activation_grad_kernel.h 10.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.

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

    http://www.apache.org/licenses/LICENSE-2.0

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. */

#pragma once

Y
YuanRisheng 已提交
17
#include "paddle/phi/common/scalar.h"
18 19
#include "paddle/phi/core/dense_tensor.h"
#include "paddle/phi/infermeta/unary.h"
20
#include "paddle/utils/optional.h"
21 22 23

namespace phi {

Y
YuanRisheng 已提交
24
#define DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(name) \
25 26 27 28 29 30
  template <typename T, typename Context>         \
  void name##GradKernel(const Context& dev_ctx,   \
                        const DenseTensor& x,     \
                        const DenseTensor& dout,  \
                        DenseTensor* dx);

Y
YuanRisheng 已提交
31
#define DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(name, attr) \
Y
YuanRisheng 已提交
32 33 34 35 36 37 38
  template <typename T, typename Context>                       \
  void name##GradKernel(const Context& dev_ctx,                 \
                        const DenseTensor& x,                   \
                        const DenseTensor& dout,                \
                        float attr,                             \
                        DenseTensor* dx);

Y
YuanRisheng 已提交
39
#define DECLARE_ACT_GRAD_KERNEL_WITH_TWO_ATTRS_DEPX(name, attr1, attr2) \
Y
YuanRisheng 已提交
40 41 42 43 44 45 46 47
  template <typename T, typename Context>                               \
  void name##GradKernel(const Context& dev_ctx,                         \
                        const DenseTensor& x,                           \
                        const DenseTensor& dout,                        \
                        float attr1,                                    \
                        float attr2,                                    \
                        DenseTensor* dx);

Y
YuanRisheng 已提交
48
#define DECLARE_ACTIVATION_GRAD_KERNEL_DEPOUT(name) \
49 50 51 52 53 54
  template <typename T, typename Context>           \
  void name##GradKernel(const Context& dev_ctx,     \
                        const DenseTensor& out,     \
                        const DenseTensor& dout,    \
                        DenseTensor* dx);

Y
YuanRisheng 已提交
55 56 57 58 59
#define DECLARE_ACTIVATION_GRAD_KERNEL_NODEP(name) \
  template <typename T, typename Context>          \
  void name##GradKernel(                           \
      const Context& dev_ctx, const DenseTensor& dout, DenseTensor* dx);

Y
YuanRisheng 已提交
60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
#define DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPOUT(name, attr) \
  template <typename T, typename Context>                         \
  void name##GradKernel(const Context& dev_ctx,                   \
                        const DenseTensor& out,                   \
                        const DenseTensor& dout,                  \
                        float attr,                               \
                        DenseTensor* dx);

#define DECLARE_ACT_GRAD_KERNEL_WITH_TWO_ATTRS_DEPOUT(name, attr1, attr2) \
  template <typename T, typename Context>                                 \
  void name##GradKernel(const Context& dev_ctx,                           \
                        const DenseTensor& out,                           \
                        const DenseTensor& dout,                          \
                        float attr1,                                      \
                        float attr2,                                      \
Y
YuanRisheng 已提交
75 76
                        DenseTensor* dx);

77 78 79 80 81 82
template <typename T, typename Context>
void ReluDoubleGradKernel(const Context& dev_ctx,
                          const DenseTensor& out,
                          const DenseTensor& ddx,
                          DenseTensor* ddout);

83 84 85 86
template <typename T, typename Context>
void TanhDoubleGradKernel(const Context& dev_ctx,
                          const DenseTensor& out,
                          const DenseTensor& dout,
87
                          const DenseTensor& ddx,
88 89 90 91 92 93 94
                          DenseTensor* dout_new,
                          DenseTensor* ddout);

template <typename T, typename Context>
void TanhTripleGradKernel(const Context& dev_ctx,
                          const DenseTensor& out,
                          const DenseTensor& dout,
95
                          const DenseTensor& ddx,
96
                          const DenseTensor& d_dout_new,
97
                          const DenseTensor& d_ddout,
98 99 100 101 102 103 104 105 106 107 108 109
                          DenseTensor* d_out_new,
                          DenseTensor* d_dout,
                          DenseTensor* d_ddx);

template <typename T, typename Context>
void LeakyReluDoubleGradKernel(const Context& dev_ctx,
                               const DenseTensor& x,
                               const DenseTensor& ddx,
                               float alpha,
                               DenseTensor* ddout);

template <typename T, typename Context>
Y
YuanRisheng 已提交
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
void EluGradKernel(const Context& dev_ctx,
                   const DenseTensor& x,
                   const DenseTensor& out,
                   const DenseTensor& dout,
                   float alpha,
                   DenseTensor* dx);

template <typename T, typename Context>
void EluDoubleGradKernel(const Context& dev_ctx,
                         const DenseTensor& x,
                         const DenseTensor& dout,
                         const DenseTensor& ddx,
                         float alpha,
                         DenseTensor* dx,
                         DenseTensor* ddout);
125

Y
YuanRisheng 已提交
126 127 128 129
template <typename T, typename Context>
void SigmoidDoubleGradKernel(const Context& dev_ctx,
                             const DenseTensor& out,
                             const DenseTensor& dout,
130
                             const DenseTensor& ddx,
Y
YuanRisheng 已提交
131 132 133 134 135 136 137
                             DenseTensor* dout_new,
                             DenseTensor* ddout);

template <typename T, typename Context>
void SigmoidTripleGradKernel(const Context& dev_ctx,
                             const DenseTensor& out,
                             const DenseTensor& dout,
138
                             const DenseTensor& ddx,
Y
YuanRisheng 已提交
139
                             const DenseTensor& d_dout_new,
140
                             paddle::optional<const DenseTensor&> d_ddout,
Y
YuanRisheng 已提交
141 142 143 144
                             DenseTensor* d_out_new,
                             DenseTensor* d_dout,
                             DenseTensor* d_ddx);

145 146 147 148 149 150 151 152
template <typename T, typename Context>
void LogDoubleGradKernel(const Context& dev_ctx,
                         const DenseTensor& x,
                         const DenseTensor& dout,
                         const DenseTensor& ddx,
                         DenseTensor* dx,
                         DenseTensor* ddout);

Y
YuanRisheng 已提交
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
template <typename T, typename Context>
void SqrtDoubleGradKernel(const Context& dev_ctx,
                          const DenseTensor& out,
                          const DenseTensor& dx,
                          const DenseTensor& ddx,
                          DenseTensor* dout,
                          DenseTensor* ddout);

template <typename T, typename Context>
void RsqrtDoubleGradKernel(const Context& dev_ctx,
                           const DenseTensor& out,
                           const DenseTensor& dx,
                           const DenseTensor& ddx,
                           DenseTensor* dout,
                           DenseTensor* ddout);

template <typename T, typename Context>
void CeluDoubleGradKernel(const Context& dev_ctx,
                          const DenseTensor& x,
                          const DenseTensor& dout,
                          const DenseTensor& ddx,
                          float alpha,
                          DenseTensor* dx,
                          DenseTensor* ddout);

template <typename T, typename Context>
void SquareDoubleGradKernel(const Context& dev_ctx,
                            const DenseTensor& x,
                            const DenseTensor& dout,
                            const DenseTensor& ddx,
                            DenseTensor* dx,
                            DenseTensor* ddout);

Y
YuanRisheng 已提交
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
template <typename T, typename Context>
void HardSwishGradKernel(const Context& dev_ctx,
                         const DenseTensor& x,
                         const DenseTensor& dout,
                         float threshold,
                         float scale,
                         float offset,
                         DenseTensor* dx);

template <typename T, typename Context>
void PowGradKernel(const Context& dev_ctx,
                   const DenseTensor& x,
                   const DenseTensor& dout,
                   const Scalar& factor,
                   DenseTensor* dx);

Y
YuanRisheng 已提交
202 203 204 205 206 207 208 209 210 211 212 213 214 215
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Cos);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Tan);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Acos);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Sin);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Asin);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Atan);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Sinh);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Cosh);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Asinh);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Acosh);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Atanh);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(TanhShrink);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Silu);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(LogSigmoid);
216 217 218 219
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Log);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Log2);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Log10);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPX(Log1p);
Y
YuanRisheng 已提交
220 221 222 223

DECLARE_ACTIVATION_GRAD_KERNEL_DEPOUT(Relu);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPOUT(Tanh);
DECLARE_ACTIVATION_GRAD_KERNEL_DEPOUT(Sigmoid);
224
DECLARE_ACTIVATION_GRAD_KERNEL_DEPOUT(Sqrt);
Y
YuanRisheng 已提交
225

Y
YuanRisheng 已提交
226 227 228 229
DECLARE_ACTIVATION_GRAD_KERNEL_NODEP(Round);
DECLARE_ACTIVATION_GRAD_KERNEL_NODEP(Floor);
DECLARE_ACTIVATION_GRAD_KERNEL_NODEP(Ceil);

Y
YuanRisheng 已提交
230 231 232 233
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(LeakyRelu, alpha);
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(ThresholdedRelu, threshold);
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(SoftShrink, lambda);
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(HardShrink, threshold);
Y
YuanRisheng 已提交
234
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(Swish, beta);
235
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(Logit, eps);
Y
YuanRisheng 已提交
236
DECLARE_ACT_GRAD_KERNEL_WITH_ONE_ATTRS_DEPX(Celu, alpha);
Y
YuanRisheng 已提交
237 238 239 240

DECLARE_ACT_GRAD_KERNEL_WITH_TWO_ATTRS_DEPX(BRelu, t_min, t_max);

DECLARE_ACT_GRAD_KERNEL_WITH_TWO_ATTRS_DEPOUT(HardSigmoid, slope, offset);
Y
YuanRisheng 已提交
241

242
}  // namespace phi