded with advanced params, advanced streaming,                 165915
-github.tar,                         level -5,                           advanced streaming,                 46747
-github.tar,                         level -5 with dict,                 advanced streaming,                 44440
-github.tar,                         level -3,                           advanced streaming,                 43537
-github.tar,                         level -3 with dict,                 advanced streaming,                 41112
-github.tar,                         level -1,                           advanced streaming,                 42465
-github.tar,                         level -1 with dict,                 advanced streaming,                 41196
-github.tar,                         level 0,                            advanced streaming,                 38441
+github,                             uncompressed literals,              advanced streaming,                 165911
+github,                             uncompressed literals optimal,      advanced streaming,                 152667
+github,                             huffman literals,                   advanced streaming,                 142365
+github,                             multithreaded with advanced params, advanced streaming,                 165911
+github.tar,                         level -5,                           advanced streaming,                 52152
+github.tar,                         level -5 with dict,                 advanced streaming,                 51045
+github.tar,                         level -3,                           advanced streaming,                 45678
+github.tar,                         level -3 with dict,                 advanced streaming,                 44656
+github.tar,                         level -1,                           advanced streaming,                 42560
+github.tar,                         level -1 with dict,                 advanced streaming,                 41151
+github.tar,                         level 0,                            advanced streaming,                 38831
 github.tar,                         level 0 with dict,                  advanced streaming,                 37995
 github.tar,                         level 0 with dict dms,              advanced streaming,                 38003
 github.tar,                         level 0 with dict dds,              advanced streaming,                 38003
 github.tar,                         level 0 with dict copy,             advanced streaming,                 37995
 github.tar,                         level 0 with dict load,             advanced streaming,                 37956
-github.tar,                         level 1,                            advanced streaming,                 39342
-github.tar,                         level 1 with dict,                  advanced streaming,                 38293
-github.tar,                         level 1 with dict dms,              advanced streaming,                 38303
-github.tar,                         level 1 with dict dds,              advanced streaming,                 38303
-github.tar,                         level 1 with dict copy,             advanced streaming,                 38293
-github.tar,                         level 1 with dict load,             advanced streaming,                 38766
-github.tar,                         level 3,                            advanced streaming,                 38441
+github.tar,                         level 1,                            advanced streaming,                 39200
+github.tar,                         level 1 with dict,                  advanced streaming,                 38089
+github.tar,                         level 1 with dict dms,              advanced streaming,                 38294
+github.tar,                         level 1 with dict dds,              advanced streaming,                 38294
+github.tar,                         level 1 with dict copy,             advanced streaming,                 38089
+github.tar,                         level 1 with dict load,             advanced streaming,                 38364
+github.tar,                         level 3,                            advanced streaming,                 38831
 github.tar,                         level 3 with dict,                  advanced streaming,                 37995
 github.tar,                         level 3 with dict dms,              advanced streaming,                 38003
 github.tar,                         level 3 with dict dds,              advanced streaming,                 38003
 github.tar,                         level 3 with dict copy,             advanced streaming,                 37995
 github.tar,                         level 3 with dict load,             advanced streaming,                 37956
-github.tar,                         level 4,                            advanced streaming,                 38467
+github.tar,                         level 4,                            advanced streaming,                 38893
 github.tar,                         level 4 with dict,                  advanced streaming,                 37948
 github.tar,                         level 4 with dict dms,              advanced streaming,                 37954
 github.tar,                         level 4 with dict dds,              advanced streaming,                 37954
 github.tar,                         level 4 with dict copy,             advanced streaming,                 37948
 github.tar,                         level 4 with dict load,             advanced streaming,                 37927
-github.tar,                         level 5 row 1,                      advanced streaming,                 39788
-github.tar,                         level 5 row 1 with dict dms,        advanced streaming,                 39365
-github.tar,                         level 5 row 1 with dict dds,        advanced streaming,                 39233
-github.tar,                         level 5 row 1 with dict copy,       advanced streaming,                 39715
-github.tar,                         level 5 row 1 with dict load,       advanced streaming,                 39209
-github.tar,                         level 5 row 2,                      advanced streaming,                 39693
-github.tar,                         level 5 row 2 with dict dms,        advanced streaming,                 39024
-github.tar,                         level 5 row 2 with dict dds,        advanced streaming,                 39028
-github.tar,                         level 5 row 2 with dict copy,       advanced streaming,                 39040
-github.tar,                         level 5 row 2 with dict load,       advanced streaming,                 39037
-github.tar,                         level 5,                            advanced streaming,                 39693
-github.tar,                         level 5 with dict,                  advanced streaming,                 39040
-github.tar,                         level 5 with dict dms,              advanced streaming,                 39024
-github.tar,                         level 5 with dict dds,              advanced streaming,                 39028
-github.tar,                         level 5 with dict copy,             advanced streaming,                 39040
-github.tar,                         level 5 with dict load,             advanced streaming,                 39037
-github.tar,                         level 6,                            advanced streaming,                 39621
-github.tar,                         level 6 with dict,                  advanced streaming,                 38622
-github.tar,                         level 6 with dict dms,              advanced streaming,                 38608
-github.tar,                         level 6 with dict dds,              advanced streaming,                 38610
-github.tar,                         level 6 with dict copy,             advanced streaming,                 38622
-github.tar,                         level 6 with dict load,             advanced streaming,                 38962
-github.tar,                         level 7 row 1,                      advanced streaming,                 39206
-github.tar,                         level 7 row 1 with dict dms,        advanced streaming,                 37954
-github.tar,                         level 7 row 1 with dict dds,        advanced streaming,                 37954
-github.tar,                         level 7 row 1 with dict copy,       advanced streaming,                 38071
-github.tar,                         level 7 row 1 with dict load,       advanced streaming,                 38584
-github.tar,                         level 7 row 2,                      advanced streaming,                 39213
-github.tar,                         level 7 row 2 with dict dms,        advanced streaming,                 37848
-github.tar,                         level 7 row 2 with dict dds,        advanced streaming,                 37867
-github.tar,                         level 7 row 2 with dict copy,       advanced streaming,                 37848
-github.tar,                         level 7 row 2 with dict load,       advanced streaming,                 38582
-github.tar,                         level 7,                            advanced streaming,                 39213
-github.tar,                         level 7 with dict,                  advanced streaming,                 37848
-github.tar,                         level 7 with dict dms,              advanced streaming,                 37848
-github.tar,                         level 7 with dict dds,              advanced streaming,                 37867
-github.tar,                         level 7 with dict copy,             advanced streaming,                 37848
-github.tar,                         level 7 with dict load,             advanced streaming,                 38582
-github.tar,                         level 9,                            advanced streaming,                 36758
-github.tar,                         level 9 with dict,                  advanced streaming,                 36457
-github.tar,                         level 9 with dict dms,              advanced streaming,                 36549
-github.tar,                         level 9 with dict dds,              advanced streaming,                 36637
-github.tar,                         level 9 with dict copy,             advanced streaming,                 36457
-github.tar,                         level 9 with dict load,             advanced streaming,                 36350
-github.tar,                         level 12 row 1,                     advanced streaming,                 36435
+github.tar,                         level 5 row 1,                      advanced streaming,                 39651
+github.tar,                         level 5 row 1 with dict dms,        advanced streaming,                 39059
+github.tar,                         level 5 row 1 with dict dds,        advanced streaming,                 39067
+github.tar,                         level 5 row 1 with dict copy,       advanced streaming,                 39082
+github.tar,                         level 5 row 1 with dict load,       advanced streaming,                 38999
+github.tar,                         level 5 row 2,                      advanced streaming,                 39701
+github.tar,                         level 5 row 2 with dict dms,        advanced streaming,                 39365
+github.tar,                         level 5 row 2 with dict dds,        advanced streaming,                 39233
+github.tar,                         level 5 row 2 with dict copy,       advanced streaming,                 39715
+github.tar,                         level 5 row 2 with dict load,       advanced streaming,                 39158
+github.tar,                         level 5,                            advanced streaming,                 39651
+github.tar,                         level 5 with dict,                  advanced streaming,                 39082
+github.tar,                         level 5 with dict dms,              advanced streaming,                 39059
+github.tar,                         level 5 with dict dds,              advanced streaming,                 39067
+github.tar,                         level 5 with dict copy,             advanced streaming,                 39082
+github.tar,                         level 5 with dict load,             advanced streaming,                 38999
+github.tar,                         level 6,                            advanced streaming,                 39282
+github.tar,                         level 6 with dict,                  advanced streaming,                 38656
+github.tar,                         level 6 with dict dms,              advanced streaming,                 38636
+github.tar,                         level 6 with dict dds,              advanced streaming,                 38634
+github.tar,                         level 6 with dict copy,             advanced streaming,                 38656
+github.tar,                         level 6 with dict load,             advanced streaming,                 38648
+github.tar,                         level 7 row 1,                      advanced streaming,                 38110
+github.tar,                         level 7 row 1 with dict dms,        advanced streaming,                 37858
+github.tar,                         level 7 row 1 with dict dds,        advanced streaming,                 37882
+github.tar,                         level 7 row 1 with dict copy,       advanced streaming,                 37865
+github.tar,                         level 7 row 1 with dict load,       advanced streaming,                 37436
+github.tar,                         level 7 row 2,                      advanced streaming,                 38077
+github.tar,                         level 7 row 2 with dict dms,        advanced streaming,                 38012
+github.tar,                         level 7 row 2 with dict dds,        advanced streaming,                 38014
+github.tar,                         level 7 row 2 with dict copy,       advanced streaming,                 38101
+github.tar,                         level 7 row 2 with dict load,       advanced streaming,                 37402
+github.tar,                         level 7,                            advanced streaming,                 38110
+github.tar,                         level 7 with dict,                  advanced streaming,                 37865
+github.tar,                         level 7 with dict dms,              advanced streaming,                 37858
+github.tar,                         level 7 with dict dds,              advanced streaming,                 37882
+github.tar,                         level 7 with dict copy,             advanced streaming,                 37865
+github.tar,                         level 7 with dict load,             advanced streaming,                 37436
+github.tar,                         level 9,                            advanced streaming,                 36760
+github.tar,                         level 9 with dict,                  advanced streaming,                 36484
+github.tar,                         level 9 with dict dms,              advanced streaming,                 36567
+github.tar,                         level 9 with dict dds,              advanced streaming,                 36628
+github.tar,                         level 9 with dict copy,             advanced streaming,                 36484
+github.tar,                         level 9 with dict load,             advanced streaming,                 36401
+github.tar,                         level 11 row 1,                     advanced streaming,                 36081
+github.tar,                         level 11 row 1 with dict dms,       advanced streaming,                 36963
+github.tar,                         level 11 row 1 with dict dds,       advanced streaming,                 36963
+github.tar,                         level 11 row 1 with dict copy,      advanced streaming,                 36557
+github.tar,                         level 11 row 1 with dict load,      advanced streaming,                 36434
+github.tar,                         level 11 row 2,                     advanced streaming,                 36110
+github.tar,                         level 11 row 2 with dict dms,       advanced streaming,                 36963
+github.tar,                         level 11 row 2 with dict dds,       advanced streaming,                 36963
+github.tar,                         level 11 row 2 with dict copy,      advanced streaming,                 36557
+github.tar,                         level 11 row 2 with dict load,      advanced streaming,                 36459
+github.tar,                         level 12 row 1,                     advanced streaming,                 36081
 github.tar,                         level 12 row 1 with dict dms,       advanced streaming,                 36986
 github.tar,                         level 12 row 1 with dict dds,       advanced streaming,                 36986
 github.tar,                         level 12 row 1 with dict copy,      advanced streaming,                 36609
-github.tar,                         level 12 row 1 with dict load,      advanced streaming,                 36419
-github.tar,                         level 12 row 2,                     advanced streaming,                 36435
+github.tar,                         level 12 row 1 with dict load,      advanced streaming,                 36434
+github.tar,                         level 12 row 2,                     advanced streaming,                 36110
 github.tar,                         level 12 row 2 with dict dms,       advanced streaming,                 36986
 github.tar,                         level 12 row 2 with dict dds,       advanced streaming,                 36986
 github.tar,                         level 12 row 2 with dict copy,      advanced streaming,                 36609
-github.tar,                         level 12 row 2 with dict load,      advanced streaming,                 36424
-github.tar,                         level 13,                           advanced streaming,                 35621
-github.tar,                         level 13 with dict,                 advanced streaming,                 38726
-github.tar,                         level 13 with dict dms,             advanced streaming,                 38903
-github.tar,                         level 13 with dict dds,             advanced streaming,                 38903
-github.tar,                         level 13 with dict copy,            advanced streaming,                 38726
-github.tar,                         level 13 with dict load,            advanced streaming,                 36372
-github.tar,                         level 16,                           advanced streaming,                 40255
-github.tar,                         level 16 with dict,                 advanced streaming,                 33639
-github.tar,                         level 16 with dict dms,             advanced streaming,                 33544
-github.tar,                         level 16 with dict dds,             advanced streaming,                 33544
-github.tar,                         level 16 with dict copy,            advanced streaming,                 33639
-github.tar,                         level 16 with dict load,            advanced streaming,                 39353
-github.tar,                         level 19,                           advanced streaming,                 32837
-github.tar,                         level 19 with dict,                 advanced streaming,                 32895
-github.tar,                         level 19 with dict dms,             advanced streaming,                 32672
-github.tar,                         level 19 with dict dds,             advanced streaming,                 32672
-github.tar,                         level 19 with dict copy,            advanced streaming,                 32895
-github.tar,                         level 19 with dict load,            advanced streaming,                 32676
-github.tar,                         no source size,                     advanced streaming,                 38438
+github.tar,                         level 12 row 2 with dict load,      advanced streaming,                 36459
+github.tar,                         level 13,                           advanced streaming,                 35501
+github.tar,                         level 13 with dict,                 advanced streaming,                 37130
+github.tar,                         level 13 with dict dms,             advanced streaming,                 37220
+github.tar,                         level 13 with dict dds,             advanced streaming,                 37220
+github.tar,                         level 13 with dict copy,            advanced streaming,                 37130
+github.tar,                         level 13 with dict load,            advanced streaming,                 36010
+github.tar,                         level 16,                           advanced streaming,                 40466
+github.tar,                         level 16 with dict,                 advanced streaming,                 33374
+github.tar,                         level 16 with dict dms,             advanced streaming,                 33206
+github.tar,                         level 16 with dict dds,             advanced streaming,                 33206
+github.tar,                         level 16 with dict copy,            advanced streaming,                 33374
+github.tar,                         level 16 with dict load,            advanced streaming,                 39081
+github.tar,                         level 19,                           advanced streaming,                 32276
+github.tar,                         level 19 with dict,                 advanced streaming,                 32712
+github.tar,                         level 19 with dict dms,             advanced streaming,                 32555
+github.tar,                         level 19 with dict dds,             advanced streaming,                 32555
+github.tar,                         level 19 with dict copy,            advanced streaming,                 32712
+github.tar,                         level 19 with dict load,            advanced streaming,                 32479
+github.tar,                         no source size,                     advanced streaming,                 38828
 github.tar,                         no source size with dict,           advanced streaming,                 38000
-github.tar,                         long distance mode,                 advanced streaming,                 39757
-github.tar,                         multithreaded,                      advanced streaming,                 38441
-github.tar,                         multithreaded long distance mode,   advanced streaming,                 39726
+github.tar,                         long distance mode,                 advanced streaming,                 40252
+github.tar,                         multithreaded,                      advanced streaming,                 38831
+github.tar,                         multithreaded long distance mode,   advanced streaming,                 40232
 github.tar,                         small window log,                   advanced streaming,                 199558
 github.tar,                         small hash log,                     advanced streaming,                 129870
 github.tar,                         small chain log,                    advanced streaming,                 41669
-github.tar,                         explicit params,                    advanced streaming,                 41227
-github.tar,                         uncompressed literals,              advanced streaming,                 41122
-github.tar,                         uncompressed literals optimal,      advanced streaming,                 35388
-github.tar,                         huffman literals,                   advanced streaming,                 38800
-github.tar,                         multithreaded with advanced params, advanced streaming,                 41122
-silesia,                            level -5,                           old streaming,                      6882505
-silesia,                            level -3,                           old streaming,                      6568376
-silesia,                            level -1,                           old streaming,                      6183403
-silesia,                            level 0,                            old streaming,                      4849552
-silesia,                            level 1,                            old streaming,                      5314162
-silesia,                            level 3,                            old streaming,                      4849552
-silesia,                            level 4,                            old streaming,                      4786970
-silesia,                            level 5,                            old streaming,                      4707794
-silesia,                            level 6,                            old streaming,                      4666383
-silesia,                            level 7,                            old streaming,                      4603381
-silesia,                            level 9,                            old streaming,                      4546001
-silesia,                            level 13,                           old streaming,                      4482135
-silesia,                            level 16,                           old streaming,                      4360251
-silesia,                            level 19,                           old streaming,                      4283237
-silesia,                            no source size,                     old streaming,                      4849516
-silesia,                            uncompressed literals,              old streaming,                      4849552
-silesia,                            uncompressed literals optimal,      old streaming,                      4283237
-silesia,                            huffman literals,                   old streaming,                      6183403
-silesia.tar,                        level -5,                           old streaming,                      6982759
-silesia.tar,                        level -3,                           old streaming,                      6641283
-silesia.tar,                        level -1,                           old streaming,                      6190795
-silesia.tar,                        level 0,                            old streaming,                      4861427
-silesia.tar,                        level 1,                            old streaming,                      5336939
-silesia.tar,                        level 3,                            old streaming,                      4861427
-silesia.tar,                        level 4,                            old streaming,                      4799630
-silesia.tar,                        level 5,                            old streaming,                      4719261
-silesia.tar,                        level 6,                            old streaming,                      4677729
-silesia.tar,                        level 7,                            old streaming,                      4613544
-silesia.tar,                        level 9,                            old streaming,                      4555432
-silesia.tar,                        level 13,                           old streaming,                      4491765
-silesia.tar,                        level 16,                           old streaming,                      4356834
-silesia.tar,                        level 19,                           old streaming,                      4264392
-silesia.tar,                        no source size,                     old streaming,                      4861423
-silesia.tar,                        uncompressed literals,              old streaming,                      4861427
-silesia.tar,                        uncompressed literals optimal,      old streaming,                      4264392
-silesia.tar,                        huffman literals,                   old streaming,                      6190795
-github,                             level -5,                           old streaming,                      205285
+github.tar,                         explicit params,                    advanced streaming,                 41385
+github.tar,                         uncompressed literals,              advanced streaming,                 41525
+github.tar,                         uncompressed literals optimal,      advanced streaming,                 35397
+github.tar,                         huffman literals,                   advanced streaming,                 38853
+github.tar,                         multithreaded with advanced params, advanced streaming,                 41525
+silesia,                            level -5,                           old streaming,                      6854744
+silesia,                            level -3,                           old streaming,                      6503319
+silesia,                            level -1,                           old streaming,                      6172207
+silesia,                            level 0,                            old streaming,                      4842075
+silesia,                            level 1,                            old streaming,                      5306388
+silesia,                            level 3,                            old streaming,                      4842075
+silesia,                            level 4,                            old streaming,                      4779186
+silesia,                            level 5,                            old streaming,                      4666323
+silesia,                            level 6,                            old streaming,                      4603066
+silesia,                            level 7,                            old streaming,                      4566984
+silesia,                            level 9,                            old streaming,                      4543018
+silesia,                            level 13,                           old streaming,                      4493990
+silesia,                            level 16,                           old streaming,                      4360041
+silesia,                            level 19,                           old streaming,                      4296055
+silesia,                            no source size,                     old streaming,                      4842039
+silesia,                            uncompressed literals,              old streaming,                      4842075
+silesia,                            uncompressed literals optimal,      old streaming,                      4296055
+silesia,                            huffman literals,                   old streaming,                      6172207
+silesia.tar,                        level -5,                           old streaming,                      6856523
+silesia.tar,                        level -3,                           old streaming,                      6505954
+silesia.tar,                        level -1,                           old streaming,                      6179056
+silesia.tar,                        level 0,                            old streaming,                      4859271
+silesia.tar,                        level 1,                            old streaming,                      5327708
+silesia.tar,                        level 3,                            old streaming,                      4859271
+silesia.tar,                        level 4,                            old streaming,                      4797470
+silesia.tar,                        level 5,                            old streaming,                      4677748
+silesia.tar,                        level 6,                            old streaming,                      4613246
+silesia.tar,                        level 7,                            old streaming,                      4576664
+silesia.tar,                        level 9,                            old streaming,                      4552900
+silesia.tar,                        level 13,                           old streaming,                      4502956
+silesia.tar,                        level 16,                           old streaming,                      4360546
+silesia.tar,                        level 19,                           old streaming,                      4265911
+silesia.tar,                        no source size,                     old streaming,                      4859267
+silesia.tar,                        uncompressed literals,              old streaming,                      4859271
+silesia.tar,                        uncompressed literals optimal,      old streaming,                      4265911
+silesia.tar,                        huffman literals,                   old streaming,                      6179056
+github,                             level -5,                           old streaming,                      204407
 github,                             level -5 with dict,                 old streaming,                      46718
-github,                             level -3,                           old streaming,                      190643
+github,                             level -3,                           old streaming,                      193253
 github,                             level -3 with dict,                 old streaming,                      45395
-github,                             level -1,                           old streaming,                      175568
+github,                             level -1,                           old streaming,                      175468
 github,                             level -1 with dict,                 old streaming,                      43170
-github,                             level 0,                            old streaming,                      136335
+github,                             level 0,                            old streaming,                      136332
 github,                             level 0 with dict,                  old streaming,                      41148
-github,                             level 1,                            old streaming,                      142465
+github,                             level 1,                            old streaming,                      142365
 github,                             level 1 with dict,                  old streaming,                      41682
-github,                             level 3,                            old streaming,                      136335
+github,                             level 3,                            old streaming,                      136332
 github,                             level 3 with dict,                  old streaming,                      41148
 github,                             level 4,                            old streaming,                      136199
 github,                             level 4 with dict,                  old streaming,                      41251
@@ -1168,131 +1240,131 @@ github,                             level 7,                            old stre
 github,                             level 7 with dict,                  old streaming,                      38758
 github,                             level 9,                            old streaming,                      135122
 github,                             level 9 with dict,                  old streaming,                      39437
-github,                             level 13,                           old streaming,                      134064
-github,                             level 13 with dict,                 old streaming,                      39743
-github,                             level 16,                           old streaming,                      134064
+github,                             level 13,                           old streaming,                      132878
+github,                             level 13 with dict,                 old streaming,                      39900
+github,                             level 16,                           old streaming,                      133209
 github,                             level 16 with dict,                 old streaming,                      37577
-github,                             level 19,                           old streaming,                      134064
+github,                             level 19,                           old streaming,                      132879
 github,                             level 19 with dict,                 old streaming,                      37576
-github,                             no source size,                     old streaming,                      140632
+github,                             no source size,                     old streaming,                      140599
 github,                             no source size with dict,           old streaming,                      40654
-github,                             uncompressed literals,              old streaming,                      136335
-github,                             uncompressed literals optimal,      old streaming,                      134064
-github,                             huffman literals,                   old streaming,                      175568
-github.tar,                         level -5,                           old streaming,                      46747
-github.tar,                         level -5 with dict,                 old streaming,                      44440
-github.tar,                         level -3,                           old streaming,                      43537
-github.tar,                         level -3 with dict,                 old streaming,                      41112
-github.tar,                         level -1,                           old streaming,                      42465
-github.tar,                         level -1 with dict,                 old streaming,                      41196
-github.tar,                         level 0,                            old streaming,                      38441
+github,                             uncompressed literals,              old streaming,                      136332
+github,                             uncompressed literals optimal,      old streaming,                      132879
+github,                             huffman literals,                   old streaming,                      175468
+github.tar,                         level -5,                           old streaming,                      52152
+github.tar,                         level -5 with dict,                 old streaming,                      51045
+github.tar,                         level -3,                           old streaming,                      45678
+github.tar,                         level -3 with dict,                 old streaming,                      44656
+github.tar,                         level -1,                           old streaming,                      42560
+github.tar,                         level -1 with dict,                 old streaming,                      41151
+github.tar,                         level 0,                            old streaming,                      38831
 github.tar,                         level 0 with dict,                  old streaming,                      37995
-github.tar,                         level 1,                            old streaming,                      39342
-github.tar,                         level 1 with dict,                  old streaming,                      38293
-github.tar,                         level 3,                            old streaming,                      38441
+github.tar,                         level 1,                            old streaming,                      39200
+github.tar,                         level 1 with dict,                  old streaming,                      38089
+github.tar,                         level 3,                            old streaming,                      38831
 github.tar,                         level 3 with dict,                  old streaming,                      37995
-github.tar,                         level 4,                            old streaming,                      38467
+github.tar,                         level 4,                            old streaming,                      38893
 github.tar,                         level 4 with dict,                  old streaming,                      37948
-github.tar,                         level 5,                            old streaming,                      39693
-github.tar,                         level 5 with dict,                  old streaming,                      39040
-github.tar,                         level 6,                            old streaming,                      39621
-github.tar,                         level 6 with dict,                  old streaming,                      38622
-github.tar,                         level 7,                            old streaming,                      39213
-github.tar,                         level 7 with dict,                  old streaming,                      37848
-github.tar,                         level 9,                            old streaming,                      36758
-github.tar,                         level 9 with dict,                  old streaming,                      36457
-github.tar,                         level 13,                           old streaming,                      35621
-github.tar,                         level 13 with dict,                 old streaming,                      38726
-github.tar,                         level 16,                           old streaming,                      40255
-github.tar,                         level 16 with dict,                 old streaming,                      33639
-github.tar,                         level 19,                           old streaming,                      32837
-github.tar,                         level 19 with dict,                 old streaming,                      32895
-github.tar,                         no source size,                     old streaming,                      38438
+github.tar,                         level 5,                            old streaming,                      39651
+github.tar,                         level 5 with dict,                  old streaming,                      39082
+github.tar,                         level 6,                            old streaming,                      39282
+github.tar,                         level 6 with dict,                  old streaming,                      38656
+github.tar,                         level 7,                            old streaming,                      38110
+github.tar,                         level 7 with dict,                  old streaming,                      37865
+github.tar,                         level 9,                            old streaming,                      36760
+github.tar,                         level 9 with dict,                  old streaming,                      36484
+github.tar,                         level 13,                           old streaming,                      35501
+github.tar,                         level 13 with dict,                 old streaming,                      37130
+github.tar,                         level 16,                           old streaming,                      40466
+github.tar,                         level 16 with dict,                 old streaming,                      33374
+github.tar,                         level 19,                           old streaming,                      32276
+github.tar,                         level 19 with dict,                 old streaming,                      32712
+github.tar,                         no source size,                     old streaming,                      38828
 github.tar,                         no source size with dict,           old streaming,                      38000
-github.tar,                         uncompressed literals,              old streaming,                      38441
-github.tar,                         uncompressed literals optimal,      old streaming,                      32837
-github.tar,                         huffman literals,                   old streaming,                      42465
-silesia,                            level -5,                           old streaming advanced,             6882505
-silesia,                            level -3,                           old streaming advanced,             6568376
-silesia,                            level -1,                           old streaming advanced,             6183403
-silesia,                            level 0,                            old streaming advanced,             4849552
-silesia,                            level 1,                            old streaming advanced,             5314162
-silesia,                            level 3,                            old streaming advanced,             4849552
-silesia,                            level 4,                            old streaming advanced,             4786970
-silesia,                            level 5,                            old streaming advanced,             4707794
-silesia,                            level 6,                            old streaming advanced,             4666383
-silesia,                            level 7,                            old streaming advanced,             4603381
-silesia,                            level 9,                            old streaming advanced,             4546001
-silesia,                            level 13,                           old streaming advanced,             4482135
-silesia,                            level 16,                           old streaming advanced,             4360251
-silesia,                            level 19,                           old streaming advanced,             4283237
-silesia,                            no source size,                     old streaming advanced,             4849516
-silesia,                            long distance mode,                 old streaming advanced,             4849552
-silesia,                            multithreaded,                      old streaming advanced,             4849552
-silesia,                            multithreaded long distance mode,   old streaming advanced,             4849552
-silesia,                            small window log,                   old streaming advanced,             7112062
+github.tar,                         uncompressed literals,              old streaming,                      38831
+github.tar,                         uncompressed literals optimal,      old streaming,                      32276
+github.tar,                         huffman literals,                   old streaming,                      42560
+silesia,                            level -5,                           old streaming advanced,             6854744
+silesia,                            level -3,                           old streaming advanced,             6503319
+silesia,                            level -1,                           old streaming advanced,             6172207
+silesia,                            level 0,                            old streaming advanced,             4842075
+silesia,                            level 1,                            old streaming advanced,             5306388
+silesia,                            level 3,                            old streaming advanced,             4842075
+silesia,                            level 4,                            old streaming advanced,             4779186
+silesia,                            level 5,                            old streaming advanced,             4666323
+silesia,                            level 6,                            old streaming advanced,             4603066
+silesia,                            level 7,                            old streaming advanced,             4566984
+silesia,                            level 9,                            old streaming advanced,             4543018
+silesia,                            level 13,                           old streaming advanced,             4493990
+silesia,                            level 16,                           old streaming advanced,             4360041
+silesia,                            level 19,                           old streaming advanced,             4296055
+silesia,                            no source size,                     old streaming advanced,             4842039
+silesia,                            long distance mode,                 old streaming advanced,             4842075
+silesia,                            multithreaded,                      old streaming advanced,             4842075
+silesia,                            multithreaded long distance mode,   old streaming advanced,             4842075
+silesia,                            small window log,                   old streaming advanced,             7111103
 silesia,                            small hash log,                     old streaming advanced,             6526141
 silesia,                            small chain log,                    old streaming advanced,             4912197
-silesia,                            explicit params,                    old streaming advanced,             4795887
-silesia,                            uncompressed literals,              old streaming advanced,             4849552
-silesia,                            uncompressed literals optimal,      old streaming advanced,             4283237
-silesia,                            huffman literals,                   old streaming advanced,             6183403
-silesia,                            multithreaded with advanced params, old streaming advanced,             4849552
-silesia.tar,                        level -5,                           old streaming advanced,             6982759
-silesia.tar,                        level -3,                           old streaming advanced,             6641283
-silesia.tar,                        level -1,                           old streaming advanced,             6190795
-silesia.tar,                        level 0,                            old streaming advanced,             4861427
-silesia.tar,                        level 1,                            old streaming advanced,             5336939
-silesia.tar,                        level 3,                            old streaming advanced,             4861427
-silesia.tar,                        level 4,                            old streaming advanced,             4799630
-silesia.tar,                        level 5,                            old streaming advanced,             4719261
-silesia.tar,                        level 6,                            old streaming advanced,             4677729
-silesia.tar,                        level 7,                            old streaming advanced,             4613544
-silesia.tar,                        level 9,                            old streaming advanced,             4555432
-silesia.tar,                        level 13,                           old streaming advanced,             4491765
-silesia.tar,                        level 16,                           old streaming advanced,             4356834
-silesia.tar,                        level 19,                           old streaming advanced,             4264392
-silesia.tar,                        no source size,                     old streaming advanced,             4861423
-silesia.tar,                        long distance mode,                 old streaming advanced,             4861427
-silesia.tar,                        multithreaded,                      old streaming advanced,             4861427
-silesia.tar,                        multithreaded long distance mode,   old streaming advanced,             4861427
-silesia.tar,                        small window log,                   old streaming advanced,             7118772
-silesia.tar,                        small hash log,                     old streaming advanced,             6529235
+silesia,                            explicit params,                    old streaming advanced,             4795452
+silesia,                            uncompressed literals,              old streaming advanced,             4842075
+silesia,                            uncompressed literals optimal,      old streaming advanced,             4296055
+silesia,                            huffman literals,                   old streaming advanced,             6172207
+silesia,                            multithreaded with advanced params, old streaming advanced,             4842075
+silesia.tar,                        level -5,                           old streaming advanced,             6856523
+silesia.tar,                        level -3,                           old streaming advanced,             6505954
+silesia.tar,                        level -1,                           old streaming advanced,             6179056
+silesia.tar,                        level 0,                            old streaming advanced,             4859271
+silesia.tar,                        level 1,                            old streaming advanced,             5327708
+silesia.tar,                        level 3,                            old streaming advanced,             4859271
+silesia.tar,                        level 4,                            old streaming advanced,             4797470
+silesia.tar,                        level 5,                            old streaming advanced,             4677748
+silesia.tar,                        level 6,                            old streaming advanced,             4613246
+silesia.tar,                        level 7,                            old streaming advanced,             4576664
+silesia.tar,                        level 9,                            old streaming advanced,             4552900
+silesia.tar,                        level 13,                           old streaming advanced,             4502956
+silesia.tar,                        level 16,                           old streaming advanced,             4360546
+silesia.tar,                        level 19,                           old streaming advanced,             4265911
+silesia.tar,                        no source size,                     old streaming advanced,             4859267
+silesia.tar,                        long distance mode,                 old streaming advanced,             4859271
+silesia.tar,                        multithreaded,                      old streaming advanced,             4859271
+silesia.tar,                        multithreaded long distance mode,   old streaming advanced,             4859271
+silesia.tar,                        small window log,                   old streaming advanced,             7117562
+silesia.tar,                        small hash log,                     old streaming advanced,             6529209
 silesia.tar,                        small chain log,                    old streaming advanced,             4917021
-silesia.tar,                        explicit params,                    old streaming advanced,             4807401
-silesia.tar,                        uncompressed literals,              old streaming advanced,             4861427
-silesia.tar,                        uncompressed literals optimal,      old streaming advanced,             4264392
-silesia.tar,                        huffman literals,                   old streaming advanced,             6190795
-silesia.tar,                        multithreaded with advanced params, old streaming advanced,             4861427
-github,                             level -5,                           old streaming advanced,             216734
+silesia.tar,                        explicit params,                    old streaming advanced,             4806873
+silesia.tar,                        uncompressed literals,              old streaming advanced,             4859271
+silesia.tar,                        uncompressed literals optimal,      old streaming advanced,             4265911
+silesia.tar,                        huffman literals,                   old streaming advanced,             6179056
+silesia.tar,                        multithreaded with advanced params, old streaming advanced,             4859271
+github,                             level -5,                           old streaming advanced,             213265
 github,                             level -5 with dict,                 old streaming advanced,             49562
-github,                             level -3,                           old streaming advanced,             192160
+github,                             level -3,                           old streaming advanced,             196126
 github,                             level -3 with dict,                 old streaming advanced,             44956
-github,                             level -1,                           old streaming advanced,             181108
+github,                             level -1,                           old streaming advanced,             181107
 github,                             level -1 with dict,                 old streaming advanced,             42383
 github,                             level 0,                            old streaming advanced,             141104
 github,                             level 0 with dict,                  old streaming advanced,             41113
-github,                             level 1,                            old streaming advanced,             143692
+github,                             level 1,                            old streaming advanced,             143693
 github,                             level 1 with dict,                  old streaming advanced,             42430
 github,                             level 3,                            old streaming advanced,             141104
 github,                             level 3 with dict,                  old streaming advanced,             41113
 github,                             level 4,                            old streaming advanced,             141104
 github,                             level 4 with dict,                  old streaming advanced,             41084
-github,                             level 5,                            old streaming advanced,             139399
-github,                             level 5 with dict,                  old streaming advanced,             38633
-github,                             level 6,                            old streaming advanced,             139402
-github,                             level 6 with dict,                  old streaming advanced,             38723
+github,                             level 5,                            old streaming advanced,             139402
+github,                             level 5 with dict,                  old streaming advanced,             38723
+github,                             level 6,                            old streaming advanced,             138676
+github,                             level 6 with dict,                  old streaming advanced,             38744
 github,                             level 7,                            old streaming advanced,             138676
-github,                             level 7 with dict,                  old streaming advanced,             38744
+github,                             level 7 with dict,                  old streaming advanced,             38924
 github,                             level 9,                            old streaming advanced,             138676
 github,                             level 9 with dict,                  old streaming advanced,             38981
 github,                             level 13,                           old streaming advanced,             138676
-github,                             level 13 with dict,                 old streaming advanced,             39731
-github,                             level 16,                           old streaming advanced,             138676
+github,                             level 13 with dict,                 old streaming advanced,             39725
+github,                             level 16,                           old streaming advanced,             138575
 github,                             level 16 with dict,                 old streaming advanced,             40789
-github,                             level 19,                           old streaming advanced,             134064
+github,                             level 19,                           old streaming advanced,             132879
 github,                             level 19 with dict,                 old streaming advanced,             37576
-github,                             no source size,                     old streaming advanced,             140632
+github,                             no source size,                     old streaming advanced,             140599
 github,                             no source size with dict,           old streaming advanced,             40608
 github,                             long distance mode,                 old streaming advanced,             141104
 github,                             multithreaded,                      old streaming advanced,             141104
@@ -1302,50 +1374,50 @@ github,                             small hash log,                     old stre
 github,                             small chain log,                    old streaming advanced,             139275
 github,                             explicit params,                    old streaming advanced,             140937
 github,                             uncompressed literals,              old streaming advanced,             141104
-github,                             uncompressed literals optimal,      old streaming advanced,             134064
-github,                             huffman literals,                   old streaming advanced,             181108
+github,                             uncompressed literals optimal,      old streaming advanced,             132879
+github,                             huffman literals,                   old streaming advanced,             181107
 github,                             multithreaded with advanced params, old streaming advanced,             141104
-github.tar,                         level -5,                           old streaming advanced,             46747
-github.tar,                         level -5 with dict,                 old streaming advanced,             44824
-github.tar,                         level -3,                           old streaming advanced,             43537
-github.tar,                         level -3 with dict,                 old streaming advanced,             41800
-github.tar,                         level -1,                           old streaming advanced,             42465
-github.tar,                         level -1 with dict,                 old streaming advanced,             41471
-github.tar,                         level 0,                            old streaming advanced,             38441
+github.tar,                         level -5,                           old streaming advanced,             52152
+github.tar,                         level -5 with dict,                 old streaming advanced,             50988
+github.tar,                         level -3,                           old streaming advanced,             45678
+github.tar,                         level -3 with dict,                 old streaming advanced,             44729
+github.tar,                         level -1,                           old streaming advanced,             42560
+github.tar,                         level -1 with dict,                 old streaming advanced,             41589
+github.tar,                         level 0,                            old streaming advanced,             38831
 github.tar,                         level 0 with dict,                  old streaming advanced,             38013
-github.tar,                         level 1,                            old streaming advanced,             39342
-github.tar,                         level 1 with dict,                  old streaming advanced,             38940
-github.tar,                         level 3,                            old streaming advanced,             38441
+github.tar,                         level 1,                            old streaming advanced,             39200
+github.tar,                         level 1 with dict,                  old streaming advanced,             38359
+github.tar,                         level 3,                            old streaming advanced,             38831
 github.tar,                         level 3 with dict,                  old streaming advanced,             38013
-github.tar,                         level 4,                            old streaming advanced,             38467
+github.tar,                         level 4,                            old streaming advanced,             38893
 github.tar,                         level 4 with dict,                  old streaming advanced,             38063
-github.tar,                         level 5,                            old streaming advanced,             39693
-github.tar,                         level 5 with dict,                  old streaming advanced,             39049
-github.tar,                         level 6,                            old streaming advanced,             39621
-github.tar,                         level 6 with dict,                  old streaming advanced,             38959
-github.tar,                         level 7,                            old streaming advanced,             39213
-github.tar,                         level 7 with dict,                  old streaming advanced,             38573
-github.tar,                         level 9,                            old streaming advanced,             36758
-github.tar,                         level 9 with dict,                  old streaming advanced,             36233
-github.tar,                         level 13,                           old streaming advanced,             35621
-github.tar,                         level 13 with dict,                 old streaming advanced,             36035
-github.tar,                         level 16,                           old streaming advanced,             40255
-github.tar,                         level 16 with dict,                 old streaming advanced,             38736
-github.tar,                         level 19,                           old streaming advanced,             32837
-github.tar,                         level 19 with dict,                 old streaming advanced,             32876
-github.tar,                         no source size,                     old streaming advanced,             38438
+github.tar,                         level 5,                            old streaming advanced,             39651
+github.tar,                         level 5 with dict,                  old streaming advanced,             38997
+github.tar,                         level 6,                            old streaming advanced,             39282
+github.tar,                         level 6 with dict,                  old streaming advanced,             38640
+github.tar,                         level 7,                            old streaming advanced,             38110
+github.tar,                         level 7 with dict,                  old streaming advanced,             37387
+github.tar,                         level 9,                            old streaming advanced,             36760
+github.tar,                         level 9 with dict,                  old streaming advanced,             36312
+github.tar,                         level 13,                           old streaming advanced,             35501
+github.tar,                         level 13 with dict,                 old streaming advanced,             35807
+github.tar,                         level 16,                           old streaming advanced,             40466
+github.tar,                         level 16 with dict,                 old streaming advanced,             38578
+github.tar,                         level 19,                           old streaming advanced,             32276
+github.tar,                         level 19 with dict,                 old streaming advanced,             32704
+github.tar,                         no source size,                     old streaming advanced,             38828
 github.tar,                         no source size with dict,           old streaming advanced,             38015
-github.tar,                         long distance mode,                 old streaming advanced,             38441
-github.tar,                         multithreaded,                      old streaming advanced,             38441
-github.tar,                         multithreaded long distance mode,   old streaming advanced,             38441
+github.tar,                         long distance mode,                 old streaming advanced,             38831
+github.tar,                         multithreaded,                      old streaming advanced,             38831
+github.tar,                         multithreaded long distance mode,   old streaming advanced,             38831
 github.tar,                         small window log,                   old streaming advanced,             199561
 github.tar,                         small hash log,                     old streaming advanced,             129870
 github.tar,                         small chain log,                    old streaming advanced,             41669
-github.tar,                         explicit params,                    old streaming advanced,             41227
-github.tar,                         uncompressed literals,              old streaming advanced,             38441
-github.tar,                         uncompressed literals optimal,      old streaming advanced,             32837
-github.tar,                         huffman literals,                   old streaming advanced,             42465
-github.tar,                         multithreaded with advanced params, old streaming advanced,             38441
+github.tar,                         explicit params,                    old streaming advanced,             41385
+github.tar,                         uncompressed literals,              old streaming advanced,             38831
+github.tar,                         uncompressed literals optimal,      old streaming advanced,             32276
+github.tar,                         huffman literals,                   old streaming advanced,             42560
+github.tar,                         multithreaded with advanced params, old streaming advanced,             38831
 github,                             level -5 with dict,                 old streaming cdict,                46718
 github,                             level -3 with dict,                 old streaming cdict,                45395
 github,                             level -1 with dict,                 old streaming cdict,                43170
@@ -1357,24 +1429,24 @@ github,                             level 5 with dict,                  old stre
 github,                             level 6 with dict,                  old streaming cdict,                38671
 github,                             level 7 with dict,                  old streaming cdict,                38758
 github,                             level 9 with dict,                  old streaming cdict,                39437
-github,                             level 13 with dict,                 old streaming cdict,                39743
+github,                             level 13 with dict,                 old streaming cdict,                39900
 github,                             level 16 with dict,                 old streaming cdict,                37577
 github,                             level 19 with dict,                 old streaming cdict,                37576
 github,                             no source size with dict,           old streaming cdict,                40654
-github.tar,                         level -5 with dict,                 old streaming cdict,                45018
-github.tar,                         level -3 with dict,                 old streaming cdict,                41886
-github.tar,                         level -1 with dict,                 old streaming cdict,                41636
+github.tar,                         level -5 with dict,                 old streaming cdict,                51191
+github.tar,                         level -3 with dict,                 old streaming cdict,                44821
+github.tar,                         level -1 with dict,                 old streaming cdict,                41775
 github.tar,                         level 0 with dict,                  old streaming cdict,                37956
-github.tar,                         level 1 with dict,                  old streaming cdict,                38766
+github.tar,                         level 1 with dict,                  old streaming cdict,                38364
 github.tar,                         level 3 with dict,                  old streaming cdict,                37956
 github.tar,                         level 4 with dict,                  old streaming cdict,                37927
-github.tar,                         level 5 with dict,                  old streaming cdict,                39037
-github.tar,                         level 6 with dict,                  old streaming cdict,                38962
-github.tar,                         level 7 with dict,                  old streaming cdict,                38582
-github.tar,                         level 9 with dict,                  old streaming cdict,                36350
-github.tar,                         level 13 with dict,                 old streaming cdict,                36372
-github.tar,                         level 16 with dict,                 old streaming cdict,                39353
-github.tar,                         level 19 with dict,                 old streaming cdict,                32676
+github.tar,                         level 5 with dict,                  old streaming cdict,                38999
+github.tar,                         level 6 with dict,                  old streaming cdict,                38648
+github.tar,                         level 7 with dict,                  old streaming cdict,                37436
+github.tar,                         level 9 with dict,                  old streaming cdict,                36401
+github.tar,                         level 13 with dict,                 old streaming cdict,                36010
+github.tar,                         level 16 with dict,                 old streaming cdict,                39081
+github.tar,                         level 19 with dict,                 old streaming cdict,                32479
 github.tar,                         no source size with dict,           old streaming cdict,                38000
 github,                             level -5 with dict,                 old streaming advanced cdict,       49562
 github,                             level -3 with dict,                 old streaming advanced cdict,       44956
@@ -1383,26 +1455,26 @@ github,                             level 0 with dict,                  old stre
 github,                             level 1 with dict,                  old streaming advanced cdict,       42430
 github,                             level 3 with dict,                  old streaming advanced cdict,       41113
 github,                             level 4 with dict,                  old streaming advanced cdict,       41084
-github,                             level 5 with dict,                  old streaming advanced cdict,       38633
-github,                             level 6 with dict,                  old streaming advanced cdict,       38723
-github,                             level 7 with dict,                  old streaming advanced cdict,       38744
+github,                             level 5 with dict,                  old streaming advanced cdict,       38723
+github,                             level 6 with dict,                  old streaming advanced cdict,       38744
+github,                             level 7 with dict,                  old streaming advanced cdict,       38924
 github,                             level 9 with dict,                  old streaming advanced cdict,       38981
-github,                             level 13 with dict,                 old streaming advanced cdict,       39731
+github,                             level 13 with dict,                 old streaming advanced cdict,       39725
 github,                             level 16 with dict,                 old streaming advanced cdict,       40789
 github,                             level 19 with dict,                 old streaming advanced cdict,       37576
 github,                             no source size with dict,           old streaming advanced cdict,       40608
-github.tar,                         level -5 with dict,                 old streaming advanced cdict,       44307
-github.tar,                         level -3 with dict,                 old streaming advanced cdict,       41359
-github.tar,                         level -1 with dict,                 old streaming advanced cdict,       41322
+github.tar,                         level -5 with dict,                 old streaming advanced cdict,       50854
+github.tar,                         level -3 with dict,                 old streaming advanced cdict,       44571
+github.tar,                         level -1 with dict,                 old streaming advanced cdict,       41477
 github.tar,                         level 0 with dict,                  old streaming advanced cdict,       38013
-github.tar,                         level 1 with dict,                  old streaming advanced cdict,       39002
+github.tar,                         level 1 with dict,                  old streaming advanced cdict,       38168
 github.tar,                         level 3 with dict,                  old streaming advanced cdict,       38013
 github.tar,                         level 4 with dict,                  old streaming advanced cdict,       38063
-github.tar,                         level 5 with dict,                  old streaming advanced cdict,       39049
-github.tar,                         level 6 with dict,                  old streaming advanced cdict,       38959
-github.tar,                         level 7 with dict,                  old streaming advanced cdict,       38573
-github.tar,                         level 9 with dict,                  old streaming advanced cdict,       36233
-github.tar,                         level 13 with dict,                 old streaming advanced cdict,       36035
-github.tar,                         level 16 with dict,                 old streaming advanced cdict,       38736
-github.tar,                         level 19 with dict,                 old streaming advanced cdict,       32876
+github.tar,                         level 5 with dict,                  old streaming advanced cdict,       38997
+github.tar,                         level 6 with dict,                  old streaming advanced cdict,       38640
+github.tar,                         level 7 with dict,                  old streaming advanced cdict,       37387
+github.tar,                         level 9 with dict,                  old streaming advanced cdict,       36312
+github.tar,                         level 13 with dict,                 old streaming advanced cdict,       35807
+github.tar,                         level 16 with dict,                 old streaming advanced cdict,       38578
+github.tar,                         level 19 with dict,                 old streaming advanced cdict,       32704
 github.tar,                         no source size with dict,           old streaming advanced cdict,       38015
diff --git a/priv/zstd/tests/regression/test.c b/priv/zstd/tests/regression/test.c
index 1de6be8..07600be 100644
--- a/priv/zstd/tests/regression/test.c
+++ b/priv/zstd/tests/regression/test.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/roundTripCrash.c b/priv/zstd/tests/roundTripCrash.c
index 9aa208c..77411cd 100644
--- a/priv/zstd/tests/roundTripCrash.c
+++ b/priv/zstd/tests/roundTripCrash.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/seqgen.c b/priv/zstd/tests/seqgen.c
index 1e340c8..0d8a766 100644
--- a/priv/zstd/tests/seqgen.c
+++ b/priv/zstd/tests/seqgen.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/seqgen.h b/priv/zstd/tests/seqgen.h
index cea3f55..df17398 100644
--- a/priv/zstd/tests/seqgen.h
+++ b/priv/zstd/tests/seqgen.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
diff --git a/priv/zstd/tests/test-license.py b/priv/zstd/tests/test-license.py
index 2247765..d54c164 100755
--- a/priv/zstd/tests/test-license.py
+++ b/priv/zstd/tests/test-license.py
@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 
 # ################################################################
-# Copyright (c) Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -43,6 +43,7 @@ def to_abs(d):
     "Makefile",
     ".mk",
     ".py",
+    ".S",
 ]
 
 # License should certainly be in the first 10 KB.
@@ -82,8 +83,8 @@ def valid_copyright(lines):
             continue
         if "present" in line:
             return (False, f"Copyright line '{line}' contains 'present'!")
-        if "Facebook, Inc" not in line:
-            return (False, f"Copyright line '{line}' does not contain 'Facebook, Inc'")
+        if "Meta Platforms, Inc" not in line:
+            return (False, f"Copyright line '{line}' does not contain 'Meta Platforms, Inc'")
         year = YEAR_REGEX.search(line)
         if year is not None:
             return (False, f"Copyright line '{line}' contains {year.group(0)}; it should be yearless")
diff --git a/priv/zstd/tests/test-variants.sh b/priv/zstd/tests/test-variants.sh
new file mode 100755
index 0000000..f3a9e06
--- /dev/null
+++ b/priv/zstd/tests/test-variants.sh
@@ -0,0 +1,115 @@
+#!/bin/sh
+set -e
+set -u
+set -x
+
+
+SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
+PROG_DIR="$SCRIPT_DIR/../programs"
+
+ZSTD="$PROG_DIR/zstd"
+ZSTD_COMPRESS="$PROG_DIR/zstd-compress"
+ZSTD_DECOMPRESS="$PROG_DIR/zstd-decompress"
+ZSTD_NOLEGACY="$PROG_DIR/zstd-nolegacy"
+ZSTD_DICTBUILDER="$PROG_DIR/zstd-dictBuilder"
+ZSTD_FRUGAL="$PROG_DIR/zstd-frugal"
+ZSTD_NOMT="$PROG_DIR/zstd-nomt"
+
+println() {
+    printf '%b\n' "${*}"
+}
+
+die() {
+    println "$@" 1>&2
+    exit 1
+}
+
+symbol_present() {
+	(nm $1 || echo "symbol_present $@ failed") | grep $2
+}
+
+symbol_not_present() {
+	symbol_present $@ && die "Binary '$1' mistakenly contains symbol '$2'" ||:
+}
+
+compress_not_present() {
+	symbol_not_present "$1" ZSTD_compress
+}
+
+decompress_not_present() {
+	symbol_not_present "$1" ZSTD_decompress
+}
+
+dict_not_present() {
+	symbol_not_present "$1" ZDICT_
+	symbol_not_present "$1" COVER_
+}
+
+cliextra_not_present() {
+	symbol_not_present "$1" TRACE_
+	symbol_not_present "$1" BMK_
+}
+
+legacy_not_present() {
+	symbol_not_present "$1" ZSTDv0
+}
+
+test_help() {
+	"$1" --help | grep -- "$2"
+}
+
+test_no_help() {
+	test_help $@ && die "'$1' supports '$2' when it shouldn't" ||:
+}
+
+extras_not_present() {
+	dict_not_present $@
+	legacy_not_present $@
+	cliextra_not_present $@
+	test_no_help $@ "--train"
+	test_no_help $@ "-b#"
+}
+
+test_compress() {
+	echo "hello" | "$1" | "$ZSTD" -t
+}
+
+test_decompress() {
+	echo "hello" | "$ZSTD" | "$1" -t
+}
+
+test_zstd() {
+	test_compress $@
+	test_decompress $@
+}
+
+extras_not_present "$ZSTD_FRUGAL"
+extras_not_present "$ZSTD_COMPRESS"
+extras_not_present "$ZSTD_DECOMPRESS"
+
+compress_not_present "$ZSTD_DECOMPRESS"
+
+decompress_not_present "$ZSTD_COMPRESS"
+decompress_not_present "$ZSTD_DICTBUILDER"
+
+cliextra_not_present "$ZSTD_DICTBUILDER"
+
+legacy_not_present "$ZSTD_DICTBUILDER"
+legacy_not_present "$ZSTD_NOLEGACY"
+
+symbol_not_present "$ZSTD" ZSTDv01
+symbol_not_present "$ZSTD" ZSTDv02
+symbol_not_present "$ZSTD" ZSTDv03
+symbol_not_present "$ZSTD" ZSTDv04
+
+test_compress "$ZSTD_COMPRESS"
+test_decompress "$ZSTD_DECOMPRESS"
+
+test_zstd "$ZSTD_FRUGAL"
+test_zstd "$ZSTD_NOLEGACY"
+
+test_help "$ZSTD" '-b#'
+test_help "$ZSTD" --train
+test_help "$ZSTD_DICTBUILDER" --train
+
+println "Success!"
diff --git a/priv/zstd/tests/test-zstd-versions.py b/priv/zstd/tests/test-zstd-versions.py
index c86af7d..1bcf39e 100755
--- a/priv/zstd/tests/test-zstd-versions.py
+++ b/priv/zstd/tests/test-zstd-versions.py
@@ -2,7 +2,7 @@
 """Test zstd interoperability between versions"""
 
 # ################################################################
-# Copyright (c) Yann Collet, Facebook, Inc.
+# Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
 #
 # This source code is licensed under both the BSD-style license (found in the
@@ -23,13 +23,25 @@
 repo_url = 'https://github.com/facebook/zstd.git'
 tmp_dir_name = 'tests/versionsTest'
 make_cmd = 'make'
+make_args = ['-j','CFLAGS=-O0']
 git_cmd = 'git'
 test_dat_src = 'README.md'
 test_dat = 'test_dat'
 head = 'vdevel'
 dict_source = 'dict_source'
-dict_files = './zstd/programs/*.c ./zstd/lib/common/*.c ./zstd/lib/compress/*.c ./zstd/lib/decompress/*.c ./zstd/lib/dictBuilder/*.c ./zstd/lib/legacy/*.c '
-dict_files += './zstd/programs/*.h ./zstd/lib/common/*.h ./zstd/lib/compress/*.h ./zstd/lib/dictBuilder/*.h ./zstd/lib/legacy/*.h'
+dict_globs = [
+    'programs/*.c',
+    'lib/common/*.c',
+    'lib/compress/*.c',
+    'lib/decompress/*.c',
+    'lib/dictBuilder/*.c',
+    'lib/legacy/*.c',
+    'programs/*.h',
+    'lib/common/*.h',
+    'lib/compress/*.h',
+    'lib/dictBuilder/*.h',
+    'lib/legacy/*.h'
+]
 
 
 def execute(command, print_output=False, print_error=True, param_shell=False):
@@ -56,8 +68,11 @@ def proc(cmd_args, pipe=True, dummy=False):
     return subproc.communicate()
 
 
-def make(args, pipe=True):
-    return proc([make_cmd] + args, pipe)
+def make(targets, pipe=True):
+    cmd = [make_cmd] + make_args + targets
+    cmd_str = str(cmd)
+    print('compilation command : ' + cmd_str)
+    return proc(cmd, pipe)
 
 
 def git(args, pipe=True):
@@ -70,59 +85,85 @@ def get_git_tags():
     return tags
 
 
-def create_dict(tag, dict_source_path):
+def dict_ok(tag, dict_name, sample):
+    if not os.path.isfile(dict_name):
+        return False
+    try:
+        cmd = ['./zstd.' + tag, '-D', dict_name]
+        with open(sample, "rb") as i:
+            subprocess.check_call(cmd, stdin=i, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+        return True
+    except:
+        return False
+
+
+def create_dict(tag, dict_source_path, fallback_tag=None):
     dict_name = 'dict.' + tag
     if not os.path.isfile(dict_name):
         cFiles = glob.glob(dict_source_path + "/*.c")
         hFiles = glob.glob(dict_source_path + "/*.h")
+        # Ensure the dictionary builder is deterministic
+        files = sorted(cFiles + hFiles)
         if tag == 'v0.5.0':
-            result = execute('./dictBuilder.' + tag + ' ' + ' '.join(cFiles) + ' ' + ' '.join(hFiles) + ' -o ' + dict_name, print_output=False, param_shell=True)
+            result = execute('./dictBuilder.' + tag + ' ' + ' '.join(files) + ' -o ' + dict_name, print_output=False, param_shell=True)
         else:
-            result = execute('./zstd.' + tag + ' -f --train ' + ' '.join(cFiles) + ' ' + ' '.join(hFiles) + ' -o ' + dict_name, print_output=False, param_shell=True)
-        if result == 0:
+            result = execute('./zstd.' + tag + ' -f --train ' + ' '.join(files) + ' -o ' + dict_name, print_output=False, param_shell=True)
+        if result == 0 and dict_ok(tag, dict_name, files[0]):
             print(dict_name + ' created')
+        elif fallback_tag is not None:
+            fallback_dict_name = 'dict.' + fallback_tag
+            print('creating dictionary ' + dict_name + ' failed, falling back to ' + fallback_dict_name)
+            shutil.copy(fallback_dict_name, dict_name)
         else:
-            print('ERROR: creating of ' + dict_name + ' failed')
+            raise RuntimeError('ERROR: creating of ' + dict_name + ' failed')
     else:
         print(dict_name + ' already exists')
 
 
+def zstd(tag, args, input_file, output_file):
+    """
+    Zstd compress input_file to output_file.
+    Need this helper because 0.5.0 is broken when stdout is not a TTY.
+    Throws an exception if the command returns non-zero.
+    """
+    with open(input_file, "rb") as i:
+        with open(output_file, "wb") as o:
+            cmd = ['./zstd.' + tag] + args
+            print("Running: '{}', input={}, output={}" .format(
+                ' '.join(cmd), input_file, output_file
+            ))
+            result = subprocess.run(cmd, stdin=i, stdout=o, stderr=subprocess.PIPE)
+            print("Stderr: {}".format(result.stderr.decode("ascii")))
+            result.check_returncode()
+
+
 def dict_compress_sample(tag, sample):
     dict_name = 'dict.' + tag
-    DEVNULL = open(os.devnull, 'wb')
-    if subprocess.call(['./zstd.' + tag, '-D', dict_name, '-f',   sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_01_64_' + tag + '_dictio.zst')
-    if subprocess.call(['./zstd.' + tag, '-D', dict_name, '-5f',  sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_05_64_' + tag + '_dictio.zst')
-    if subprocess.call(['./zstd.' + tag, '-D', dict_name, '-9f',  sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_09_64_' + tag + '_dictio.zst')
-    if subprocess.call(['./zstd.' + tag, '-D', dict_name, '-15f', sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_15_64_' + tag + '_dictio.zst')
-    if subprocess.call(['./zstd.' + tag, '-D', dict_name, '-18f', sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_18_64_' + tag + '_dictio.zst')
+    verbose = ['-v', '-v', '-v']
+    zstd(tag, ['-D', dict_name, '-1'] + verbose, sample, sample + '_01_64_' + tag + '_dictio.zst')
+    zstd(tag, ['-D', dict_name, '-3'], sample, sample + '_03_64_' + tag + '_dictio.zst')
+    zstd(tag, ['-D', dict_name, '-5'], sample, sample + '_05_64_' + tag + '_dictio.zst')
+    zstd(tag, ['-D', dict_name, '-9'], sample, sample + '_09_64_' + tag + '_dictio.zst')
+    zstd(tag, ['-D', dict_name, '-15'], sample, sample + '_15_64_' + tag + '_dictio.zst')
+    zstd(tag, ['-D', dict_name, '-18'], sample, sample + '_18_64_' + tag + '_dictio.zst')
     # zstdFiles = glob.glob("*.zst*")
     # print(zstdFiles)
     print(tag + " : dict compression completed")
 
 
 def compress_sample(tag, sample):
-    DEVNULL = open(os.devnull, 'wb')
-    if subprocess.call(['./zstd.' + tag, '-f',   sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_01_64_' + tag + '_nodict.zst')
-    if subprocess.call(['./zstd.' + tag, '-5f',  sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_05_64_' + tag + '_nodict.zst')
-    if subprocess.call(['./zstd.' + tag, '-9f',  sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_09_64_' + tag + '_nodict.zst')
-    if subprocess.call(['./zstd.' + tag, '-15f', sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_15_64_' + tag + '_nodict.zst')
-    if subprocess.call(['./zstd.' + tag, '-18f', sample], stderr=DEVNULL) == 0:
-        os.rename(sample + '.zst', sample + '_18_64_' + tag + '_nodict.zst')
+    zstd(tag, ['-1'], sample, sample + '_01_64_' + tag + '_nodict.zst')
+    zstd(tag, ['-3'], sample, sample + '_03_64_' + tag + '_nodict.zst')
+    zstd(tag, ['-5'], sample, sample + '_05_64_' + tag + '_nodict.zst')
+    zstd(tag, ['-9'], sample, sample + '_09_64_' + tag + '_nodict.zst')
+    zstd(tag, ['-15'], sample, sample + '_15_64_' + tag + '_nodict.zst')
+    zstd(tag, ['-18'], sample, sample + '_18_64_' + tag + '_nodict.zst')
     # zstdFiles = glob.glob("*.zst*")
     # print(zstdFiles)
     print(tag + " : compression completed")
 
 
-# http://stackoverflow.com/a/19711609/2132223
+# https://stackoverflow.com/a/19711609/2132223
 def sha1_of_file(filepath):
     with open(filepath, 'rb') as f:
         return hashlib.sha1(f.read()).hexdigest()
@@ -146,23 +187,13 @@ def decompress_zst(tag):
     dec_error = 0
     list_zst = sorted(glob.glob('*_nodict.zst'))
     for file_zst in list_zst:
-        print(file_zst, end=' ')
-        print(tag, end=' ')
+        print(file_zst + ' ' + tag)
         file_dec = file_zst + '_d64_' + tag + '.dec'
-        if tag <= 'v0.5.0':
-            params = ['./zstd.' + tag, '-df', file_zst, file_dec]
-        else:
-            params = ['./zstd.' + tag, '-df', file_zst, '-o', file_dec]
-        if execute(params) == 0:
-            if not filecmp.cmp(file_dec, test_dat):
-                print('ERR !! ')
-                dec_error = 1
-            else:
-                print('OK     ')
+        zstd(tag, ['-d'], file_zst, file_dec)
+        if not filecmp.cmp(file_dec, test_dat):
+            raise RuntimeError('Decompression failed: tag={} file={}'.format(tag, file_zst))
         else:
-            print('command does not work')
-            dec_error = 1
-    return dec_error
+            print('OK     ')
 
 
 def decompress_dict(tag):
@@ -177,22 +208,13 @@ def decompress_dict(tag):
         if tag == 'v0.6.0' and dict_tag < 'v0.6.0':
             continue
         dict_name = 'dict.' + dict_tag
-        print(file_zst + ' ' + tag + ' dict=' + dict_tag, end=' ')
+        print(file_zst + ' ' + tag + ' dict=' + dict_tag)
         file_dec = file_zst + '_d64_' + tag + '.dec'
-        if tag <= 'v0.5.0':
-            params = ['./zstd.' + tag, '-D', dict_name, '-df', file_zst, file_dec]
+        zstd(tag, ['-D', dict_name, '-d'], file_zst, file_dec)
+        if not filecmp.cmp(file_dec, test_dat):
+            raise RuntimeError('Decompression failed: tag={} file={}'.format(tag, file_zst))
         else:
-            params = ['./zstd.' + tag, '-D', dict_name, '-df', file_zst, '-o', file_dec]
-        if execute(params) == 0:
-            if not filecmp.cmp(file_dec, test_dat):
-                print('ERR !! ')
-                dec_error = 1
-            else:
-                print('OK     ')
-        else:
-            print('command does not work')
-            dec_error = 1
-    return dec_error
+            print('OK     ')
 
 
 if __name__ == '__main__':
@@ -223,20 +245,28 @@ def decompress_dict(tag):
         dst_zstd = '{}/zstd.{}'.format(tmp_dir, tag)  # /path/to/zstd/tests/versionsTest/zstd.
         if not os.path.isfile(dst_zstd) or tag == head:
             if tag != head:
+                print('-----------------------------------------------')
+                print('compiling ' + tag)
+                print('-----------------------------------------------')
                 r_dir = '{}/{}'.format(tmp_dir, tag)  # /path/to/zstd/tests/versionsTest/
                 os.makedirs(r_dir, exist_ok=True)
                 os.chdir(clone_dir)
                 git(['--work-tree=' + r_dir, 'checkout', tag, '--', '.'], False)
                 if tag == 'v0.5.0':
                     os.chdir(r_dir + '/dictBuilder')  # /path/to/zstd/tests/versionsTest/v0.5.0/dictBuilder
-                    make(['clean', 'dictBuilder'], False)
+                    make(['clean'], False)   # separate 'clean' target to allow parallel build
+                    make(['dictBuilder'], False)
                     shutil.copy2('dictBuilder', '{}/dictBuilder.{}'.format(tmp_dir, tag))
                 os.chdir(r_dir + '/programs')  # /path/to/zstd/tests/versionsTest//programs
-                make(['clean', 'zstd'], False)
+                make(['clean'], False)  # separate 'clean' target to allow parallel build
+                make(['zstd'], False)
             else:
                 os.chdir(programs_dir)
+                print('-----------------------------------------------')
+                print('compiling head')
+                print('-----------------------------------------------')
                 make(['zstd'], False)
-            shutil.copy2('zstd',   dst_zstd)
+            shutil.copy2('zstd', dst_zstd)
 
     # remove any remaining *.zst and *.dec from previous test
     os.chdir(tmp_dir)
@@ -248,30 +278,31 @@ def decompress_dict(tag):
     # copy *.c and *.h to a temporary directory ("dict_source")
     if not os.path.isdir(dict_source_path):
         os.mkdir(dict_source_path)
-        print('cp ' + dict_files + ' ' + dict_source_path)
-        execute('cp ' + dict_files + ' ' + dict_source_path, param_shell=True)
-
+        for dict_glob in dict_globs:
+            files = glob.glob(dict_glob, root_dir=base_dir)
+            for file in files:
+                file = os.path.join(base_dir, file)
+                print("copying " + file + " to " + dict_source_path)
+                shutil.copy(file, dict_source_path)
+
+    print('-----------------------------------------------')
     print('Compress test.dat by all released zstd')
+    print('-----------------------------------------------')
 
-    error_code = 0
+    create_dict(head, dict_source_path)
     for tag in tags:
         print(tag)
         if tag >= 'v0.5.0':
-            create_dict(tag, dict_source_path)
+            create_dict(tag, dict_source_path, head)
             dict_compress_sample(tag, test_dat)
             remove_duplicates()
-            error_code += decompress_dict(tag)
+            decompress_dict(tag)
         compress_sample(tag, test_dat)
         remove_duplicates()
-        error_code += decompress_zst(tag)
+        decompress_zst(tag)
 
     print('')
     print('Enumerate different compressed files')
     zstds = sorted(glob.glob('*.zst'))
     for zstd in zstds:
         print(zstd + ' : ' + repr(os.path.getsize(zstd)) + ', ' + sha1_of_file(zstd))
-
-    if error_code != 0:
-        print('======  ERROR !!!  =======')
-
-    sys.exit(error_code)
diff --git a/priv/zstd/tests/zstreamtest.c b/priv/zstd/tests/zstreamtest.c
index bbef903..14c4af8 100644
--- a/priv/zstd/tests/zstreamtest.c
+++ b/priv/zstd/tests/zstreamtest.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -25,6 +25,7 @@
 #include        /* free */
 #include         /* fgets, sscanf */
 #include        /* strcmp */
+#include          /* time_t, time(), to randomize seed */
 #include        /* assert */
 #include "timefn.h"       /* UTIL_time_t, UTIL_getTime */
 #include "mem.h"
@@ -39,7 +40,7 @@
 #include "seqgen.h"
 #include "util.h"
 #include "timefn.h"       /* UTIL_time_t, UTIL_clockSpanMicro, UTIL_getTime */
-
+#include "external_matchfinder.h"   /* zstreamSequenceProducer, EMF_testCase */
 
 /*-************************************
  *  Constants
@@ -259,7 +260,7 @@ static U32 badParameters(ZSTD_CCtx* zc, ZSTD_parameters const savedParams)
     return 0;
 }
 
-static int basicUnitTests(U32 seed, double compressibility)
+static int basicUnitTests(U32 seed, double compressibility, int bigTests)
 {
     size_t const CNBufferSize = COMPRESSIBLE_NOISE_LENGTH;
     void* CNBuffer = malloc(CNBufferSize);
@@ -424,6 +425,15 @@ static int basicUnitTests(U32 seed, double compressibility)
     }   }
     DISPLAYLEVEL(3, "OK \n");
 
+    /* check decompression fails early if first bytes are wrong */
+    DISPLAYLEVEL(3, "test%3i : early decompression error if first bytes are incorrect : ", testNb++);
+    {   const char buf[3] = { 0 };  /* too short, not enough to start decoding header */
+        ZSTD_inBuffer inb = { buf, sizeof(buf), 0 };
+        size_t const remaining = ZSTD_decompressStream(zd, &outBuff, &inb);
+        if (!ZSTD_isError(remaining)) goto _output_error; /* should have errored out immediately (note: this does not test the exact error code) */
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
     /* context size functions */
     DISPLAYLEVEL(3, "test%3i : estimate DStream size : ", testNb++);
     {   ZSTD_frameHeader fhi;
@@ -513,7 +523,7 @@ static int basicUnitTests(U32 seed, double compressibility)
     }
     DISPLAYLEVEL(3, "OK \n");
 
-    DISPLAYLEVEL(3, "test%3i : NULL buffers : ", testNb++);
+    DISPLAYLEVEL(3, "test%3i : NULL output and NULL input : ", testNb++);
     inBuff.src = NULL;
     inBuff.size = 0;
     inBuff.pos = 0;
@@ -539,6 +549,36 @@ static int basicUnitTests(U32 seed, double compressibility)
     {   size_t const ret = ZSTD_decompressStream(zd, &outBuff, &inBuff);
         if (ret != 0) goto _output_error;
     }
+    DISPLAYLEVEL(3, "OK\n");
+
+    DISPLAYLEVEL(3, "test%3i : NULL output buffer with non-NULL input : ", testNb++);
+    {
+        const char* test = "aa";
+        inBuff.src = test;
+        inBuff.size = 2;
+        inBuff.pos = 0;
+        outBuff.dst = NULL;
+        outBuff.size = 0;
+        outBuff.pos = 0;
+        CHECK_Z( ZSTD_compressStream(zc, &outBuff, &inBuff) );
+        CHECK(inBuff.pos != inBuff.size, "Entire input should be consumed");
+        CHECK_Z( ZSTD_endStream(zc, &outBuff) );
+        outBuff.dst = (char*)(compressedBuffer);
+        outBuff.size = compressedBufferSize;
+        outBuff.pos = 0;
+        {   size_t const r = ZSTD_endStream(zc, &outBuff);
+            CHECK(r != 0, "Error or some data not flushed (ret=%zu)", r);
+        }
+        inBuff.src = outBuff.dst;
+        inBuff.size = outBuff.pos;
+        inBuff.pos = 0;
+        outBuff.dst = NULL;
+        outBuff.size = 0;
+        outBuff.pos = 0;
+        CHECK_Z( ZSTD_initDStream(zd) );
+        CHECK_Z(ZSTD_decompressStream(zd, &outBuff, &inBuff));
+    }
+
     DISPLAYLEVEL(3, "OK\n");
     /* _srcSize compression test */
     DISPLAYLEVEL(3, "test%3i : compress_srcSize %u bytes : ", testNb++, COMPRESSIBLE_NOISE_LENGTH);
@@ -613,7 +653,7 @@ static int basicUnitTests(U32 seed, double compressibility)
             DISPLAYLEVEL(3, "OK (error detected : %s) \n", ZSTD_getErrorName(r));
     }   }
 
-    /* Complex context re-use scenario */
+    /* Compression state re-use scenario */
     DISPLAYLEVEL(3, "test%3i : context re-use : ", testNb++);
     ZSTD_freeCStream(zc);
     zc = ZSTD_createCStream();
@@ -634,8 +674,7 @@ static int basicUnitTests(U32 seed, double compressibility)
         CHECK_Z( ZSTD_compressStream(zc, &outBuff, &inBuff) );
         if (inBuff.pos != inBuff.size) goto _output_error;   /* entire input should be consumed */
         DISPLAYLEVEL(5, "end1 ");
-        { size_t const r = ZSTD_endStream(zc, &outBuff);
-            if (r != 0) goto _output_error; }  /* error, or some data not flushed */
+        if (ZSTD_endStream(zc, &outBuff) != 0) goto _output_error;  /* error, or some data not flushed */
     }
     /* use 2 */
     {   size_t const inSize = 1025;   /* will not continue, because tables auto-adjust and are therefore different size */
@@ -653,8 +692,7 @@ static int basicUnitTests(U32 seed, double compressibility)
         CHECK_Z( ZSTD_compressStream(zc, &outBuff, &inBuff) );
         if (inBuff.pos != inBuff.size) goto _output_error;   /* entire input should be consumed */
         DISPLAYLEVEL(5, "end2 ");
-        { size_t const r = ZSTD_endStream(zc, &outBuff);
-            if (r != 0) goto _output_error; }  /* error, or some data not flushed */
+        if (ZSTD_endStream(zc, &outBuff) != 0) goto _output_error;   /* error, or some data not flushed */
     }
     DISPLAYLEVEL(3, "OK \n");
 
@@ -771,11 +809,12 @@ static int basicUnitTests(U32 seed, double compressibility)
     }
 
     /* Compression with ZSTD_c_stable{In,Out}Buffer */
-    {   ZSTD_CCtx* cctx = ZSTD_createCCtx();
+    {   ZSTD_CCtx* const cctx = ZSTD_createCCtx();
         ZSTD_inBuffer in;
         ZSTD_outBuffer out;
         size_t cctxSize1;
         size_t cctxSize2;
+        assert(cctx != NULL);
         in.src = CNBuffer;
         in.size = CNBufferSize;
         out.dst = compressedBuffer;
@@ -786,30 +825,33 @@ static int basicUnitTests(U32 seed, double compressibility)
         CHECK(!(cSize < ZSTD_compressBound(CNBufferSize)), "cSize too large for test");
         CHECK_Z(cSize = ZSTD_compress2(cctx, compressedBuffer, cSize + 4, CNBuffer, CNBufferSize));
         CHECK_Z(cctxSize1 = ZSTD_sizeof_CCtx(cctx));
-        {   ZSTD_CCtx* cctx2 = ZSTD_createCCtx();
+        /* @cctxSize2 : sizeof_CCtx when doing full streaming (no stable in/out) */
+        {   ZSTD_CCtx* const cctx2 = ZSTD_createCCtx();
+            assert(cctx2 != NULL);
             in.pos = out.pos = 0;
             CHECK_Z(ZSTD_compressStream2(cctx2, &out, &in, ZSTD_e_continue));
             CHECK(!(ZSTD_compressStream2(cctx2, &out, &in, ZSTD_e_end) == 0), "Not finished");
             CHECK_Z(cctxSize2 = ZSTD_sizeof_CCtx(cctx2));
             ZSTD_freeCCtx(cctx2);
         }
-        {   ZSTD_CCtx* cctx3 = ZSTD_createCCtx();
+        /* @cctxSize1 : sizeof_CCtx when doing single-shot compression (no streaming) */
+        {   ZSTD_CCtx* const cctx1 = ZSTD_createCCtx();
             ZSTD_parameters params = ZSTD_getParams(0, CNBufferSize, 0);
             size_t cSize3;
+            assert(cctx1 != NULL);
             params.fParams.checksumFlag = 1;
-            cSize3 = ZSTD_compress_advanced(cctx3, compressedBuffer, compressedBufferSize, CNBuffer, CNBufferSize, NULL, 0, params);
+            cSize3 = ZSTD_compress_advanced(cctx1, compressedBuffer, compressedBufferSize, CNBuffer, CNBufferSize, NULL, 0, params);
             CHECK_Z(cSize3);
             CHECK(!(cSize == cSize3), "Must be same compressed size");
-            CHECK(!(cctxSize1 == ZSTD_sizeof_CCtx(cctx3)), "Must be same CCtx size");
-            ZSTD_freeCCtx(cctx3);
+            CHECK(!(cctxSize1 == ZSTD_sizeof_CCtx(cctx1)), "Must be same CCtx size");
+            ZSTD_freeCCtx(cctx1);
         }
         CHECK(!(cctxSize1 < cctxSize2), "Stable buffers means less allocated size");
         CHECK_Z(ZSTD_decompress(decodedBuffer, CNBufferSize, compressedBuffer, cSize));
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : ZSTD_compress2() doesn't modify user parameters : ", testNb++);
-        {
-            int stableInBuffer;
+        {   int stableInBuffer;
             int stableOutBuffer;
             CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_c_stableInBuffer, &stableInBuffer));
             CHECK_Z(ZSTD_CCtx_getParameter(cctx, ZSTD_c_stableOutBuffer, &stableOutBuffer));
@@ -866,30 +908,73 @@ static int basicUnitTests(U32 seed, double compressibility)
         in.pos = 0;
         {   size_t const ret = ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_end);
             CHECK(!ZSTD_isError(ret), "Must error");
-            CHECK(!(ZSTD_getErrorCode(ret) == ZSTD_error_srcBuffer_wrong), "Must be this error");
+            CHECK(!(ZSTD_getErrorCode(ret) == ZSTD_error_stabilityCondition_notRespected), "Must be this error");
         }
         DISPLAYLEVEL(3, "OK \n");
 
-        DISPLAYLEVEL(3, "test%3i : ZSTD_compressStream2() ZSTD_c_stableInBuffer with continue and flush : ", testNb++);
-        in.src = CNBuffer;
-        in.size = CNBufferSize;
-        in.pos = 0;
-        out.pos = 0;
-        out.size = compressedBufferSize;
-        CHECK_Z(ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only));
-        {   size_t const ret = ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_continue);
-            CHECK(!ZSTD_isError(ret), "Must error");
-            CHECK(!(ZSTD_getErrorCode(ret) == ZSTD_error_srcBuffer_wrong), "Must be this error");
-        }
-        CHECK_Z(ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only));
-        {   size_t const ret = ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_flush);
-            CHECK(!ZSTD_isError(ret), "Must error");
-            CHECK(!(ZSTD_getErrorCode(ret) == ZSTD_error_srcBuffer_wrong), "Must be this error");
-        }
+        /* stableSrc + streaming */
+        DISPLAYLEVEL(3, "test%3i : ZSTD_c_stableInBuffer compatibility with compressStream, flushStream and endStream : ", testNb++);
+        CHECK_Z( ZSTD_initCStream(cctx, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_stableInBuffer, 1) );
+        {   ZSTD_inBuffer inBuf;
+            ZSTD_outBuffer outBuf;
+            const size_t nonZeroStartPos = 18;
+            const size_t inputSize = 500;
+            inBuf.src = CNBuffer;
+            inBuf.size = 100;
+            inBuf.pos = nonZeroStartPos;
+            outBuf.dst = (char*)(compressedBuffer)+cSize;
+            outBuf.size = ZSTD_compressBound(inputSize);
+            outBuf.pos = 0;
+            CHECK_Z( ZSTD_compressStream(cctx, &outBuf, &inBuf) );
+            inBuf.size = 200;
+            CHECK_Z( ZSTD_compressStream(cctx, &outBuf, &inBuf) );
+            CHECK_Z( ZSTD_flushStream(cctx, &outBuf) );
+            inBuf.size = nonZeroStartPos + inputSize;
+            CHECK_Z( ZSTD_compressStream(cctx, &outBuf, &inBuf) );
+            CHECK(ZSTD_endStream(cctx, &outBuf) != 0, "compression should be successful and fully flushed");
+            {   const void* const realSrcStart = (const char*)inBuf.src + nonZeroStartPos;
+                void* const verifBuf = (char*)outBuf.dst + outBuf.pos;
+                const size_t decSize = ZSTD_decompress(verifBuf, inputSize, outBuf.dst, outBuf.pos);
+                CHECK_Z(decSize);
+                CHECK(decSize != inputSize, "regenerated %zu bytes, instead of %zu", decSize, inputSize);
+                CHECK(memcmp(realSrcStart, verifBuf, inputSize) != 0, "regenerated data different from original");
+        }   }
         DISPLAYLEVEL(3, "OK \n");
 
-        DISPLAYLEVEL(3, "test%3i : ZSTD_compressStream2() ZSTD_c_stableInBuffer allocated size : ", testNb++);
+        /* stableSrc + streaming */
+        DISPLAYLEVEL(3, "test%3i : ZSTD_c_stableInBuffer compatibility with compressStream2, using different end directives : ", testNb++);
+        CHECK_Z( ZSTD_initCStream(cctx, 1) );
+        CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_c_stableInBuffer, 1) );
+        {   ZSTD_inBuffer inBuf;
+            ZSTD_outBuffer outBuf;
+            const size_t nonZeroStartPos = 18;
+            const size_t inputSize = 500;
+            inBuf.src = CNBuffer;
+            inBuf.size = 100;
+            inBuf.pos = nonZeroStartPos;
+            outBuf.dst = (char*)(compressedBuffer)+cSize;
+            outBuf.size = ZSTD_compressBound(inputSize);
+            outBuf.pos = 0;
+            CHECK_Z( ZSTD_compressStream2(cctx, &outBuf, &inBuf, ZSTD_e_continue) );
+            inBuf.size = 200;
+            CHECK_Z( ZSTD_compressStream2(cctx, &outBuf, &inBuf, ZSTD_e_continue) );
+            CHECK_Z( ZSTD_compressStream2(cctx, &outBuf, &inBuf, ZSTD_e_flush) );
+            inBuf.size = nonZeroStartPos + inputSize;
+            CHECK_Z( ZSTD_compressStream2(cctx, &outBuf, &inBuf, ZSTD_e_continue) );
+            CHECK( ZSTD_compressStream2(cctx, &outBuf, &inBuf, ZSTD_e_end) != 0, "compression should be successful and fully flushed");
+            {   const void* const realSrcStart = (const char*)inBuf.src + nonZeroStartPos;
+                void* const verifBuf = (char*)outBuf.dst + outBuf.pos;
+                const size_t decSize = ZSTD_decompress(verifBuf, inputSize, outBuf.dst, outBuf.pos);
+                CHECK_Z(decSize);
+                CHECK(decSize != inputSize, "regenerated %zu bytes, instead of %zu", decSize, inputSize);
+                CHECK(memcmp(realSrcStart, verifBuf, inputSize) != 0, "regenerated data different from original");
+        }   }
+        DISPLAYLEVEL(3, "OK \n");
+
+        DISPLAYLEVEL(3, "test%3i : ZSTD_compressStream2() with ZSTD_c_stableInBuffer: context size : ", testNb++);
         {   size_t const cctxSize = ZSTD_sizeof_CCtx(cctx);
+            DISPLAYLEVEL(4, "cctxSize1=%zu; cctxSize=%zu; cctxSize2=%zu : ", cctxSize1, cctxSize, cctxSize2);
             CHECK(!(cctxSize1 < cctxSize), "Must be bigger than single-pass");
             CHECK(!(cctxSize < cctxSize2), "Must be smaller than streaming");
             cctxSize1 = cctxSize;
@@ -900,13 +985,15 @@ static int basicUnitTests(U32 seed, double compressibility)
         CHECK_Z(ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters));
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_checksumFlag, 1));
         CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_stableOutBuffer, 1));
+        in.src = CNBuffer;
         in.pos = out.pos = 0;
         in.size = MIN(CNBufferSize, 10);
+        out.size = compressedBufferSize;
         CHECK_Z(ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_flush));
         in.pos = 0;
         in.size = CNBufferSize - in.size;
         CHECK(!(ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_end) == 0), "Not finished");
-        CHECK_Z(ZSTD_decompress(decodedBuffer, CNBufferSize, compressedBuffer, cSize));
+        CHECK_Z(ZSTD_decompress(decodedBuffer, CNBufferSize, compressedBuffer, out.pos));
         DISPLAYLEVEL(3, "OK \n");
 
         DISPLAYLEVEL(3, "test%3i : ZSTD_compressStream2() ZSTD_c_stableOutBuffer modify buffer : ", testNb++);
@@ -916,12 +1003,13 @@ static int basicUnitTests(U32 seed, double compressibility)
         in.pos = out.pos = 0;
         {   size_t const ret = ZSTD_compressStream2(cctx, &out, &in, ZSTD_e_continue);
             CHECK(!ZSTD_isError(ret), "Must have errored");
-            CHECK(!(ZSTD_getErrorCode(ret) == ZSTD_error_dstBuffer_wrong), "Must be this error");
+            CHECK(!(ZSTD_getErrorCode(ret) == ZSTD_error_stabilityCondition_notRespected), "Must be this error");
         }
         DISPLAYLEVEL(3, "OK \n");
 
-        DISPLAYLEVEL(3, "test%3i : ZSTD_compressStream2() ZSTD_c_stableOutBuffer allocated size : ", testNb++);
+        DISPLAYLEVEL(3, "test%3i : ZSTD_compressStream2() with ZSTD_c_stableOutBuffer: context size : ", testNb++);
         {   size_t const cctxSize = ZSTD_sizeof_CCtx(cctx);
+            DISPLAYLEVEL(4, "cctxSize1=%zu; cctxSize=%zu; cctxSize2=%zu : ", cctxSize1, cctxSize, cctxSize2);
             CHECK(!(cctxSize1 < cctxSize), "Must be bigger than single-pass and stableInBuffer");
             CHECK(!(cctxSize < cctxSize2), "Must be smaller than streaming");
         }
@@ -1063,7 +1151,7 @@ static int basicUnitTests(U32 seed, double compressibility)
             if (ZSTD_decompressStream(dctx, &out, &in) != 0) goto _output_error;
             if (in.pos != in.size) goto _output_error;
         }
-        /* The dictionary should presist across calls. */
+        /* The dictionary should persist across calls. */
         {   ZSTD_outBuffer out = {decodedBuffer, decodedBufferSize, 0};
             ZSTD_inBuffer in = {compressedBuffer, cSize, 0};
             if (ZSTD_decompressStream(dctx, &out, &in) != 0) goto _output_error;
@@ -1128,7 +1216,7 @@ static int basicUnitTests(U32 seed, double compressibility)
             if (ZSTD_decompressStream(dctx, &out, &in) != 0) goto _output_error;
             if (in.pos != in.size) goto _output_error;
         }
-        /* The ddict should presist across calls. */
+        /* The ddict should persist across calls. */
         {   ZSTD_outBuffer out = {decodedBuffer, decodedBufferSize, 0};
             ZSTD_inBuffer in = {compressedBuffer, cSize, 0};
             if (ZSTD_decompressStream(dctx, &out, &in) != 0) goto _output_error;
@@ -1175,12 +1263,12 @@ static int basicUnitTests(U32 seed, double compressibility)
         /* We should succeed to decompress with the dictionary. */
         CHECK_Z( ZSTD_initDStream_usingDict(dctx, dictionary.start, dictionary.filled) );
         CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, decodedBufferSize, compressedBuffer, cSize) );
-        /* The dictionary should presist across calls. */
+        /* The dictionary should persist across calls. */
         CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, decodedBufferSize, compressedBuffer, cSize) );
         /* We should succeed to decompress with the ddict. */
         CHECK_Z( ZSTD_initDStream_usingDDict(dctx, ddict) );
         CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, decodedBufferSize, compressedBuffer, cSize) );
-        /* The ddict should presist across calls. */
+        /* The ddict should persist across calls. */
         CHECK_Z( ZSTD_decompressDCtx(dctx, decodedBuffer, decodedBufferSize, compressedBuffer, cSize) );
         /* When we reset the context the ddict is cleared. */
         CHECK_Z( ZSTD_initDStream(dctx) );
@@ -1478,6 +1566,84 @@ static int basicUnitTests(U32 seed, double compressibility)
     CHECK(!ZSTD_isError(ZSTD_CCtx_setParameter(zc, ZSTD_c_srcSizeHint, -1)), "Out of range doesn't error");
     DISPLAYLEVEL(3, "OK \n");
 
+    DISPLAYLEVEL(3, "test%3i : ZSTD_lazy compress with hashLog = 29 and searchLog = 4 : ", testNb++);
+    if (MEM_64bits()) {
+        ZSTD_outBuffer out = { compressedBuffer, compressedBufferSize, 0 };
+        ZSTD_inBuffer in = { CNBuffer, CNBufferSize, 0 };
+        CHECK_Z(ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters));
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_strategy, ZSTD_lazy));
+        /* Force enable the row based match finder */
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_useRowMatchFinder, ZSTD_ps_enable));
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_searchLog, 4));
+        /* Set windowLog to 29 so the hashLog doesn't get sized down */
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_windowLog, 29));
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_hashLog, 29));
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_checksumFlag, 1));
+        /* Compress with continue first so the hashLog doesn't get sized down */
+        CHECK_Z(ZSTD_compressStream2(zc, &out, &in, ZSTD_e_continue));
+        CHECK_Z(ZSTD_compressStream2(zc, &out, &in, ZSTD_e_end));
+        cSize = out.pos;
+        CHECK_Z(ZSTD_decompress(decodedBuffer, CNBufferSize, compressedBuffer, cSize));
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+    DISPLAYLEVEL(3, "test%3i : Test offset == windowSize : ", testNb++);
+    {
+        int windowLog;
+        int const kMaxWindowLog = bigTests ? 29 : 26;
+        size_t const kNbSequences = 10000;
+        size_t const kMaxSrcSize = (1u << kMaxWindowLog) + 10 * kNbSequences;
+        char* src = calloc(kMaxSrcSize, 1);
+        ZSTD_Sequence* sequences = malloc(sizeof(ZSTD_Sequence) * kNbSequences);
+        for (windowLog = ZSTD_WINDOWLOG_MIN; windowLog <= kMaxWindowLog; ++windowLog) {
+            size_t const srcSize = ((size_t)1 << windowLog) + 10 * (kNbSequences - 1);
+
+            sequences[0].offset = 32;
+            sequences[0].litLength = 32;
+            sequences[0].matchLength = (1u << windowLog) - 32;
+            sequences[0].rep = 0;
+            {
+                size_t i;
+                for (i = 1; i < kNbSequences; ++i) {
+                    sequences[i].offset = (1u << windowLog) - (FUZ_rand(&seed) % 8);
+                    sequences[i].litLength = FUZ_rand(&seed) & 7;
+                    sequences[i].matchLength = 10 - sequences[i].litLength;
+                    sequences[i].rep = 0;
+                }
+            }
+
+            CHECK_Z(ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters));
+            CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_checksumFlag, 1));
+            CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_minMatch, 3));
+            CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_validateSequences, 1));
+            CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_windowLog, windowLog));
+            assert(srcSize <= kMaxSrcSize);
+            cSize = ZSTD_compressSequences(zc, compressedBuffer, compressedBufferSize, sequences, kNbSequences, src, srcSize);
+            CHECK_Z(cSize);
+            CHECK_Z(ZSTD_DCtx_reset(zd, ZSTD_reset_session_and_parameters));
+            CHECK_Z(ZSTD_DCtx_setParameter(zd, ZSTD_d_windowLogMax, windowLog))
+            {
+                ZSTD_inBuffer in = {compressedBuffer, cSize, 0};
+                size_t decompressedBytes = 0;
+                for (;;) {
+                    ZSTD_outBuffer out = {decodedBuffer, decodedBufferSize, 0};
+                    size_t const ret = ZSTD_decompressStream(zd, &out, &in);
+                    CHECK_Z(ret);
+                    CHECK(decompressedBytes + out.pos > srcSize, "Output too large");
+                    CHECK(memcmp(out.dst, src + decompressedBytes, out.pos), "Corrupted");
+                    decompressedBytes += out.pos;
+                    if (ret == 0) {
+                        break;
+                    }
+                }
+                CHECK(decompressedBytes != srcSize, "Output wrong size");
+            }
+        }
+        free(sequences);
+        free(src);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
     /* Overlen overwriting window data bug */
     DISPLAYLEVEL(3, "test%3i : wildcopy doesn't overwrite potential match data : ", testNb++);
     {   /* This test has a window size of 1024 bytes and consists of 3 blocks:
@@ -1690,6 +1856,445 @@ static int basicUnitTests(U32 seed, double compressibility)
     }
     DISPLAYLEVEL(3, "OK \n");
 
+    DISPLAYLEVEL(3, "test%3i : Block-Level External Sequence Producer API: ", testNb++);
+    {
+        size_t const dstBufSize = ZSTD_compressBound(CNBufferSize);
+        BYTE* const dstBuf = (BYTE*)malloc(ZSTD_compressBound(dstBufSize));
+        size_t const checkBufSize = CNBufferSize;
+        BYTE* const checkBuf = (BYTE*)malloc(checkBufSize);
+        int enableFallback;
+        EMF_testCase sequenceProducerState;
+
+        CHECK(dstBuf == NULL || checkBuf == NULL, "allocation failed");
+
+        CHECK_Z(ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters));
+
+        /* Reference external matchfinder outside the test loop to
+         * check that the reference is preserved across compressions */
+        ZSTD_registerSequenceProducer(zc, &sequenceProducerState, zstreamSequenceProducer);
+
+        for (enableFallback = 0; enableFallback <= 1; enableFallback++) {
+            size_t testCaseId;
+            size_t const numTestCases = 9;
+
+            EMF_testCase const testCases[] = {
+                EMF_ONE_BIG_SEQ,
+                EMF_LOTS_OF_SEQS,
+                EMF_ZERO_SEQS,
+                EMF_BIG_ERROR,
+                EMF_SMALL_ERROR,
+                EMF_INVALID_OFFSET,
+                EMF_INVALID_MATCHLEN,
+                EMF_INVALID_LITLEN,
+                EMF_INVALID_LAST_LITS
+            };
+
+            ZSTD_ErrorCode const errorCodes[] = {
+                ZSTD_error_no_error,
+                ZSTD_error_no_error,
+                ZSTD_error_sequenceProducer_failed,
+                ZSTD_error_sequenceProducer_failed,
+                ZSTD_error_sequenceProducer_failed,
+                ZSTD_error_externalSequences_invalid,
+                ZSTD_error_externalSequences_invalid,
+                ZSTD_error_externalSequences_invalid,
+                ZSTD_error_externalSequences_invalid
+            };
+
+            for (testCaseId = 0; testCaseId < numTestCases; testCaseId++) {
+                size_t res;
+
+                int const compressionShouldSucceed = (
+                    (errorCodes[testCaseId] == ZSTD_error_no_error) ||
+                    (enableFallback && errorCodes[testCaseId] == ZSTD_error_sequenceProducer_failed)
+                );
+
+                int const testWithSequenceValidation = (
+                    testCases[testCaseId] == EMF_INVALID_OFFSET
+                );
+
+                sequenceProducerState = testCases[testCaseId];
+
+                ZSTD_CCtx_reset(zc, ZSTD_reset_session_only);
+                CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_validateSequences, testWithSequenceValidation));
+                CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_enableSeqProducerFallback, enableFallback));
+                res = ZSTD_compress2(zc, dstBuf, dstBufSize, CNBuffer, CNBufferSize);
+
+                if (compressionShouldSucceed) {
+                    CHECK(ZSTD_isError(res), "EMF: Compression error: %s", ZSTD_getErrorName(res));
+                    CHECK_Z(ZSTD_decompress(checkBuf, checkBufSize, dstBuf, res));
+                    CHECK(memcmp(CNBuffer, checkBuf, CNBufferSize) != 0, "EMF: Corruption!");
+                } else {
+                    CHECK(!ZSTD_isError(res), "EMF: Should have raised an error!");
+                    CHECK(
+                        ZSTD_getErrorCode(res) != errorCodes[testCaseId],
+                        "EMF: Wrong error code: %s", ZSTD_getErrorName(res)
+                    );
+                }
+            }
+
+            /* Test compression with external matchfinder + empty src buffer */
+            {
+                size_t res;
+                sequenceProducerState = EMF_ZERO_SEQS;
+                ZSTD_CCtx_reset(zc, ZSTD_reset_session_only);
+                CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_enableSeqProducerFallback, enableFallback));
+                res = ZSTD_compress2(zc, dstBuf, dstBufSize, CNBuffer, 0);
+                CHECK(ZSTD_isError(res), "EMF: Compression error: %s", ZSTD_getErrorName(res));
+                CHECK(ZSTD_decompress(checkBuf, checkBufSize, dstBuf, res) != 0, "EMF: Empty src round trip failed!");
+            }
+        }
+
+        /* Test that reset clears the external matchfinder */
+        CHECK_Z(ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters));
+        sequenceProducerState = EMF_BIG_ERROR; /* ensure zstd will fail if the matchfinder wasn't cleared */
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_enableSeqProducerFallback, 0));
+        CHECK_Z(ZSTD_compress2(zc, dstBuf, dstBufSize, CNBuffer, CNBufferSize));
+
+        /* Test that registering mFinder == NULL clears the external matchfinder */
+        ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters);
+        ZSTD_registerSequenceProducer(zc, &sequenceProducerState, zstreamSequenceProducer);
+        sequenceProducerState = EMF_BIG_ERROR; /* ensure zstd will fail if the matchfinder wasn't cleared */
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_enableSeqProducerFallback, 0));
+        ZSTD_registerSequenceProducer(zc, NULL, NULL); /* clear the external matchfinder */
+        CHECK_Z(ZSTD_compress2(zc, dstBuf, dstBufSize, CNBuffer, CNBufferSize));
+
+        /* Test that external matchfinder doesn't interact with older APIs */
+        ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters);
+        ZSTD_registerSequenceProducer(zc, &sequenceProducerState, zstreamSequenceProducer);
+        sequenceProducerState = EMF_BIG_ERROR; /* ensure zstd will fail if the matchfinder is used */
+        CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_enableSeqProducerFallback, 0));
+        CHECK_Z(ZSTD_compressCCtx(zc, dstBuf, dstBufSize, CNBuffer, CNBufferSize, 3));
+
+        /* Test that compression returns the correct error with LDM */
+        CHECK_Z(ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters));
+        {
+            size_t res;
+            ZSTD_registerSequenceProducer(zc, &sequenceProducerState, zstreamSequenceProducer);
+            CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_enableLongDistanceMatching, ZSTD_ps_enable));
+            res = ZSTD_compress2(zc, dstBuf, dstBufSize, CNBuffer, CNBufferSize);
+            CHECK(!ZSTD_isError(res), "EMF: Should have raised an error!");
+            CHECK(
+                ZSTD_getErrorCode(res) != ZSTD_error_parameter_combination_unsupported,
+                "EMF: Wrong error code: %s", ZSTD_getErrorName(res)
+            );
+        }
+
+#ifdef ZSTD_MULTITHREAD
+        /* Test that compression returns the correct error with nbWorkers > 0 */
+        CHECK_Z(ZSTD_CCtx_reset(zc, ZSTD_reset_session_and_parameters));
+        {
+            size_t res;
+            ZSTD_registerSequenceProducer(zc, &sequenceProducerState, zstreamSequenceProducer);
+            CHECK_Z(ZSTD_CCtx_setParameter(zc, ZSTD_c_nbWorkers, 1));
+            res = ZSTD_compress2(zc, dstBuf, dstBufSize, CNBuffer, CNBufferSize);
+            CHECK(!ZSTD_isError(res), "EMF: Should have raised an error!");
+            CHECK(
+                ZSTD_getErrorCode(res) != ZSTD_error_parameter_combination_unsupported,
+                "EMF: Wrong error code: %s", ZSTD_getErrorName(res)
+            );
+        }
+#endif
+
+        free(dstBuf);
+        free(checkBuf);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+
+    /* Test maxBlockSize cctx param functionality */
+    DISPLAYLEVEL(3, "test%3i : Testing maxBlockSize PR#3418: ", testNb++);
+    {
+        ZSTD_CCtx* cctx = ZSTD_createCCtx();
+
+        /* Quick test to make sure maxBlockSize bounds are enforced */
+        assert(ZSTD_isError(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, ZSTD_BLOCKSIZE_MAX_MIN - 1)));
+        assert(ZSTD_isError(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, ZSTD_BLOCKSIZE_MAX + 1)));
+
+        /* Test maxBlockSize < windowSize and windowSize < maxBlockSize*/
+        {
+            size_t srcSize = 2 << 10;
+            void* const src = CNBuffer;
+            size_t dstSize = ZSTD_compressBound(srcSize);
+            void* const dst1 = compressedBuffer;
+            void* const dst2 = (BYTE*)compressedBuffer + dstSize;
+            size_t size1, size2;
+            void* const checkBuf = malloc(srcSize);
+            memset(src, 'x', srcSize);
+
+            /* maxBlockSize = 1KB */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, 1u << 10));
+            size1 = ZSTD_compress2(cctx, dst1, dstSize, src, srcSize);
+
+            if (ZSTD_isError(size1)) goto _output_error;
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst1, size1));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            /* maxBlockSize = 3KB */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, 3u << 10));
+            size2 = ZSTD_compress2(cctx, dst2, dstSize, src, srcSize);
+
+            if (ZSTD_isError(size2)) goto _output_error;
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst2, size2));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            assert(size1 - size2 == 4); /* We add another RLE block with header + character */
+            assert(memcmp(dst1, dst2, size2) != 0); /* Compressed output should not be equal */
+
+            /* maxBlockSize = 1KB, windowLog = 10 */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, 1u << 10));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, 10));
+            size1 = ZSTD_compress2(cctx, dst1, dstSize, src, srcSize);
+
+            if (ZSTD_isError(size1)) goto _output_error;
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst1, size1));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            /* maxBlockSize = 3KB, windowLog = 10 */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, 3u << 10));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_windowLog, 10));
+            size2 = ZSTD_compress2(cctx, dst2, dstSize, src, srcSize);
+
+            if (ZSTD_isError(size2)) goto _output_error;
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst2, size2));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            assert(size1 == size2);
+            assert(memcmp(dst1, dst2, size1) == 0); /* Compressed output should be equal */
+
+            free(checkBuf);
+        }
+
+        ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+
+        /* Test maxBlockSize = 0 is valid */
+        {   size_t srcSize = 256 << 10;
+            void* const src = CNBuffer;
+            size_t dstSize = ZSTD_compressBound(srcSize);
+            void* const dst1 = compressedBuffer;
+            void* const dst2 = (BYTE*)compressedBuffer + dstSize;
+            size_t size1, size2;
+            void* const checkBuf = malloc(srcSize);
+
+            /* maxBlockSize = 0 */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, 0));
+            size1 = ZSTD_compress2(cctx, dst1, dstSize, src, srcSize);
+
+            if (ZSTD_isError(size1)) goto _output_error;
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst1, size1));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            /* maxBlockSize = ZSTD_BLOCKSIZE_MAX */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_maxBlockSize, ZSTD_BLOCKSIZE_MAX));
+            size2 = ZSTD_compress2(cctx, dst2, dstSize, src, srcSize);
+
+            if (ZSTD_isError(size2)) goto _output_error;
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst2, size2));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            assert(size1 == size2);
+            assert(memcmp(dst1, dst2, size1) == 0); /* Compressed output should be equal */
+            free(checkBuf);
+        }
+        ZSTD_freeCCtx(cctx);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+    /* Test Sequence Validation */
+    DISPLAYLEVEL(3, "test%3i : Testing sequence validation: ", testNb++);
+    {
+        ZSTD_CCtx* cctx = ZSTD_createCCtx();
+
+        /* Test minMatch >= 4, matchLength < 4 */
+        {
+            size_t srcSize = 11;
+            void* const src = CNBuffer;
+            size_t dstSize = ZSTD_compressBound(srcSize);
+            void* const dst = compressedBuffer;
+            size_t const kNbSequences = 4;
+            ZSTD_Sequence* sequences = malloc(sizeof(ZSTD_Sequence) * kNbSequences);
+
+            memset(src, 'x', srcSize);
+
+            sequences[0] = (ZSTD_Sequence) {1, 1, 3, 0};
+            sequences[1] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[2] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[3] = (ZSTD_Sequence) {0, 1, 0, 0};
+
+            /* Test with sequence validation */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, 5));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_explicitBlockDelimiters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 1));
+
+            cSize = ZSTD_compressSequences(cctx, dst, dstSize,
+                                   sequences, kNbSequences,
+                                   src, srcSize);
+
+            CHECK(!ZSTD_isError(cSize), "Should throw an error"); /* maxNbSeq is too small and an assert will fail */
+            CHECK(ZSTD_getErrorCode(cSize) != ZSTD_error_externalSequences_invalid, "Wrong error code: %s", ZSTD_getErrorName(cSize)); /* fails sequence validation */
+
+            ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+
+            /* Test without sequence validation */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, 5));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_explicitBlockDelimiters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 0));
+
+            cSize = ZSTD_compressSequences(cctx, dst, dstSize,
+                                   sequences, kNbSequences,
+                                   src, srcSize);
+
+            CHECK(!ZSTD_isError(cSize), "Should throw an error"); /* maxNbSeq is too small and an assert will fail */
+            CHECK(ZSTD_getErrorCode(cSize) != ZSTD_error_externalSequences_invalid, "Wrong error code: %s", ZSTD_getErrorName(cSize)); /* fails sequence validation */
+
+            free(sequences);
+        }
+
+        ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+
+
+        /* Test with no block delim */
+        {
+            size_t srcSize = 4;
+            void* const src = CNBuffer;
+            size_t dstSize = ZSTD_compressBound(srcSize);
+            void* const dst = compressedBuffer;
+            size_t const kNbSequences = 1;
+            ZSTD_Sequence* sequences = malloc(sizeof(ZSTD_Sequence) * kNbSequences);
+            void* const checkBuf = malloc(srcSize);
+
+            memset(src, 'x', srcSize);
+
+            sequences[0] = (ZSTD_Sequence) {1, 1, 3, 0};
+
+            /* Test with sequence validation */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, 3));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_noBlockDelimiters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 1));
+
+            cSize = ZSTD_compressSequences(cctx, dst, dstSize,
+                                   sequences, kNbSequences,
+                                   src, srcSize);
+
+            CHECK(ZSTD_isError(cSize), "Should not throw an error");
+            CHECK_Z(ZSTD_decompress(checkBuf, srcSize, dst, cSize));
+            CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+
+            free(sequences);
+            free(checkBuf);
+        }
+
+        ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+
+        { /* Test case with two additional sequences */
+            size_t srcSize = 19;
+            void* const src = CNBuffer;
+            size_t dstSize = ZSTD_compressBound(srcSize);
+            void* const dst = compressedBuffer;
+            size_t const kNbSequences = 7;
+            ZSTD_Sequence* sequences = malloc(sizeof(ZSTD_Sequence) * kNbSequences);
+
+            memset(src, 'x', srcSize);
+
+            sequences[0] = (ZSTD_Sequence) {1, 1, 3, 0};
+            sequences[1] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[2] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[3] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[4] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[5] = (ZSTD_Sequence) {1, 0, 3, 0};
+            sequences[6] = (ZSTD_Sequence) {0, 0, 0, 0};
+
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, 5));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_explicitBlockDelimiters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 1));
+
+            cSize = ZSTD_compressSequences(cctx, dst, dstSize,
+                                   sequences, kNbSequences,
+                                   src, srcSize);
+
+            CHECK(!ZSTD_isError(cSize), "Should throw an error"); /* maxNbSeq is too small and an assert will fail */
+            CHECK(ZSTD_getErrorCode(cSize) != ZSTD_error_externalSequences_invalid, "Wrong error code: %s", ZSTD_getErrorName(cSize)); /* fails sequence validation */
+
+            ZSTD_CCtx_reset(cctx, ZSTD_reset_session_and_parameters);
+
+            /* Test without sequence validation */
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_minMatch, 5));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_blockDelimiters, ZSTD_sf_explicitBlockDelimiters));
+            CHECK_Z(ZSTD_CCtx_setParameter(cctx, ZSTD_c_validateSequences, 0));
+
+            cSize = ZSTD_compressSequences(cctx, dst, dstSize,
+                                   sequences, kNbSequences,
+                                   src, srcSize);
+
+            CHECK(!ZSTD_isError(cSize), "Should throw an error"); /* maxNbSeq is too small and an assert will fail */
+            CHECK(ZSTD_getErrorCode(cSize) != ZSTD_error_externalSequences_invalid, "Wrong error code: %s", ZSTD_getErrorName(cSize)); /* fails sequence validation */
+
+            free(sequences);
+        }
+        ZSTD_freeCCtx(cctx);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
+
+    DISPLAYLEVEL(3, "test%3i : Testing large offset with small window size: ", testNb++);
+    {
+        ZSTD_CCtx* cctx = ZSTD_createCCtx();
+        ZSTD_DCtx* dctx = ZSTD_createDCtx();
+
+        /* Test large offset, small window size*/
+        {
+            size_t srcSize = 21;
+            void* const src = CNBuffer;
+            size_t dstSize = ZSTD_compressBound(srcSize);
+            void* const dst = compressedBuffer;
+            size_t const kNbSequences = 4;
+            ZSTD_Sequence* sequences = malloc(sizeof(ZSTD_Sequence) * kNbSequences);
+            void* const checkBuf = malloc(srcSize);
+            const size_t largeDictSize = 1 << 25;
+            ZSTD_CDict* cdict = NULL;
+            ZSTD_DDict* ddict = NULL;
+
+            /* Generate large dictionary */
+            void* dictBuffer = calloc(largeDictSize, 1);
+            ZSTD_compressionParameters cParams = ZSTD_getCParams(1, srcSize, largeDictSize);
+            cParams.minMatch = ZSTD_MINMATCH_MIN;
+            cParams.hashLog = ZSTD_HASHLOG_MIN;
+            cParams.chainLog = ZSTD_CHAINLOG_MIN;
+
+            cdict = ZSTD_createCDict_advanced(dictBuffer, largeDictSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, cParams, ZSTD_defaultCMem);
+            ddict = ZSTD_createDDict_advanced(dictBuffer, largeDictSize, ZSTD_dlm_byRef, ZSTD_dct_rawContent, ZSTD_defaultCMem);
+
+            ZSTD_CCtx_refCDict(cctx, cdict);
+            ZSTD_DCtx_refDDict(dctx, ddict);
+
+            sequences[0] = (ZSTD_Sequence) {3, 3, 3, 0};
+            sequences[1] = (ZSTD_Sequence) {1 << 25, 0, 3, 0};
+            sequences[2] = (ZSTD_Sequence) {1 << 25, 0, 9, 0};
+            sequences[3] = (ZSTD_Sequence) {3, 0, 3, 0};
+
+            cSize = ZSTD_compressSequences(cctx, dst, dstSize,
+                                   sequences, kNbSequences,
+                                   src, srcSize);
+
+            CHECK(ZSTD_isError(cSize), "Should not throw an error");
+
+            {
+                size_t dSize = ZSTD_decompressDCtx(dctx, checkBuf, srcSize, dst, cSize);
+                CHECK(ZSTD_isError(dSize), "Should not throw an error");
+                CHECK(memcmp(src, checkBuf, srcSize) != 0, "Corruption!");
+            }
+
+            free(sequences);
+            free(checkBuf);
+            free(dictBuffer);
+            ZSTD_freeCDict(cdict);
+            ZSTD_freeDDict(ddict);
+        }
+        ZSTD_freeCCtx(cctx);
+        ZSTD_freeDCtx(dctx);
+    }
+    DISPLAYLEVEL(3, "OK \n");
+
 _end:
     FUZ_freeDictionary(dictionary);
     ZSTD_freeCStream(zc);
@@ -1855,7 +2460,7 @@ static int fuzzerTests(U32 seed, unsigned nbTests, unsigned startTest, double co
             && oldTestLog /* at least one test happened */ && resetAllowed) {
             maxTestSize = FUZ_randomLength(&lseed, oldTestLog+2);
             maxTestSize = MIN(maxTestSize, srcBufferSize-16);
-            {   U64 const pledgedSrcSize = (FUZ_rand(&lseed) & 3) ? ZSTD_CONTENTSIZE_UNKNOWN : maxTestSize; 
+            {   U64 const pledgedSrcSize = (FUZ_rand(&lseed) & 3) ? ZSTD_CONTENTSIZE_UNKNOWN : maxTestSize;
                 CHECK_Z( ZSTD_CCtx_reset(zc, ZSTD_reset_session_only) );
                 CHECK_Z( ZSTD_CCtx_setPledgedSrcSize(zc, pledgedSrcSize) );
             }
@@ -1887,8 +2492,9 @@ static int fuzzerTests(U32 seed, unsigned nbTests, unsigned startTest, double co
         /* multi-segments compression test */
         XXH64_reset(&xxhState, 0);
         {   ZSTD_outBuffer outBuff = { cBuffer, cBufferSize, 0 } ;
-            U32 n;
-            for (n=0, cSize=0, totalTestSize=0 ; totalTestSize < maxTestSize ; n++) {
+            cSize=0;
+            totalTestSize=0;
+            while(totalTestSize < maxTestSize) {
                 /* compress random chunks into randomly sized dst buffers */
                 {   size_t const randomSrcSize = FUZ_randomLength(&lseed, maxSampleLog);
                     size_t const srcSize = MIN(maxTestSize-totalTestSize, randomSrcSize);
@@ -2193,7 +2799,7 @@ static int fuzzerTests_newAPI(U32 seed, int nbTests, int startTest,
 
                 /* mess with long distance matching parameters */
                 if (bigTests) {
-                    if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_c_enableLongDistanceMatching, FUZ_rand(&lseed) & 63, opaqueAPI) );
+                    if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_c_enableLongDistanceMatching, FUZ_randomClampedLength(&lseed, ZSTD_ps_auto, ZSTD_ps_disable), opaqueAPI) );
                     if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_c_ldmHashLog, FUZ_randomClampedLength(&lseed, ZSTD_HASHLOG_MIN, 23), opaqueAPI) );
                     if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_c_ldmMinMatch, FUZ_randomClampedLength(&lseed, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX), opaqueAPI) );
                     if (FUZ_rand(&lseed) & 3) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_c_ldmBucketSizeLog, FUZ_randomClampedLength(&lseed, ZSTD_LDM_BUCKETSIZELOG_MIN, ZSTD_LDM_BUCKETSIZELOG_MAX), opaqueAPI) );
@@ -2277,7 +2883,7 @@ static int fuzzerTests_newAPI(U32 seed, int nbTests, int startTest,
                     CHECK_Z( ZSTD_CCtx_refPrefix(zc, dict, dictSize) );
                 }
 
-                /* Adjust number of workers occassionally - result must be deterministic independent of nbWorkers */
+                /* Adjust number of workers occasionally - result must be deterministic independent of nbWorkers */
                 CHECK_Z(ZSTD_CCtx_getParameter(zc, ZSTD_c_nbWorkers, &nbWorkers));
                 if (nbWorkers > 0 && (FUZ_rand(&lseed) & 7) == 0) {
                     DISPLAYLEVEL(6, "t%u: Modify nbWorkers: %d -> %d \n", testNb, nbWorkers, nbWorkers + iter);
@@ -2356,14 +2962,20 @@ static int fuzzerTests_newAPI(U32 seed, int nbTests, int startTest,
         CHECK(badParameters(zc, savedParams), "CCtx params are wrong");
 
         /* multi - fragments decompression test */
+        if (FUZ_rand(&lseed) & 1) {
+            CHECK_Z(ZSTD_DCtx_reset(zd, ZSTD_reset_session_and_parameters));
+        }
         if (!dictSize /* don't reset if dictionary : could be different */ && (FUZ_rand(&lseed) & 1)) {
-            DISPLAYLEVEL(5, "resetting DCtx (dict:%p) \n", dict);
+            DISPLAYLEVEL(5, "resetting DCtx (dict:%p) \n", (void const*)dict);
             CHECK_Z( ZSTD_resetDStream(zd) );
         } else {
             if (dictSize)
                 DISPLAYLEVEL(5, "using dictionary of size %zu \n", dictSize);
             CHECK_Z( ZSTD_initDStream_usingDict(zd, dict, dictSize) );
         }
+        if (FUZ_rand(&lseed) & 1) {
+            CHECK_Z(ZSTD_DCtx_setParameter(zd, ZSTD_d_disableHuffmanAssembly, FUZ_rand(&lseed) & 1));
+        }
         {   size_t decompressionResult = 1;
             ZSTD_inBuffer  inBuff = { cBuffer, cSize, 0 };
             ZSTD_outBuffer outBuff= { dstBuffer, dstBufferSize, 0 };
@@ -2405,7 +3017,14 @@ static int fuzzerTests_newAPI(U32 seed, int nbTests, int startTest,
         }   }
 
         /* try decompression on noisy data */
-        CHECK_Z( ZSTD_initDStream(zd_noise) );   /* note : no dictionary */
+        if (FUZ_rand(&lseed) & 1) {
+            CHECK_Z(ZSTD_DCtx_reset(zd_noise, ZSTD_reset_session_and_parameters));
+        } else {
+            CHECK_Z(ZSTD_DCtx_reset(zd_noise, ZSTD_reset_session_only));
+        }
+        if (FUZ_rand(&lseed) & 1) {
+            CHECK_Z(ZSTD_DCtx_setParameter(zd_noise, ZSTD_d_disableHuffmanAssembly, FUZ_rand(&lseed) & 1));
+        }
         {   ZSTD_inBuffer  inBuff = { cBuffer, cSize, 0 };
             ZSTD_outBuffer outBuff= { dstBuffer, dstBufferSize, 0 };
             while (outBuff.pos < dstBufferSize) {
@@ -2488,6 +3107,7 @@ int main(int argc, const char** argv)
 
             if (!strcmp(argument, "--newapi")) { selected_api=advanced_api; testNb += !testNb; continue; }
             if (!strcmp(argument, "--no-big-tests")) { bigTests=0; continue; }
+            if (!strcmp(argument, "--big-tests")) { bigTests=1; continue; }
 
             argument++;
             while (*argument!=0) {
@@ -2589,7 +3209,7 @@ int main(int argc, const char** argv)
     if (nbTests<=0) nbTests=1;
 
     if (testNb==0) {
-        result = basicUnitTests(0, ((double)proba) / 100);  /* constant seed for predictability */
+        result = basicUnitTests(0, ((double)proba) / 100, bigTests);  /* constant seed for predictability */
     }
 
     if (!result) {
diff --git a/priv/zstd/zlibWrapper/Makefile b/priv/zstd/zlibWrapper/Makefile
index 6fd5ac3..830b294 100644
--- a/priv/zstd/zlibWrapper/Makefile
+++ b/priv/zstd/zlibWrapper/Makefile
@@ -66,9 +66,10 @@ test: example fitblk example_zstd fitblk_zstd zwrapbench minigzip minigzip_zstd
 	./zwrapbench -qi1b3B1K $(TEST_FILE)
 	./zwrapbench -rqi1b1e3 ../lib
 
-#valgrindTest: ZSTDLIBRARY = $(ZSTDLIBDIR)/libzstd.so
-valgrindTest: VALGRIND = LD_LIBRARY_PATH=$(ZSTDLIBDIR) valgrind --track-origins=yes --leak-check=full --error-exitcode=1
-valgrindTest: clean example fitblk example_zstd fitblk_zstd zwrapbench
+.PHONY: test-valgrind
+#test-valgrind: ZSTDLIBRARY = $(ZSTDLIBDIR)/libzstd.so
+test-valgrind: VALGRIND = LD_LIBRARY_PATH=$(ZSTDLIBDIR) valgrind --track-origins=yes --leak-check=full --error-exitcode=1
+test-valgrind: clean example fitblk example_zstd fitblk_zstd zwrapbench
 	@echo "\n ---- valgrind tests ----"
 	$(VALGRIND) ./example
 	$(VALGRIND) ./example_zstd
diff --git a/priv/zstd/zlibWrapper/README.md b/priv/zstd/zlibWrapper/README.md
index e61767c..311e1db 100644
--- a/priv/zstd/zlibWrapper/README.md
+++ b/priv/zstd/zlibWrapper/README.md
@@ -1,7 +1,7 @@
 Zstandard wrapper for zlib
 ================================
 
-The main objective of creating a zstd wrapper for [zlib](http://zlib.net/) is to allow a quick and smooth transition to zstd for projects already using zlib.
+The main objective of creating a zstd wrapper for [zlib](https://zlib.net/) is to allow a quick and smooth transition to zstd for projects already using zlib.
 
 #### Required files
 
@@ -43,7 +43,7 @@ This behavior can be changed using `ZWRAP_setDecompressionType(ZWRAP_FORCE_ZLIB)
 
 
 #### Example
-We have take the file `test/example.c` from [the zlib library distribution](http://zlib.net/) and copied it to [zlibWrapper/examples/example.c](examples/example.c).
+We have taken the file `test/example.c` from [the zlib library distribution](https://zlib.net/) and copied it to [zlibWrapper/examples/example.c](examples/example.c).
 After compilation and execution it shows the following results: 
 ```
 zlib version 1.2.8 = 0x1280, compile flags = 0x65
diff --git a/priv/zstd/zlibWrapper/examples/example.c b/priv/zstd/zlibWrapper/examples/example.c
index 9000f7a..d7590e3 100644
--- a/priv/zstd/zlibWrapper/examples/example.c
+++ b/priv/zstd/zlibWrapper/examples/example.c
@@ -56,26 +56,26 @@ z_const char hello[] = "hello, hello! I said hello, hello!";
 const char dictionary[] = "hello, hello!";
 uLong dictId; /* Adler32 value of the dictionary */
 
-void test_deflate       OF((Byte *compr, uLong comprLen));
-void test_inflate       OF((Byte *compr, uLong comprLen,
+void test_deflate       _Z_OF((Byte *compr, uLong comprLen));
+void test_inflate       _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_large_deflate OF((Byte *compr, uLong comprLen,
+void test_large_deflate _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_large_inflate OF((Byte *compr, uLong comprLen,
+void test_large_inflate _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_flush         OF((Byte *compr, uLong *comprLen));
-void test_sync          OF((Byte *compr, uLong comprLen,
+void test_flush         _Z_OF((Byte *compr, uLong *comprLen));
+void test_sync          _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_dict_deflate  OF((Byte *compr, uLong comprLen));
-void test_dict_inflate  OF((Byte *compr, uLong comprLen,
+void test_dict_deflate  _Z_OF((Byte *compr, uLong comprLen));
+void test_dict_inflate  _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-int  main               OF((int argc, char *argv[]));
+int  main               _Z_OF((int argc, char *argv[]));
 
 
 #ifdef Z_SOLO
 
-void *myalloc OF((void *, unsigned, unsigned));
-void myfree OF((void *, void *));
+void *myalloc _Z_OF((void *, unsigned, unsigned));
+void myfree _Z_OF((void *, void *));
 
 void *myalloc(q, n, m)
     void *q;
@@ -102,9 +102,9 @@ static free_func zfree = myfree;
 static alloc_func zalloc = (alloc_func)0;
 static free_func zfree = (free_func)0;
 
-void test_compress      OF((Byte *compr, uLong comprLen,
+void test_compress      _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_gzio          OF((const char *fname,
+void test_gzio          _Z_OF((const char *fname,
                             Byte *uncompr, uLong uncomprLen));
 
 /* ===========================================================================
diff --git a/priv/zstd/zlibWrapper/examples/example_original.c b/priv/zstd/zlibWrapper/examples/example_original.c
index 649882b..5b4e4d1 100644
--- a/priv/zstd/zlibWrapper/examples/example_original.c
+++ b/priv/zstd/zlibWrapper/examples/example_original.c
@@ -51,26 +51,26 @@ z_const char hello[] = "hello, hello!";
 const char dictionary[] = "hello";
 uLong dictId; /* Adler32 value of the dictionary */
 
-void test_deflate       OF((Byte *compr, uLong comprLen));
-void test_inflate       OF((Byte *compr, uLong comprLen,
+void test_deflate       _Z_OF((Byte *compr, uLong comprLen));
+void test_inflate       _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_large_deflate OF((Byte *compr, uLong comprLen,
+void test_large_deflate _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_large_inflate OF((Byte *compr, uLong comprLen,
+void test_large_inflate _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_flush         OF((Byte *compr, uLong *comprLen));
-void test_sync          OF((Byte *compr, uLong comprLen,
+void test_flush         _Z_OF((Byte *compr, uLong *comprLen));
+void test_sync          _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_dict_deflate  OF((Byte *compr, uLong comprLen));
-void test_dict_inflate  OF((Byte *compr, uLong comprLen,
+void test_dict_deflate  _Z_OF((Byte *compr, uLong comprLen));
+void test_dict_inflate  _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-int  main               OF((int argc, char *argv[]));
+int  main               _Z_OF((int argc, char *argv[]));
 
 
 #ifdef Z_SOLO
 
-void *myalloc OF((void *, unsigned, unsigned));
-void myfree OF((void *, void *));
+void *myalloc _Z_OF((void *, unsigned, unsigned));
+void myfree _Z_OF((void *, void *));
 
 void *myalloc(q, n, m)
     void *q;
@@ -94,9 +94,9 @@ static free_func zfree = myfree;
 static alloc_func zalloc = (alloc_func)0;
 static free_func zfree = (free_func)0;
 
-void test_compress      OF((Byte *compr, uLong comprLen,
+void test_compress      _Z_OF((Byte *compr, uLong comprLen,
                             Byte *uncompr, uLong uncomprLen));
-void test_gzio          OF((const char *fname,
+void test_gzio          _Z_OF((const char *fname,
                             Byte *uncompr, uLong uncomprLen));
 
 /* ===========================================================================
diff --git a/priv/zstd/zlibWrapper/examples/fitblk.c b/priv/zstd/zlibWrapper/examples/fitblk.c
index 669b176..8dc7071 100644
--- a/priv/zstd/zlibWrapper/examples/fitblk.c
+++ b/priv/zstd/zlibWrapper/examples/fitblk.c
@@ -119,7 +119,7 @@ local int recompress(z_streamp inf, z_streamp def)
         if (ret == Z_MEM_ERROR)
             return ret;
 
-        /* compress what was decompresed until done or no room */
+        /* compress what was decompressed until done or no room */
         def->avail_in = RAWLEN - inf->avail_out;
         def->next_in = raw;
         if (inf->avail_out != 0)
diff --git a/priv/zstd/zlibWrapper/examples/fitblk_original.c b/priv/zstd/zlibWrapper/examples/fitblk_original.c
index 20f351b..723dc00 100644
--- a/priv/zstd/zlibWrapper/examples/fitblk_original.c
+++ b/priv/zstd/zlibWrapper/examples/fitblk_original.c
@@ -109,7 +109,7 @@ local int recompress(z_streamp inf, z_streamp def)
         if (ret == Z_MEM_ERROR)
             return ret;
 
-        /* compress what was decompresed until done or no room */
+        /* compress what was decompressed until done or no room */
         def->avail_in = RAWLEN - inf->avail_out;
         def->next_in = raw;
         if (inf->avail_out != 0)
diff --git a/priv/zstd/zlibWrapper/examples/minigzip.c b/priv/zstd/zlibWrapper/examples/minigzip.c
index f67be09..717a94d 100644
--- a/priv/zstd/zlibWrapper/examples/minigzip.c
+++ b/priv/zstd/zlibWrapper/examples/minigzip.c
@@ -3,7 +3,7 @@
 
 /* minigzip.c -- simulate gzip using the zlib compression library
  * Copyright (C) 1995-2006, 2010, 2011 Jean-loup Gailly.
- * For conditions of distribution and use, see http://www.zlib.net/zlib_license.html
+ * For conditions of distribution and use, see https://www.zlib.net/zlib_license.html
  */
 
 /*
@@ -64,7 +64,7 @@
 
 #if !defined(Z_HAVE_UNISTD_H) && !defined(_LARGEFILE64_SOURCE)
 #ifndef WIN32 /* unlink already in stdio.h for WIN32 */
-  extern int unlink OF((const char *));
+  extern int unlink _Z_OF((const char *));
 #endif
 #endif
 
@@ -154,8 +154,8 @@ static void pwinerror (s)
 #  include        /* for unlink() */
 #endif
 
-void *myalloc OF((void *, unsigned, unsigned));
-void myfree OF((void *, void *));
+void *myalloc _Z_OF((void *, unsigned, unsigned));
+void myfree _Z_OF((void *, void *));
 
 void *myalloc(q, n, m)
     void *q;
@@ -180,9 +180,9 @@ typedef struct gzFile_s {
     z_stream strm;
 } *gzFile;
 
-gzFile gzopen OF((const char *, const char *));
-gzFile gzdopen OF((int, const char *));
-gzFile gz_open OF((const char *, int, const char *));
+gzFile gzopen _Z_OF((const char *, const char *));
+gzFile gzdopen _Z_OF((int, const char *));
+gzFile gz_open _Z_OF((const char *, int, const char *));
 
 gzFile gzopen(path, mode)
 const char *path;
@@ -236,7 +236,7 @@ gzFile gz_open(path, fd, mode)
     return gz;
 }
 
-int gzwrite OF((gzFile, const void *, unsigned));
+int gzwrite _Z_OF((gzFile, const void *, unsigned));
 
 int gzwrite(gz, buf, len)
     gzFile gz;
@@ -260,7 +260,7 @@ int gzwrite(gz, buf, len)
     return len;
 }
 
-int gzread OF((gzFile, void *, unsigned));
+int gzread _Z_OF((gzFile, void *, unsigned));
 
 int gzread(gz, buf, len)
     gzFile gz;
@@ -297,7 +297,7 @@ int gzread(gz, buf, len)
     return len - strm->avail_out;
 }
 
-int gzclose OF((gzFile));
+int gzclose _Z_OF((gzFile));
 
 int gzclose(gz)
     gzFile gz;
@@ -326,7 +326,7 @@ int gzclose(gz)
     return Z_OK;
 }
 
-const char *gzerror OF((gzFile, int *));
+const char *gzerror _Z_OF((gzFile, int *));
 
 const char *gzerror(gz, err)
     gzFile gz;
@@ -340,15 +340,15 @@ const char *gzerror(gz, err)
 
 char *prog;
 
-void error            OF((const char *msg));
-void gz_compress      OF((FILE   *in, gzFile out));
+void error            _Z_OF((const char *msg));
+void gz_compress      _Z_OF((FILE   *in, gzFile out));
 #ifdef USE_MMAP
-int  gz_compress_mmap OF((FILE   *in, gzFile out));
+int  gz_compress_mmap _Z_OF((FILE   *in, gzFile out));
 #endif
-void gz_uncompress    OF((gzFile in, FILE   *out));
-void file_compress    OF((char  *file, char *mode));
-void file_uncompress  OF((char  *file));
-int  main             OF((int argc, char *argv[]));
+void gz_uncompress    _Z_OF((gzFile in, FILE   *out));
+void file_compress    _Z_OF((char  *file, char *mode));
+void file_uncompress  _Z_OF((char  *file));
+int  main             _Z_OF((int argc, char *argv[]));
 
 /* ===========================================================================
  * Display error message and exit
diff --git a/priv/zstd/zlibWrapper/examples/zwrapbench.c b/priv/zstd/zlibWrapper/examples/zwrapbench.c
index 5993e51..3bc9a1a 100644
--- a/priv/zstd/zlibWrapper/examples/zwrapbench.c
+++ b/priv/zstd/zlibWrapper/examples/zwrapbench.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -388,7 +388,7 @@ static int BMK_benchMem(z_const void* srcBuffer, size_t srcSize,
             markNb = (markNb+1) % NB_MARKS;
             DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.3f),%6.1f MB/s\r",
                     marks[markNb], displayName, (unsigned)srcSize, (unsigned)cSize, ratio,
-                    (double)srcSize / fastestC );
+                    (double)srcSize / (double)fastestC );
 
             (void)fastestD; (void)crcOrig;   /*  unused when decompression disabled */
 #if 1
@@ -428,8 +428,10 @@ static int BMK_benchMem(z_const void* srcBuffer, size_t srcSize,
                     ZSTD_DStream* zbd = ZSTD_createDStream();
                     size_t rSize;
                     if (zbd == NULL) EXM_THROW(1, "ZSTD_createDStream() allocation failure");
-                    rSize = ZSTD_initDStream_usingDict(zbd, dictBuffer, dictBufferSize);
-                    if (ZSTD_isError(rSize)) EXM_THROW(1, "ZSTD_initDStream() failed : %s", ZSTD_getErrorName(rSize));
+                    rSize = ZSTD_DCtx_reset(zbd, ZSTD_reset_session_only);
+                    if (ZSTD_isError(rSize)) EXM_THROW(1, "ZSTD_DCtx_reset() failed : %s", ZSTD_getErrorName(rSize));
+                    rSize = ZSTD_DCtx_loadDictionary(zbd, dictBuffer, dictBufferSize);
+                    if (ZSTD_isError(rSize)) EXM_THROW(1, "ZSTD_DCtx_loadDictionary() failed : %s", ZSTD_getErrorName(rSize));
                     do {
                         U32 blockNb;
                         for (blockNb=0; blockNb%10u (%5.3f),%6.1f MB/s ,%6.1f MB/s\r",
                     marks[markNb], displayName, (unsigned)srcSize, (unsigned)cSize, ratio,
-                    (double)srcSize / fastestC,
-                    (double)srcSize / fastestD );
+                    (double)srcSize / (double)fastestC,
+                    (double)srcSize / (double)fastestD );
 
             /* CRC Checking */
             {   U64 const crcCheck = XXH64(resultBuffer, srcSize, 0);
@@ -558,8 +560,8 @@ static int BMK_benchMem(z_const void* srcBuffer, size_t srcSize,
         }   /* for (testNb = 1; testNb <= (g_nbIterations + !g_nbIterations); testNb++) */
 
         if (g_displayLevel == 1) {
-            double cSpeed = (double)srcSize / fastestC;
-            double dSpeed = (double)srcSize / fastestD;
+            double cSpeed = (double)srcSize / (double)fastestC;
+            double dSpeed = (double)srcSize / (double)fastestD;
             if (g_additionalParam)
                 DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, g_additionalParam);
             else
@@ -836,7 +838,7 @@ static int usage(const char* programName)
     DISPLAY( " -b#    : benchmark file(s), using # compression level (default : %d) \n", ZSTDCLI_CLEVEL_DEFAULT);
     DISPLAY( " -e#    : test all compression levels from -bX to # (default: %d)\n", ZSTDCLI_CLEVEL_DEFAULT);
     DISPLAY( " -i#    : minimum evaluation time in seconds (default : 3s)\n");
-    DISPLAY( " -B#    : cut file into independent blocks of size # (default: no block)\n");
+    DISPLAY( " -B#    : cut file into independent chunks of size # (default: no chunking)\n");
     return 0;
 }
 
diff --git a/priv/zstd/zlibWrapper/gzclose.c b/priv/zstd/zlibWrapper/gzclose.c
index 25d3789..ba43b8c 100644
--- a/priv/zstd/zlibWrapper/gzclose.c
+++ b/priv/zstd/zlibWrapper/gzclose.c
@@ -3,7 +3,7 @@
 
 /* gzclose.c -- zlib gzclose() function
  * Copyright (C) 2004, 2010 Mark Adler
- * For conditions of distribution and use, see http://www.zlib.net/zlib_license.html
+ * For conditions of distribution and use, see https://www.zlib.net/zlib_license.html
  */
 
 #include "gzguts.h"
diff --git a/priv/zstd/zlibWrapper/gzcompatibility.h b/priv/zstd/zlibWrapper/gzcompatibility.h
index c1aa2b8..9d11b98 100644
--- a/priv/zstd/zlibWrapper/gzcompatibility.h
+++ b/priv/zstd/zlibWrapper/gzcompatibility.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,10 +11,10 @@
 
 
 #if ZLIB_VERNUM <= 0x1240
-ZEXTERN int ZEXPORT gzclose_r OF((gzFile file));
-ZEXTERN int ZEXPORT gzclose_w OF((gzFile file));
-ZEXTERN int ZEXPORT gzbuffer OF((gzFile file, unsigned size));
-ZEXTERN z_off_t ZEXPORT gzoffset OF((gzFile file));
+ZEXTERN int ZEXPORT gzclose_r _Z_OF((gzFile file));
+ZEXTERN int ZEXPORT gzclose_w _Z_OF((gzFile file));
+ZEXTERN int ZEXPORT gzbuffer _Z_OF((gzFile file, unsigned size));
+ZEXTERN z_off_t ZEXPORT gzoffset _Z_OF((gzFile file));
 
 #if !defined(_WIN32) && defined(Z_LARGE64)
 #  define z_off64_t off64_t
@@ -40,7 +40,7 @@ struct gzFile_s {
 #if ZLIB_VERNUM <= 0x1270
 #if defined(_WIN32) && !defined(Z_SOLO)
 #    include          /* for wchar_t */
-ZEXTERN gzFile         ZEXPORT gzopen_w OF((const wchar_t *path,
+ZEXTERN gzFile         ZEXPORT gzopen_w _Z_OF((const wchar_t *path,
                                             const char *mode));
 #endif
 #endif
@@ -61,8 +61,8 @@ ZEXTERN gzFile         ZEXPORT gzopen_w OF((const wchar_t *path,
 #  endif
 #  undef z_longlong
 #endif
-ZEXTERN z_size_t ZEXPORT gzfread OF((voidp buf, z_size_t size, z_size_t nitems,
+ZEXTERN z_size_t ZEXPORT gzfread _Z_OF((voidp buf, z_size_t size, z_size_t nitems,
                                      gzFile file));
-ZEXTERN z_size_t ZEXPORT gzfwrite OF((voidpc buf, z_size_t size,
+ZEXTERN z_size_t ZEXPORT gzfwrite _Z_OF((voidpc buf, z_size_t size,
                                       z_size_t nitems, gzFile file));
 #endif
diff --git a/priv/zstd/zlibWrapper/gzguts.h b/priv/zstd/zlibWrapper/gzguts.h
index b639b4b..70a609d 100644
--- a/priv/zstd/zlibWrapper/gzguts.h
+++ b/priv/zstd/zlibWrapper/gzguts.h
@@ -4,7 +4,7 @@
 
 /* gzguts.h -- zlib internal header definitions for gz* operations
  * Copyright (C) 2004, 2005, 2010, 2011, 2012, 2013, 2016 Mark Adler
- * For conditions of distribution and use, see http://www.zlib.net/zlib_license.html
+ * For conditions of distribution and use, see https://www.zlib.net/zlib_license.html
  */
 
 #ifdef _LARGEFILE64_SOURCE
@@ -126,8 +126,8 @@
 
 /* gz* functions always use library allocation functions */
 #ifndef STDC
-  extern voidp  malloc OF((uInt size));
-  extern void   free   OF((voidpf ptr));
+  extern voidp  malloc _Z_OF((uInt size));
+  extern void   free   _Z_OF((voidpf ptr));
 #endif
 
 /* get errno and strerror definition */
@@ -145,10 +145,10 @@
 
 /* provide prototypes for these when building zlib without LFS */
 #if !defined(_LARGEFILE64_SOURCE) || _LFS64_LARGEFILE-0 == 0
-    ZEXTERN gzFile ZEXPORT gzopen64 OF((const char *, const char *));
-    ZEXTERN z_off64_t ZEXPORT gzseek64 OF((gzFile, z_off64_t, int));
-    ZEXTERN z_off64_t ZEXPORT gztell64 OF((gzFile));
-    ZEXTERN z_off64_t ZEXPORT gzoffset64 OF((gzFile));
+    ZEXTERN gzFile ZEXPORT gzopen64 _Z_OF((const char *, const char *));
+    ZEXTERN z_off64_t ZEXPORT gzseek64 _Z_OF((gzFile, z_off64_t, int));
+    ZEXTERN z_off64_t ZEXPORT gztell64 _Z_OF((gzFile));
+    ZEXTERN z_off64_t ZEXPORT gzoffset64 _Z_OF((gzFile));
 #endif
 
 /* default memLevel */
@@ -213,9 +213,9 @@ typedef union {
 } gz_statep;
 
 /* shared functions */
-void ZLIB_INTERNAL gz_error OF((gz_statep, int, const char *));
+void ZLIB_INTERNAL gz_error _Z_OF((gz_statep, int, const char *));
 #if defined UNDER_CE
-char ZLIB_INTERNAL *gz_strwinerror OF((DWORD error));
+char ZLIB_INTERNAL *gz_strwinerror _Z_OF((DWORD error));
 #endif
 
 /* GT_OFF(x), where x is an unsigned value, is true if x > maximum z_off64_t
@@ -224,6 +224,6 @@ char ZLIB_INTERNAL *gz_strwinerror OF((DWORD error));
 #ifdef INT_MAX
 #  define GT_OFF(x) (sizeof(int) == sizeof(z_off64_t) && (x) > INT_MAX)
 #else
-unsigned ZLIB_INTERNAL gz_intmax OF((void));
+unsigned ZLIB_INTERNAL gz_intmax _Z_OF((void));
 #  define GT_OFF(x) (sizeof(int) == sizeof(z_off64_t) && (x) > gz_intmax())
 #endif
diff --git a/priv/zstd/zlibWrapper/gzlib.c b/priv/zstd/zlibWrapper/gzlib.c
index b1fb985..eea480a 100644
--- a/priv/zstd/zlibWrapper/gzlib.c
+++ b/priv/zstd/zlibWrapper/gzlib.c
@@ -3,7 +3,7 @@
 
 /* gzlib.c -- zlib functions common to reading and writing gzip files
  * Copyright (C) 2004-2017 Mark Adler
- * For conditions of distribution and use, see http://www.zlib.net/zlib_license.html
+ * For conditions of distribution and use, see https://www.zlib.net/zlib_license.html
  */
 
 #include "gzguts.h"
@@ -19,8 +19,8 @@
 #endif
 
 /* Local functions */
-local void gz_reset OF((gz_statep));
-local gzFile gz_open OF((const void *, int, const char *));
+local void gz_reset _Z_OF((gz_statep));
+local gzFile gz_open _Z_OF((const void *, int, const char *));
 
 #if defined UNDER_CE
 
diff --git a/priv/zstd/zlibWrapper/gzread.c b/priv/zstd/zlibWrapper/gzread.c
index 359d178..584fad1 100644
--- a/priv/zstd/zlibWrapper/gzread.c
+++ b/priv/zstd/zlibWrapper/gzread.c
@@ -3,7 +3,7 @@
 
  /* gzread.c -- zlib functions for reading gzip files
  * Copyright (C) 2004, 2005, 2010, 2011, 2012, 2013, 2016 Mark Adler
- * For conditions of distribution and use, see http://www.zlib.net/zlib_license.html
+ * For conditions of distribution and use, see https://www.zlib.net/zlib_license.html
  */
 
 #include "gzguts.h"
@@ -17,13 +17,13 @@
 
 
 /* Local functions */
-local int gz_load OF((gz_statep, unsigned char *, unsigned, unsigned *));
-local int gz_avail OF((gz_statep));
-local int gz_look OF((gz_statep));
-local int gz_decomp OF((gz_statep));
-local int gz_fetch OF((gz_statep));
-local int gz_skip OF((gz_statep, z_off64_t));
-local z_size_t gz_read OF((gz_statep, voidp, z_size_t));
+local int gz_load _Z_OF((gz_statep, unsigned char *, unsigned, unsigned *));
+local int gz_avail _Z_OF((gz_statep));
+local int gz_look _Z_OF((gz_statep));
+local int gz_decomp _Z_OF((gz_statep));
+local int gz_fetch _Z_OF((gz_statep));
+local int gz_skip _Z_OF((gz_statep, z_off64_t));
+local z_size_t gz_read _Z_OF((gz_statep, voidp, z_size_t));
 
 /* Use read() to load a buffer -- return -1 on error, otherwise 0.  Read from
    state.state->fd, and update state.state->eof, state.state->err, and state.state->msg as appropriate.
@@ -464,8 +464,8 @@ z_size_t ZEXPORT gzfread(buf, size, nitems, file)
 #endif
 
 #if ZLIB_VERNUM <= 0x1250
-ZEXTERN int ZEXPORT gzgetc OF((gzFile file));
-ZEXTERN int ZEXPORT gzgetc_ OF((gzFile file));
+ZEXTERN int ZEXPORT gzgetc _Z_OF((gzFile file));
+ZEXTERN int ZEXPORT gzgetc_ _Z_OF((gzFile file));
 #endif
 
 int ZEXPORT gzgetc(file)
diff --git a/priv/zstd/zlibWrapper/gzwrite.c b/priv/zstd/zlibWrapper/gzwrite.c
index 422ff17..ccd4f71 100644
--- a/priv/zstd/zlibWrapper/gzwrite.c
+++ b/priv/zstd/zlibWrapper/gzwrite.c
@@ -3,7 +3,7 @@
 
  /* gzwrite.c -- zlib functions for writing gzip files
  * Copyright (C) 2004-2017 Mark Adler
- * For conditions of distribution and use, see http://www.zlib.net/zlib_license.html
+ * For conditions of distribution and use, see https://www.zlib.net/zlib_license.html
  */
 
 #include 
@@ -11,10 +11,10 @@
 #include "gzguts.h"
 
 /* Local functions */
-local int gz_init OF((gz_statep));
-local int gz_comp OF((gz_statep, int));
-local int gz_zero OF((gz_statep, z_off64_t));
-local z_size_t gz_write OF((gz_statep, voidpc, z_size_t));
+local int gz_init _Z_OF((gz_statep));
+local int gz_comp _Z_OF((gz_statep, int));
+local int gz_zero _Z_OF((gz_statep, z_off64_t));
+local z_size_t gz_write _Z_OF((gz_statep, voidpc, z_size_t));
 
 /* Initialize state for writing a gzip file.  Mark initialization by setting
    state.state->size to non-zero.  Return -1 on a memory allocation failure, or 0 on
diff --git a/priv/zstd/zlibWrapper/zstd_zlibwrapper.c b/priv/zstd/zlibWrapper/zstd_zlibwrapper.c
index ceb2393..479ddd4 100644
--- a/priv/zstd/zlibWrapper/zstd_zlibwrapper.c
+++ b/priv/zstd/zlibWrapper/zstd_zlibwrapper.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -101,7 +101,7 @@ ZWRAP_decompress_type ZWRAP_getDecompressionType(void) { return g_ZWRAPdecompres
 
 const char * zstdVersion(void) { return ZSTD_VERSION_STRING; }
 
-ZEXTERN const char * ZEXPORT z_zlibVersion OF((void)) { return zlibVersion();  }
+ZEXTERN const char * ZEXPORT z_zlibVersion _Z_OF((void)) { return zlibVersion();  }
 
 static void* ZWRAP_allocFunction(void* opaque, size_t size)
 {
@@ -260,7 +260,7 @@ static struct internal_state* convert_into_sis(void* ptr)
     return (struct internal_state*) ptr;
 }
 
-ZEXTERN int ZEXPORT z_deflateInit_ OF((z_streamp strm, int level,
+ZEXTERN int ZEXPORT z_deflateInit_ _Z_OF((z_streamp strm, int level,
                                      const char *version, int stream_size))
 {
     ZWRAP_CCtx* zwc;
@@ -287,7 +287,7 @@ ZEXTERN int ZEXPORT z_deflateInit_ OF((z_streamp strm, int level,
 }
 
 
-ZEXTERN int ZEXPORT z_deflateInit2_ OF((z_streamp strm, int level, int method,
+ZEXTERN int ZEXPORT z_deflateInit2_ _Z_OF((z_streamp strm, int level, int method,
                                       int windowBits, int memLevel,
                                       int strategy, const char *version,
                                       int stream_size))
@@ -319,7 +319,7 @@ int ZWRAP_deflateReset_keepDict(z_streamp strm)
 }
 
 
-ZEXTERN int ZEXPORT z_deflateReset OF((z_streamp strm))
+ZEXTERN int ZEXPORT z_deflateReset _Z_OF((z_streamp strm))
 {
     LOG_WRAPPERC("- deflateReset\n");
     if (!g_ZWRAP_useZSTDcompression)
@@ -334,7 +334,7 @@ ZEXTERN int ZEXPORT z_deflateReset OF((z_streamp strm))
 }
 
 
-ZEXTERN int ZEXPORT z_deflateSetDictionary OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_deflateSetDictionary _Z_OF((z_streamp strm,
                                              const Bytef *dictionary,
                                              uInt  dictLength))
 {
@@ -359,7 +359,7 @@ ZEXTERN int ZEXPORT z_deflateSetDictionary OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_deflate OF((z_streamp strm, int flush))
+ZEXTERN int ZEXPORT z_deflate _Z_OF((z_streamp strm, int flush))
 {
     ZWRAP_CCtx* zwc;
 
@@ -465,7 +465,7 @@ ZEXTERN int ZEXPORT z_deflate OF((z_streamp strm, int flush))
 }
 
 
-ZEXTERN int ZEXPORT z_deflateEnd OF((z_streamp strm))
+ZEXTERN int ZEXPORT z_deflateEnd _Z_OF((z_streamp strm))
 {
     if (!g_ZWRAP_useZSTDcompression) {
         LOG_WRAPPERC("- deflateEnd\n");
@@ -483,7 +483,7 @@ ZEXTERN int ZEXPORT z_deflateEnd OF((z_streamp strm))
 }
 
 
-ZEXTERN uLong ZEXPORT z_deflateBound OF((z_streamp strm,
+ZEXTERN uLong ZEXPORT z_deflateBound _Z_OF((z_streamp strm,
                                        uLong sourceLen))
 {
     if (!g_ZWRAP_useZSTDcompression)
@@ -493,7 +493,7 @@ ZEXTERN uLong ZEXPORT z_deflateBound OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_deflateParams OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_deflateParams _Z_OF((z_streamp strm,
                                       int level,
                                       int strategy))
 {
@@ -594,7 +594,7 @@ static int ZWRAPD_finishWithErrorMsg(z_streamp strm, char* message)
 }
 
 
-ZEXTERN int ZEXPORT z_inflateInit_ OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflateInit_ _Z_OF((z_streamp strm,
                                      const char* version, int stream_size))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB) {
@@ -623,7 +623,7 @@ ZEXTERN int ZEXPORT z_inflateInit_ OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_inflateInit2_ OF((z_streamp strm, int  windowBits,
+ZEXTERN int ZEXPORT z_inflateInit2_ _Z_OF((z_streamp strm, int  windowBits,
                                       const char *version, int stream_size))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB) {
@@ -660,7 +660,7 @@ int ZWRAP_inflateReset_keepDict(z_streamp strm)
 }
 
 
-ZEXTERN int ZEXPORT z_inflateReset OF((z_streamp strm))
+ZEXTERN int ZEXPORT z_inflateReset _Z_OF((z_streamp strm))
 {
     LOG_WRAPPERD("- inflateReset\n");
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved)
@@ -678,7 +678,7 @@ ZEXTERN int ZEXPORT z_inflateReset OF((z_streamp strm))
 
 
 #if ZLIB_VERNUM >= 0x1240
-ZEXTERN int ZEXPORT z_inflateReset2 OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflateReset2 _Z_OF((z_streamp strm,
                                       int windowBits))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved)
@@ -696,7 +696,7 @@ ZEXTERN int ZEXPORT z_inflateReset2 OF((z_streamp strm,
 #endif
 
 
-ZEXTERN int ZEXPORT z_inflateSetDictionary OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflateSetDictionary _Z_OF((z_streamp strm,
                                              const Bytef *dictionary,
                                              uInt  dictLength))
 {
@@ -706,8 +706,10 @@ ZEXTERN int ZEXPORT z_inflateSetDictionary OF((z_streamp strm,
 
     {   ZWRAP_DCtx* const zwd = (ZWRAP_DCtx*) strm->state;
         if (zwd == NULL || zwd->zbd == NULL) return Z_STREAM_ERROR;
-        { size_t const initErr = ZSTD_initDStream_usingDict(zwd->zbd, dictionary, dictLength);
-          if (ZSTD_isError(initErr)) return ZWRAPD_finishWithError(zwd, strm, 0); }
+        { size_t const resetErr = ZSTD_DCtx_reset(zwd->zbd, ZSTD_reset_session_only);
+          if (ZSTD_isError(resetErr)) return ZWRAPD_finishWithError(zwd, strm, 0); }
+        { size_t const loadErr = ZSTD_DCtx_loadDictionary(zwd->zbd, dictionary, dictLength);
+          if (ZSTD_isError(loadErr)) return ZWRAPD_finishWithError(zwd, strm, 0); }
         zwd->decompState = ZWRAP_useReset;
 
         if (zwd->totalInBytes == ZSTD_HEADERSIZE) {
@@ -730,7 +732,7 @@ ZEXTERN int ZEXPORT z_inflateSetDictionary OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_inflate OF((z_streamp strm, int flush))
+ZEXTERN int ZEXPORT z_inflate _Z_OF((z_streamp strm, int flush))
 {
     ZWRAP_DCtx* zwd;
 
@@ -926,7 +928,7 @@ ZEXTERN int ZEXPORT z_inflate OF((z_streamp strm, int flush))
 }
 
 
-ZEXTERN int ZEXPORT z_inflateEnd OF((z_streamp strm))
+ZEXTERN int ZEXPORT z_inflateEnd _Z_OF((z_streamp strm))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved)
         return inflateEnd(strm);
@@ -943,7 +945,7 @@ ZEXTERN int ZEXPORT z_inflateEnd OF((z_streamp strm))
 }
 
 
-ZEXTERN int ZEXPORT z_inflateSync OF((z_streamp strm))
+ZEXTERN int ZEXPORT z_inflateSync _Z_OF((z_streamp strm))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved) {
         return inflateSync(strm);
@@ -955,7 +957,7 @@ ZEXTERN int ZEXPORT z_inflateSync OF((z_streamp strm))
 
 
 /* Advanced compression functions */
-ZEXTERN int ZEXPORT z_deflateCopy OF((z_streamp dest,
+ZEXTERN int ZEXPORT z_deflateCopy _Z_OF((z_streamp dest,
                                     z_streamp source))
 {
     if (!g_ZWRAP_useZSTDcompression)
@@ -964,7 +966,7 @@ ZEXTERN int ZEXPORT z_deflateCopy OF((z_streamp dest,
 }
 
 
-ZEXTERN int ZEXPORT z_deflateTune OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_deflateTune _Z_OF((z_streamp strm,
                                     int good_length,
                                     int max_lazy,
                                     int nice_length,
@@ -977,7 +979,7 @@ ZEXTERN int ZEXPORT z_deflateTune OF((z_streamp strm,
 
 
 #if ZLIB_VERNUM >= 0x1260
-ZEXTERN int ZEXPORT z_deflatePending OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_deflatePending _Z_OF((z_streamp strm,
                                        unsigned *pending,
                                        int *bits))
 {
@@ -988,7 +990,7 @@ ZEXTERN int ZEXPORT z_deflatePending OF((z_streamp strm,
 #endif
 
 
-ZEXTERN int ZEXPORT z_deflatePrime OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_deflatePrime _Z_OF((z_streamp strm,
                                      int bits,
                                      int value))
 {
@@ -998,7 +1000,7 @@ ZEXTERN int ZEXPORT z_deflatePrime OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_deflateSetHeader OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_deflateSetHeader _Z_OF((z_streamp strm,
                                          gz_headerp head))
 {
     if (!g_ZWRAP_useZSTDcompression)
@@ -1011,7 +1013,7 @@ ZEXTERN int ZEXPORT z_deflateSetHeader OF((z_streamp strm,
 
 /* Advanced decompression functions */
 #if ZLIB_VERNUM >= 0x1280
-ZEXTERN int ZEXPORT z_inflateGetDictionary OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflateGetDictionary _Z_OF((z_streamp strm,
                                              Bytef *dictionary,
                                              uInt  *dictLength))
 {
@@ -1022,7 +1024,7 @@ ZEXTERN int ZEXPORT z_inflateGetDictionary OF((z_streamp strm,
 #endif
 
 
-ZEXTERN int ZEXPORT z_inflateCopy OF((z_streamp dest,
+ZEXTERN int ZEXPORT z_inflateCopy _Z_OF((z_streamp dest,
                                     z_streamp source))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !source->reserved)
@@ -1032,7 +1034,7 @@ ZEXTERN int ZEXPORT z_inflateCopy OF((z_streamp dest,
 
 
 #if ZLIB_VERNUM >= 0x1240
-ZEXTERN long ZEXPORT z_inflateMark OF((z_streamp strm))
+ZEXTERN long ZEXPORT z_inflateMark _Z_OF((z_streamp strm))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved)
         return inflateMark(strm);
@@ -1041,7 +1043,7 @@ ZEXTERN long ZEXPORT z_inflateMark OF((z_streamp strm))
 #endif
 
 
-ZEXTERN int ZEXPORT z_inflatePrime OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflatePrime _Z_OF((z_streamp strm,
                                      int bits,
                                      int value))
 {
@@ -1051,7 +1053,7 @@ ZEXTERN int ZEXPORT z_inflatePrime OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_inflateGetHeader OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflateGetHeader _Z_OF((z_streamp strm,
                                          gz_headerp head))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved)
@@ -1060,7 +1062,7 @@ ZEXTERN int ZEXPORT z_inflateGetHeader OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_inflateBackInit_ OF((z_streamp strm, int windowBits,
+ZEXTERN int ZEXPORT z_inflateBackInit_ _Z_OF((z_streamp strm, int windowBits,
                                          unsigned char FAR *window,
                                          const char *version,
                                          int stream_size))
@@ -1071,7 +1073,7 @@ ZEXTERN int ZEXPORT z_inflateBackInit_ OF((z_streamp strm, int windowBits,
 }
 
 
-ZEXTERN int ZEXPORT z_inflateBack OF((z_streamp strm,
+ZEXTERN int ZEXPORT z_inflateBack _Z_OF((z_streamp strm,
                                     in_func in, void FAR *in_desc,
                                     out_func out, void FAR *out_desc))
 {
@@ -1081,7 +1083,7 @@ ZEXTERN int ZEXPORT z_inflateBack OF((z_streamp strm,
 }
 
 
-ZEXTERN int ZEXPORT z_inflateBackEnd OF((z_streamp strm))
+ZEXTERN int ZEXPORT z_inflateBackEnd _Z_OF((z_streamp strm))
 {
     if (g_ZWRAPdecompressionType == ZWRAP_FORCE_ZLIB || !strm->reserved)
         return inflateBackEnd(strm);
@@ -1089,14 +1091,14 @@ ZEXTERN int ZEXPORT z_inflateBackEnd OF((z_streamp strm))
 }
 
 
-ZEXTERN uLong ZEXPORT z_zlibCompileFlags OF((void)) { return zlibCompileFlags(); }
+ZEXTERN uLong ZEXPORT z_zlibCompileFlags _Z_OF((void)) { return zlibCompileFlags(); }
 
 
 
                     /* ===   utility functions  === */
 #ifndef Z_SOLO
 
-ZEXTERN int ZEXPORT z_compress OF((Bytef *dest,   uLongf *destLen,
+ZEXTERN int ZEXPORT z_compress _Z_OF((Bytef *dest,   uLongf *destLen,
                                  const Bytef *source, uLong sourceLen))
 {
     if (!g_ZWRAP_useZSTDcompression)
@@ -1115,7 +1117,7 @@ ZEXTERN int ZEXPORT z_compress OF((Bytef *dest,   uLongf *destLen,
 }
 
 
-ZEXTERN int ZEXPORT z_compress2 OF((Bytef *dest,   uLongf *destLen,
+ZEXTERN int ZEXPORT z_compress2 _Z_OF((Bytef *dest,   uLongf *destLen,
                                   const Bytef *source, uLong sourceLen,
                                   int level))
 {
@@ -1131,7 +1133,7 @@ ZEXTERN int ZEXPORT z_compress2 OF((Bytef *dest,   uLongf *destLen,
 }
 
 
-ZEXTERN uLong ZEXPORT z_compressBound OF((uLong sourceLen))
+ZEXTERN uLong ZEXPORT z_compressBound _Z_OF((uLong sourceLen))
 {
     if (!g_ZWRAP_useZSTDcompression)
         return compressBound(sourceLen);
@@ -1140,7 +1142,7 @@ ZEXTERN uLong ZEXPORT z_compressBound OF((uLong sourceLen))
 }
 
 
-ZEXTERN int ZEXPORT z_uncompress OF((Bytef *dest,   uLongf *destLen,
+ZEXTERN int ZEXPORT z_uncompress _Z_OF((Bytef *dest,   uLongf *destLen,
                                    const Bytef *source, uLong sourceLen))
 {
     if (!ZSTD_isFrame(source, sourceLen))
@@ -1159,24 +1161,24 @@ ZEXTERN int ZEXPORT z_uncompress OF((Bytef *dest,   uLongf *destLen,
 
                         /* checksum functions */
 
-ZEXTERN uLong ZEXPORT z_adler32 OF((uLong adler, const Bytef *buf, uInt len))
+ZEXTERN uLong ZEXPORT z_adler32 _Z_OF((uLong adler, const Bytef *buf, uInt len))
 {
     return adler32(adler, buf, len);
 }
 
-ZEXTERN uLong ZEXPORT z_crc32   OF((uLong crc, const Bytef *buf, uInt len))
+ZEXTERN uLong ZEXPORT z_crc32   _Z_OF((uLong crc, const Bytef *buf, uInt len))
 {
     return crc32(crc, buf, len);
 }
 
 
 #if ZLIB_VERNUM >= 0x12B0
-ZEXTERN uLong ZEXPORT z_adler32_z OF((uLong adler, const Bytef *buf, z_size_t len))
+ZEXTERN uLong ZEXPORT z_adler32_z _Z_OF((uLong adler, const Bytef *buf, z_size_t len))
 {
     return adler32_z(adler, buf, len);
 }
 
-ZEXTERN uLong ZEXPORT z_crc32_z OF((uLong crc, const Bytef *buf, z_size_t len))
+ZEXTERN uLong ZEXPORT z_crc32_z _Z_OF((uLong crc, const Bytef *buf, z_size_t len))
 {
     return crc32_z(crc, buf, len);
 }
@@ -1184,8 +1186,15 @@ ZEXTERN uLong ZEXPORT z_crc32_z OF((uLong crc, const Bytef *buf, z_size_t len))
 
 
 #if ZLIB_VERNUM >= 0x1270
-ZEXTERN const z_crc_t FAR * ZEXPORT z_get_crc_table    OF((void))
+ZEXTERN const z_crc_t FAR * ZEXPORT z_get_crc_table    _Z_OF((void))
 {
     return get_crc_table();
 }
 #endif
+
+                        /* Error function */
+ZEXTERN const char * ZEXPORT z_zError _Z_OF((int err))
+{
+    /* Just use zlib Error function */
+    return zError(err);
+}
diff --git a/priv/zstd/zlibWrapper/zstd_zlibwrapper.h b/priv/zstd/zlibWrapper/zstd_zlibwrapper.h
index 042ab9f..230bf84 100644
--- a/priv/zstd/zlibWrapper/zstd_zlibwrapper.h
+++ b/priv/zstd/zlibWrapper/zstd_zlibwrapper.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -25,6 +25,9 @@ extern "C" {
     #define z_const
 #endif
 
+#if !defined(_Z_OF)
+    #define _Z_OF OF
+#endif
 
 /* returns a string with version of zstd library */
 const char * zstdVersion(void);