log_softmax_op.cc 4.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */

#include <string>
#include <unordered_map>
17

18 19
#include "paddle/fluid/framework/infershape_utils.h"
#include "paddle/fluid/framework/op_registry.h"
20
#include "paddle/fluid/operators/common_infer_shape_functions.h"
21 22
#include "paddle/phi/core/infermeta_utils.h"
#include "paddle/phi/infermeta/unary.h"
23 24 25 26 27 28 29 30 31 32 33

namespace paddle {
namespace operators {

class LogSoftmaxOp : public framework::OperatorWithKernel {
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

 protected:
  framework::OpKernelType GetExpectedKernelType(
      const framework::ExecutionContext& ctx) const override {
34 35 36
    auto input_data_type =
        framework::OperatorWithKernel::IndicateVarDataType(ctx, "X");
    return framework::OpKernelType(input_data_type, ctx.GetPlace());
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
  }
};

class LogSoftmaxOpMaker : public framework::OpProtoAndCheckerMaker {
 public:
  void Make() override {
    AddInput("X",
             "The input tensor of softmax, "
             "whose dimension :attr:`axis` is the input_feature_dimensions.");
    AddOutput("Out", "The normalized values with the same shape as X.");
    AddAttr<int>("axis",
                 "The dimension index of Input(x) to perform log_softmax,"
                 "default -1 for last dimension")
        .SetDefault(-1);
    AddComment(R"DOC(
LogSoftmax Operator.

)DOC");
  }
};

class LogSoftmaxOpInferVarType
    : public framework::PassInDtypeAndVarTypeToOutput {
 protected:
  std::unordered_map<std::string, std::string>& GetInputOutputWithSameType()
      const override {
    static std::unordered_map<std::string, std::string> m{{"X", /*->*/ "Out"}};
    return m;
  }
};

class LogSoftmaxGradOp : public framework::OperatorWithKernel {
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

  void InferShape(framework::InferShapeContext* ctx) const override {
    OP_INOUT_CHECK(ctx->HasInput("Out"), "Input", "Out", "log_softmax_grad");
74 75 76 77
    OP_INOUT_CHECK(ctx->HasInput(framework::GradVarName("Out")),
                   "Input",
                   "Out@grad",
                   "log_softmax_grad");
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115
    PADDLE_ENFORCE_EQ(
        ctx->GetInputDim("Out"),
        ctx->GetInputDim(framework::GradVarName("Out")),
        platform::errors::InvalidArgument("Input(Out) and its gradients "
                                          "should have the same shape."));

    ctx->SetOutputDim(framework::GradVarName("X"),
                      ctx->GetInputDim(framework::GradVarName("Out")));
  }

 protected:
  framework::OpKernelType GetExpectedKernelType(
      const framework::ExecutionContext& ctx) const override {
    return framework::OpKernelType(OperatorWithKernel::IndicateVarDataType(
                                       ctx, framework::GradVarName("Out")),
                                   ctx.device_context());
  }
};

template <typename T>
class LogSoftmaxGradOpMaker : public framework::SingleGradOpMaker<T> {
 public:
  using framework::SingleGradOpMaker<T>::SingleGradOpMaker;

 protected:
  void Apply(GradOpPtr<T> op) const override {
    op->SetType("log_softmax_grad");
    op->SetInput("Out", this->Output("Out"));
    op->SetInput(framework::GradVarName("Out"), this->OutputGrad("Out"));
    op->SetOutput(framework::GradVarName("X"), this->InputGrad("X"));
    op->SetAttrMap(this->Attrs());
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
116 117
DECLARE_INFER_SHAPE_FUNCTOR(log_softmax,
                            LogSoftmaxInferShapeFunctor,
118
                            PD_INFER_META(phi::UnchangedInferMetaCheckAxis));
119 120 121
REGISTER_OPERATOR(log_softmax,
                  ops::LogSoftmaxOp,
                  ops::LogSoftmaxOpMaker,
122 123
                  ops::LogSoftmaxOpInferVarType,
                  ops::LogSoftmaxGradOpMaker<paddle::framework::OpDesc>,
124 125
                  ops::LogSoftmaxGradOpMaker<paddle::imperative::OpBase>,
                  LogSoftmaxInferShapeFunctor);
126
REGISTER_OPERATOR(log_softmax_grad, ops::LogSoftmaxGradOp);