scale_op.cc 4.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Y
Yu Yang 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
Yu Yang 已提交
6

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

15
#include <string>
16

17 18
#include "paddle/fluid/framework/infershape_utils.h"
#include "paddle/fluid/framework/op_registry.h"
19
#include "paddle/fluid/platform/float16.h"
20 21
#include "paddle/phi/core/infermeta_utils.h"
#include "paddle/phi/infermeta/unary.h"
W
wanghuancoder 已提交
22

Y
Yu Yang 已提交
23 24 25 26 27
namespace paddle {
namespace operators {

class ScaleOp : public framework::OperatorWithKernel {
 public:
28
  using framework::OperatorWithKernel::OperatorWithKernel;
29 30 31 32 33 34 35 36

  framework::OpKernelType GetExpectedKernelType(
      const framework::ExecutionContext &ctx) const override {
    auto input_data_type =
        framework::OperatorWithKernel::IndicateVarDataType(ctx, "X");

#ifdef PADDLE_WITH_MKLDNN
    if (this->CanMKLDNNBeUsed(ctx, input_data_type)) {
37 38
      return framework::OpKernelType(input_data_type,
                                     ctx.GetPlace(),
39 40 41 42 43 44
                                     framework::DataLayout::kMKLDNN,
                                     framework::LibraryType::kMKLDNN);
    }
#endif
    return framework::OpKernelType(input_data_type, ctx.GetPlace());
  }
Y
Yu Yang 已提交
45 46 47 48
};

class ScaleOpMaker : public framework::OpProtoAndCheckerMaker {
 public:
Y
Yu Yang 已提交
49
  void Make() override {
50
    AddInput("X", "(Tensor) Input tensor of scale operator.");
51 52 53 54 55
    AddInput("ScaleTensor",
             "(Tensor) If provided, use this as "
             "scale factor, this has a higher priority than "
             "attr(scale), the shape of this tensor MUST BE 1.")
        .AsDispensable();
56 57
    AddOutput("Out", "(Tensor) Output tensor of scale operator.");
    AddComment(R"DOC(
Y
yi.wu 已提交
58 59
**Scale operator**

S
sneaxiy 已提交
60
Apply scaling and bias addition to the input tensor.
Y
Yu Yang 已提交
61

S
sneaxiy 已提交
62 63 64 65 66 67 68
if bias_after_scale=True:

$$Out = scale*X + bias$$

else:

$$Out = scale*(X + bias)$$
Y
Yu Yang 已提交
69
)DOC");
Y
yi.wu 已提交
70
    AddAttr<float>("scale", "The scaling factor of the scale operator.")
C
caoying03 已提交
71
        .SetDefault(1.0);
S
sneaxiy 已提交
72
    AddAttr<float>("bias", "The bias of the scale operator.").SetDefault(0.0);
S
sneaxiy 已提交
73 74 75 76 77
    AddAttr<bool>(
        "bias_after_scale",
        "Apply bias addition after or before scaling. It is useful for "
        "numeric stability in some circumstances.")
        .SetDefault(true);
78 79
    AddAttr<bool>("use_mkldnn",
                  "(bool, default false) Only used in mkldnn kernel")
80 81
        .SetDefault(false)
        .AsExtra();
Y
Yu Yang 已提交
82 83 84
  }
};

85 86
class ScaleOpVarTypeInference : public framework::VarTypeInference {
 public:
M
minqiyang 已提交
87
  void operator()(framework::InferVarTypeContext *ctx) const override {
88
    ctx->SyncTypeAndDataType("X", "Out");
89 90 91
  }
};

H
hong 已提交
92 93
template <typename T>
class ScaleGradMaker : public framework::SingleGradOpMaker<T> {
Y
Yu Yang 已提交
94
 public:
H
hong 已提交
95
  using framework::SingleGradOpMaker<T>::SingleGradOpMaker;
96

97
  void Apply(GradOpPtr<T> grad_op) const override {
Y
Yu Yang 已提交
98
    grad_op->SetType("scale");
H
hong 已提交
99
    grad_op->SetInput("X", this->OutputGrad("Out"));
100 101 102
    if (this->HasInput("ScaleTensor") > 0) {
      grad_op->SetInput("ScaleTensor", this->Input("ScaleTensor"));
    }
H
hong 已提交
103
    grad_op->SetOutput("Out", this->InputGrad("X"));
J
Jiabin Yang 已提交
104
    VLOG(6) << "Finish SetOutput";
H
hong 已提交
105
    grad_op->SetAttr("scale", this->GetAttr("scale"));
J
Jiabin Yang 已提交
106
    VLOG(6) << "Finish Set Attr scale";
S
sneaxiy 已提交
107
    grad_op->SetAttr("bias", 0.0f);
J
Jiabin Yang 已提交
108
    VLOG(6) << "Finish Set Attr bias";
S
sneaxiy 已提交
109
    grad_op->SetAttr("bias_after_scale", true);
J
Jiabin Yang 已提交
110 111 112
    VLOG(6) << "Finish Set Attr bias_after_scale";
    if (grad_op->HasAttr("use_mkldnn")) {
      VLOG(6) << "Finish Check Attr use_mkldnn";
113
      grad_op->SetAttr("use_mkldnn", this->GetAttr("use_mkldnn"));
J
Jiabin Yang 已提交
114 115 116
      VLOG(6) << "Finish Set Attr use_mkldnn";
    }
    VLOG(6) << "Finish Apply";
Y
Yu Yang 已提交
117 118 119
  }
};

120
DECLARE_INPLACE_OP_INFERER(ScaleOpInplaceInferer, {"X", "Out"});
Y
Yu Yang 已提交
121 122 123 124 125
}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;

126 127
DECLARE_INFER_SHAPE_FUNCTOR(scale,
                            ScaleInferShapeFunctor,
128
                            PD_INFER_META(phi::UnchangedInferMeta));
129 130 131
REGISTER_OPERATOR(scale,
                  ops::ScaleOp,
                  ops::ScaleOpMaker,
H
hong 已提交
132 133
                  ops::ScaleGradMaker<paddle::framework::OpDesc>,
                  ops::ScaleGradMaker<paddle::imperative::OpBase>,
134 135
                  ScaleInferShapeFunctor,
                  ops::ScaleOpVarTypeInference,
136
                  ops::ScaleOpInplaceInferer);