Merge pull request #72556 from ClickHouse/devirtualize_serialize_call

Devirtualize serialize call in AggregateFunction

src/AggregateFunctions/AggregateFunctionGroupArrayIntersect.cpp

@@ -192,7 +192,7 @@ struct AggregateFunctionGroupArrayIntersectGenericData
  *  For such columns GroupArrayIntersect() can be implemented more efficiently (especially for small numeric arrays).
  */
 template <bool is_plain_column = false>
-class AggregateFunctionGroupArrayIntersectGeneric
+class AggregateFunctionGroupArrayIntersectGeneric final
     : public IAggregateFunctionDataHelper<AggregateFunctionGroupArrayIntersectGenericData,
         AggregateFunctionGroupArrayIntersectGeneric<is_plain_column>>
 {
@@ -353,7 +353,7 @@ namespace
 {
 
 /// Substitute return type for Date and DateTime
-class AggregateFunctionGroupArrayIntersectDate : public AggregateFunctionGroupArrayIntersect<DataTypeDate::FieldType>
+class AggregateFunctionGroupArrayIntersectDate final : public AggregateFunctionGroupArrayIntersect<DataTypeDate::FieldType>
 {
 public:
     explicit AggregateFunctionGroupArrayIntersectDate(const DataTypePtr & argument_type, const Array & parameters_)
@@ -361,7 +361,7 @@ public:
     static DataTypePtr createResultType() { return std::make_shared<DataTypeArray>(std::make_shared<DataTypeDate>()); }
 };
 
-class AggregateFunctionGroupArrayIntersectDateTime : public AggregateFunctionGroupArrayIntersect<DataTypeDateTime::FieldType>
+class AggregateFunctionGroupArrayIntersectDateTime final : public AggregateFunctionGroupArrayIntersect<DataTypeDateTime::FieldType>
 {
 public:
     explicit AggregateFunctionGroupArrayIntersectDateTime(const DataTypePtr & argument_type, const Array & parameters_)
@@ -369,7 +369,7 @@ public:
     static DataTypePtr createResultType() { return std::make_shared<DataTypeArray>(std::make_shared<DataTypeDateTime>()); }
 };
 
-class AggregateFunctionGroupArrayIntersectDate32 : public AggregateFunctionGroupArrayIntersect<DataTypeDate32::FieldType>
+class AggregateFunctionGroupArrayIntersectDate32 final : public AggregateFunctionGroupArrayIntersect<DataTypeDate32::FieldType>
 {
 public:
     explicit AggregateFunctionGroupArrayIntersectDate32(const DataTypePtr & argument_type, const Array & parameters_)

src/AggregateFunctions/AggregateFunctionGroupUniqArray.cpp

@@ -153,7 +153,7 @@ static void deserializeAndInsertImpl(StringRef str, IColumn & data_to);
  *  For such columns groupUniqArray() can be implemented more efficiently (especially for small numeric arrays).
  */
 template <bool is_plain_column = false, typename LimitNumElems = std::false_type>
-class AggregateFunctionGroupUniqArrayGeneric
+class AggregateFunctionGroupUniqArrayGeneric final
     : public IAggregateFunctionDataHelper<AggregateFunctionGroupUniqArrayGenericData,
         AggregateFunctionGroupUniqArrayGeneric<is_plain_column, LimitNumElems>>
 {
@@ -245,7 +245,7 @@ public:
 
 /// Substitute return type for Date and DateTime
 template <typename HasLimit>
-class AggregateFunctionGroupUniqArrayDate : public AggregateFunctionGroupUniqArray<DataTypeDate::FieldType, HasLimit>
+class AggregateFunctionGroupUniqArrayDate final : public AggregateFunctionGroupUniqArray<DataTypeDate::FieldType, HasLimit>
 {
 public:
     explicit AggregateFunctionGroupUniqArrayDate(const DataTypePtr & argument_type, const Array & parameters_, UInt64 max_elems_ = std::numeric_limits<UInt64>::max())
@@ -254,7 +254,7 @@ public:
 };
 
 template <typename HasLimit>
-class AggregateFunctionGroupUniqArrayDateTime : public AggregateFunctionGroupUniqArray<DataTypeDateTime::FieldType, HasLimit>
+class AggregateFunctionGroupUniqArrayDateTime final : public AggregateFunctionGroupUniqArray<DataTypeDateTime::FieldType, HasLimit>
 {
 public:
     explicit AggregateFunctionGroupUniqArrayDateTime(const DataTypePtr & argument_type, const Array & parameters_, UInt64 max_elems_ = std::numeric_limits<UInt64>::max())
@@ -263,7 +263,7 @@ public:
 };
 
 template <typename HasLimit>
-class AggregateFunctionGroupUniqArrayIPv4 : public AggregateFunctionGroupUniqArray<DataTypeIPv4::FieldType, HasLimit>
+class AggregateFunctionGroupUniqArrayIPv4 final : public AggregateFunctionGroupUniqArray<DataTypeIPv4::FieldType, HasLimit>
 {
 public:
     explicit AggregateFunctionGroupUniqArrayIPv4(const DataTypePtr & argument_type, const Array & parameters_, UInt64 max_elems_ = std::numeric_limits<UInt64>::max())

src/AggregateFunctions/AggregateFunctionMeanZTest.cpp

@@ -28,7 +28,7 @@ namespace
 
 /// Returns tuple of (z-statistic, p-value, confidence-interval-low, confidence-interval-high)
 template <typename Data>
-class AggregateFunctionMeanZTest :
+class AggregateFunctionMeanZTest final:
     public IAggregateFunctionDataHelper<Data, AggregateFunctionMeanZTest<Data>>
 {
 private:

src/AggregateFunctions/AggregateFunctionRankCorrelation.cpp

@@ -47,7 +47,7 @@ struct RankCorrelationData : public StatisticalSample<Float64, Float64>
     }
 };
 
-class AggregateFunctionRankCorrelation :
+class AggregateFunctionRankCorrelation final :
     public IAggregateFunctionDataHelper<RankCorrelationData, AggregateFunctionRankCorrelation>
 {
 public:

src/AggregateFunctions/AggregateFunctionTTest.h

@@ -38,7 +38,7 @@ namespace ErrorCodes
 /// Returns tuple of (t-statistic, p-value)
 /// https://cpb-us-w2.wpmucdn.com/voices.uchicago.edu/dist/9/1193/files/2016/01/05b-TandP.pdf
 template <typename Data>
-class AggregateFunctionTTest :
+class AggregateFunctionTTest final:
     public IAggregateFunctionDataHelper<Data, AggregateFunctionTTest<Data>>
 {
 private:

src/AggregateFunctions/AggregateFunctionTopK.cpp

@@ -203,7 +203,7 @@ struct AggregateFunctionTopKGenericData
  *  For such columns topK() can be implemented more efficiently (especially for small numeric arrays).
  */
 template <bool is_plain_column, bool is_weighted>
-class AggregateFunctionTopKGeneric
+class AggregateFunctionTopKGeneric final
     : public IAggregateFunctionDataHelper<AggregateFunctionTopKGenericData, AggregateFunctionTopKGeneric<is_plain_column, is_weighted>>
 {
 private:
@@ -367,7 +367,7 @@ public:
 
 /// Substitute return type for Date and DateTime
 template <bool is_weighted>
-class AggregateFunctionTopKDate : public AggregateFunctionTopK<DataTypeDate::FieldType, is_weighted>
+class AggregateFunctionTopKDate final : public AggregateFunctionTopK<DataTypeDate::FieldType, is_weighted>
 {
 public:
     using AggregateFunctionTopK<DataTypeDate::FieldType, is_weighted>::AggregateFunctionTopK;
@@ -384,7 +384,7 @@ public:
 };
 
 template <bool is_weighted>
-class AggregateFunctionTopKDateTime : public AggregateFunctionTopK<DataTypeDateTime::FieldType, is_weighted>
+class AggregateFunctionTopKDateTime final : public AggregateFunctionTopK<DataTypeDateTime::FieldType, is_weighted>
 {
 public:
     using AggregateFunctionTopK<DataTypeDateTime::FieldType, is_weighted>::AggregateFunctionTopK;
@@ -401,7 +401,7 @@ public:
 };
 
 template <bool is_weighted>
-class AggregateFunctionTopKIPv4 : public AggregateFunctionTopK<DataTypeIPv4::FieldType, is_weighted>
+class AggregateFunctionTopKIPv4 final : public AggregateFunctionTopK<DataTypeIPv4::FieldType, is_weighted>
 {
 public:
     using AggregateFunctionTopK<DataTypeIPv4::FieldType, is_weighted>::AggregateFunctionTopK;

src/AggregateFunctions/Combinators/AggregateFunctionDistinct.h

@@ -150,7 +150,7 @@ struct AggregateFunctionDistinctMultipleGenericData : public AggregateFunctionDi
   * Adding -Distinct suffix to aggregate function
 **/
 template <typename Data>
-class AggregateFunctionDistinct : public IAggregateFunctionDataHelper<Data, AggregateFunctionDistinct<Data>>
+class AggregateFunctionDistinct final : public IAggregateFunctionDataHelper<Data, AggregateFunctionDistinct<Data>>
 {
 private:
     AggregateFunctionPtr nested_func;

src/AggregateFunctions/CrossTab.h

@@ -114,7 +114,7 @@ struct CrossTabData
 
 
 template <typename Data>
-class AggregateFunctionCrossTab : public IAggregateFunctionDataHelper<Data, AggregateFunctionCrossTab<Data>>
+class AggregateFunctionCrossTab final : public IAggregateFunctionDataHelper<Data, AggregateFunctionCrossTab<Data>>
 {
 public:
     explicit AggregateFunctionCrossTab(const DataTypes & arguments)

src/AggregateFunctions/IAggregateFunction.h

@@ -176,6 +176,8 @@ public:
     /// Serializes state (to transmit it over the network, for example).
     virtual void serialize(ConstAggregateDataPtr __restrict place, WriteBuffer & buf, std::optional<size_t> version = std::nullopt) const = 0; /// NOLINT
 
+    virtual void serializeBatch(const PaddedPODArray<AggregateDataPtr> & data, size_t start, size_t size, WriteBuffer & buf, std::optional<size_t> version = std::nullopt) const = 0; /// NOLINT
+
     /// Deserializes state. This function is called only for empty (just created) states.
     virtual void deserialize(AggregateDataPtr __restrict place, ReadBuffer & buf, std::optional<size_t> version = std::nullopt, Arena * arena = nullptr) const = 0; /// NOLINT
 
@@ -471,6 +473,12 @@ public:
         }
     }
 
+    void serializeBatch(const PaddedPODArray<AggregateDataPtr> & data, size_t start, size_t size, WriteBuffer & buf, std::optional<size_t> version) const override // NOLINT
+    {
+        for (size_t i = start; i < size; ++i)
+            static_cast<const Derived *>(this)->serialize(data[i], buf, version);
+    }
+
     void addBatchSparse(
         size_t row_begin,
         size_t row_end,

src/Columns/ColumnAggregateFunction.cpp

@@ -418,9 +418,10 @@ WeakHash32 ColumnAggregateFunction::getWeakHash32() const
 
 void ColumnAggregateFunction::updateHashFast(SipHash & hash) const
 {
-    /// Fallback to per-element hashing, as there is no faster way
-    for (size_t i = 0; i < size(); ++i)
-        updateHashWithValue(i, hash);
+    WriteBufferFromOwnString wbuf;
+    const ColumnAggregateFunction::Container & vec = getData();
+    func->serializeBatch(vec, 0, size(), wbuf);
+    hash.update(wbuf.str().c_str(), wbuf.str().size());
 }
 
 /// The returned size is less than real size. The reason is that some parts of

src/DataTypes/Serializations/SerializationAggregateFunction.cpp

@@ -60,12 +60,11 @@ void SerializationAggregateFunction::serializeBinaryBulk(const IColumn & column,
     const ColumnAggregateFunction & real_column = typeid_cast<const ColumnAggregateFunction &>(column);
     const ColumnAggregateFunction::Container & vec = real_column.getData();
 
-    ColumnAggregateFunction::Container::const_iterator it = vec.begin() + offset;
-    ColumnAggregateFunction::Container::const_iterator end = limit ? it + limit : vec.end();
+    size_t end = vec.size();
+    if (limit)
+        end = std::min(end, offset + limit);
 
-    end = std::min(end, vec.end());
-    for (; it != end; ++it)
-        function->serialize(*it, ostr, version);
+    function->serializeBatch(vec, offset, end, ostr, version);
 }
 
 void SerializationAggregateFunction::deserializeBinaryBulk(IColumn & column, ReadBuffer & istr, size_t limit, double /*avg_value_size_hint*/) const

tests/performance/avg_serialization.xml

@@ -0,0 +1,10 @@
+<test>
+    <create_query>DROP TABLE IF EXISTS test_avg_insert</create_query>
+    <create_query>
+        CREATE TABLE test_avg_insert (key UInt64, value AggregateFunction(avg, UInt8)) ENGINE = Memory()
+    </create_query>
+
+    <query>INSERT INTO test_avg_insert with initializeAggregation('avgState', 1) as s select number, s AS value FROM numbers(200000000)</query>
+
+    <drop_query>DROP TABLE IF EXISTS test_avg_insert</drop_query>
+</test>