pad_op.cc 4.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
W
wanghaoshuang 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
W
wanghaoshuang 已提交
6

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

Y
Yi Wang 已提交
15
#include "paddle/fluid/operators/pad_op.h"
S
sneaxiy 已提交
16
#include <memory>
W
wanghaoshuang 已提交
17 18 19 20 21 22 23 24 25 26

namespace paddle {
namespace operators {

using framework::Tensor;

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

27
  void InferShape(framework::InferShapeContext* ctx) const override {
Q
Qiao Longfei 已提交
28 29 30 31 32
    PADDLE_ENFORCE(ctx->HasInput("X"), "Input(X) of PadOp should not be null.");
    PADDLE_ENFORCE(ctx->HasOutput("Out"),
                   "Output(Out) of PadOp should not be null.");

    auto x_dim = ctx->GetInputDim("X");
S
sneaxiy 已提交
33
    auto& paddings = ctx->Attrs().Get<std::vector<int>>("paddings");
W
wanghaoshuang 已提交
34
    PADDLE_ENFORCE_EQ(x_dim.size() * 2, int64_t(paddings.size()),
W
wanghaoshuang 已提交
35 36
                      "Size of paddings should be equal to 2 * dimension size "
                      "of input tensor.");
W
wanghaoshuang 已提交
37
    std::vector<int64_t> out_dims(x_dim.size());
W
wanghaoshuang 已提交
38
    for (int i = 0; i < x_dim.size(); ++i) {
39 40 41 42 43
      if ((!ctx->IsRuntime()) && (x_dim[i] == -1)) {
        out_dims[i] = -1;
      } else {
        out_dims[i] = x_dim[i] + paddings[i * 2] + paddings[i * 2 + 1];
      }
W
wanghaoshuang 已提交
44
    }
Q
Qiao Longfei 已提交
45
    ctx->SetOutputDim("Out", framework::make_ddim(out_dims));
D
Fix bug  
dangqingqing 已提交
46 47 48
    if (out_dims[0] == x_dim[0]) {
      // Only pass LoD when the first dimension is equal between
      // output and input.
Q
Qiao Longfei 已提交
49
      ctx->ShareLoD("X", /*->*/ "Out");
D
Fix bug  
dangqingqing 已提交
50
    }
W
wanghaoshuang 已提交
51 52 53
  }
};

W
wanghaoshuang 已提交
54
class PadOpMaker : public framework::OpProtoAndCheckerMaker {
W
wanghaoshuang 已提交
55
 public:
Y
Yu Yang 已提交
56
  void Make() override {
W
wanghaoshuang 已提交
57 58 59 60
    AddInput("X",
             "The input of pad op. "
             "The input should be a k-D tensor(k > 0 and k < 7)");
    AddOutput("Out",
K
kexinzhao 已提交
61
              "The output of pad op. "
62
              "A tensor with the same shape as X.");
K
kexinzhao 已提交
63 64 65 66 67 68 69 70 71 72 73 74
    AddAttr<std::vector<int>>(
        "paddings",
        "(vector<int>) "
        "A list<int> to describe the padding rules for each dimension. "
        "For 2-D image tensor, paddings=[0, 1, 2, 3] means "
        "padding 0 row to top, 1 row to bottom, 2 columns to left "
        "and 3 columns to right. Size of paddings should be equal to "
        "2 * dimension size of the input tensor.");
    AddAttr<float>("pad_value",
                   "(float, default 0.0) "
                   "The value to fill the padded areas.")
        .SetDefault(0.0f);
W
wanghaoshuang 已提交
75
    AddComment(R"DOC(
K
kexinzhao 已提交
76 77 78 79
Pad Operator.

Pad input into output, as specified by paddings and pad_value. 
The input should be a k-D tensor(k > 0 and k < 7). As an example:
W
wanghaoshuang 已提交
80 81 82 83

Given:

X = [[1, 2],
K
kexinzhao 已提交
84
     [3, 4]],
W
wanghaoshuang 已提交
85

K
kexinzhao 已提交
86
paddings = [0, 1, 1, 2],
W
wanghaoshuang 已提交
87 88 89

and

K
kexinzhao 已提交
90
pad_value = 0,
Q
Qiao Longfei 已提交
91

K
kexinzhao 已提交
92
we have:
W
wanghaoshuang 已提交
93 94 95 96

Out = [[0, 1, 2, 0, 0]
       [0, 3, 4, 0, 0]
       [0, 0, 0, 0, 0]]
K
kexinzhao 已提交
97

W
wanghaoshuang 已提交
98 99 100 101 102 103 104 105
)DOC");
  }
};

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

106
  void InferShape(framework::InferShapeContext* ctx) const override {
Q
Qiao Longfei 已提交
107 108
    auto x_grad_name = framework::GradVarName("X");
    if (ctx->HasOutput(x_grad_name)) {
S
sneaxiy 已提交
109 110 111
      auto dout_dims = ctx->GetInputDim(framework::GradVarName("Out"));
      auto& paddings = ctx->Attrs().Get<std::vector<int>>("paddings");
      for (int i = 0; i < dout_dims.size(); ++i) {
112 113 114
        if (ctx->IsRuntime() || (dout_dims[i] != -1)) {
          dout_dims[i] -= (paddings[i * 2] + paddings[i * 2 + 1]);
        }
S
sneaxiy 已提交
115 116
      }
      ctx->SetOutputDim(x_grad_name, dout_dims);
W
wanghaoshuang 已提交
117
    }
W
wanghaoshuang 已提交
118 119 120
  }
};

121 122 123
class PadOpGradMaker : public framework::SingleGradOpDescMaker {
 public:
  using framework::SingleGradOpDescMaker::SingleGradOpDescMaker;
Y
Yu Yang 已提交
124 125

 protected:
Y
Yu Yang 已提交
126 127
  std::unique_ptr<framework::OpDesc> Apply() const override {
    auto* bind = new framework::OpDesc();
Y
Yu Yang 已提交
128 129 130
    bind->SetInput(framework::GradVarName("Out"), OutputGrad("Out"));
    bind->SetOutput(framework::GradVarName("X"), InputGrad("X"));
    bind->SetAttrMap(Attrs());
Y
Yu Yang 已提交
131
    bind->SetType("pad_grad");
Y
Yu Yang 已提交
132
    return std::unique_ptr<framework::OpDesc>(bind);
Y
Yu Yang 已提交
133
  }
134 135
};

W
wanghaoshuang 已提交
136 137 138 139
}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
140 141 142

REGISTER_OPERATOR(pad, ops::PadOp, ops::PadOpMaker, ops::PadOpGradMaker);
REGISTER_OPERATOR(pad_grad, ops::PadOpGrad);
Q
QI JUN 已提交
143 144 145 146
REGISTER_OP_CPU_KERNEL(
    pad, ops::PadKernel<paddle::platform::CPUDeviceContext, float>);
REGISTER_OP_CPU_KERNEL(
    pad_grad, ops::PadGradKernel<paddle::platform::CPUDeviceContext, float>);