parallel_do_op.cc 5.4 KB
Newer Older
Y
Yang Yang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

   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 <vector>
#include "paddle/framework/executor.h"
#include "paddle/framework/op_registry.h"
#include "paddle/framework/operator.h"

namespace paddle {
namespace operators {

constexpr char kInputs[] = "inputs";
constexpr char kParameters[] = "parameters";
constexpr char kPlaces[] = "places";
Y
Yang Yang 已提交
26

Y
Yang Yang 已提交
27
constexpr char kOutputs[] = "outputs";
Y
Yang Yang 已提交
28 29 30
constexpr char kParallelScopes[] = "parallel_scopes";

constexpr char kParallelBlock[] = "sub_block";
Y
Yang Yang 已提交
31 32 33 34 35 36 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 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
// #define GRAD_SUFFIX "@GRAD"
// constexpr char kInputGrads[] = "inputs" GRAD_SUFFIX;
// constexpr char kOutputGrads[] = "outputs" GRAD_SUFFIX;
// constexpr char kParamGrads[] = "parameters" GRAD_SUFFIX;

using ParallelScopeVar = std::vector<framework::Scope *>;
using OperatorBase = framework::OperatorBase;

class ParallelDoOp : public OperatorBase {
 public:
  ParallelDoOp(const std::string &type,
               const framework::VariableNameMap &inputs,
               const framework::VariableNameMap &outputs,
               const framework::AttributeMap &attrs)
      : OperatorBase(type, inputs, outputs, attrs) {}

  void Run(const framework::Scope &scope,
           const platform::DeviceContext &dev_ctx) const override {
    // create scope
    // copy parameters
  }
};

class ParallelDoGradOp : public OperatorBase {
 public:
  ParallelDoGradOp(const std::string &type,
                   const framework::VariableNameMap &inputs,
                   const framework::VariableNameMap &outputs,
                   const framework::AttributeMap &attrs)
      : OperatorBase(type, inputs, outputs, attrs) {}

  void Run(const framework::Scope &scope,
           const platform::DeviceContext &dev_ctx) const override {}
};

class ParallelDoOpProtoMaker : public framework::OpProtoAndCheckerMaker {
 public:
  ParallelDoOpProtoMaker(framework::OpProto *proto,
                         framework::OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput(kInputs, "").AsDuplicable();
    AddInput(kParameters, "").AsDuplicable();
    AddInput(kPlaces, "");
    AddOutput(kOutputs, "").AsDuplicable();
    AddOutput(kParallelScopes, "");
    AddAttr<framework::BlockDescBind *>(kParallelBlock, "");
    AddComment(R"DOC(
ParallelDo Operator.
)DOC");
  }
};

class ParallelDoGradOpDescMaker : public framework::SingleGradOpDescMaker {
 public:
  using framework::SingleGradOpDescMaker::SingleGradOpDescMaker;

 protected:
  virtual std::unique_ptr<framework::OpDescBind> Apply() const {
    PADDLE_THROW("Not Implemented");
    auto *grad = new framework::OpDescBind();
    grad->SetType("recurrent_grad");
    for (auto &input_param : this->InputNames()) {
      grad->SetInput(input_param, this->Input(input_param));
      grad->SetOutput(framework::GradVarName(input_param),
                      this->InputGrad(input_param));
    }

    for (auto &output_param : this->OutputNames()) {
      if (output_param == kParallelScopes) {
        grad->SetInput(output_param, this->Output(output_param));
        grad->SetInput(framework::GradVarName(output_param),
                       this->Output(output_param));
      } else {
        grad->SetInput(output_param, this->Output(output_param));
        grad->SetInput(framework::GradVarName(output_param),
                       this->OutputGrad(output_param));
      }
    }
    grad->SetAttrMap(this->Attrs());
    grad->SetBlockAttr(kParallelBlock, *grad_block_[0]);

    return std::unique_ptr<framework::OpDescBind>(grad);
  }
};

class ParallelDoGradOpShapeInference : public framework::InferShapeBase {
 public:
  void operator()(framework::InferShapeContext *ctx) const override {
    PADDLE_THROW("Not Implemented");
    // std::vector<std::string> input{kInputs};
    // std::vector<std::string> output{kOutputs};
    // for (auto &s : input) {
    //   PADDLE_ENFORCE(ctx->HasInputs(s));
    //   PADDLE_ENFORCE(ctx->HasOutputs(framework::GradVarName(s)),
    //                  "Cannot find the gradient variable %s",
    //                  framework::GradVarName(s));
    // }
    // for (auto &s : output) {
    //   PADDLE_ENFORCE(ctx->HasInputs(s));
    // }
    // for (auto &s : input) {
    //   ctx->SetOutputsDim(framework::GradVarName(s), ctx->GetInputsDim(s));
    // }
    // if (ctx->HasInputs(kParameters)) {
    //   PADDLE_ENFORCE(ctx->HasOutputs(framework::GradVarName(kParameters)));
    //   ctx->SetOutputsDim(framework::GradVarName(kParameters),
    //                      ctx->GetInputsDim(kParameters));
    // }
  }
};

}  // namespace operators
}  // namespace paddle

REGISTER_OPERATOR(parallel_do, paddle::operators::ParallelDoOp,
                  paddle::operators::ParallelDoOpProtoMaker,
                  paddle::operators::ParallelDoGradOpDescMaker);
REGISTER_OPERATOR(parallel_do_grad, paddle::operators::ParallelDoGradOp,
                  paddle::operators::ParallelDoGradOpShapeInference);