From 5f9bd7d39b8b491a6ffa28f6edce85f9c1f3fc64 Mon Sep 17 00:00:00 2001
From: Kapil Singh <kapilsingh@microsoft.com>
Date: Fri, 8 Dec 2023 12:13:48 +0530
Subject: [PATCH] Add support to read plain encoded INT96 timestamp from
 Parquet file

---
 velox/dwio/common/DirectDecoder.h             |  12 +++++-
 velox/dwio/common/IntDecoder.h                |  36 +++++++++++++++++-
 .../examples/timestamp_dict_int96.parquet     | Bin 0 -> 467 bytes
 .../examples/timestamp_plain_int96.parquet    | Bin 0 -> 429 bytes
 .../parquet/tests/reader/E2EFilterTest.cpp    |  14 +++++++
 .../tests/reader/ParquetTableScanTest.cpp     |  28 ++++++++++++++
 6 files changed, 88 insertions(+), 2 deletions(-)
 create mode 100644 velox/dwio/parquet/tests/examples/timestamp_dict_int96.parquet
 create mode 100644 velox/dwio/parquet/tests/examples/timestamp_plain_int96.parquet

diff --git a/velox/dwio/common/DirectDecoder.h b/velox/dwio/common/DirectDecoder.h
index 4cd9396d0936b..c175ffebbe7b2 100644
--- a/velox/dwio/common/DirectDecoder.h
+++ b/velox/dwio/common/DirectDecoder.h
@@ -92,7 +92,17 @@ class DirectDecoder : public IntDecoder<isSigned> {
       } else if constexpr (std::is_same_v<
                                typename Visitor::DataType,
                                int128_t>) {
-        toSkip = visitor.process(super::template readInt<int128_t>(), atEnd);
+        if (super::numBytes != 12) {
+          toSkip = visitor.process(super::template readInt<int128_t>(), atEnd);
+        } else {
+          // Reads INT96 timestamp as int128_t type and extracts the days and
+          // nanos.
+          const int128_t encoded = super::template readInt<int128_t>();
+          const int32_t days = encoded & ((1ULL << 32) - 1);
+          const uint64_t nanos = static_cast<uint64_t>(encoded >> 32);
+          auto ts = Timestamp::fromDaysAndNanos(days, nanos);
+          toSkip = visitor.process(reinterpret_cast<int128_t&>(ts), atEnd);
+        }
       } else {
         toSkip = visitor.process(super::template readInt<int64_t>(), atEnd);
       }
diff --git a/velox/dwio/common/IntDecoder.h b/velox/dwio/common/IntDecoder.h
index e5e37b429663d..8bbc0751b3169 100644
--- a/velox/dwio/common/IntDecoder.h
+++ b/velox/dwio/common/IntDecoder.h
@@ -154,6 +154,9 @@ class IntDecoder {
   template <typename T>
   T readInt();
 
+  // Reads Int96 timestamp composed of days and nanos as int128_t.
+  int128_t readInt96();
+
   template <typename T>
   T readVInt();
 
@@ -438,12 +441,43 @@ inline T IntDecoder<isSigned>::readInt() {
     return readLittleEndianFromBigEndian<T>();
   } else {
     if constexpr (std::is_same_v<T, int128_t>) {
-      VELOX_NYI();
+      if (numBytes == 12) {
+        VELOX_DCHECK(!useVInts, "Int96 should not be VInt encoded.");
+        return readInt96();
+      } else {
+        VELOX_NYI();
+      }
     }
     return readLongLE();
   }
 }
 
+template <bool isSigned>
+inline int128_t IntDecoder<isSigned>::readInt96() {
+  int64_t offset = 0;
+  unsigned char ch;
+
+  // Read 8 unsigned bytes.
+  uint64_t part1 = 0;
+  for (uint32_t i = 0; i < 8; ++i) {
+    ch = readByte();
+    part1 |= (ch & BASE_256_MASK) << offset;
+    offset += 8;
+  }
+
+  // Read 4 signed bytes.
+  int32_t part2 = 0;
+  offset = 0;
+  for (uint32_t i = 0; i < 4; ++i) {
+    ch = readByte();
+    part2 |= (ch & BASE_256_MASK) << offset;
+    offset += 8;
+  }
+
+  int128_t result = part1;
+  return (result << 32) | part2;
+}
+
 template <bool isSigned>
 template <typename T>
 inline T IntDecoder<isSigned>::readVInt() {
diff --git a/velox/dwio/parquet/tests/examples/timestamp_dict_int96.parquet b/velox/dwio/parquet/tests/examples/timestamp_dict_int96.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..661cb7a285227f8ead6c89fc829fc7cc99cbf37c
GIT binary patch
literal 467
zcmZXRze~eF6vtnFSSmP4Uq~QBIB2m@kEUsAsX7P_;#6E!<Sw~tG3k#awMgmW>?jTn
zx`=;_Aa3Gc;_RiilQ+HZd!P5d@9w$>Cobw($0mN>y}oFBQs@BY8N5I3f;{*8Wq=2N
zx{48S6Ak?M{`%4E9L53`0oEN=5}KxHs%0A-4-F|%wQ5Hj>4+uQ7in-MVumVUp{$yE
z){g}kx@B8-h1*!L{DNMz{_?0Ay78zO&q{??igpXYsx6#kGsr+X8E*=fAcyoyQG%tM
zG0kCeN!f@=E4px{HzG;<!-2WxQIEP{R{s&{IHGdF@kC@a8BS>|G8VFonddW--Hrt5
zkTgpsK}Kq%*B4QklFsRj3|O3(@@vS#63kVav3Mj?gGm%I9!W@ciyB$YC7s!Pem)2D
zc<H(wC(d$B$E9wAdd?P~^rO(}wZlM!Zo3gQ1GmRR5wwDUdv4qJy&dkex+_{DZ1JjP
N+w!#s3P4wW=P&+$b_oCg

literal 0
HcmV?d00001

diff --git a/velox/dwio/parquet/tests/examples/timestamp_plain_int96.parquet b/velox/dwio/parquet/tests/examples/timestamp_plain_int96.parquet
new file mode 100644
index 0000000000000000000000000000000000000000..f2aa666b7d7105249ce99d1ef3c549eefbffaa1f
GIT binary patch
literal 429
zcmZWmK}y3w6#b^LL~z*&2@DbjS}erDX-q7&g6_msaVte9$yB3BnkJJLDZPP9SDwL@
z$M6>3!2>wex^nm5e{bG<{285JF$8Fc&d}S}`{xcux`It)0PIN(02&SWcs)@AJfF4!
zUi`)1Y-7_vxv3!1Z1-ZfDtP&LQ$_b8<498`ZOzhUk_j+%8{4>JVAQnzCK&eb9%-g-
zKA3sO7zO|~R>8^&z?`ENC_?WTX5LbaYgyCsj`D&lbNZ7*=c25VX>M(M)T0cn?mtB}
zOR1_ftA(WHbU`yAdCVoZ#!Dif3PA>>l4Tu9;*v=s(zqgn>m|u~wx-s_6|oVnQ6+g+
zs8wF4DGyU6Qm=(e{(Q+`xmt}^U|nuf53<tP)N~kSJ?c68VV$J0GwH{Xh*`fE9Yt&s
i#v%%$DD+t0_r2rL=Z8!LA`Zf?jj_7#?H0gDebhJPYH$Gn

literal 0
HcmV?d00001

diff --git a/velox/dwio/parquet/tests/reader/E2EFilterTest.cpp b/velox/dwio/parquet/tests/reader/E2EFilterTest.cpp
index 0b27395f9a00d..1634da0556c2f 100644
--- a/velox/dwio/parquet/tests/reader/E2EFilterTest.cpp
+++ b/velox/dwio/parquet/tests/reader/E2EFilterTest.cpp
@@ -256,6 +256,20 @@ TEST_F(E2EFilterTest, integerDictionary) {
       20);
 }
 
+TEST_F(E2EFilterTest, timestampDirect) {
+  options_.enableDictionary = false;
+  options_.dataPageSize = 4 * 1024;
+  options_.writeInt96AsTimestamp = true;
+
+  testWithTypes(
+      "timestamp_val_0:timestamp,"
+      "timestamp_val_1:timestamp",
+      [&]() {},
+      true,
+      {"timestamp_val_0", "timestamp_val_1"},
+      20);
+}
+
 TEST_F(E2EFilterTest, timestampDictionary) {
   options_.dataPageSize = 4 * 1024;
   options_.writeInt96AsTimestamp = true;
diff --git a/velox/dwio/parquet/tests/reader/ParquetTableScanTest.cpp b/velox/dwio/parquet/tests/reader/ParquetTableScanTest.cpp
index 4261ee7022490..0a96227f19867 100644
--- a/velox/dwio/parquet/tests/reader/ParquetTableScanTest.cpp
+++ b/velox/dwio/parquet/tests/reader/ParquetTableScanTest.cpp
@@ -836,6 +836,34 @@ TEST_F(ParquetTableScanTest, timestampPrecisionMicrosecond) {
   assertEqualResults({expected}, result.second);
 }
 
+
+TEST_F(ParquetTableScanTest, timestampINT96) {
+  auto a = makeFlatVector<Timestamp>({Timestamp(1, 0), Timestamp(2, 0)});
+  auto expected = makeRowVector({"time"}, {a});
+  createDuckDbTable("expected", {expected});
+
+  auto vector = makeArrayVector<Timestamp>({{}});
+  loadData(
+      getExampleFilePath("timestamp_dict_int96.parquet"),
+      ROW({"time"}, {TIMESTAMP()}),
+      makeRowVector(
+          {"time"},
+          {
+              vector,
+          }));
+  assertSelect({"time"}, "SELECT time from expected");
+
+  loadData(
+      getExampleFilePath("timestamp_plain_int96.parquet"),
+      ROW({"time"}, {TIMESTAMP()}),
+      makeRowVector(
+          {"time"},
+          {
+              vector,
+          }));
+  assertSelect({"time"}, "SELECT time from expected");
+}
+
 int main(int argc, char** argv) {
   testing::InitGoogleTest(&argc, argv);
   folly::Init init{&argc, &argv, false};