elementwise_add_op.h 14.2 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. */
F
fengjiayi 已提交
14 15
#pragma once

16 17
#include <algorithm>
#include <utility>
W
Wu Yi 已提交
18
#include "paddle/fluid/operators/elementwise/elementwise_op.h"
19
#include "paddle/fluid/operators/elementwise/elementwise_op_function.cu.h"
W
Wu Yi 已提交
20
#include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
21
#include "paddle/fluid/operators/math/blas.h"
W
wanghuancoder 已提交
22

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

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

42 43 44 45 46 47
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);
};
48

Q
QI JUN 已提交
49
template <typename DeviceContext, typename T>
Y
Yu Yang 已提交
50
class ElementwiseAddKernel : public framework::OpKernel<T> {
G
gongweibao 已提交
51
 public:
C
chengduo 已提交
52 53 54 55
  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
56
    z->mutable_data<T>(ctx.GetPlace());
57
    auto dims_equal = x->dims() == y->dims();
58
    if (dims_equal) {
59 60
      SameDimsElemwiseAdd<DeviceContext, T> same_dims_add;
      same_dims_add(ctx, x, y, z);
61
    } else {
62
      default_elementwise_add<DeviceContext, T>(ctx, x, y, z);
63
    }
G
gongweibao 已提交
64 65 66 67
  }
};

template <typename T>
Y
Yu Yang 已提交
68 69
struct IdentityGrad {
  HOSTDEVICE T operator()(T x, T y, T out, T dout) const { return dout; }
G
gongweibao 已提交
70 71
};

72
template <typename DeviceContext, typename T>
73 74 75 76 77 78 79
void 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) {
80 81
  int axis = ctx.Attr<int>("axis");

82 83 84 85
  ElemwiseExplicitGradCompute<DeviceContext, T, IdentityGrad<T>,
                              IdentityGrad<T>>(ctx, *x, *y, *out, *dout, axis,
                                               dx, dy, IdentityGrad<T>(),
                                               IdentityGrad<T>());
86 87
}

88
template <typename DeviceContext, typename T>
89 90 91
typename std::enable_if<
    std::is_floating_point<T>::value &&
    std::is_same<DeviceContext, platform::CPUDeviceContext>::value>::type
92 93 94 95 96
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) {
97 98 99 100 101 102 103 104 105 106 107 108
  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()));
  }
}

109
template <typename DeviceContext, typename T>
110
typename std::enable_if<
111 112
    !std::is_floating_point<T>::value &&
    std::is_same<DeviceContext, platform::CPUDeviceContext>::value>::type
113 114 115 116 117 118
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);
119 120
}

121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249
#ifdef PADDLE_WITH_CUDA
#ifdef __NVCC__

template <typename T, int BLOCK_W, int BLOCK_H>
__global__ void MatrixColReduce(const T *__restrict__ in, T *__restrict__ out,
                                size_t width, size_t height) {
  __shared__ T sdata[BLOCK_H][BLOCK_W + 1];
  size_t idx = threadIdx.x + blockDim.x * blockIdx.x;
  size_t width_stride = gridDim.x * blockDim.x;
  size_t full_width = (width & (~((uint64_t)(BLOCK_W - 1)))) +
                      ((width & (BLOCK_W - 1)) ? BLOCK_W : 0);

#pragma unroll
  for (size_t w = idx; w < full_width; w += width_stride) {
    sdata[threadIdx.y][threadIdx.x] = 0;
    __syncthreads();
    size_t offset = w + threadIdx.y * width;
#pragma unroll
    for (size_t h = threadIdx.y; h < height;
         h += BLOCK_H) {  // block-stride loop across matrix height
      sdata[threadIdx.y][threadIdx.x] +=
          (w < width) ? in[offset] : (static_cast<T>(0));
      offset += width * BLOCK_H;
    }
    __syncthreads();

    T val = sdata[threadIdx.x][threadIdx.y];
    for (int i = warpSize >> 1; i > 0; i >>= 1)
      val += platform::CudaShuffleXorSync(0xFFFFFFFF, val, i);

    __syncthreads();
    if (threadIdx.x == 0) sdata[0][threadIdx.y] = val;
    __syncthreads();
    if ((threadIdx.y == 0) && ((w) < width)) out[w] = sdata[0][threadIdx.x];
  }
}

template <int BLOCK_W, int BLOCK_H>
__global__ void FP16MatrixColReduce(
    const paddle::platform::float16 *__restrict__ in,
    paddle::platform::float16 *__restrict__ out, size_t width, size_t height) {
  constexpr int repeats = BLOCK_H / BLOCK_W;
  __shared__ paddle::platform::float16 sdata[BLOCK_H][BLOCK_W + 1];
  size_t idx = threadIdx.x + blockDim.x * blockIdx.x;
  size_t width_stride = gridDim.x * blockDim.x;
  size_t full_width = (width & (~((uint64_t)(BLOCK_W - 1)))) +
                      ((width & (BLOCK_W - 1)) ? BLOCK_W : 0);

#pragma unroll
  for (size_t w = idx; w < full_width; w += width_stride) {
    for (int r = 0; r < repeats; r++) {
      sdata[threadIdx.y + r * BLOCK_W][threadIdx.x] = 0;
    }
    __syncthreads();
    for (int r = 0; r < repeats; r++) {
      size_t offset = w + (r * BLOCK_W + threadIdx.y) * width;
#pragma unroll
      for (size_t h = r * BLOCK_H + threadIdx.y; h < height;
           h += BLOCK_H) {  // block-stride loop across matrix height
        sdata[r * BLOCK_W + threadIdx.y][threadIdx.x] +=
            (w < width) ? in[offset + r * BLOCK_W * width]
                        : (static_cast<paddle::platform::float16>(0));
        offset += width * BLOCK_H;
      }
    }
    __syncthreads();

    paddle::platform::float16 result =
        static_cast<paddle::platform::float16>(0);
    for (int r = 0; r < repeats; r++) {
      paddle::platform::float16 val =
          sdata[threadIdx.x + r * BLOCK_W][threadIdx.y];
      for (int i = warpSize >> 1; i > 0; i >>= 1)
        val += platform::CudaShuffleXorSync(0xFFFFFFFF, val, i);
      __syncthreads();
      result += val;
    }
    if (threadIdx.x == 0) sdata[0][threadIdx.y] = result;
    __syncthreads();
    if ((threadIdx.y == 0) && ((w) < width)) out[w] = sdata[0][threadIdx.x];
  }
}
#endif
#endif
bool static RunSpecialDims(const framework::DDim &dx_dims,
                           const framework::DDim &dy_dims,
                           const framework::DDim &dout_dims, int axis) {
  auto smaller_dims = dx_dims;
  auto bigger_dims = dy_dims;
  auto smaller_dims_size = smaller_dims.size();
  auto bigger_dims_size = bigger_dims.size();
  int smaller_ignore_size = 0;
  int bigger_ignore_size = 0;
  for (int i = 0; i < smaller_dims_size; i++) {
    if (smaller_dims[i] == 1)
      smaller_ignore_size++;
    else
      break;
  }
  for (int i = 0; i < bigger_dims_size; i++) {
    if (bigger_dims[i] == 1)
      bigger_ignore_size++;
    else
      break;
  }

  int smaller_real_size = smaller_dims.size() - smaller_ignore_size;
  int bigger_real_size = bigger_dims.size() - bigger_ignore_size;

  if (smaller_real_size == bigger_real_size) return false;

  if (bigger_real_size < smaller_real_size) {
    smaller_dims = dy_dims;
    bigger_dims = dx_dims;
    std::swap(smaller_real_size, bigger_real_size);
  }
  int big_size = bigger_dims.size();
  int small_size = smaller_dims.size();
  for (int i = 1; i <= smaller_real_size; i++) {
    if (bigger_dims[big_size - i] != smaller_dims[small_size - i]) return false;
  }

  if (axis != -1 && (axis != (bigger_real_size - smaller_real_size))) {
    return false;
  }

  return true;
}

250 251 252 253 254
#ifdef PADDLE_WITH_CUDA
// cuda definition
template <typename DeviceContext, typename T>
typename std::enable_if<
    std::is_same<DeviceContext, platform::CUDADeviceContext>::value>::type
255 256 257 258 259
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);
260 261
#endif

Q
QI JUN 已提交
262
template <typename DeviceContext, typename T>
263
class ElementwiseAddGradKernel : public ElemwiseGradKernel<T> {
G
gongweibao 已提交
264
 public:
C
chengduo 已提交
265
  void Compute(const framework::ExecutionContext &ctx) const override {
266 267
    ElemwiseGradKernel<T>::Compute(ctx);

C
chengduoZH 已提交
268 269
    using Tensor = framework::Tensor;

270 271
    auto *x = ctx.Input<Tensor>("X");
    auto *y = ctx.Input<Tensor>("Y");
C
chengduo 已提交
272 273 274
    auto *dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
    auto *dx = ctx.Output<Tensor>(framework::GradVarName("X"));
    auto *dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
275
    // skip out
C
chengduo 已提交
276
    auto *out = dout;
277

278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
#ifdef PADDLE_WITH_CUDA
#ifdef __NVCC__

    int axis = ctx.Attr<int>("axis");
    if (ctx.GetPlace() == platform::CUDAPlace() && dx != nullptr &&
        dy != nullptr && dout != nullptr && dx->numel() != dy->numel() &&
        RunSpecialDims(dx->dims(), dy->dims(), dout->dims(), axis)) {
      auto *dx_data = dx->mutable_data<T>(ctx.GetPlace());
      auto *dy_data = dy->mutable_data<T>(ctx.GetPlace());
      auto *dout_data = dout->data<T>();
      auto stream = ctx.cuda_device_context().stream();
      auto *out_data = dx_data;
      int width = dx->numel();
      int height = dout->numel() / width;
      if (dx->dims() == dout->dims()) {
        width = dy->numel();
        height = dout->numel() / width;
        out_data = dy_data;
        framework::TensorCopy(
            *dout, ctx.GetPlace(),
            ctx.template device_context<platform::DeviceContext>(), dx);
      } else {
        framework::TensorCopy(
            *dout, ctx.GetPlace(),
            ctx.template device_context<platform::DeviceContext>(), dy);
      }

      constexpr int block_x = 32;
      constexpr int block_y = 32;
      dim3 blocks(block_x, block_y);

      int max_physical_threads =
          ctx.cuda_device_context().GetMaxPhysicalThreadCount();
      int max_blocks = std::max(max_physical_threads / (block_x * block_y), 1);
      int theory_block = (width + blocks.x - 1) / blocks.x;
      dim3 grids(std::min(theory_block, max_blocks));
      if (std::is_same<T, paddle::platform::float16>::value) {
        const paddle::platform::float16 *ptr1 =
            reinterpret_cast<const paddle::platform::float16 *>(dout_data);
        paddle::platform::float16 *ptr2 =
            reinterpret_cast<paddle::platform::float16 *>(out_data);
        if (height <= 32) {
          FP16MatrixColReduce<32, 32><<<grids, blocks, 0, stream>>>(
              ptr1, ptr2, width, height);
        } else {
          FP16MatrixColReduce<32, 64><<<grids, blocks, 0, stream>>>(
              ptr1, ptr2, width, height);
        }
        return;
      }
      MatrixColReduce<T, block_x, block_y><<<grids, blocks, 0, stream>>>(
          dout_data, out_data, width, height);
      return;
    }

#endif
#endif
335 336 337 338 339 340 341 342 343 344 345 346 347 348
    // 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())) {
349
      elementwise_add_grad<DeviceContext, T>(ctx, x, y, out, dout, dx, dy);
350
    } else {
351 352
      default_elementwise_add_grad<DeviceContext, T>(ctx, x, y, out, dout, dx,
                                                     dy);
353
    }
G
gongweibao 已提交
354 355 356
  }
};

357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
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());
377 378
      default_elementwise_add<DeviceContext, T>(ctx, &ddx_safe, &ddy_safe,
                                                ddout);
379 380 381 382
    }
  }
};

G
gongweibao 已提交
383 384
}  // namespace operators
}  // namespace paddle