lod_reset_op.cc 6.0 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
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
6

L
Luo Tao 已提交
7
    http://www.apache.org/licenses/LICENSE-2.0
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. */
14

Y
Yi Wang 已提交
15
#include "paddle/fluid/operators/lod_reset_op.h"
16 17 18 19 20 21 22 23 24 25 26 27 28

namespace paddle {
namespace operators {

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

  void InferShape(framework::InferShapeContext *ctx) const override {
    PADDLE_ENFORCE(ctx->HasInput("X"),
                   "Input(X) of LoDResetOp should not be null.");
    PADDLE_ENFORCE(ctx->HasOutput("Out"),
                   "Output(Out) of LoDResetOp should not be null.");
29 30

    if (!ctx->HasInput("Y")) {
31
      auto level0 = ctx->Attrs().Get<std::vector<int>>("target_lod");
32
      PADDLE_ENFORCE_GT(level0.size(), 1,
Y
yangyaming 已提交
33
                        "If Input(Y) not provided, the target lod should be "
34
                        "specified by attribute `target_lod`.");
H
Hongyu Liu 已提交
35
    } else {
H
Hongyu Liu 已提交
36
      ctx->ShareLoD("Y", "Out");
P
phlrain 已提交
37 38
    }

39 40 41 42
    ctx->SetOutputDim("Out", ctx->GetInputDim("X"));
  }

 protected:
43
  framework::OpKernelType GetExpectedKernelType(
44
      const framework::ExecutionContext &ctx) const override {
Y
Yu Yang 已提交
45 46
    return framework::OpKernelType(ctx.Input<framework::LoDTensor>("X")->type(),
                                   ctx.device_context());
47 48 49 50 51
  }
};

class LoDResetOpMaker : public framework::OpProtoAndCheckerMaker {
 public:
Y
Yu Yang 已提交
52
  void Make() override {
53 54 55 56 57
    AddInput("X",
             "(Tensor, LoDTensor) Input variable of LoDResetOp which "
             "could be a Tensor or LoDTensor, where the data of output "
             "variable inherits from.");
    AddInput("Y",
Y
yangyaming 已提交
58 59 60 61
             "(Tensor, LoDTensor, optional) If provided and Y is LoDTensor, "
             "lod of Input(Y) would be considered as the target lod first, "
             "otherwise data of Input(Y) would be considered as the "
             "target lod.")
62
        .AsDispensable();
63 64 65
    AddOutput("Out",
              "(LoDTensor) Output variable of LoDResetOp which should be a "
              "LoDTensor.");
66 67 68 69 70
    AddAttr<std::vector<int>>("target_lod",
                              "The target level 0 LoD from Attr().")
        .SetDefault(std::vector<int>{});
    AddComment(R"DOC(LoDReset operator

71
Set LoD of `X` to a new one specified by `Y` or attribute `target_lod`. When `Y`
Y
yangyaming 已提交
72 73 74 75 76
provided and `Y` is a LoDTensor, `Y.lod` would be considered as target LoD
first, otherwise `Y.data` would be considered as target LoD. If `Y` is not
provided, target LoD should be specified by attribute `target_lod`.
If target LoD is specified by `Y.data` or `target_lod`, only one level LoD
is supported.
77

Y
yangyaming 已提交
78
Example 1:
79

Y
yangyaming 已提交
80 81
Given a 1-level LoDTensor input(X):
    X.lod =  [[ 0,     2,                   5      6 ]]
82 83
    X.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
    X.dims = [6, 1]
84

Y
yangyaming 已提交
85
attr(target_lod): [0, 4, 6]
86

Y
yangyaming 已提交
87
then we get a 1-level LoDTensor:
88 89 90
    Out.lod =  [[ 0,                   4,            6 ]]
    Out.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
    Out.dims = [6, 1]
91

Y
yangyaming 已提交
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
Example 2:

Given a 1-level LoDTensor input(X):
    X.lod =  [[ 0,     2,                   5      6 ]]
    X.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
    X.dims = [6, 1]

input(Y) is a Tensor:
    Y.data = [[0, 2, 6]]
    Y.dims = [1, 3]

then we get a 1-level LoDTensor:
    Out.lod =  [[ 0,     2,                          6 ]]
    Out.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
    Out.dims = [6, 1]

Example 3:

Given a 1-level LoDTensor input(X):
    X.lod =  [[ 0,      2,                   5     6 ]]
    X.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
    X.dims = [6, 1]

input(Y) is a 2-level LoDTensor:
    Y.lod =  [[0, 2, 4], [0, 2, 5, 6]]
    Y.data = [[1.1], [2.1], [3.1], [4.1], [5.1], [6.1]]
    Y.dims = [6, 1]

then we get a 2-level LoDTensor:
    Out.lod =  [[0, 2, 4], [0, 2, 5, 6]]
    Out.data = [[1.0], [2.0], [3.0], [4.0], [5.0], [6.0]]
    Out.dims = [6, 1]

125 126 127 128 129 130 131 132 133
)DOC");
  }
};

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

  void InferShape(framework::InferShapeContext *ctx) const override {
134 135
    PADDLE_ENFORCE(ctx->HasInput("X"),
                   "Input(X) of LoDResetGradOp should not be null.");
136
    PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Out")),
137 138 139 140 141 142 143
                   "Input(Out@Grad) of LoDResetGradOp should not be null.");

    auto x_grad_name = framework::GradVarName("X");
    if (ctx->HasOutput(x_grad_name)) {
      ctx->SetOutputDim(x_grad_name, ctx->GetInputDim("X"));
      ctx->ShareLoD("X", /*->*/ x_grad_name);
    }
144 145 146
  }

 protected:
147
  framework::OpKernelType GetExpectedKernelType(
148
      const framework::ExecutionContext &ctx) const override {
Y
Yu Yang 已提交
149 150
    return framework::OpKernelType(ctx.Input<framework::LoDTensor>("X")->type(),
                                   ctx.device_context());
151 152 153 154 155 156 157
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
Y
Yang Yang 已提交
158
REGISTER_OPERATOR(lod_reset, ops::LoDResetOp, ops::LoDResetOpMaker,
159 160
                  paddle::framework::DefaultGradOpDescMaker<true>);
REGISTER_OPERATOR(lod_reset_grad, ops::LoDResetGradOp);
161 162 163 164 165
REGISTER_OP_CPU_KERNEL(
    lod_reset, ops::LoDResetKernel<paddle::platform::CPUPlace, float>,
    ops::LoDResetKernel<paddle::platform::CPUPlace, double>,
    ops::LoDResetKernel<paddle::platform::CPUPlace, int>,
    ops::LoDResetKernel<paddle::platform::CPUPlace, int64_t>);
166 167
REGISTER_OP_CPU_KERNEL(
    lod_reset_grad, ops::LoDResetGradKernel<paddle::platform::CPUPlace, float>,
168 169 170
    ops::LoDResetGradKernel<paddle::platform::CPUPlace, double>,
    ops::LoDResetGradKernel<paddle::platform::CPUPlace, int>,
    ops::LoDResetGradKernel<paddle::platform::CPUPlace, int64_t>);