Add support to read plain encoded INT96 timestamp from Parquet file (o…

…ap-project#456)

velox/dwio/common/DirectDecoder.h

@@ -98,7 +98,17 @@ 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 /* INT96 */) {
+          int128_t encoded = super::template readInt<int128_t>();
+          int32_t days = encoded & ((1ULL << 32) - 1);
+          uint64_t nanos = static_cast<uint64_t>(encoded >> 32);
+
+          auto timestamp = Timestamp::fromDaysAndNanos(days, nanos);
+          toSkip =
+              visitor.process(*reinterpret_cast<int128_t*>(&timestamp), atEnd);
+        } else {
+          toSkip = visitor.process(super::template readInt<int128_t>(), atEnd);
+        }
       } else {
         toSkip = visitor.process(super::template readInt<int64_t>(), atEnd);
       }

velox/dwio/common/IntDecoder.h

@@ -167,6 +167,9 @@ class IntDecoder {
   template <typename T>
   T readInt();
 
+  template <typename T>
+  T readInt96();
+
   template <typename T>
   T readVInt();
 
@@ -453,12 +456,44 @@ inline T IntDecoder<isSigned>::readInt() {
     return readLittleEndianFromBigEndian<T>();
   } else {
     if constexpr (std::is_same_v<T, int128_t>) {
-      VELOX_NYI();
+      if (numBytes == 12) {
+        // TODO:: Do we need to handle useVInts case?
+        return readInt96<T>();
+      } else {
+        VELOX_NYI();
+      }
     }
     return readLongLE();
   }
 }
 
+template <bool isSigned>
+template <typename T>
+inline T IntDecoder<isSigned>::readInt96() {
+  int64_t offset = 0;
+  unsigned char ch;
+
+  // read unsigned byte 64
+  uint64_t part1 = 0;
+  for (uint32_t i = 0; i < 8; ++i) {
+    ch = readByte();
+    part1 |= (ch & BASE_256_MASK) << offset;
+    offset += 8;
+  }
+
+  // read signed byte 32
+  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() {

velox/dwio/parquet/reader/PageReader.cpp

@@ -414,11 +414,7 @@ void PageReader::prepareDictionary(const PageHeader& pageHeader) {
       // We start from the end to allow in-place expansion.
       auto values = dictionary_.values->asMutable<Timestamp>();
       auto parquetValues = dictionary_.values->asMutable<char>();
-      static constexpr int64_t kJulianToUnixEpochDays = 2440588LL;
-      static constexpr int64_t kSecondsPerDay = 86400LL;
-      static constexpr int64_t kNanosPerSecond =
-          Timestamp::kNanosecondsInMillisecond *
-          Timestamp::kMillisecondsInSecond;
+
       for (auto i = dictionary_.numValues - 1; i >= 0; --i) {
         // Convert the timestamp into seconds and nanos since the Unix epoch,
         // 00:00:00.000000 on 1 January 1970.
@@ -432,12 +428,8 @@ void PageReader::prepareDictionary(const PageHeader& pageHeader) {
             &days,
             parquetValues + i * sizeof(Int96Timestamp) + sizeof(uint64_t),
             sizeof(int32_t));
-        int64_t seconds = (days - kJulianToUnixEpochDays) * kSecondsPerDay;
-        if (nanos > Timestamp::kMaxNanos) {
-          seconds += nanos / kNanosPerSecond;
-          nanos -= (nanos / kNanosPerSecond) * kNanosPerSecond;
-        }
-        values[i] = Timestamp(seconds, nanos);
+
+        values[i] = Timestamp::fromDaysAndNanos(days, nanos);
       }
       break;
     }

velox/dwio/parquet/tests/reader/ParquetTableScanTest.cpp

@@ -625,6 +625,33 @@ TEST_F(ParquetTableScanTest, timestampFilter) {
       "testInt128() is not supported");
 }
 
+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(&argc, &argv, false);

velox/functions/sparksql/tests/DecimalUtilTest.cpp

@@ -30,7 +30,7 @@ class DecimalUtilTest : public testing::Test {
       R expectedResult,
       bool expectedOverflow) {
     R r;
-    bool overflow;
+    bool overflow = false;
     DecimalUtil::divideWithRoundUp<R, A, B>(r, a, b, aRescale, overflow);
     ASSERT_EQ(overflow, expectedOverflow);
     ASSERT_EQ(r, expectedResult);

velox/type/Timestamp.h

@@ -100,6 +100,21 @@ struct Timestamp {
     VELOX_USER_DCHECK_LE(nanos, kMaxNanos, "Timestamp nanos out of range");
   }
 
+  static Timestamp fromDaysAndNanos(int32_t days, uint64_t nanos) {
+    static constexpr int64_t kJulianToUnixEpochDays = 2440588LL;
+    static constexpr int64_t kSecondsPerDay = 86400LL;
+    static constexpr int64_t kNanosPerSecond =
+        Timestamp::kNanosecondsInMillisecond * Timestamp::kMillisecondsInSecond;
+
+    int64_t seconds = (days - kJulianToUnixEpochDays) * kSecondsPerDay;
+    if (nanos > Timestamp::kMaxNanos) {
+      seconds += nanos / kNanosPerSecond;
+      nanos -= (nanos / kNanosPerSecond) * kNanosPerSecond;
+    }
+
+    return Timestamp(seconds, nanos);
+  }
+
   // Returns the current unix timestamp (ms precision).
   static Timestamp now();