remove virtual function call

marin-ma · Apr 26, 2024 · e9abef7 · e9abef7
1 parent dbb1a92
commit e9abef7
Showing 1 changed file with 57 additions and 30 deletions.
diff --git a/velox/functions/sparksql/Hash.cpp b/velox/functions/sparksql/Hash.cpp
@@ -82,6 +82,31 @@ ReturnType hashOne(StringView input, SeedType seed) {
   return HashClass::hashBytes(input, seed);
 }
 
+template <
+    typename HashClass,
+    TypeKind kind,
+    typename SeedType = typename HashTraits<HashClass>::SeedType,
+    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
+class PrimitiveVectorHasher;
+
+template <
+    typename HashClass,
+    typename SeedType = typename HashTraits<HashClass>::SeedType,
+    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
+class ArrayVectorHasher;
+
+template <
+    typename HashClass,
+    typename SeedType = typename HashTraits<HashClass>::SeedType,
+    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
+class MapVectorHasher;
+
+template <
+    typename HashClass,
+    typename SeedType = typename HashTraits<HashClass>::SeedType,
+    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
+class RowVectorHasher;
+
 // Class to compute hashes identical to one produced by Spark.
 // Hashes are computed using the algorithm implemented in HashClass.
 template <
@@ -103,36 +128,38 @@ class SparkVectorHasher {
   }
 
   // Compute the hash value of input vector at index for non-null values.
-  virtual ReturnType hashNotNullAt(vector_size_t index, SeedType seed) = 0;
+  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) {
+    switch (decoded_.base()->typeKind()) {
+      case TypeKind::ARRAY:
+        return static_cast<ArrayVectorHasher<HashClass, SeedType, ReturnType>*>(
+                   this)
+            ->hashNotNullAt(index, seed);
+      case TypeKind::MAP:
+        return static_cast<MapVectorHasher<HashClass, SeedType, ReturnType>*>(
+                   this)
+            ->hashNotNullAt(index, seed);
+      case TypeKind::ROW:
+        return static_cast<RowVectorHasher<HashClass, SeedType, ReturnType>*>(
+                   this)
+            ->hashNotNullAt(index, seed);
+      default:
+        return VELOX_DYNAMIC_SCALAR_TYPE_DISPATCH(
+            hashPrimitive, decoded_.base()->typeKind(), index, seed);
+    }
+  }
 
  protected:
   const DecodedVector& decoded_;
-};
 
-template <
-    typename HashClass,
-    TypeKind kind,
-    typename SeedType = typename HashTraits<HashClass>::SeedType,
-    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
-class PrimitiveVectorHasher;
-
-template <
-    typename HashClass,
-    typename SeedType = typename HashTraits<HashClass>::SeedType,
-    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
-class ArrayVectorHasher;
-
-template <
-    typename HashClass,
-    typename SeedType = typename HashTraits<HashClass>::SeedType,
-    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
-class MapVectorHasher;
-
-template <
-    typename HashClass,
-    typename SeedType = typename HashTraits<HashClass>::SeedType,
-    typename ReturnType = typename HashTraits<HashClass>::ReturnType>
-class RowVectorHasher;
+ private:
+  template <TypeKind kind>
+  ReturnType hashPrimitive(vector_size_t index, SeedType seed) {
+    return static_cast<
+               PrimitiveVectorHasher<HashClass, kind, SeedType, ReturnType>*>(
+               this)
+        ->hashNotNullAt(index, seed);
+  }
+};
 
 template <typename HashClass, TypeKind kind>
 std::shared_ptr<SparkVectorHasher<HashClass>> createPrimitiveVectorHasher(
@@ -169,7 +196,7 @@ class PrimitiveVectorHasher : public SparkVectorHasher<HashClass> {
   PrimitiveVectorHasher(DecodedVector& decoded)
       : SparkVectorHasher<HashClass>(decoded) {}
 
-  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) override {
+  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) {
     return hashOne<HashClass, SeedType, ReturnType>(
         this->decoded_.template valueAt<typename TypeTraits<kind>::NativeType>(
             index),
@@ -190,7 +217,7 @@ class ArrayVectorHasher : public SparkVectorHasher<HashClass> {
     elementHasher_ = createVectorHasher<HashClass>(decodedElements_);
   }
 
-  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) override {
+  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) {
     auto size = base_->sizeAt(indices_[index]);
     auto offset = base_->offsetAt(indices_[index]);
 
@@ -223,7 +250,7 @@ class MapVectorHasher : public SparkVectorHasher<HashClass> {
     valueHasher_ = createVectorHasher<HashClass>(decodedValues_);
   }
 
-  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) override {
+  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) {
     auto size = base_->sizeAt(indices_[index]);
     auto offset = base_->offsetAt(indices_[index]);
 
@@ -261,7 +288,7 @@ class RowVectorHasher : public SparkVectorHasher<HashClass> {
     }
   }
 
-  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) override {
+  ReturnType hashNotNullAt(vector_size_t index, SeedType seed) {
     ReturnType result = seed;
     for (auto i = 0; i < base_->childrenSize(); ++i) {
       result = hashers_[i]->hashAt(indices_[index], result);