scale_op.cc 4.5 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
#include "paddle/fluid/framework/infershape_utils.h"
#include "paddle/fluid/framework/op_registry.h"
18
#include "paddle/fluid/platform/float16.h"
19 20
#include "paddle/phi/core/infermeta_utils.h"
#include "paddle/phi/infermeta/unary.h"
W
wanghuancoder 已提交
21

Y
Yu Yang 已提交
22 23 24 25 26
namespace paddle {
namespace operators {

class ScaleOp : public framework::OperatorWithKernel {
 public:
27
  using framework::OperatorWithKernel::OperatorWithKernel;
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42

  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)) {
      return framework::OpKernelType(input_data_type, ctx.GetPlace(),
                                     framework::DataLayout::kMKLDNN,
                                     framework::LibraryType::kMKLDNN);
    }
#endif
    return framework::OpKernelType(input_data_type, ctx.GetPlace());
  }
Y
Yu Yang 已提交
43 44 45 46
};

class ScaleOpMaker : public framework::OpProtoAndCheckerMaker {
 public:
Y
Yu Yang 已提交
47
  void Make() override {
48
    AddInput("X", "(Tensor) Input tensor of scale operator.");
49 50 51 52 53
    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();
54 55
    AddOutput("Out", "(Tensor) Output tensor of scale operator.");
    AddComment(R"DOC(
Y
yi.wu 已提交
56 57
**Scale operator**

S
sneaxiy 已提交
58
Apply scaling and bias addition to the input tensor.
Y
Yu Yang 已提交
59

S
sneaxiy 已提交
60 61 62 63 64 65 66
if bias_after_scale=True:

$$Out = scale*X + bias$$

else:

$$Out = scale*(X + bias)$$
Y
Yu Yang 已提交
67
)DOC");
Y
yi.wu 已提交
68
    AddAttr<float>("scale", "The scaling factor of the scale operator.")
C
caoying03 已提交
69
        .SetDefault(1.0);
S
sneaxiy 已提交
70
    AddAttr<float>("bias", "The bias of the scale operator.").SetDefault(0.0);
S
sneaxiy 已提交
71 72 73 74 75
    AddAttr<bool>(
        "bias_after_scale",
        "Apply bias addition after or before scaling. It is useful for "
        "numeric stability in some circumstances.")
        .SetDefault(true);
76 77
    AddAttr<bool>("use_mkldnn",
                  "(bool, default false) Only used in mkldnn kernel")
78 79
        .SetDefault(false)
        .AsExtra();
Y
Yu Yang 已提交
80 81 82
  }
};

83 84
class ScaleOpVarTypeInference : public framework::VarTypeInference {
 public:
M
minqiyang 已提交
85
  void operator()(framework::InferVarTypeContext *ctx) const override {
86
    ctx->SyncTypeAndDataType("X", "Out");
87 88 89
  }
};

H
hong 已提交
90 91
template <typename T>
class ScaleGradMaker : public framework::SingleGradOpMaker<T> {
Y
Yu Yang 已提交
92
 public:
H
hong 已提交
93
  using framework::SingleGradOpMaker<T>::SingleGradOpMaker;
94

95
  void Apply(GradOpPtr<T> grad_op) const override {
Y
Yu Yang 已提交
96
    grad_op->SetType("scale");
H
hong 已提交
97
    grad_op->SetInput("X", this->OutputGrad("Out"));
98 99 100
    if (this->HasInput("ScaleTensor") > 0) {
      grad_op->SetInput("ScaleTensor", this->Input("ScaleTensor"));
    }
H
hong 已提交
101
    grad_op->SetOutput("Out", this->InputGrad("X"));
J
Jiabin Yang 已提交
102
    VLOG(6) << "Finish SetOutput";
H
hong 已提交
103
    grad_op->SetAttr("scale", this->GetAttr("scale"));
J
Jiabin Yang 已提交
104
    VLOG(6) << "Finish Set Attr scale";
S
sneaxiy 已提交
105
    grad_op->SetAttr("bias", 0.0f);
J
Jiabin Yang 已提交
106
    VLOG(6) << "Finish Set Attr bias";
S
sneaxiy 已提交
107
    grad_op->SetAttr("bias_after_scale", true);
J
Jiabin Yang 已提交
108 109 110
    VLOG(6) << "Finish Set Attr bias_after_scale";
    if (grad_op->HasAttr("use_mkldnn")) {
      VLOG(6) << "Finish Check Attr use_mkldnn";
111
      grad_op->SetAttr("use_mkldnn", this->GetAttr("use_mkldnn"));
J
Jiabin Yang 已提交
112 113 114
      VLOG(6) << "Finish Set Attr use_mkldnn";
    }
    VLOG(6) << "Finish Apply";
Y
Yu Yang 已提交
115 116 117
  }
};

118
DECLARE_INPLACE_OP_INFERER(ScaleOpInplaceInferer, {"X", "Out"});
Y
Yu Yang 已提交
119 120 121 122 123
}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;

124
DELCARE_INFER_SHAPE_FUNCTOR(scale, ScaleInferShapeFunctor,
125
                            PT_INFER_META(phi::UnchangedInferMeta));
H
hong 已提交
126 127 128
REGISTER_OPERATOR(scale, ops::ScaleOp, ops::ScaleOpMaker,
                  ops::ScaleGradMaker<paddle::framework::OpDesc>,
                  ops::ScaleGradMaker<paddle::imperative::OpBase>,
129 130
                  ScaleInferShapeFunctor, ops::ScaleOpVarTypeInference,
                  ops::ScaleOpInplaceInferer);