Perform scoring for 4 and 7 bit quantized vectors off-heap

Author: Kaival Parikh

lucene/CHANGES.txt

@@ -136,6 +136,8 @@ Optimizations
 * GITHUB#15151: Use `SimScorer#score` bulk API to compute impact scores per
   block of postings. (Adrien Grand)
 
+* GITHUB#14863: Perform scoring for 4 and 7 bit quantized vectors off-heap. (Kaival Parikh)
+
 Bug Fixes
 ---------------------
 * GITHUB#14161: PointInSetQuery's constructor now throws IllegalArgumentException

lucene/benchmark-jmh/src/java/org/apache/lucene/benchmark/jmh/VectorUtilBenchmark.java

@@ -54,11 +54,13 @@ static void compressBytes(byte[] raw, byte[] compressed) {
   private byte[] bytesA;
   private byte[] bytesB;
   private byte[] halfBytesA;
+  private byte[] halfBytesAPacked;
   private byte[] halfBytesB;
   private byte[] halfBytesBPacked;
   private float[] floatsA;
   private float[] floatsB;
-  private int expectedhalfByteDotProduct;
+  private int expectedHalfByteDotProduct;
+  private int expectedHalfByteSquareDistance;
 
   @Param({"1", "128", "207", "256", "300", "512", "702", "1024"})
   int size;
@@ -74,16 +76,23 @@ public void init() {
     random.nextBytes(bytesB);
     // random half byte arrays for binary methods
     // this means that all values must be between 0 and 15
-    expectedhalfByteDotProduct = 0;
+    expectedHalfByteDotProduct = 0;
+    expectedHalfByteSquareDistance = 0;
     halfBytesA = new byte[size];
     halfBytesB = new byte[size];
     for (int i = 0; i < size; ++i) {
       halfBytesA[i] = (byte) random.nextInt(16);
       halfBytesB[i] = (byte) random.nextInt(16);
-      expectedhalfByteDotProduct += halfBytesA[i] * halfBytesB[i];
+      expectedHalfByteDotProduct += halfBytesA[i] * halfBytesB[i];
+
+      int diff = halfBytesA[i] - halfBytesB[i];
+      expectedHalfByteSquareDistance += diff * diff;
     }
     // pack the half byte arrays
     if (size % 2 == 0) {
+      halfBytesAPacked = new byte[(size + 1) >> 1];
+      compressBytes(halfBytesA, halfBytesAPacked);
+
       halfBytesBPacked = new byte[(size + 1) >> 1];
       compressBytes(halfBytesB, halfBytesBPacked);
     }
@@ -108,6 +117,74 @@ public float binaryCosineVector() {
     return VectorUtil.cosine(bytesA, bytesB);
   }
 
+  @Benchmark
+  public int binarySquareScalar() {
+    return VectorUtil.squareDistance(bytesA, bytesB);
+  }
+
+  @Benchmark
+  @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
+  public int binarySquareVector() {
+    return VectorUtil.squareDistance(bytesA, bytesB);
+  }
+
+  @Benchmark
+  public int binaryHalfByteSquareScalar() {
+    int v = VectorUtil.int4SquareDistance(halfBytesA, halfBytesB);
+    if (v != expectedHalfByteSquareDistance) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
+  }
+
+  @Benchmark
+  @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
+  public int binaryHalfByteSquareVector() {
+    int v = VectorUtil.int4SquareDistance(halfBytesA, halfBytesB);
+    if (v != expectedHalfByteSquareDistance) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
+  }
+
+  @Benchmark
+  public int binaryHalfByteSquareSinglePackedScalar() {
+    int v = VectorUtil.int4SquareDistanceSinglePacked(halfBytesA, halfBytesBPacked);
+    if (v != expectedHalfByteSquareDistance) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
+  }
+
+  @Benchmark
+  @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
+  public int binaryHalfByteSquareSinglePackedVector() {
+    int v = VectorUtil.int4SquareDistanceSinglePacked(halfBytesA, halfBytesBPacked);
+    if (v != expectedHalfByteSquareDistance) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
+  }
+
+  @Benchmark
+  public int binaryHalfByteSquareBothPackedScalar() {
+    int v = VectorUtil.int4SquareDistanceBothPacked(halfBytesAPacked, halfBytesBPacked);
+    if (v != expectedHalfByteSquareDistance) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
+  }
+
+  @Benchmark
+  @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
+  public int binaryHalfByteSquareBothPackedVector() {
+    int v = VectorUtil.int4SquareDistanceBothPacked(halfBytesAPacked, halfBytesBPacked);
+    if (v != expectedHalfByteSquareDistance) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
+  }
+
   @Benchmark
   public int binaryDotProductScalar() {
     return VectorUtil.dotProduct(bytesA, bytesB);
@@ -131,14 +208,22 @@ public int binaryDotProductUint8Vector() {
   }
 
   @Benchmark
-  public int binarySquareScalar() {
-    return VectorUtil.squareDistance(bytesA, bytesB);
+  public int binaryHalfByteDotProductScalar() {
+    int v = VectorUtil.int4DotProduct(halfBytesA, halfBytesB);
+    if (v != expectedHalfByteDotProduct) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
   }
 
   @Benchmark
   @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
-  public int binarySquareVector() {
-    return VectorUtil.squareDistance(bytesA, bytesB);
+  public int binaryHalfByteDotProductVector() {
+    int v = VectorUtil.int4DotProduct(halfBytesA, halfBytesB);
+    if (v != expectedHalfByteDotProduct) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
   }
 
   @Benchmark
@@ -153,37 +238,39 @@ public int binarySquareUint8Vector() {
   }
 
   @Benchmark
-  public int binaryHalfByteScalar() {
-    return VectorUtil.int4DotProduct(halfBytesA, halfBytesB);
+  public int binaryHalfByteDotProductSinglePackedScalar() {
+    int v = VectorUtil.int4DotProductSinglePacked(halfBytesA, halfBytesBPacked);
+    if (v != expectedHalfByteDotProduct) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
   }
 
   @Benchmark
   @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
-  public int binaryHalfByteVector() {
-    return VectorUtil.int4DotProduct(halfBytesA, halfBytesB);
+  public int binaryHalfByteDotProductSinglePackedVector() {
+    int v = VectorUtil.int4DotProductSinglePacked(halfBytesA, halfBytesBPacked);
+    if (v != expectedHalfByteDotProduct) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
+    }
+    return v;
   }
 
   @Benchmark
-  public int binaryHalfByteScalarPacked() {
-    if (size % 2 != 0) {
-      throw new RuntimeException("Size must be even for this benchmark");
-    }
-    int v = VectorUtil.int4DotProductPacked(halfBytesA, halfBytesBPacked);
-    if (v != expectedhalfByteDotProduct) {
-      throw new RuntimeException("Expected " + expectedhalfByteDotProduct + " but got " + v);
+  public int binaryHalfByteDotProductBothPackedScalar() {
+    int v = VectorUtil.int4DotProductBothPacked(halfBytesAPacked, halfBytesBPacked);
+    if (v != expectedHalfByteDotProduct) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
     }
     return v;
   }
 
   @Benchmark
   @Fork(jvmArgsPrepend = {"--add-modules=jdk.incubator.vector"})
-  public int binaryHalfByteVectorPacked() {
-    if (size % 2 != 0) {
-      throw new RuntimeException("Size must be even for this benchmark");
-    }
-    int v = VectorUtil.int4DotProductPacked(halfBytesA, halfBytesBPacked);
-    if (v != expectedhalfByteDotProduct) {
-      throw new RuntimeException("Expected " + expectedhalfByteDotProduct + " but got " + v);
+  public int binaryHalfByteDotProductBothPackedVector() {
+    int v = VectorUtil.int4DotProductBothPacked(halfBytesAPacked, halfBytesBPacked);
+    if (v != expectedHalfByteDotProduct) {
+      throw new RuntimeException("Expected " + expectedHalfByteDotProduct + " but got " + v);
     }
     return v;
   }

lucene/core/src/java/org/apache/lucene/codecs/hnsw/FlatVectorScorerUtil.java

@@ -37,4 +37,8 @@ private FlatVectorScorerUtil() {}
   public static FlatVectorsScorer getLucene99FlatVectorsScorer() {
     return IMPL.getLucene99FlatVectorsScorer();
   }
+
+  public static FlatVectorsScorer getLucene99ScalarQuantizedVectorsScorer() {
+    return IMPL.getLucene99ScalarQuantizedVectorsScorer();
+  }
 }

lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99ScalarQuantizedVectorScorer.java

@@ -23,6 +23,7 @@
 import org.apache.lucene.codecs.hnsw.FlatVectorsScorer;
 import org.apache.lucene.index.KnnVectorValues;
 import org.apache.lucene.index.VectorSimilarityFunction;
+import org.apache.lucene.util.FloatToFloatFunction;
 import org.apache.lucene.util.VectorUtil;
 import org.apache.lucene.util.hnsw.RandomVectorScorer;
 import org.apache.lucene.util.hnsw.RandomVectorScorerSupplier;
@@ -245,7 +246,7 @@ public float score(int vectorOrdinal) throws IOException {
       values.getSlice().seek((long) vectorOrdinal * (values.getVectorByteLength() + Float.BYTES));
       values.getSlice().readBytes(compressedVector, 0, compressedVector.length);
       float vectorOffset = values.getScoreCorrectionConstant(vectorOrdinal);
-      int dotProduct = VectorUtil.int4DotProductPacked(targetBytes, compressedVector);
+      int dotProduct = VectorUtil.int4DotProductSinglePacked(targetBytes, compressedVector);
       // For the current implementation of scalar quantization, all dotproducts should
       // be >= 0;
       assert dotProduct >= 0;
@@ -301,11 +302,6 @@ public void setScoringOrdinal(int node) throws IOException {
     }
   }
 
-  @FunctionalInterface
-  private interface FloatToFloatFunction {
-    float apply(float f);
-  }
-
   private static final class ScalarQuantizedRandomVectorScorerSupplier
       implements RandomVectorScorerSupplier {
 

lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99ScalarQuantizedVectorsFormat.java

@@ -18,10 +18,10 @@
 package org.apache.lucene.codecs.lucene99;
 
 import java.io.IOException;
-import org.apache.lucene.codecs.hnsw.DefaultFlatVectorScorer;
 import org.apache.lucene.codecs.hnsw.FlatVectorScorerUtil;
 import org.apache.lucene.codecs.hnsw.FlatVectorsFormat;
 import org.apache.lucene.codecs.hnsw.FlatVectorsReader;
+import org.apache.lucene.codecs.hnsw.FlatVectorsScorer;
 import org.apache.lucene.codecs.hnsw.FlatVectorsWriter;
 import org.apache.lucene.index.SegmentReadState;
 import org.apache.lucene.index.SegmentWriteState;
@@ -68,7 +68,7 @@ public class Lucene99ScalarQuantizedVectorsFormat extends FlatVectorsFormat {
 
   final byte bits;
   final boolean compress;
-  final Lucene99ScalarQuantizedVectorScorer flatVectorScorer;
+  final FlatVectorsScorer flatVectorScorer;
 
   /** Constructs a format using default graph construction parameters */
   public Lucene99ScalarQuantizedVectorsFormat() {
@@ -115,8 +115,7 @@ public Lucene99ScalarQuantizedVectorsFormat(
     this.bits = (byte) bits;
     this.confidenceInterval = confidenceInterval;
     this.compress = compress;
-    this.flatVectorScorer =
-        new Lucene99ScalarQuantizedVectorScorer(DefaultFlatVectorScorer.INSTANCE);
+    this.flatVectorScorer = FlatVectorScorerUtil.getLucene99ScalarQuantizedVectorsScorer();
   }
 
   public static float calculateDefaultConfidenceInterval(int vectorDimension) {

lucene/core/src/java/org/apache/lucene/internal/vectorization/DefaultVectorUtilSupport.java

@@ -164,24 +164,35 @@ public int uint8DotProduct(byte[] a, byte[] b) {
   }
 
   @Override
-  public int int4DotProduct(byte[] a, boolean apacked, byte[] b, boolean bpacked) {
-    assert (apacked && bpacked) == false;
-    if (apacked || bpacked) {
-      byte[] packed = apacked ? a : b;
-      byte[] unpacked = apacked ? b : a;
-      int total = 0;
-      for (int i = 0; i < packed.length; i++) {
-        byte packedByte = packed[i];
-        byte unpacked1 = unpacked[i];
-        byte unpacked2 = unpacked[i + packed.length];
-        total += (packedByte & 0x0F) * unpacked2;
-        total += ((packedByte & 0xFF) >> 4) * unpacked1;
-      }
-      return total;
-    }
+  public int int4DotProduct(byte[] a, byte[] b) {
     return dotProduct(a, b);
   }
 
+  @Override
+  public int int4DotProductSinglePacked(byte[] unpacked, byte[] packed) {
+    int total = 0;
+    for (int i = 0; i < packed.length; i++) {
+      byte packedByte = packed[i];
+      byte unpacked1 = unpacked[i];
+      byte unpacked2 = unpacked[i + packed.length];
+      total += (packedByte & 0x0F) * unpacked2;
+      total += ((packedByte & 0xFF) >> 4) * unpacked1;
+    }
+    return total;
+  }
+
+  @Override
+  public int int4DotProductBothPacked(byte[] a, byte[] b) {
+    int total = 0;
+    for (int i = 0; i < a.length; i++) {
+      byte aByte = a[i];
+      byte bByte = b[i];
+      total += (aByte & 0x0F) * (bByte & 0x0F);
+      total += ((aByte & 0xFF) >> 4) * ((bByte & 0xFF) >> 4);
+    }
+    return total;
+  }
+
   @Override
   public float cosine(byte[] a, byte[] b) {
     // Note: this will not overflow if dim < 2^18, since max(byte * byte) = 2^14.
@@ -210,6 +221,42 @@ public int squareDistance(byte[] a, byte[] b) {
     return squareSum;
   }
 
+  @Override
+  public int int4SquareDistance(byte[] a, byte[] b) {
+    return squareDistance(a, b);
+  }
+
+  @Override
+  public int int4SquareDistanceSinglePacked(byte[] unpacked, byte[] packed) {
+    int total = 0;
+    for (int i = 0; i < packed.length; i++) {
+      byte packedByte = packed[i];
+      byte unpacked1 = unpacked[i];
+      byte unpacked2 = unpacked[i + packed.length];
+
+      int diff1 = (packedByte & 0x0F) - unpacked2;
+      int diff2 = ((packedByte & 0xFF) >> 4) - unpacked1;
+
+      total += diff1 * diff1 + diff2 * diff2;
+    }
+    return total;
+  }
+
+  @Override
+  public int int4SquareDistanceBothPacked(byte[] a, byte[] b) {
+    int total = 0;
+    for (int i = 0; i < a.length; i++) {
+      byte aByte = a[i];
+      byte bByte = b[i];
+
+      int diff1 = (aByte & 0x0F) - (bByte & 0x0F);
+      int diff2 = ((aByte & 0xFF) >> 4) - ((bByte & 0xFF) >> 4);
+
+      total += diff1 * diff1 + diff2 * diff2;
+    }
+    return total;
+  }
+
   @Override
   public int uint8SquareDistance(byte[] a, byte[] b) {
     // Note: this will not overflow if dim < 2^16, since max(ubyte * ubyte) = 2^16.

lucene/core/src/java/org/apache/lucene/internal/vectorization/DefaultVectorizationProvider.java

@@ -19,6 +19,7 @@
 
 import org.apache.lucene.codecs.hnsw.DefaultFlatVectorScorer;
 import org.apache.lucene.codecs.hnsw.FlatVectorsScorer;
+import org.apache.lucene.codecs.lucene99.Lucene99ScalarQuantizedVectorScorer;
 import org.apache.lucene.store.IndexInput;
 
 /** Default provider returning scalar implementations. */
@@ -40,6 +41,11 @@ public FlatVectorsScorer getLucene99FlatVectorsScorer() {
     return DefaultFlatVectorScorer.INSTANCE;
   }
 
+  @Override
+  public FlatVectorsScorer getLucene99ScalarQuantizedVectorsScorer() {
+    return new Lucene99ScalarQuantizedVectorScorer(DefaultFlatVectorScorer.INSTANCE);
+  }
+
   @Override
   public PostingDecodingUtil newPostingDecodingUtil(IndexInput input) {
     return new PostingDecodingUtil(input);