Use lz4-java and xxhash32

2 files changed, 27 insertions, 13 deletions

diff --git a/vespajlib/src/main/java/com/yahoo/tensor/TensorAddress.java b/vespajlib/src/main/java/com/yahoo/tensor/TensorAddress.java
index f841b7757fb..1346ee7cf46 100644
--- a/vespajlib/src/main/java/com/yahoo/tensor/TensorAddress.java
+++ b/vespajlib/src/main/java/com/yahoo/tensor/TensorAddress.java
@@ -3,7 +3,10 @@ package com.yahoo.tensor;
 
 import com.yahoo.tensor.impl.NumericTensorAddress;
 import com.yahoo.tensor.impl.StringTensorAddress;
+import net.jpountz.xxhash.XXHash32;
+import net.jpountz.xxhash.XXHashFactory;
 
+import java.nio.charset.StandardCharsets;
 import java.util.Arrays;
 import java.util.Objects;
 import java.util.Optional;
@@ -16,6 +19,8 @@ import java.util.Optional;
  */
 public abstract class TensorAddress implements Comparable<TensorAddress> {
 
+    private static final XXHash32 hasher = XXHashFactory.fastestJavaInstance().hash32();
+
     public static TensorAddress of(String[] labels) {
         return StringTensorAddress.of(labels);
     }
@@ -28,6 +33,8 @@ public abstract class TensorAddress implements Comparable<TensorAddress> {
         return NumericTensorAddress.of(labels);
     }
 
+    private int cached_hash = 0;
+
     /** Returns the number of labels in this */
     public abstract int size();
 
@@ -62,12 +69,17 @@ public abstract class TensorAddress implements Comparable<TensorAddress> {
 
     @Override
     public int hashCode() {
-        int result = 1;
+        if (cached_hash != 0) return cached_hash;
+
+        int hash = 0;
         for (int i = 0; i < size(); i++) {
-            if (label(i) != null)
-                result = 31 * result + label(i).hashCode();
+            String label = label(i);
+            if (label != null) {
+                byte [] buf = label.getBytes(StandardCharsets.UTF_8);
+                hash = hasher.hash(buf, 0, buf.length, hash);
+            }
         }
-        return result;
+        return cached_hash = hash;
     }
 
     @Override
diff --git a/vespajlib/src/test/java/com/yahoo/tensor/TensorFunctionBenchmark.java b/vespajlib/src/test/java/com/yahoo/tensor/TensorFunctionBenchmark.java
index 37c0fb87be0..74237a218fb 100644
--- a/vespajlib/src/test/java/com/yahoo/tensor/TensorFunctionBenchmark.java
+++ b/vespajlib/src/test/java/com/yahoo/tensor/TensorFunctionBenchmark.java
@@ -9,8 +9,10 @@ import com.yahoo.tensor.functions.Join;
 import com.yahoo.tensor.functions.Reduce;
 import com.yahoo.tensor.functions.TensorFunction;
 
-import java.util.*;
-import java.util.stream.Collectors;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Random;
+
 
 /**
  * Microbenchmark of tensor operations.
@@ -91,7 +93,7 @@ public class TensorFunctionBenchmark {
                         .value(random.nextDouble());
             }
         }
-        return Collections.singletonList(builder.build());
+        return List.of(builder.build());
     }
 
     private static TensorType vectorType(TensorType.Builder builder, String name, TensorType.Dimension.Type type, int size) {
@@ -136,22 +138,22 @@ public class TensorFunctionBenchmark {
         // ---------------- Mapped with extra space (sidesteps current special-case optimizations):
         time = new TensorFunctionBenchmark().benchmark(1000, vectors(100, 300, TensorType.Dimension.Type.mapped), TensorType.Dimension.Type.mapped, true);
         System.out.printf("Mapped vectors, x space  time per join: %1$8.3f ms\n", time);
+
         time = new TensorFunctionBenchmark().benchmark(1000, matrix(100, 300, TensorType.Dimension.Type.mapped), TensorType.Dimension.Type.mapped, true);
         System.out.printf("Mapped matrix, x space   time per join: %1$8.3f ms\n", time);
 
-        /** 2.4Ghz Intel Core i9, Macbook Pro 2019
+        /* 2.4Ghz Intel Core i9, Macbook Pro 2019
          * Indexed unbound vectors, time per join:    0,067 ms
          * Indexed unbound matrix,  time per join:    0,107 ms
          * Indexed bound vectors,   time per join:    0,068 ms
          * Indexed bound matrix,    time per join:    0,105 ms
-         * Mapped vectors,          time per join:    1,780 ms
-         * Mapped matrix,           time per join:    5,339 ms
+         * Mapped vectors,          time per join:    1,342 ms
+         * Mapped matrix,           time per join:    3,448 ms
          * Indexed vectors, x space time per join:    6,398 ms
          * Indexed matrix, x space  time per join:    3,220 ms
-         * Mapped vectors, x space  time per join:   13,026 ms
-         * Mapped matrix, x space   time per join:   28,259 ms
+         * Mapped vectors, x space  time per join:   14,984 ms
+         * Mapped matrix, x space   time per join:   19,873 ms
          */
-
     }
 
 }