elementwise_add_op.h 9.9 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
G
gongweibao 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
G
gongweibao 已提交
6

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

F
fengjiayi 已提交
15 16
#pragma once

17 18
#include <algorithm>
#include <utility>
W
Wu Yi 已提交
19
#include "paddle/fluid/operators/elementwise/elementwise_op.h"
20

21
#include "paddle/pten/kernels/math_kernel.h"
22

G
gongweibao 已提交
23 24 25
namespace paddle {
namespace operators {

26
template <typename DeviceContext, typename T>
27 28 29 30
void LaunchBroadcastElementwiseCpuKernel(const framework::ExecutionContext &ctx,
                                         const framework::Tensor *x,
                                         const framework::Tensor *y,
                                         framework::Tensor *z) {
31
  int axis = ctx.Attr<int>("axis");
32 33 34
  auto x_dims = x->dims();
  auto y_dims = y->dims();
  if (x_dims.size() >= y_dims.size()) {
35 36 37 38 39 40
    ElementwiseComputeEx<AddFunctor<T>, DeviceContext, T>(ctx, x, y, axis,
                                                          AddFunctor<T>(), z);
  } else {
    ElementwiseComputeEx<InverseAddFunctor<T>, DeviceContext, T>(
        ctx, x, y, axis, InverseAddFunctor<T>(), z);
  }
41 42
}

43 44 45 46 47 48
template <typename DeviceContext, typename T, class Enable = void>
struct SameDimsElemwiseAdd {
  void operator()(const framework::ExecutionContext &ctx,
                  const framework::Tensor *x, const framework::Tensor *y,
                  framework::Tensor *z);
};
49

Q
QI JUN 已提交
50
template <typename DeviceContext, typename T>
Y
Yu Yang 已提交
51
class ElementwiseAddKernel : public framework::OpKernel<T> {
G
gongweibao 已提交
52
 public:
C
chengduo 已提交
53 54 55 56
  void Compute(const framework::ExecutionContext &ctx) const override {
    auto *x = ctx.Input<framework::LoDTensor>("X");
    auto *y = ctx.Input<framework::LoDTensor>("Y");
    auto *z = ctx.Output<framework::LoDTensor>("Out");
C
chengduoZH 已提交
57
    z->mutable_data<T>(ctx.GetPlace());
58 59 60 61 62 63

    auto &dev_ctx = ctx.device_context<DeviceContext>();
    int axis = ctx.Attr<int>("axis");
    auto pt_x = paddle::experimental::MakePtenDenseTensor(*x);
    auto pt_y = paddle::experimental::MakePtenDenseTensor(*y);
    auto pt_z = paddle::experimental::MakePtenDenseTensor(*z);
64
    pten::AddKernel<T>(dev_ctx, *pt_x.get(), *pt_y.get(), axis, pt_z.get());
G
gongweibao 已提交
65 66 67 68
  }
};

template <typename T>
Y
Yu Yang 已提交
69 70
struct IdentityGrad {
  HOSTDEVICE T operator()(T x, T y, T out, T dout) const { return dout; }
G
gongweibao 已提交
71 72
};

73
template <typename DeviceContext, typename T>
74 75 76 77 78 79 80 81
typename std::enable_if<
    std::is_same<DeviceContext, platform::CPUDeviceContext>::value>::type
default_elementwise_add_grad(const framework::ExecutionContext &ctx,
                             const framework::Tensor *x,
                             const framework::Tensor *y,
                             const framework::Tensor *out,
                             const framework::Tensor *dout,
                             framework::Tensor *dx, framework::Tensor *dy) {
82 83
  int axis = ctx.Attr<int>("axis");

84 85 86 87
  ElemwiseExplicitGradCompute<DeviceContext, T, IdentityGrad<T>,
                              IdentityGrad<T>>(ctx, *x, *y, *out, *dout, axis,
                                               dx, dy, IdentityGrad<T>(),
                                               IdentityGrad<T>());
88 89
}

90
template <typename DeviceContext, typename T>
91 92 93
typename std::enable_if<
    std::is_floating_point<T>::value &&
    std::is_same<DeviceContext, platform::CPUDeviceContext>::value>::type
94 95 96 97 98
elementwise_add_grad(const framework::ExecutionContext &ctx,
                     const framework::Tensor *x, const framework::Tensor *y,
                     const framework::Tensor *out,
                     const framework::Tensor *dout, framework::Tensor *dx,
                     framework::Tensor *dy) {
99 100 101 102 103 104 105 106 107 108 109 110
  auto blas = math::GetBlas<DeviceContext, T>(ctx);
  if (dx) {
    blas.VCOPY(dout->numel(), dout->data<T>(),
               dx->mutable_data<T>(ctx.GetPlace()));
  }

  if (dy) {
    blas.VCOPY(dout->numel(), dout->data<T>(),
               dy->mutable_data<T>(ctx.GetPlace()));
  }
}

111
template <typename DeviceContext, typename T>
112
typename std::enable_if<
113 114
    !std::is_floating_point<T>::value &&
    std::is_same<DeviceContext, platform::CPUDeviceContext>::value>::type
115 116 117 118 119 120
elementwise_add_grad(const framework::ExecutionContext &ctx,
                     const framework::Tensor *x, const framework::Tensor *y,
                     const framework::Tensor *out,
                     const framework::Tensor *dout, framework::Tensor *dx,
                     framework::Tensor *dy) {
  default_elementwise_add_grad<DeviceContext, T>(ctx, x, y, out, dout, dx, dy);
121 122
}

123
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
124 125 126 127
// cuda definition
template <typename DeviceContext, typename T>
typename std::enable_if<
    std::is_same<DeviceContext, platform::CUDADeviceContext>::value>::type
128 129 130 131 132
elementwise_add_grad(const framework::ExecutionContext &ctx,
                     const framework::Tensor *x, const framework::Tensor *y,
                     const framework::Tensor *out,
                     const framework::Tensor *dout, framework::Tensor *dx,
                     framework::Tensor *dy);
133 134 135 136 137 138 139 140 141 142

template <typename DeviceContext, typename T>
typename std::enable_if<
    std::is_same<DeviceContext, platform::CUDADeviceContext>::value>::type
default_elementwise_add_grad(const framework::ExecutionContext &ctx,
                             const framework::Tensor *x,
                             const framework::Tensor *y,
                             const framework::Tensor *out,
                             const framework::Tensor *dout,
                             framework::Tensor *dx, framework::Tensor *dy);
143 144
#endif

Q
QI JUN 已提交
145
template <typename DeviceContext, typename T>
146
class ElementwiseAddGradKernel : public ElemwiseGradKernel<T> {
G
gongweibao 已提交
147
 public:
C
chengduo 已提交
148
  void Compute(const framework::ExecutionContext &ctx) const override {
149 150
    ElemwiseGradKernel<T>::Compute(ctx);

C
chengduoZH 已提交
151 152
    using Tensor = framework::Tensor;

153 154
    auto *x = ctx.Input<Tensor>("X");
    auto *y = ctx.Input<Tensor>("Y");
C
chengduo 已提交
155 156 157
    auto *dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
    auto *dx = ctx.Output<Tensor>(framework::GradVarName("X"));
    auto *dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
158
    // skip out
C
chengduo 已提交
159
    auto *out = dout;
160

161 162 163 164 165 166 167 168 169 170 171 172 173 174
    // Special case when dy is not needed and dx doesn't reduce
    if (dx != nullptr && dy == nullptr && dx->dims() == dout->dims()) {
      VLOG(4) << "Special case when dy is not needed and dx doesn't "
                 "reduce";
      framework::TensorCopy(
          *dout, ctx.GetPlace(),
          ctx.template device_context<platform::DeviceContext>(), dx);
    } else if (dx == nullptr && dy != nullptr && dy->dims() == dout->dims()) {
      VLOG(4) << "Special case when dx is not needed and dy doesn't "
                 "reduce";
      framework::TensorCopy(
          *dout, ctx.GetPlace(),
          ctx.template device_context<platform::DeviceContext>(), dy);
    } else if (dx != nullptr && dy != nullptr && (dx->dims() == dy->dims())) {
175
      elementwise_add_grad<DeviceContext, T>(ctx, x, y, out, dout, dx, dy);
176
    } else {
177 178
      default_elementwise_add_grad<DeviceContext, T>(ctx, x, y, out, dout, dx,
                                                     dy);
179
    }
G
gongweibao 已提交
180 181 182
  }
};

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202
template <typename DeviceContext, typename T>
class ElementwiseAddDoubleGradKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext &ctx) const override {
    using Tensor = framework::Tensor;

    auto *y = ctx.Input<Tensor>("Y");
    auto *dout = ctx.Input<Tensor>("DOut");
    auto *ddx = ctx.Input<Tensor>("DDX");
    auto *ddy = ctx.Input<Tensor>("DDY");

    auto *ddout = ctx.Output<Tensor>("DDOut");

    // ddOut = ddx + ddy
    if (ddout) {
      Tensor ddx_safe, ddy_safe;
      GetDoubleGradSafeTensor<DeviceContext, T>(ctx, dout, ddx, &ddx_safe);
      GetDoubleGradSafeTensor<DeviceContext, T>(ctx, y, ddy, &ddy_safe);

      ddout->mutable_data<T>(ctx.GetPlace());
203 204
      LaunchBroadcastElementwiseCpuKernel<DeviceContext, T>(ctx, &ddx_safe,
                                                            &ddy_safe, ddout);
205 206 207 208
    }
  }
};

209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
template <typename DeviceContext, typename T>
class ElementwiseAddTripleGradKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext &ctx) const override {
    using Tensor = framework::Tensor;
    auto *ddx = ctx.Input<Tensor>("DDX");
    auto *ddy = ctx.Input<Tensor>("DDY");
    auto *d_ddout = ctx.Input<Tensor>("D_DDOut");
    auto *d_ddx = ctx.Output<Tensor>("D_DDX");
    auto *d_ddy = ctx.Output<Tensor>("D_DDY");
    // skip out
    auto *out = d_ddout;

    // Special case when d_ddy is not needed and d_ddx doesn't reduce
    if (d_ddx != nullptr && d_ddy == nullptr &&
        d_ddx->dims() == d_ddout->dims()) {
      VLOG(4) << "Special case when d_ddy is not needed and d_ddx doesn't "
                 "reduce";
      framework::TensorCopy(
          *d_ddout, ctx.GetPlace(),
          ctx.template device_context<platform::DeviceContext>(), d_ddx);
    } else if (d_ddx == nullptr && d_ddy != nullptr &&
               d_ddy->dims() == d_ddout->dims()) {
      VLOG(4) << "Special case when d_ddx is not needed and d_ddy doesn't "
                 "reduce";
      framework::TensorCopy(
          *d_ddout, ctx.GetPlace(),
          ctx.template device_context<platform::DeviceContext>(), d_ddy);
    } else if (d_ddx != nullptr && d_ddy != nullptr &&
               (d_ddx->dims() == d_ddy->dims())) {
      elementwise_add_grad<DeviceContext, T>(ctx, ddx, ddy, out, d_ddout, d_ddx,
                                             d_ddy);
    } else {
      default_elementwise_add_grad<DeviceContext, T>(ctx, ddx, ddy, out,
                                                     d_ddout, d_ddx, d_ddy);
    }
  }
};

G
gongweibao 已提交
248 249
}  // namespace operators
}  // namespace paddle