diff --git a/compiler/rustc_codegen_llvm/messages.ftl b/compiler/rustc_codegen_llvm/messages.ftl index aed4a8f3c85ff..ddaff36f24b1d 100644 --- a/compiler/rustc_codegen_llvm/messages.ftl +++ b/compiler/rustc_codegen_llvm/messages.ftl @@ -83,6 +83,8 @@ codegen_llvm_unknown_ctarget_feature_prefix = unknown feature specified for `-Ctarget-feature`: `{$feature}` .note = features must begin with a `+` to enable or `-` to disable it +codegen_llvm_unknown_debuginfo_compression = unknown debuginfo compression algorithm {$algorithm} - will fall back to uncompressed debuginfo + codegen_llvm_write_bytecode = failed to write bytecode to {$path}: {$err} codegen_llvm_write_ir = failed to write LLVM IR to {$path} diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index c1d3392386cad..5cf83b1accb16 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -1,4 +1,6 @@ -use crate::back::write::{self, save_temp_bitcode, CodegenDiagnosticsStage, DiagnosticHandlers}; +use crate::back::write::{ + self, bitcode_section_name, save_temp_bitcode, CodegenDiagnosticsStage, DiagnosticHandlers, +}; use crate::errors::{ DynamicLinkingWithLTO, LlvmError, LtoBitcodeFromRlib, LtoDisallowed, LtoDylib, }; @@ -120,6 +122,7 @@ fn prepare_lto( info!("adding bitcode from {}", name); match get_bitcode_slice_from_object_data( child.data(&*archive_data).expect("corrupt rlib"), + cgcx, ) { Ok(data) => { let module = SerializedModule::FromRlib(data.to_vec()); @@ -141,10 +144,29 @@ fn prepare_lto( Ok((symbols_below_threshold, upstream_modules)) } -fn get_bitcode_slice_from_object_data(obj: &[u8]) -> Result<&[u8], LtoBitcodeFromRlib> { +fn get_bitcode_slice_from_object_data<'a>( + obj: &'a [u8], + cgcx: &CodegenContext, +) -> Result<&'a [u8], LtoBitcodeFromRlib> { + // We're about to assume the data here is an object file with sections, but if it's raw LLVM IR that + // won't work. Fortunately, if that's what we have we can just return the object directly, so we sniff + // the relevant magic strings here and return. + if obj.starts_with(b"\xDE\xC0\x17\x0B") || obj.starts_with(b"BC\xC0\xDE") { + return Ok(obj); + } + // We drop the "__LLVM," prefix here because on Apple platforms there's a notion of "segment name" + // which in the public API for sections gets treated as part of the section name, but internally + // in MachOObjectFile.cpp gets treated separately. + let section_name = bitcode_section_name(cgcx).trim_start_matches("__LLVM,"); let mut len = 0; - let data = - unsafe { llvm::LLVMRustGetBitcodeSliceFromObjectData(obj.as_ptr(), obj.len(), &mut len) }; + let data = unsafe { + llvm::LLVMRustGetSliceFromObjectDataByName( + obj.as_ptr(), + obj.len(), + section_name.as_ptr(), + &mut len, + ) + }; if !data.is_null() { assert!(len != 0); let bc = unsafe { slice::from_raw_parts(data, len) }; diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index c49d322fc0000..1f394a7335c7e 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -5,13 +5,17 @@ use crate::back::profiling::{ use crate::base; use crate::common; use crate::errors::{ - CopyBitcode, FromLlvmDiag, FromLlvmOptimizationDiag, LlvmError, WithLlvmError, WriteBytecode, + CopyBitcode, FromLlvmDiag, FromLlvmOptimizationDiag, LlvmError, UnknownCompression, + WithLlvmError, WriteBytecode, }; use crate::llvm::{self, DiagnosticInfo, PassManager}; use crate::llvm_util; use crate::type_::Type; use crate::LlvmCodegenBackend; use crate::ModuleLlvm; +use llvm::{ + LLVMRustLLVMHasZlibCompressionForDebugSymbols, LLVMRustLLVMHasZstdCompressionForDebugSymbols, +}; use rustc_codegen_ssa::back::link::ensure_removed; use rustc_codegen_ssa::back::write::{ BitcodeSection, CodegenContext, EmitObj, ModuleConfig, TargetMachineFactoryConfig, @@ -234,6 +238,22 @@ pub fn target_machine_factory( args_cstr_buff }; + let debuginfo_compression = sess.opts.debuginfo_compression.to_string(); + match sess.opts.debuginfo_compression { + rustc_session::config::DebugInfoCompression::Zlib => { + if !unsafe { LLVMRustLLVMHasZlibCompressionForDebugSymbols() } { + sess.emit_warning(UnknownCompression { algorithm: "zlib" }); + } + } + rustc_session::config::DebugInfoCompression::Zstd => { + if !unsafe { LLVMRustLLVMHasZstdCompressionForDebugSymbols() } { + sess.emit_warning(UnknownCompression { algorithm: "zstd" }); + } + } + rustc_session::config::DebugInfoCompression::None => {} + }; + let debuginfo_compression = SmallCStr::new(&debuginfo_compression); + Arc::new(move |config: TargetMachineFactoryConfig| { let split_dwarf_file = path_mapping.map_prefix(config.split_dwarf_file.unwrap_or_default()).0; @@ -259,6 +279,7 @@ pub fn target_machine_factory( relax_elf_relocations, use_init_array, split_dwarf_file.as_ptr(), + debuginfo_compression.as_ptr(), force_emulated_tls, args_cstr_buff.as_ptr() as *const c_char, args_cstr_buff.len(), @@ -873,6 +894,27 @@ fn create_section_with_flags_asm(section_name: &str, section_flags: &str, data: asm } +fn target_is_apple(cgcx: &CodegenContext) -> bool { + cgcx.opts.target_triple.triple().contains("-ios") + || cgcx.opts.target_triple.triple().contains("-darwin") + || cgcx.opts.target_triple.triple().contains("-tvos") + || cgcx.opts.target_triple.triple().contains("-watchos") +} + +fn target_is_aix(cgcx: &CodegenContext) -> bool { + cgcx.opts.target_triple.triple().contains("-aix") +} + +pub(crate) fn bitcode_section_name(cgcx: &CodegenContext) -> &'static str { + if target_is_apple(cgcx) { + "__LLVM,__bitcode\0" + } else if target_is_aix(cgcx) { + ".ipa\0" + } else { + ".llvmbc\0" + } +} + /// Embed the bitcode of an LLVM module in the LLVM module itself. /// /// This is done primarily for iOS where it appears to be standard to compile C @@ -933,11 +975,8 @@ unsafe fn embed_bitcode( // Unfortunately, LLVM provides no way to set custom section flags. For ELF // and COFF we emit the sections using module level inline assembly for that // reason (see issue #90326 for historical background). - let is_aix = cgcx.opts.target_triple.triple().contains("-aix"); - let is_apple = cgcx.opts.target_triple.triple().contains("-ios") - || cgcx.opts.target_triple.triple().contains("-darwin") - || cgcx.opts.target_triple.triple().contains("-tvos") - || cgcx.opts.target_triple.triple().contains("-watchos"); + let is_aix = target_is_aix(cgcx); + let is_apple = target_is_apple(cgcx); if is_apple || is_aix || cgcx.opts.target_triple.triple().starts_with("wasm") @@ -952,13 +991,7 @@ unsafe fn embed_bitcode( ); llvm::LLVMSetInitializer(llglobal, llconst); - let section = if is_apple { - "__LLVM,__bitcode\0" - } else if is_aix { - ".ipa\0" - } else { - ".llvmbc\0" - }; + let section = bitcode_section_name(cgcx); llvm::LLVMSetSection(llglobal, section.as_ptr().cast()); llvm::LLVMRustSetLinkage(llglobal, llvm::Linkage::PrivateLinkage); llvm::LLVMSetGlobalConstant(llglobal, llvm::True); diff --git a/compiler/rustc_codegen_llvm/src/errors.rs b/compiler/rustc_codegen_llvm/src/errors.rs index fced6d504d26e..264c273ba3072 100644 --- a/compiler/rustc_codegen_llvm/src/errors.rs +++ b/compiler/rustc_codegen_llvm/src/errors.rs @@ -226,3 +226,9 @@ pub(crate) struct WriteBytecode<'a> { pub(crate) struct CopyBitcode { pub err: std::io::Error, } + +#[derive(Diagnostic)] +#[diag(codegen_llvm_unknown_debuginfo_compression)] +pub struct UnknownCompression { + pub algorithm: &'static str, +} diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs index 1f03c932cfdc5..2ebfdae39e805 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs @@ -2131,6 +2131,7 @@ extern "C" { RelaxELFRelocations: bool, UseInitArray: bool, SplitDwarfFile: *const c_char, + DebugInfoCompression: *const c_char, ForceEmulatedTls: bool, ArgsCstrBuff: *const c_char, ArgsCstrBuffLen: usize, @@ -2322,6 +2323,12 @@ extern "C" { len: usize, out_len: &mut usize, ) -> *const u8; + pub fn LLVMRustGetSliceFromObjectDataByName( + data: *const u8, + len: usize, + name: *const u8, + out_len: &mut usize, + ) -> *const u8; pub fn LLVMRustLinkerNew(M: &Module) -> &mut Linker<'_>; pub fn LLVMRustLinkerAdd( @@ -2360,6 +2367,10 @@ extern "C" { pub fn LLVMRustIsBitcode(ptr: *const u8, len: usize) -> bool; + pub fn LLVMRustLLVMHasZlibCompressionForDebugSymbols() -> bool; + + pub fn LLVMRustLLVMHasZstdCompressionForDebugSymbols() -> bool; + pub fn LLVMRustGetSymbols( buf_ptr: *const u8, buf_len: usize, diff --git a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp index 4ed086ff42fa8..87ef74b9264a3 100644 --- a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp +++ b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp @@ -1,5 +1,6 @@ #include +#include #include #include #include @@ -406,6 +407,7 @@ extern "C" LLVMTargetMachineRef LLVMRustCreateTargetMachine( bool RelaxELFRelocations, bool UseInitArray, const char *SplitDwarfFile, + const char *DebugInfoCompression, bool ForceEmulatedTls, const char *ArgsCstrBuff, size_t ArgsCstrBuffLen) { @@ -437,6 +439,16 @@ extern "C" LLVMTargetMachineRef LLVMRustCreateTargetMachine( if (SplitDwarfFile) { Options.MCOptions.SplitDwarfFile = SplitDwarfFile; } +#if LLVM_VERSION_GE(16, 0) + if (!strcmp("zlib", DebugInfoCompression) && llvm::compression::zlib::isAvailable()) { + Options.CompressDebugSections = DebugCompressionType::Zlib; + } else if (!strcmp("zstd", DebugInfoCompression) && llvm::compression::zstd::isAvailable()) { + Options.CompressDebugSections = DebugCompressionType::Zstd; + } else if (!strcmp("none", DebugInfoCompression)) { + Options.CompressDebugSections = DebugCompressionType::None; + } +#endif + Options.RelaxELFRelocations = RelaxELFRelocations; Options.UseInitArray = UseInitArray; @@ -1555,6 +1567,38 @@ LLVMRustGetBitcodeSliceFromObjectData(const char *data, return BitcodeOrError->getBufferStart(); } +// Find a section of an object file by name. Fail if the section is missing or +// empty. +extern "C" const char *LLVMRustGetSliceFromObjectDataByName(const char *data, + size_t len, + const char *name, + size_t *out_len) { + *out_len = 0; + StringRef Data(data, len); + MemoryBufferRef Buffer(Data, ""); // The id is unused. + file_magic Type = identify_magic(Buffer.getBuffer()); + Expected> ObjFileOrError = + object::ObjectFile::createObjectFile(Buffer, Type); + if (!ObjFileOrError) { + LLVMRustSetLastError(toString(ObjFileOrError.takeError()).c_str()); + return nullptr; + } + for (const object::SectionRef &Sec : (*ObjFileOrError)->sections()) { + Expected Name = Sec.getName(); + if (Name && *Name == name) { + Expected SectionOrError = Sec.getContents(); + if (!SectionOrError) { + LLVMRustSetLastError(toString(SectionOrError.takeError()).c_str()); + return nullptr; + } + *out_len = SectionOrError->size(); + return SectionOrError->data(); + } + } + LLVMRustSetLastError("could not find requested section"); + return nullptr; +} + // Computes the LTO cache key for the provided 'ModId' in the given 'Data', // storing the result in 'KeyOut'. // Currently, this cache key is a SHA-1 hash of anything that could affect diff --git a/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp index 70cdf3d6d2395..4390486b0deb1 100644 --- a/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp +++ b/compiler/rustc_llvm/llvm-wrapper/RustWrapper.cpp @@ -2044,3 +2044,19 @@ extern "C" bool LLVMRustIsNonGVFunctionPointerTy(LLVMValueRef V) { } return false; } + +extern "C" bool LLVMRustLLVMHasZlibCompressionForDebugSymbols() { +#if LLVM_VERSION_GE(16, 0) + return llvm::compression::zlib::isAvailable(); +#else + return false; +#endif +} + +extern "C" bool LLVMRustLLVMHasZstdCompressionForDebugSymbols() { +#if LLVM_VERSION_GE(16, 0) + return llvm::compression::zstd::isAvailable(); +#else + return false; +#endif +} diff --git a/compiler/rustc_session/src/config.rs b/compiler/rustc_session/src/config.rs index 302e4e55d597d..86a03124fe182 100644 --- a/compiler/rustc_session/src/config.rs +++ b/compiler/rustc_session/src/config.rs @@ -381,6 +381,24 @@ pub enum DebugInfo { Full, } +#[derive(Clone, Copy, Debug, PartialEq, Hash)] +pub enum DebugInfoCompression { + None, + Zlib, + Zstd, +} + +impl ToString for DebugInfoCompression { + fn to_string(&self) -> String { + match self { + DebugInfoCompression::None => "none", + DebugInfoCompression::Zlib => "zlib", + DebugInfoCompression::Zstd => "zstd", + } + .to_owned() + } +} + /// Split debug-information is enabled by `-C split-debuginfo`, this enum is only used if split /// debug-information is enabled (in either `Packed` or `Unpacked` modes), and the platform /// uses DWARF for debug-information. @@ -1015,6 +1033,7 @@ impl Default for Options { crate_types: Vec::new(), optimize: OptLevel::No, debuginfo: DebugInfo::None, + debuginfo_compression: DebugInfoCompression::None, lint_opts: Vec::new(), lint_cap: None, describe_lints: false, @@ -2277,6 +2296,13 @@ fn select_debuginfo(matches: &getopts::Matches, cg: &CodegenOptions) -> DebugInf if max_g > max_c { DebugInfo::Full } else { cg.debuginfo } } +fn select_debuginfo_compression( + _handler: &EarlyErrorHandler, + unstable_opts: &UnstableOptions, +) -> DebugInfoCompression { + unstable_opts.debuginfo_compression +} + pub(crate) fn parse_assert_incr_state( handler: &EarlyErrorHandler, opt_assertion: &Option, @@ -2752,6 +2778,8 @@ pub fn build_session_options( // for more details. let debug_assertions = cg.debug_assertions.unwrap_or(opt_level == OptLevel::No); let debuginfo = select_debuginfo(matches, &cg); + let debuginfo_compression: DebugInfoCompression = + select_debuginfo_compression(handler, &unstable_opts); let mut search_paths = vec![]; for s in &matches.opt_strs("L") { @@ -2828,6 +2856,7 @@ pub fn build_session_options( crate_types, optimize: opt_level, debuginfo, + debuginfo_compression, lint_opts, lint_cap, describe_lints, @@ -3113,11 +3142,11 @@ impl PpMode { /// how the hash should be calculated when adding a new command-line argument. pub(crate) mod dep_tracking { use super::{ - BranchProtection, CFGuard, CFProtection, CrateType, DebugInfo, ErrorOutputType, - InstrumentCoverage, InstrumentXRay, LdImpl, LinkerPluginLto, LocationDetail, LtoCli, - OomStrategy, OptLevel, OutFileName, OutputType, OutputTypes, Passes, ResolveDocLinks, - SourceFileHashAlgorithm, SplitDwarfKind, SwitchWithOptPath, SymbolManglingVersion, - TraitSolver, TrimmedDefPaths, + BranchProtection, CFGuard, CFProtection, CrateType, DebugInfo, DebugInfoCompression, + ErrorOutputType, InstrumentCoverage, InstrumentXRay, LdImpl, LinkerPluginLto, + LocationDetail, LtoCli, OomStrategy, OptLevel, OutFileName, OutputType, OutputTypes, + Passes, ResolveDocLinks, SourceFileHashAlgorithm, SplitDwarfKind, SwitchWithOptPath, + SymbolManglingVersion, TraitSolver, TrimmedDefPaths, }; use crate::lint; use crate::options::WasiExecModel; @@ -3195,6 +3224,7 @@ pub(crate) mod dep_tracking { OptLevel, LtoCli, DebugInfo, + DebugInfoCompression, UnstableFeatures, NativeLib, NativeLibKind, diff --git a/compiler/rustc_session/src/options.rs b/compiler/rustc_session/src/options.rs index 40099de707be6..6c26859228c23 100644 --- a/compiler/rustc_session/src/options.rs +++ b/compiler/rustc_session/src/options.rs @@ -139,6 +139,7 @@ top_level_options!( /// can influence whether overflow checks are done or not. debug_assertions: bool [TRACKED], debuginfo: DebugInfo [TRACKED], + debuginfo_compression: DebugInfoCompression [TRACKED], lint_opts: Vec<(String, lint::Level)> [TRACKED_NO_CRATE_HASH], lint_cap: Option [TRACKED_NO_CRATE_HASH], describe_lints: bool [UNTRACKED], @@ -376,6 +377,7 @@ mod desc { "either a boolean (`yes`, `no`, `on`, `off`, etc), `checks`, or `nochecks`"; pub const parse_cfprotection: &str = "`none`|`no`|`n` (default), `branch`, `return`, or `full`|`yes`|`y` (equivalent to `branch` and `return`)"; pub const parse_debuginfo: &str = "either an integer (0, 1, 2), `none`, `line-directives-only`, `line-tables-only`, `limited`, or `full`"; + pub const parse_debuginfo_compression: &str = "one of `none`, `zlib`, or `zstd`"; pub const parse_strip: &str = "either `none`, `debuginfo`, or `symbols`"; pub const parse_linker_flavor: &str = ::rustc_target::spec::LinkerFlavorCli::one_of(); pub const parse_optimization_fuel: &str = "crate=integer"; @@ -782,6 +784,19 @@ mod parse { true } + pub(crate) fn parse_debuginfo_compression( + slot: &mut DebugInfoCompression, + v: Option<&str>, + ) -> bool { + match v { + Some("none") => *slot = DebugInfoCompression::None, + Some("zlib") => *slot = DebugInfoCompression::Zlib, + Some("zstd") => *slot = DebugInfoCompression::Zstd, + _ => return false, + }; + true + } + pub(crate) fn parse_linker_flavor(slot: &mut Option, v: Option<&str>) -> bool { match v.and_then(LinkerFlavorCli::from_str) { Some(lf) => *slot = Some(lf), @@ -1424,6 +1439,8 @@ options! { "emit discriminators and other data necessary for AutoFDO"), debug_macros: bool = (false, parse_bool, [TRACKED], "emit line numbers debug info inside macros (default: no)"), + debuginfo_compression: DebugInfoCompression = (DebugInfoCompression::None, parse_debuginfo_compression, [TRACKED], + "compress debug info sections (none, zlib, zstd, default: none)"), deduplicate_diagnostics: bool = (true, parse_bool, [UNTRACKED], "deduplicate identical diagnostics (default: yes)"), dep_info_omit_d_target: bool = (false, parse_bool, [TRACKED], diff --git a/tests/run-make/compressed-debuginfo/Makefile b/tests/run-make/compressed-debuginfo/Makefile new file mode 100644 index 0000000000000..f9e4927d008e4 --- /dev/null +++ b/tests/run-make/compressed-debuginfo/Makefile @@ -0,0 +1,15 @@ +# ignore-cross-compile +include ../tools.mk + +# only-linux +# min-llvm-version: 16.0 +# +# This tests debuginfo-compression. + +all: zlib zstandard + +zlib: + test "`$(RUSTC) --crate-name=foo --crate-type=lib --emit=obj -C debuginfo=full -Z debuginfo-compression=zlib foo.rs 2>&1 | sed 's/.*unknown.*zlib.*/missing/' | head -n 1`" = missing || readelf -t $(TMPDIR)/foo.o | grep -q ZLIB + +zstandard: + test "`$(RUSTC) --crate-name=foo --crate-type=lib --emit=obj -C debuginfo=full -Z debuginfo-compression=zstd foo.rs 2>&1 | sed 's/.*unknown.*zstd.*/missing/' | head -n 1`" = missing || readelf -t $(TMPDIR)/foo.o | grep -q ZST diff --git a/tests/run-make/compressed-debuginfo/foo.rs b/tests/run-make/compressed-debuginfo/foo.rs new file mode 100644 index 0000000000000..185ce22450c90 --- /dev/null +++ b/tests/run-make/compressed-debuginfo/foo.rs @@ -0,0 +1,3 @@ +pub fn foo() -> i32 { + 42 +}