optimize

Author: marin-ma

velox/functions/sparksql/Hash.cpp

@@ -133,9 +133,21 @@ template <
 class VectorHasher {
  public:
   // Compute the hash value of input vector at index.
-  virtual ReturnType hashAt(vector_size_t index, SeedType seed) = 0;
+  ReturnType hashAt(vector_size_t index, SeedType seed) {
+    if (decoded_.isNullAt(index)) {
+      return seed;
+    }
+    return hashNotNull(index, seed);
+  }
+
+  virtual ReturnType hashNotNull(vector_size_t index, SeedType seed) = 0;
+
+  VectorHasher(DecodedVector& decoded) : decoded_(decoded) {}
 
   virtual ~VectorHasher() = default;
+
+ protected:
+  const DecodedVector& decoded_;
 };
 
 template <typename HashClass, TypeKind kind>
@@ -169,24 +181,23 @@ template <
 class PrimitiveVectorHasher
     : public VectorHasher<HashClass, SeedType, ReturnType> {
  public:
-  PrimitiveVectorHasher(DecodedVector& decoded) : decoded_(decoded) {}
+  using T = typename TypeTraits<kind>::NativeType;
 
-  ReturnType hashAt(vector_size_t index, SeedType seed) override {
-    if (decoded_.isNullAt(index)) {
-      return seed;
-    }
+  PrimitiveVectorHasher(DecodedVector& decoded)
+      : VectorHasher<HashClass, SeedType, ReturnType>(decoded) {}
+
+  ReturnType hashNotNull(vector_size_t index, SeedType seed) override {
     return hashOne<HashClass>(
-        decoded_.valueAt<typename TypeTraits<kind>::NativeType>(index), seed);
+        this->decoded_.template valueAt<typename TypeTraits<kind>::NativeType>(
+            index),
+        seed);
   }
-
- private:
-  const DecodedVector& decoded_;
 };
 
 template <typename HashClass, typename SeedType, typename ReturnType>
 class ArrayVectorHasher : public VectorHasher<HashClass> {
  public:
-  ArrayVectorHasher(DecodedVector& decoded) {
+  ArrayVectorHasher(DecodedVector& decoded) : VectorHasher<HashClass>(decoded) {
     base_ = decoded.base()->as<ArrayVector>();
     indices_ = decoded.indices();
 
@@ -195,10 +206,7 @@ class ArrayVectorHasher : public VectorHasher<HashClass> {
     elementHasher_ = createVectorHasher<HashClass>(decodedElements_);
   }
 
-  ReturnType hashAt(vector_size_t index, SeedType seed) override {
-    if (base_->isNullAt(indices_[index])) {
-      return seed;
-    }
+  ReturnType hashNotNull(vector_size_t index, SeedType seed) override {
     auto size = base_->sizeAt(indices_[index]);
     auto offset = base_->offsetAt(indices_[index]);
 
@@ -219,7 +227,7 @@ class ArrayVectorHasher : public VectorHasher<HashClass> {
 template <typename HashClass, typename SeedType, typename ReturnType>
 class MapVectorHasher : public VectorHasher<HashClass, SeedType, ReturnType> {
  public:
-  MapVectorHasher(DecodedVector& decoded) {
+  MapVectorHasher(DecodedVector& decoded) : VectorHasher<HashClass>(decoded) {
     base_ = decoded.base()->as<MapVector>();
     indices_ = decoded.indices();
 
@@ -230,11 +238,7 @@ class MapVectorHasher : public VectorHasher<HashClass, SeedType, ReturnType> {
     valueHasher_ = createVectorHasher<HashClass>(decodedValues_);
   }
 
-  ReturnType hashAt(vector_size_t index, SeedType seed) override {
-    if (base_->isNullAt(indices_[index])) {
-      return seed;
-    }
-
+  ReturnType hashNotNull(vector_size_t index, SeedType seed) override {
     auto size = base_->sizeAt(indices_[index]);
     auto offset = base_->offsetAt(indices_[index]);
 
@@ -258,7 +262,7 @@ class MapVectorHasher : public VectorHasher<HashClass, SeedType, ReturnType> {
 template <typename HashClass, typename SeedType, typename ReturnType>
 class RowVectorHasher : public VectorHasher<HashClass, SeedType, ReturnType> {
  public:
-  RowVectorHasher(DecodedVector& decoded) {
+  RowVectorHasher(DecodedVector& decoded) : VectorHasher<HashClass>(decoded) {
     base_ = decoded.base()->as<RowVector>();
     indices_ = decoded.indices();
 
@@ -271,14 +275,10 @@ class RowVectorHasher : public VectorHasher<HashClass, SeedType, ReturnType> {
     }
   }
 
-  ReturnType hashAt(vector_size_t index, SeedType seed) override {
-    if (base_->isNullAt(indices_[index])) {
-      return seed;
-    }
-
+  ReturnType hashNotNull(vector_size_t index, SeedType seed) override {
     ReturnType result = seed;
     for (auto i = 0; i < base_->childrenSize(); ++i) {
-      result = hashers_[i]->hashAt(index, result);
+      result = hashers_[i]->hashAt(indices_[index], result);
     }
     return result;
   }
@@ -323,12 +323,13 @@ void applyWithType(
 
     auto hasher = createVectorHasher<HashClass>(*decoded);
     selected->applyToSelected([&](int row) {
-      result.set(row, hasher->hashAt(row, result.valueAt(row)));
+      result.set(row, hasher->hashNotNull(row, result.valueAt(row)));
     });
   }
 }
 
-// Derived from src/main/java/org/apache/spark/unsafe/hash/Murmur3_x86_32.java.
+// Derived from
+// src/main/java/org/apache/spark/unsafe/hash/Murmur3_x86_32.java.
 //
 // Spark's Murmur3 seems slightly different from the original from Austin
 // Appleby: in particular the fmix function's first line is different. The