shrink_rnn_memory_op.cc 7.2 KB
Newer Older
L
Luo Tao 已提交
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Y
Yang Yu 已提交
2

L
Luo Tao 已提交
3 4 5
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
Y
Yang Yu 已提交
6

L
Luo Tao 已提交
7
    http://www.apache.org/licenses/LICENSE-2.0
Y
Yang Yu 已提交
8

L
Luo Tao 已提交
9 10 11 12 13
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. */
Y
Yang Yu 已提交
14
#include "paddle/framework/lod_rank_table.h"
Y
yangyaming 已提交
15
#include "paddle/framework/lod_tensor.h"
Y
Yang Yu 已提交
16 17 18 19 20 21
#include "paddle/operators/array_operator.h"
#include "paddle/operators/math/math_function.h"

namespace paddle {
namespace operators {

Y
Yang Yu 已提交
22
class ShrinkRNNMemoryOp : public ArrayOp {
Y
Yang Yu 已提交
23
 public:
Y
Yang Yu 已提交
24 25 26 27
  ShrinkRNNMemoryOp(const std::string &type,
                    const framework::VariableNameMap &inputs,
                    const framework::VariableNameMap &outputs,
                    const framework::AttributeMap &attrs)
Y
Yang Yu 已提交
28 29 30
      : ArrayOp(type, inputs, outputs, attrs) {}

  void Run(const framework::Scope &scope,
D
dzhwinter 已提交
31
           const platform::Place &place) const override {
Y
Yang Yu 已提交
32 33 34
    auto *x_var = scope.FindVar(Input("X"));
    PADDLE_ENFORCE(x_var != nullptr, "Input X must be set");
    auto &x_tensor = x_var->Get<framework::LoDTensor>();
D
dzhwinter 已提交
35
    size_t offset = this->GetOffset(scope, place);
Y
Yang Yu 已提交
36 37 38 39
    auto *rank_table_var = scope.FindVar(Input("RankTable"));
    PADDLE_ENFORCE(rank_table_var != nullptr, "RankTable must be set");
    auto &rank_table = rank_table_var->Get<framework::LoDRankTable>();

Y
Yang Yu 已提交
40 41 42 43 44 45
    auto &rank_items = rank_table.items();
    int dst_num_rows =
        std::lower_bound(rank_items.begin(), rank_items.end(), offset,
                         [](const framework::LoDRankTable::TableItem &a,
                            size_t b) { return a.length > b; }) -
        rank_items.begin();
Y
Yang Yu 已提交
46 47 48 49

    auto *out_var = scope.FindVar(Output("Out"));
    PADDLE_ENFORCE(out_var != nullptr, "Output Out must be set");
    auto &out_tensor = *out_var->GetMutable<framework::LoDTensor>();
Y
yangyaming 已提交
50 51

    size_t height = dst_num_rows;
Y
yangyaming 已提交
52

53 54 55 56 57
    // do shrink for the top level LoD
    if (x_tensor.lod().size() > 0 &&
        x_tensor.lod()[0].size() > static_cast<size_t>(dst_num_rows)) {
      auto lod_offset = framework::GetSubLoDAndAbsoluteOffset(x_tensor.lod(), 0,
                                                              dst_num_rows, 0);
Y
yangyaming 已提交
58
      height = lod_offset.second.second;
Y
yangyaming 已提交
59
      auto out_lod = out_tensor.mutable_lod();
60
      framework::AppendLoD(out_lod, lod_offset.first);
Y
yangyaming 已提交
61 62
    }

Y
Yang Yu 已提交
63
    if (dst_num_rows != 0) {
Y
yangyaming 已提交
64
      out_tensor.ShareDataWith(x_tensor.Slice(0, height));
Y
Yang Yu 已提交
65 66 67 68
    }
  }
};

Y
Yang Yu 已提交
69
class ShrinkRNNMemoryOpProtoMaker : public framework::OpProtoAndCheckerMaker {
Y
Yang Yu 已提交
70
 public:
71
  ShrinkRNNMemoryOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker)
Y
Yang Yu 已提交
72
      : OpProtoAndCheckerMaker(proto, op_checker) {
73 74 75 76 77 78 79 80
    AddInput("X", "(LoDTensor) The RNN step memory to be shrinked.");
    AddInput("RankTable", "(LoDRankTable) The lod_rank_table of dynamic RNN.");
    AddInput("I",
             "(LoDTensor) The step index. The RNN step memory 'X' will be "
             "shrinked to match the size of the input of the index'th step.");
    AddOutput("Out", "(LoDTensor) The shrinked RNN step memory.");
    AddComment(
        R"DOC(
Y
yangyaming 已提交
81 82
        In dynamic RNN, we are able to handle sequences of different lengths.
        Because of the multiple lengths, the size of each step input can be
83
        different, which may lead to a mismatching between the input of
Y
yangyaming 已提交
84 85
        the current step and the memory generated by the previous one. This
        operator shrinks memory according to the size of the next step input,
86 87
        to make sure that they can match each other.
        )DOC");
Y
Yang Yu 已提交
88 89 90
  }
};

Y
Yang Yu 已提交
91
class ShrinkRNNMemoryInferShape : public framework::InferShapeBase {
Y
Yang Yu 已提交
92 93 94 95 96 97 98 99 100
 public:
  void operator()(framework::InferShapeContext *context) const override {
    PADDLE_ENFORCE(context->HasInput("X"));
    PADDLE_ENFORCE(context->HasInput("I"));
    PADDLE_ENFORCE(context->HasInput("RankTable"));
    context->SetOutputDim("Out", context->GetInputDim("X"));
  }
};

Y
Yang Yu 已提交
101
class ShrinkRNNMemoryGradOp : public ArrayOp {
Y
Yang Yu 已提交
102
 public:
Y
Yang Yu 已提交
103 104 105 106
  ShrinkRNNMemoryGradOp(const std::string &type,
                        const framework::VariableNameMap &inputs,
                        const framework::VariableNameMap &outputs,
                        const framework::AttributeMap &attrs)
Y
Yang Yu 已提交
107 108 109
      : ArrayOp(type, inputs, outputs, attrs) {}

  void Run(const framework::Scope &scope,
D
dzhwinter 已提交
110
           const platform::Place &place) const override {
Y
Yang Yu 已提交
111
    auto *dout_var = scope.FindVar(Input(framework::GradVarName("Out")));
Y
Yang Yu 已提交
112
    auto *dx_var = scope.FindVar(Output(framework::GradVarName("X")));
Y
Yang Yu 已提交
113 114 115 116 117 118 119 120 121
    PADDLE_ENFORCE(dx_var != nullptr, "Input Gradient should not be nullptr");
    auto *x_var = scope.FindVar(Input("X"));
    PADDLE_ENFORCE(x_var != nullptr);

    auto &x_tensor = x_var->Get<framework::LoDTensor>();
    auto &dx_tensor = *dx_var->GetMutable<framework::LoDTensor>();
    dx_tensor.Resize(x_tensor.dims());
    dx_tensor.mutable_data(x_tensor.place(), x_tensor.type());

D
dzhwinter 已提交
122
    // get device context from pool
Y
Yang Yu 已提交
123 124
    platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
    auto &dev_ctx = *pool.Get(place);
D
dzhwinter 已提交
125

Y
Yang Yu 已提交
126 127 128 129 130
    if (dout_var == nullptr) {  // dx_tensor fill zero
      math::set_constant(dev_ctx, &dx_tensor, 0.0f);
    } else {
      auto &dout_tensor = dout_var->Get<framework::LoDTensor>();
      auto height = dout_tensor.dims()[0];
D
dzhwinter 已提交
131
      auto slice = dx_tensor.Slice(0, static_cast<int>(height));
132
      framework::Copy(dout_tensor, dout_tensor.place(), dev_ctx, &slice);
Y
Refine  
Yang Yu 已提交
133
      if (dx_tensor.dims()[0] > height) {
Y
Yang Yu 已提交
134
        auto rest_tensor = dx_tensor.Slice(
Y
Refine  
Yang Yu 已提交
135
            static_cast<int>(height), static_cast<int>(dx_tensor.dims()[0]));
Y
Yang Yu 已提交
136 137 138 139 140 141
        math::set_constant(dev_ctx, &rest_tensor, 0.0f);
      }
    }
  }
};

Y
Yang Yu 已提交
142
class ShrinkRNNMemoryGradInferShape : public framework::InferShapeBase {
Y
Yang Yu 已提交
143 144 145 146 147 148
 public:
  void operator()(framework::InferShapeContext *context) const override {
    PADDLE_ENFORCE(context->HasInput("X"));
    PADDLE_ENFORCE(context->HasOutput(framework::GradVarName("X")));
    context->SetOutputDim(framework::GradVarName("X"),
                          context->GetInputDim("X"));
149
    context->ShareLoD("X", framework::GradVarName("X"));
Y
Yang Yu 已提交
150 151 152
  }
};

Y
Yang Yu 已提交
153
class ShrinkRNNGradOpMaker : public framework::SingleGradOpDescMaker {
Y
Yang Yu 已提交
154 155 156 157
 public:
  using framework::SingleGradOpDescMaker::SingleGradOpDescMaker;

 protected:
Y
Yu Yang 已提交
158 159
  std::unique_ptr<framework::OpDesc> Apply() const override {
    auto *op = new framework::OpDesc();
Y
Yang Yu 已提交
160
    op->SetType("shrink_rnn_memory_grad");
Y
Yang Yu 已提交
161 162 163 164
    op->SetInput("X", Input("X"));
    op->SetInput(framework::GradVarName("Out"), OutputGrad("Out"));
    op->SetOutput(framework::GradVarName("X"), InputGrad("X"));
    op->SetAttrMap(Attrs());
Y
Yu Yang 已提交
165
    return std::unique_ptr<framework::OpDesc>(op);
Y
Yang Yu 已提交
166 167 168 169 170 171 172
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
Y
Yang Yu 已提交
173 174 175 176 177
REGISTER_OPERATOR(shrink_rnn_memory, ops::ShrinkRNNMemoryOp,
                  ops::ShrinkRNNMemoryInferShape,
                  ops::ShrinkRNNMemoryOpProtoMaker, ops::ShrinkRNNGradOpMaker);
REGISTER_OPERATOR(shrink_rnn_memory_grad, ops::ShrinkRNNMemoryGradOp,
                  ops::ShrinkRNNMemoryGradInferShape);