ngraph/core/include/ngraph/op/quantized_dot.hpp

   1 //*****************************************************************************
   2 // Copyright 2017-2020 Intel Corporation
   3 //
   4 // Licensed under the Apache License, Version 2.0 (the "License");
   5 // you may not use this file except in compliance with the License.
   6 // You may obtain a copy of the License at
   7 //
   8 //     http://www.apache.org/licenses/LICENSE-2.0
   9 //
  10 // Unless required by applicable law or agreed to in writing, software
  11 // distributed under the License is distributed on an "AS IS" BASIS,
  12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 // See the License for the specific language governing permissions and
  14 // limitations under the License.
  15 //*****************************************************************************
  16
  17 #pragma once
  18
  19 #include "ngraph/op/op.hpp"
  20
  21 namespace ngraph
  22 {
  23     namespace op
  24     {
  25         namespace v0
  26         {
  27             class NGRAPH_DEPRECATED(
  28                 "This operation is deprecated and will be removed soon. Please do not use it.")
  29                 NGRAPH_API QuantizedDot : public Op
  30             {
  31                 NGRAPH_SUPPRESS_DEPRECATED_START
  32             public:
  33                 static constexpr NodeTypeInfo type_info{"QuantizedDot", 0};
  34                 const NodeTypeInfo& get_type_info() const override { return type_info; }
  35                 QuantizedDot() = default;
  36                 /// \brief Constructs a quantized convolution operation.
  37                 ///
  38                 /// \param input0 The node producing the input data batch tensor.
  39                 /// \param input1 The node producing the filters tensor.
  40                 /// \param input0_scale Scale to transform the input
  41                 /// \param input0_zero_point Zero point used for mapping
  42                 /// \param input1_scale Scale to transform the filters
  43                 /// \param input1_zero_point Zero point used for mapping
  44                 /// \param output_scale Scale to transform the output
  45                 /// \param output_zero_point Zero point used for mapping
  46                 /// \param output_type Output element type
  47                 /// \param input0_axes Input0 axes set for channel wise quantization
  48                 /// \param input1_axes Input1 axes set for channel wise quantization
  49                 /// \param output_axes Output axes set for channel wise quantization
  50                 QuantizedDot(const Output<Node>& input0,
  51                              const Output<Node>& input1,
  52                              size_t reduction_axes_count,
  53                              const Output<Node>& input0_scale,
  54                              const Output<Node>& input0_zero_point,
  55                              const Output<Node>& input1_scale,
  56                              const Output<Node>& input1_zero_point,
  57                              const Output<Node>& output_scale,
  58                              const Output<Node>& output_zero_point,
  59                              const element::Type& output_type,
  60                              const AxisSet& input0_axes = ngraph::AxisSet{},
  61                              const AxisSet& input1_axes = ngraph::AxisSet{},
  62                              const AxisSet& output_axes = ngraph::AxisSet{});
  63
  64                 std::shared_ptr<Node> get_input0() { return input_value(0).get_node_shared_ptr(); }
  65                 std::shared_ptr<Node> get_input1() { return input_value(1).get_node_shared_ptr(); }
  66                 const ngraph::element::Type& get_output_type() const { return m_output_type; }
  67                 const ngraph::AxisSet& get_input0_axes() const { return m_input0_axes; }
  68                 const ngraph::AxisSet& get_input1_axes() const { return m_input1_axes; }
  69                 const ngraph::AxisSet& get_output_axes() const { return m_output_axes; }
  70                 size_t get_reduction_axes_count() const { return m_reduction_axes_count; }
  71                 void set_reduction_axes_count(size_t reduction_axes_count)
  72                 {
  73                     m_reduction_axes_count = reduction_axes_count;
  74                 }
  75                 void validate_and_infer_types() override;
  76                 virtual std::shared_ptr<Node>
  77                     clone_with_new_inputs(const OutputVector& new_args) const override;
  78
  79             protected:
  80                 size_t m_reduction_axes_count;
  81                 bool m_has_reduction_axes_count;
  82                 ngraph::element::Type m_output_type;
  83                 ngraph::AxisSet m_input0_axes;
  84                 ngraph::AxisSet m_input1_axes;
  85                 ngraph::AxisSet m_output_axes;
  86                 NGRAPH_SUPPRESS_DEPRECATED_END
  87             };
  88         }
  89         NGRAPH_SUPPRESS_DEPRECATED_START
  90         using v0::QuantizedDot;
  91         NGRAPH_SUPPRESS_DEPRECATED_END
  92     }
  93 }