elementwise_op_function.h 111.4 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
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
6

L
Luo Tao 已提交
7
    http://www.apache.org/licenses/LICENSE-2.0
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 15

#pragma once
16

17
#include <glog/logging.h>
18

19
#include <algorithm>
20
#include <functional>  // for multiplies
D
dzhwinter 已提交
21
#include <iterator>
22
#include <vector>
23

Y
Yi Wang 已提交
24 25 26
#include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/operator.h"
27 28 29
#include "paddle/fluid/memory/malloc.h"
#include "paddle/fluid/operators/elementwise/elementwise_op_function.cu.h"
#include "paddle/fluid/platform/gpu_info.h"
Y
Yi Wang 已提交
30
#include "paddle/fluid/platform/transform.h"
31

32
#if defined(__NVCC__) || defined(__HIPCC__)
C
chengduoZH 已提交
33
#ifdef __NVCC__
34
#include <cuda.h>
35 36 37
#elif defined(__HIPCC__)
#include <hip/hip_runtime.h>
#endif
C
chengduoZH 已提交
38
#include <thrust/iterator/iterator_adaptor.h>
39

40
#include "paddle/fluid/platform/cuda_device_function.h"
D
dzhwinter 已提交
41
#include "paddle/fluid/platform/cuda_primitives.h"
R
ronnywang 已提交
42 43 44
#ifdef __HIPCC__
constexpr int ELEMWISE_MAX_BLOCK_DIM = 256;
#else
Y
Yu Yang 已提交
45
constexpr int ELEMWISE_MAX_BLOCK_DIM = 1024;
R
ronnywang 已提交
46
#endif
47 48
#define BLOCK_X 32
#define BLOCK_Y 32
C
chengduoZH 已提交
49 50
#endif

Y
Yi Wang 已提交
51
#include "paddle/fluid/operators/math/math_function.h"
Y
Yu Yang 已提交
52
#include "paddle/fluid/platform/for_range.h"
53 54 55 56 57 58
#define GetDivMod(dividend, divisor, div, mod) \
  do {                                         \
    const auto dividend_copy = dividend;       \
    *div = dividend_copy / divisor;            \
    *mod = dividend_copy % divisor;            \
  } while (0)
59

60 61 62 63
#define DIVUP(x, y) (((x) + (y)-1) / (y))

#define ROUNDUP(x, y) (DIVUP((x), (y)) * (y))

64 65 66
namespace paddle {
namespace operators {

67
/*
68 69 70 71 72 73 74
*  Pack input and output tensors into respective vectors with
*  consideration of varible X`s class type.
*  Input variable X is supported to be whether LoDTensor or
*  SelectedRows class type in this package function, once X
*  was SelectedRows type, a valid pointer x_for_selectedrows
*  is excepted to be passed in from op kernel for acquisition
*  of the valid address of LoDTensor created ahead in the function.
75
*/
76 77 78
template <typename OutT>
int PackTensorsIntoVector(const framework::ExecutionContext &ctx,
                          std::vector<const framework::Tensor *> *ins,
79 80
                          std::vector<framework::Tensor *> *outs,
                          framework::Tensor *x_for_selectedrows = nullptr) {
81
  int axis = -1;
82 83 84 85 86
  auto x_var = ctx.InputVar("X");
  PADDLE_ENFORCE_NOT_NULL(
      x_var, platform::errors::InvalidArgument(
                 "Unable to get input Variable X, Variable name is %s.\n",
                 ctx.InputName("X")));
87
  auto *y = ctx.Input<framework::LoDTensor>("Y");
88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
  framework::Tensor *z;

  if (x_var->IsType<framework::LoDTensor>()) {
    auto *x = ctx.Input<framework::LoDTensor>("X");
    z = ctx.Output<framework::LoDTensor>("Out");
    ins->emplace_back(x);
  } else if (x_var->IsType<framework::SelectedRows>()) {
    PADDLE_ENFORCE_EQ(y->dims().size() == 1 && y->dims()[0] == 1, true,
                      platform::errors::InvalidArgument(
                          "For elementwise_op, if X is Sparse, Y must be "
                          "scalar. But reveived the size of Y = %d.",
                          y->dims().size()));
    PADDLE_ENFORCE_NOT_NULL(
        x_for_selectedrows,
        platform::errors::InvalidArgument(
            "The parameter x_for_selectedrows is excepted to "
            "be valid, once input varible X`s class type is "
            "SelectedRows.\n"));
    auto &x_sele = x_var->Get<framework::SelectedRows>();
    auto out_sele = ctx.Output<framework::SelectedRows>("Out");
    *x_for_selectedrows = x_sele.value();
    out_sele->set_rows(x_sele.rows());
    out_sele->set_height(x_sele.height());
    out_sele->mutable_value()->Resize(x_sele.value().dims());
    out_sele->mutable_value()->mutable_data(ctx.GetPlace(),
                                            x_for_selectedrows->type());
    z = ctx.Output<framework::SelectedRows>("Out")->mutable_value();
    ins->emplace_back(x_for_selectedrows);
  } else {
    PADDLE_THROW(platform::errors::InvalidArgument(
        "X's type[%s] is not supported by elementwise_op. X's type should be "
        "LoDTensor or SelectedRows.",
        framework::ToTypeName(x_var->Type())));
  }
122
  z->mutable_data<OutT>(ctx.GetPlace());
123 124 125 126
  outs->emplace_back(z);

  if (y != nullptr) {
    ins->emplace_back(y);
127
    axis = ctx.HasAttr("axis") ? ctx.Attr<int>("axis") : -1;
128
  }
129
  return axis;
130 131
}

132 133 134 135 136 137
/*
 * Out = X ⊙ Y
 * If Y's shape does not match X' shape, they will be reshaped.
 * For example:
 * 1. shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1
 *    pre=2, n=3*4, post=5
C
chengduo 已提交
138
 *    x.shape(2, 12, 5) * y.shape(1, 12, 1).broadcast(2, 12, 5)
139 140
 * 2. shape(X) = (2, 3, 4, 5), shape(Y) = (4,5)
 *    pre=2*3, n=4*5, post=1
C
chengduo 已提交
141
 *    x.shape(6, 20, 1) * y.shape(1, 20, 1).broadcast(6, 20, 1)
142
 *
143 144
 * New parameter: *is_run_common_broadcast* is a flag to record whether to run
 * common broadcast code.
145
 */
146 147
inline void get_mid_dims(const framework::DDim &x_dims,
                         const framework::DDim &y_dims, const int axis,
148 149
                         int *pre, int *n, int *post,
                         int *is_run_common_broadcast) {
150 151 152
  *pre = 1;
  *n = 1;
  *post = 1;
153 154 155 156 157 158
  *is_run_common_broadcast = 0;
  for (int i = 0; i < axis; ++i) {
    (*pre) *= x_dims[i];
  }
  for (int i = 0; i < y_dims.size(); ++i) {
    if (x_dims[i + axis] != y_dims[i]) {
159 160 161 162 163 164 165
      PADDLE_ENFORCE_EQ(y_dims[i] == 1 || x_dims[i + axis] == 1, true,
                        platform::errors::InvalidArgument(
                            "Broadcast dimension mismatch. Operands "
                            "could not be broadcast together with the shape of "
                            "X = [%s] and the shape of Y = [%s]. Received [%d] "
                            "in X is not equal to [%d] in Y.",
                            x_dims, y_dims, x_dims[i + axis], y_dims[i]));
166 167
      *is_run_common_broadcast = 1;
      return;
168
    }
169 170 171 172 173 174
    (*n) *= y_dims[i];
  }
  for (int i = axis + y_dims.size(); i < x_dims.size(); ++i) {
    (*post) *= x_dims[i];
  }
}
175

176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
inline int GetElementwiseIndex(const int *x_dims_array, const int max_dim,
                               const int *index_array) {
  int index_ = 0;
  for (int i = 0; i < max_dim; i++) {
    if (x_dims_array[i] > 1) {
      index_ = index_ * x_dims_array[i] + index_array[i];
    }
  }
  return index_;
}

inline void UpdateElementwiseIndexArray(const int *out_dims_array,
                                        const int max_dim, int *index_array) {
  for (int i = max_dim - 1; i >= 0; --i) {
    ++index_array[i];
    if (index_array[i] >= out_dims_array[i]) {
      index_array[i] -= out_dims_array[i];
193
    } else {
194 195 196 197 198 199 200 201 202 203
      break;
    }
  }
}

inline void GetBroadcastDimsArrays(const framework::DDim &x_dims,
                                   const framework::DDim &y_dims,
                                   int *x_dims_array, int *y_dims_array,
                                   int *out_dims_array, const int max_dim,
                                   const int axis) {
204 205 206 207 208 209 210 211 212
  PADDLE_ENFORCE_GE(
      axis, 0,
      platform::errors::InvalidArgument(
          "Axis should be great than or equal to 0, but received axis is %d.",
          axis));
  PADDLE_ENFORCE_LT(axis, max_dim,
                    platform::errors::InvalidArgument(
                        "Axis should be less than %d, but received axis is %d.",
                        max_dim, axis));
213 214 215 216
  if (x_dims.size() > y_dims.size()) {
    std::fill(y_dims_array, y_dims_array + axis, 1);
    if (axis + y_dims.size() < max_dim) {
      std::fill(y_dims_array + axis + y_dims.size(), y_dims_array + max_dim, 1);
217
    }
218 219 220 221 222 223
    std::copy(x_dims.Get(), x_dims.Get() + x_dims.size(), x_dims_array);
    std::copy(y_dims.Get(), y_dims.Get() + y_dims.size(), y_dims_array + axis);
  } else {
    std::fill(x_dims_array, x_dims_array + axis, 1);
    if (axis + x_dims.size() < max_dim) {
      std::fill(x_dims_array + axis + x_dims.size(), x_dims_array + max_dim, 1);
224
    }
225 226 227
    std::copy(x_dims.Get(), x_dims.Get() + x_dims.size(), x_dims_array + axis);
    std::copy(y_dims.Get(), y_dims.Get() + y_dims.size(), y_dims_array);
  }
228

229
  for (int i = 0; i < max_dim; i++) {
230 231 232 233 234 235 236 237 238
    PADDLE_ENFORCE_EQ(
        x_dims_array[i] == y_dims_array[i] || x_dims_array[i] <= 1 ||
            y_dims_array[i] <= 1,
        true, platform::errors::InvalidArgument(
                  "Broadcast dimension mismatch. Operands could "
                  "not be broadcast together with the shape of X = [%s] and "
                  "the shape of Y = [%s]. Received [%d] in X is not equal to "
                  "[%d] in Y at i:%d.",
                  x_dims, y_dims, x_dims_array[i], y_dims_array[i], i));
239 240
    if ((x_dims_array[i] > 1 || y_dims_array[i] > 1) ||
        (x_dims_array[i] == 1 && y_dims_array[i] == 1)) {
241
      out_dims_array[i] = std::max(x_dims_array[i], y_dims_array[i]);
242 243
    } else {
      out_dims_array[i] = -1;
244
    }
245 246
  }
}
247

248 249 250 251 252 253 254 255 256 257 258
template <typename Functor, typename T, typename OutType = T>
void CommonForwardBroadcastCPU(const framework::Tensor *x,
                               const framework::Tensor *y, framework::Tensor *z,
                               int *x_dims_array, int *y_dims_array,
                               int *out_dims_array, int max_dim,
                               const platform::CPUDeviceContext &ctx,
                               Functor func,
                               const bool is_xsize_larger = true) {
  std::vector<int> index_array(max_dim, 0);
  const T *x_data = x->data<T>();
  const T *y_data = y->data<T>();
259 260 261 262
  PADDLE_ENFORCE_NOT_NULL(x_data, platform::errors::InvalidArgument(
                                      "The input X should not be empty."));
  PADDLE_ENFORCE_NOT_NULL(y_data, platform::errors::InvalidArgument(
                                      "The input Y should not be empty."));
263 264 265 266 267 268 269 270 271 272 273 274
  OutType *out_data = z->mutable_data<OutType>(ctx.GetPlace());

  const int out_size = std::accumulate(out_dims_array, out_dims_array + max_dim,
                                       1, std::multiplies<int>());
  int x_index, y_index;
  for (int out_index = 0; out_index < out_size; ++out_index) {
    x_index = GetElementwiseIndex(x_dims_array, max_dim, index_array.data());
    y_index = GetElementwiseIndex(y_dims_array, max_dim, index_array.data());
    if (is_xsize_larger) {
      out_data[out_index] = func(x_data[x_index], y_data[y_index]);
    } else {
      out_data[out_index] = func(y_data[y_index], x_data[x_index]);
275
    }
276 277

    UpdateElementwiseIndexArray(out_dims_array, max_dim, index_array.data());
278 279 280
  }
}

281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402
#if defined(__NVCC__) || defined(__HIPCC__)
template <typename Functor, typename T, typename OutType>
__global__ void ElementwiseKernel(const T *__restrict__ x_data,
                                  const T *__restrict__ y_data,
                                  OutType *__restrict__ out_data, int n,
                                  int post, const size_t total, Functor func) {
  int tid = threadIdx.x + blockDim.x * blockIdx.x;
  int stride = blockDim.x * gridDim.x;

  for (int i = tid; i < total; i += stride) {
    int idx = i / post % n;
    out_data[i] = func(x_data[i], y_data[idx]);
  }
}

template <typename Functor, typename T, typename OutType>
void ComputeElementwiseCUDA(const framework::Tensor *x,
                            const framework::Tensor *y, framework::Tensor *z,
                            int pre, int n, int post,
                            const platform::CUDADeviceContext &ctx,
                            Functor func, const bool is_xsize_larger = true) {
  const T *x_data = x->data<T>();
  const T *y_data = y->data<T>();
  OutType *out_data = z->mutable_data<OutType>(ctx.GetPlace());

  int numel = pre * n * post;
  int threads = 256;
  int blocks = (numel + threads - 1) / threads;

  if (is_xsize_larger) {
    ElementwiseKernel<Functor, T,
                      OutType><<<blocks, threads, 0, ctx.stream()>>>(
        x_data, y_data, out_data, n, post, numel, func);

  } else {
    ElementwiseKernel<Functor, T,
                      OutType><<<blocks, threads, 0, ctx.stream()>>>(
        y_data, x_data, out_data, n, post, numel, func);
  }
}

template <typename Functor, typename T, typename OutType = T>
__global__ void CommonForwardBroadcastCUDAKernel(
    const int *x_strides_array, const int *y_strides_array,
    const int *out_dims_array, const T *x, const T *y, OutType *out,
    int out_size, int max_dim, Functor func, const bool is_xsize_larger) {
  for (int out_index = blockIdx.x * blockDim.x + threadIdx.x;
       out_index < out_size; out_index += blockDim.x * gridDim.x) {
    int x_index = 0;
    int y_index = 0;
    int out_index_quotient = out_index;
    int remainder = 0;
#pragma unroll
    for (int i = max_dim - 1; i >= 0; --i) {
      GetDivMod(out_index_quotient, out_dims_array[i], &out_index_quotient,
                &remainder);
      x_index += remainder * x_strides_array[i];
      y_index += remainder * y_strides_array[i];
    }
    if (is_xsize_larger) {
      out[out_index] = func(x[x_index], y[y_index]);
    } else {
      out[out_index] = func(y[y_index], x[x_index]);
    }
  }
}

template <typename Functor, typename T, typename OutType = T>
void CommonForwardBroadcastCUDA(
    const framework::Tensor *x, const framework::Tensor *y,
    framework::Tensor *z, int *x_dims_array, int *y_dims_array,
    int *out_dims_array, int max_dim, const platform::CUDADeviceContext &ctx,
    Functor func, const bool is_xsize_larger = true) {
  const auto gplace = BOOST_GET_CONST(platform::CUDAPlace, ctx.GetPlace());
  auto cplace = platform::CPUPlace();
  const T *x_data = x->data<T>();
  const T *y_data = y->data<T>();
  OutType *out_data = z->mutable_data<OutType>(ctx.GetPlace());

  std::vector<int> x_strides_array(max_dim);
  std::vector<int> y_strides_array(max_dim);
  int x_stride = 1;
  int y_stride = 1;
  for (int i = max_dim - 1; i >= 0; i--) {
    x_strides_array[i] = x_dims_array[i] == 1 ? 0 : x_stride;
    y_strides_array[i] = y_dims_array[i] == 1 ? 0 : y_stride;
    x_stride *= x_dims_array[i];
    y_stride *= y_dims_array[i];
  }

  int bytes = max_dim * sizeof(int);
  auto x_strides_array_tmp = memory::Alloc(ctx, bytes);
  int *x_strides_array_gpu =
      reinterpret_cast<int *>(x_strides_array_tmp->ptr());
  memory::Copy(gplace, x_strides_array_gpu, cplace, x_strides_array.data(),
               bytes, ctx.stream());

  auto y_strides_array_tmp = memory::Alloc(ctx, bytes);
  int *y_strides_array_gpu =
      reinterpret_cast<int *>(y_strides_array_tmp->ptr());
  memory::Copy(gplace, y_strides_array_gpu, cplace, y_strides_array.data(),
               bytes, ctx.stream());

  auto out_dims_array_tmp = memory::Alloc(ctx, bytes);
  int *out_dims_array_gpu = reinterpret_cast<int *>(out_dims_array_tmp->ptr());
  memory::Copy(gplace, out_dims_array_gpu, cplace, out_dims_array, bytes,
               ctx.stream());

  const int out_size = std::accumulate(out_dims_array, out_dims_array + max_dim,
                                       1, std::multiplies<int>());
  dim3 gird_size = dim3(
      (out_size + PADDLE_CUDA_THREAD_SIZE - 1) / PADDLE_CUDA_THREAD_SIZE, 1);
  dim3 block_size = dim3(PADDLE_CUDA_THREAD_SIZE, 1);

  CommonForwardBroadcastCUDAKernel<
      Functor, T, OutType><<<gird_size, block_size, 0, ctx.stream()>>>(
      x_strides_array_gpu, y_strides_array_gpu, out_dims_array_gpu, x_data,
      y_data, out_data, out_size, max_dim, func, is_xsize_larger);
}

#endif  // __NVCC__ or __HIPCC__

403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472
template <typename T, typename DX_OP, typename DY_OP>
void CommonGradBroadcastCPU(
    const framework::Tensor &x, const framework::Tensor &y,
    const framework::Tensor &out, const framework::Tensor &dout,
    framework::Tensor *dx, framework::Tensor *dy, int *x_dims_array,
    int *y_dims_array, int *out_dims_array, int max_dim,
    const platform::CPUDeviceContext &ctx, DX_OP dx_op, DY_OP dy_op) {
  std::vector<int> index_array(max_dim, 0);
  const T *x_data = x.data<T>();
  const T *y_data = y.data<T>();
  const T *out_data = out.data<T>();
  const T *dout_data = dout.data<T>();
  T *dx_data = dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace());
  T *dy_data = dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace());
  if (dx_data != nullptr) {
    memset(dx_data, 0, dx->numel() * sizeof(T));
  }
  if (dy_data != nullptr) {
    memset(dy_data, 0, dy->numel() * sizeof(T));
  }
  const int out_size = std::accumulate(out_dims_array, out_dims_array + max_dim,
                                       1, std::multiplies<int>());
  int x_index, y_index;
  for (int out_index = 0; out_index < out_size; ++out_index) {
    x_index = GetElementwiseIndex(x_dims_array, max_dim, index_array.data());
    y_index = GetElementwiseIndex(y_dims_array, max_dim, index_array.data());
    if (dx_data != nullptr) {
      dx_data[x_index] += dx_op(x_data[x_index], y_data[y_index],
                                out_data[out_index], dout_data[out_index]);
    }
    if (dy_data != nullptr) {
      dy_data[y_index] += dy_op(x_data[x_index], y_data[y_index],
                                out_data[out_index], dout_data[out_index]);
    }

    UpdateElementwiseIndexArray(out_dims_array, max_dim, index_array.data());
  }
}

inline void ComputeBroadcastKernelSize(int *x_dims_array, int *out_dims_array,
                                       int *x_blocks, int *x_threads,
                                       int max_dim) {
  *x_blocks = 1;
  *x_threads = 1;
  for (int i = 0; i < max_dim; i++) {
    if (x_dims_array[i] == out_dims_array[i]) {
      *x_blocks *= x_dims_array[i];
    } else {
      *x_threads *= out_dims_array[i];
    }
  }
}

inline void ComputeBroadcastTranspositionArray(const int *x_one_indexs,
                                               int *x_trans_indexs,
                                               const int max_dim,
                                               const int x_one_size) {
  int diff = max_dim - x_one_size;
  std::copy_n(x_one_indexs, x_one_size, x_trans_indexs + diff);
  int p = 0;
  int q = diff;
  for (int i = 0; i < max_dim; ++i) {
    if (q < max_dim && i == x_trans_indexs[q]) {
      ++q;
    } else {
      x_trans_indexs[p++] = i;
    }
  }
}

473
#if defined(__NVCC__) || defined(__HIPCC__)
474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
template <typename T, typename DX_OP, typename DY_OP>
static __global__ void ElemwiseGradBroadcast1CUDAKernel(
    const T *x, const T *y, const T *out, const T *dout, int h, int w,
    bool is_xsize_larger, DX_OP dx_op, DY_OP dy_op, T *dx, T *dy) {
  int j = blockIdx.x;
  int i = threadIdx.x;
  int tid = threadIdx.x;
  T val(0);
  if (is_xsize_larger) {
    do {
      int x_offset = i * w + j;
      if (dx) {
        dx[x_offset] = dx_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
      }
      if (dy) {
        val += dy_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
      }
      i += ELEMWISE_MAX_BLOCK_DIM;
    } while (i < h);

    if (dy) {
      h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dy[j] = val;
      }
    }
  } else {  // x.dims < y.dims, broadcast for x.
    do {
      int y_offset = i * w + j;
      if (dy) {
        dy[y_offset] = dy_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
      }
      if (dx) {
        val += dx_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
      }
      i += ELEMWISE_MAX_BLOCK_DIM;
    } while (i < h);

    if (dx) {
      h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dx[j] = val;
      }
    }
  }
}

// suppose use 2D block is fast because more parallel
// and memory coalesced
template <typename T, typename DX_OP, typename DY_OP>
static __global__ void FastElemwiseGradBroadcast1CUDAKernel(
    const T *x, const T *y, const T *out, const T *dout, int h, int w,
    bool is_xsize_larger, DX_OP dx_op, DY_OP dy_op, T *dx, T *dy) {
  __shared__ T sdata[BLOCK_Y][BLOCK_X + 1];

  T val(0);
  size_t width_stride = gridDim.x * blockDim.x;
  size_t idx = threadIdx.x + blockDim.x * blockIdx.x;
  size_t full_width =
      (w & (~((uint64_t)(BLOCK_X - 1)))) + ((w & (BLOCK_X - 1)) ? BLOCK_X : 0);
  size_t full_height =
      (h & (~((uint64_t)(BLOCK_Y - 1)))) + ((h & (BLOCK_Y - 1)) ? BLOCK_Y : 0);
  if (is_xsize_larger) {
    for (int m = idx; m < full_width; m += width_stride) {
      sdata[threadIdx.y][threadIdx.x] = 0;
      for (int n = threadIdx.y; n < full_height; n += BLOCK_Y) {
        int x_offset = n * w + m;
        if (dx && m < w && n < h) {
          dx[x_offset] =
              dx_op(x[x_offset], y[m], out[x_offset], dout[x_offset]);
        }
        if (dy) {
          if (m < w && n < h) {
            T val = dy_op(x[x_offset], y[m], out[x_offset], dout[x_offset]);
            sdata[threadIdx.y][threadIdx.x] += val;
          }
          __syncthreads();
        }
      }
      if (dy) {
        T my_val = sdata[threadIdx.x][threadIdx.y];
        for (int i = warpSize >> 1; i > 0; i >>= 1)
          my_val += platform::CudaShuffleXorSync(0xFFFFFFFF, my_val, i);
        __syncthreads();
        if ((threadIdx.x == 0)) {
          sdata[0][threadIdx.y] = my_val;
        }
        __syncthreads();
        if (threadIdx.y == 0 && m < w) {
          dy[m] = sdata[0][threadIdx.x];
        }
      }
    }
  } else {  // x.dims < y.dims, broadcast for x.
    for (int m = idx; m < full_width; m += width_stride) {
      sdata[threadIdx.y][threadIdx.x] = 0;
      for (int n = threadIdx.y; n < full_height; n += BLOCK_Y) {
        int y_offset = n * w + m;
        if (dy && m < w && n < h) {
          dy[y_offset] =
              dy_op(x[m], y[y_offset], out[y_offset], dout[y_offset]);
        }
        if (dx) {
          if (m < w && n < h) {
            T val = dx_op(x[m], y[y_offset], out[y_offset], dout[y_offset]);
            sdata[threadIdx.y][threadIdx.x] += val;
          }
          __syncthreads();
        }
      }
      if (dx) {
        T my_val = sdata[threadIdx.x][threadIdx.y];
        for (int i = warpSize >> 1; i > 0; i >>= 1)
          my_val += platform::CudaShuffleXorSync(0xFFFFFFFF, my_val, i);
        __syncthreads();
        if ((threadIdx.x == 0)) {
          sdata[0][threadIdx.y] = my_val;
        }
        __syncthreads();
        if (threadIdx.y == 0 && m < w) {
          dx[m] = sdata[0][threadIdx.x];
        }
      }
    }
  }
}

603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641
template <typename T, typename DX_OP>
__global__ void CommonGradBroadcastCUDAKernel(
    const int *x_strides_array, const int *y_strides_array,
    const int *out_dims_array, const int *y_strides_order,
    const int *y_dims_order, const T *x, const T *y, const T *out,
    const T *dout, T *dx, int out_size, int max_dim, int thread_num,
    DX_OP dx_op) {
  T val(0);
  int i = blockIdx.x;
  int tid = threadIdx.x;
  for (int j = tid; j < thread_num; j += blockDim.x) {
    const int X_index = i * thread_num + j;
    int out_index = X_index;
    int C_index = 0;
    int B_index = i * thread_num + j;
    int remainder = 0;
#pragma unroll
    for (int d = max_dim - 1; d >= 0; --d) {
      GetDivMod(B_index, y_dims_order[d], &B_index, &remainder);
      C_index += remainder * y_strides_order[d];
    }
    int x_index = 0;
    int y_index = 0;
    int C_index_val = C_index;
#pragma unroll
    for (int d = max_dim - 1; d >= 0; --d) {
      GetDivMod(C_index_val, out_dims_array[d], &C_index_val, &remainder);
      x_index += remainder * x_strides_array[d];
      y_index += remainder * y_strides_array[d];
    }
    out_index = C_index;
    val += dx_op(x[x_index], y[y_index], out[out_index], dout[out_index]);
  }
  val = paddle::platform::reduceSum(val, tid, thread_num);
  if (threadIdx.x == 0) {
    dx[i] = val;
  }
}

642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798
template <typename T, typename DY_OP>
static __global__ void CommonGradBroadcast1CUDAKernelHeight(
    const T *x, const T *y, const T *out, const T *dout, int h, int w,
    DY_OP dy_op, T *dy, int x_h, int x_w, bool is_y) {
  int j = blockIdx.x;
  int i = threadIdx.x;
  int tid = threadIdx.x;
  T val(0);

  if (is_y) {
    do {
      int out_offset = i * w + j;
      int x_offset = (i % x_h) * x_w + j % x_w;
      if (dy) {
        val += dy_op(x[x_offset], y[j], out[out_offset], dout[out_offset]);
      }
      i += ELEMWISE_MAX_BLOCK_DIM;
    } while (i < h);

    if (dy) {
      h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dy[j] = val;
      }
    }
  } else {
    do {
      int out_offset = i * w + j;
      int y_offset = (i % x_h) * x_w + j % x_w;
      if (dy) {
        val += dy_op(x[j], y[y_offset], out[out_offset], dout[out_offset]);
      }
      i += ELEMWISE_MAX_BLOCK_DIM;
    } while (i < h);

    if (dy) {
      h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dy[j] = val;
      }
    }
  }
}

template <typename T, typename DY_OP>
static __global__ void FastCommonGradBroadcastCUDAKernelHeight(
    const T *x, const T *y, const T *out, const T *dout, int h, int w,
    DY_OP dy_op, T *dy, int x_h, int x_w, bool is_y) {
  __shared__ T sdata[BLOCK_Y][BLOCK_X + 1];

  T val(0);
  size_t width_stride = gridDim.x * blockDim.x;
  size_t idx = threadIdx.x + blockDim.x * blockIdx.x;
  size_t full_width =
      (w & (~((uint64_t)(BLOCK_X - 1)))) + ((w & (BLOCK_X - 1)) ? BLOCK_X : 0);
  size_t full_height =
      (h & (~((uint64_t)(BLOCK_Y - 1)))) + ((h & (BLOCK_Y - 1)) ? BLOCK_Y : 0);
  if (is_y) {
    for (int m = idx; m < full_width; m += width_stride) {
      sdata[threadIdx.y][threadIdx.x] = 0;
      for (int n = threadIdx.y; n < full_height; n += BLOCK_Y) {
        int out_offset = n * w + m;
        int x_offset = (n % x_h) * x_w + m % x_w;
        if (dy) {
          if (m < w && n < h) {
            T val = dy_op(x[x_offset], y[m], out[out_offset], dout[out_offset]);
            sdata[threadIdx.y][threadIdx.x] += val;
          }
          __syncthreads();
        }
      }
      if (dy) {
        T my_val = sdata[threadIdx.x][threadIdx.y];
        for (int i = warpSize >> 1; i > 0; i >>= 1) {
          my_val += platform::CudaShuffleXorSync(0xFFFFFFFF, my_val, i);
        }
        __syncthreads();
        if ((threadIdx.x == 0)) {
          sdata[0][threadIdx.y] = my_val;
        }
        __syncthreads();
        if (threadIdx.y == 0 && m < w) {
          dy[m] = sdata[0][threadIdx.x];
        }
      }
    }
  } else {
    for (int m = idx; m < full_width; m += width_stride) {
      sdata[threadIdx.y][threadIdx.x] = 0;
      for (int n = threadIdx.y; n < full_height; n += BLOCK_Y) {
        int out_offset = n * w + m;
        int y_offset = (n % x_h) * x_w + m % x_w;
        if (dy) {
          if (m < w && n < h) {
            T val = dy_op(x[m], y[y_offset], out[out_offset], dout[out_offset]);
            sdata[threadIdx.y][threadIdx.x] += val;
          }
          __syncthreads();
        }
      }
      if (dy) {
        T my_val = sdata[threadIdx.x][threadIdx.y];
        for (int i = warpSize >> 1; i > 0; i >>= 1) {
          my_val += platform::CudaShuffleXorSync(0xFFFFFFFF, my_val, i);
        }
        __syncthreads();
        if ((threadIdx.x == 0)) {
          sdata[0][threadIdx.y] = my_val;
        }
        __syncthreads();
        if (threadIdx.y == 0 && m < w) {
          dy[m] = sdata[0][threadIdx.x];
        }
      }
    }
  }
}

template <typename T, typename DY_OP, typename DX_OP>
static __global__ void FastCommonGradBroadcastAllCUDAKernel(
    const T *x, const T *y, const T *out, const T *dout, int pre, int n,
    int post, bool is_xsize_larger, DX_OP dx_op, DY_OP dy_op, T *dx, T *dy) {
  int tid = threadIdx.x;
  int bid = blockIdx.x;

  T val(0);
  if (is_xsize_larger) {
    for (int i = tid; i < n; i += ELEMWISE_MAX_BLOCK_DIM) {
      int b_i = bid / post;
      int b_j = bid % post;
      int x_offset = b_i * n * post + i * post + b_j;
      int y_offset = b_i * post + b_j;
      if (dx) {
        dx[x_offset] =
            dx_op(x[x_offset], y[y_offset], out[x_offset], dout[x_offset]);
      }
      if (dy) {
        val += dy_op(x[x_offset], y[y_offset], out[x_offset], dout[x_offset]);
      }
    }
    if (dy) {
      int h = n > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : n;
      val = paddle::platform::reduceSum(val, tid, h);
      if (tid == 0) {
        dy[bid] = val;
      }
    }
  } else {
    for (int i = tid; i < n; i += ELEMWISE_MAX_BLOCK_DIM) {
      int b_i = bid / post;
      int b_j = bid % post;
      int y_offset = b_i * n * post + i * post + b_j;
      int x_offset = b_i * post + b_j;
      if (dy) {
        dy[y_offset] =
799
            dy_op(x[x_offset], y[y_offset], out[y_offset], dout[y_offset]);
800 801
      }
      if (dx) {
802
        val += dx_op(x[x_offset], y[y_offset], out[y_offset], dout[y_offset]);
803 804 805 806 807 808 809 810 811 812 813 814
      }
    }
    if (dx) {
      int h = n > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : n;
      val = paddle::platform::reduceSum(val, tid, h);
      if (tid == 0) {
        dx[bid] = val;
      }
    }
  }
}

815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872
template <typename T, typename OP>
static __global__ void FastCommonGradBroadcastOneCUDAKernel(
    const T *x, const T *y, const T *out, const T *dout, int pre, int n,
    int post, int y_pre, int y_n, int y_post, bool is_xsize, OP op, T *dd) {
  int tid = threadIdx.x;
  int bid = blockIdx.x;

  T val(0);
  if (is_xsize) {
    // do reduce for x
    for (int i = tid; i < n; i += ELEMWISE_MAX_BLOCK_DIM) {
      int b_i = bid / post;
      int b_j = bid % post;
      int x_offset = b_i * n * post + b_j;
      int out_offset = b_i * n * post + i * post + b_j;

      // Get y pre rows id with x post and y_pre.
      int b_yi = bid / (post * y_pre);
      int b_yj = bid % y_post;
      int y_offset = b_yi * y_n + i * y_post + b_yj;

      if (dd) {
        val += op(x[x_offset], y[y_offset], out[out_offset], dout[out_offset]);
      }
    }
    if (dd) {
      int h = n > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : n;
      val = paddle::platform::reduceSum(val, tid, h);
      if (tid == 0) {
        dd[bid] = val;
      }
    }
  } else {
    // do reduce for y
    for (int i = tid; i < n; i += ELEMWISE_MAX_BLOCK_DIM) {
      int b_i = bid / post;
      int b_j = bid % post;
      int y_offset = b_i * n * post + b_j;
      int out_offset = b_i * n * post + i * post + b_j;

      int b_yi = bid / (post * y_pre);
      int b_yj = bid % y_post;
      int x_offset = b_yi * y_n + i * y_post + b_yj;

      if (dd) {
        val += op(x[x_offset], y[y_offset], out[out_offset], dout[out_offset]);
      }
    }
    if (dd) {
      int h = n > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : n;
      val = paddle::platform::reduceSum(val, tid, h);
      if (tid == 0) {
        dd[bid] = val;
      }
    }
  }
}

873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892
// Check input can be split into 2 parts
static inline bool SplitDims(const std::vector<int> &y_broadcast_pos,
                             int max_dim) {
  bool can_split_dim2 = true;
  // must at start or end.
  if (y_broadcast_pos[0] != 0 &&
      y_broadcast_pos[y_broadcast_pos.size() - 1] != max_dim - 1) {
    can_split_dim2 = false;
  } else {
    for (int i = 1; i < y_broadcast_pos.size(); ++i) {
      // dim must be continue
      if (y_broadcast_pos[i] != y_broadcast_pos[i - 1] + 1) {
        can_split_dim2 = false;
        break;
      }
    }
  }
  return can_split_dim2;
}

893 894 895 896 897 898 899 900 901 902
// Suppose only has contiguous dims
static inline bool CheckContiguousDims(const std::vector<int> &broadcast_pos) {
  for (int i = 1; i < broadcast_pos.size(); ++i) {
    if (broadcast_pos[i] != broadcast_pos[i - 1] + 1) {
      return false;
    }
  }
  return true;
}

903 904 905 906 907 908 909
template <typename T, typename DX_OP, typename DY_OP>
void CommonGradBroadcastCUDA(
    const framework::Tensor &x, const framework::Tensor &y,
    const framework::Tensor &out, const framework::Tensor &dout,
    framework::Tensor *dx, framework::Tensor *dy, int *x_dims_array,
    int *y_dims_array, int *out_dims_array, int max_dim,
    const platform::CUDADeviceContext &ctx, DX_OP dx_op, DY_OP dy_op) {
910
  const auto gplace = BOOST_GET_CONST(platform::CUDAPlace, ctx.GetPlace());
911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965
  auto cplace = platform::CPUPlace();
  const T *x_data = x.data<T>();
  const T *y_data = y.data<T>();
  const T *out_data = out.data<T>();
  const T *dout_data = dout.data<T>();
  T *dx_data = dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace());
  T *dy_data = dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace());

  std::vector<int> x_one_indexs;
  std::vector<int> y_one_indexs;
  for (int i = 0; i < max_dim; i++) {
    if (x_dims_array[i] != y_dims_array[i]) {
      if (x_dims_array[i] == 1) {
        x_one_indexs.push_back(i);
      }
      if (y_dims_array[i] == 1) {
        y_one_indexs.push_back(i);
      }
    }
  }

  std::vector<int> x_trans_indexs(max_dim);
  std::vector<int> y_trans_indexs(max_dim);
  ComputeBroadcastTranspositionArray(x_one_indexs.data(), x_trans_indexs.data(),
                                     max_dim, x_one_indexs.size());
  ComputeBroadcastTranspositionArray(y_one_indexs.data(), y_trans_indexs.data(),
                                     max_dim, y_one_indexs.size());

  // compute array stride for cuda kernel;
  // e.g. x.dims=[2,3,4], x_stride=[12,4,1]
  std::vector<int> x_strides_array(max_dim);
  std::vector<int> y_strides_array(max_dim);
  std::vector<int> out_strides_array(max_dim);
  int x_stride = 1;
  int y_stride = 1;
  int z_stride = 1;
  for (int i = max_dim - 1; i >= 0; i--) {
    x_strides_array[i] = x_dims_array[i] == 1 ? 0 : x_stride;
    y_strides_array[i] = y_dims_array[i] == 1 ? 0 : y_stride;
    out_strides_array[i] = z_stride;
    x_stride *= x_dims_array[i];
    y_stride *= y_dims_array[i];
    z_stride *= out_dims_array[i];
  }

  std::vector<int> x_strides_order(max_dim);
  std::vector<int> y_strides_order(max_dim);
  std::vector<int> x_dims_order(max_dim);
  std::vector<int> y_dims_order(max_dim);
  for (int i = 0; i < max_dim; ++i) {
    x_strides_order[i] = out_strides_array[x_trans_indexs[i]];
    y_strides_order[i] = out_strides_array[y_trans_indexs[i]];
    x_dims_order[i] = out_dims_array[x_trans_indexs[i]];
    y_dims_order[i] = out_dims_array[y_trans_indexs[i]];
  }
966 967 968 969 970 971 972 973 974 975 976 977 978
  std::vector<int> x_broadcast_pos;
  std::vector<int> y_broadcast_pos;

  int bytes = max_dim * sizeof(int);

  for (int i = 0; i < max_dim; ++i) {
    if (x_dims_array[i] != out_dims_array[i] && x_dims_array[i] == 1) {
      x_broadcast_pos.emplace_back(i);
    }
    if (y_dims_array[i] != out_dims_array[i] && y_dims_array[i] == 1) {
      y_broadcast_pos.emplace_back(i);
    }
  }
979

980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086
  auto stream = ctx.stream();
  bool can_split_x = false;
  bool can_split_y = false;

  auto FastCommonCUDAF = [&](const std::vector<int> &broadcast_pos, bool is_y) {
    int h =
        std::accumulate(out_dims_array, out_dims_array + broadcast_pos.size(),
                        1, std::multiplies<int>());
    int w =
        std::accumulate(out_dims_array + broadcast_pos.size(),
                        out_dims_array + max_dim, 1, std::multiplies<int>());

    VLOG(3) << "FastCommonCUDAF elementwise w:" << w << " h:" << h
            << " is_y:" << is_y;

    int split_h;
    int split_w;
    int kh = h;
    int kw = w;

    if (is_y) {
      split_h =
          std::accumulate(x_dims_array, x_dims_array + broadcast_pos.size(), 1,
                          std::multiplies<int>());
      split_w =
          std::accumulate(x_dims_array + broadcast_pos.size(),
                          x_dims_array + max_dim, 1, std::multiplies<int>());

    } else {
      split_h =
          std::accumulate(y_dims_array, y_dims_array + broadcast_pos.size(), 1,
                          std::multiplies<int>());
      split_w =
          std::accumulate(y_dims_array + broadcast_pos.size(),
                          y_dims_array + max_dim, 1, std::multiplies<int>());
    }

    if (h > split_h) kh = split_h;
    if (w > split_w) kw = split_w;

    if (is_y) {
      if (w < 16 || h < 16) {
        int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, h);
        int grid_size = w;
        CommonGradBroadcast1CUDAKernelHeight<<<grid_size, block_size, 0,
                                               stream>>>(
            x_data, y_data, out_data, dout_data, h, w, dy_op, dy_data, kh, kw,
            is_y);
      } else {
        dim3 block_size = dim3(BLOCK_X, BLOCK_Y);
        int grid_size = (w + BLOCK_X - 1) / BLOCK_X;
        FastCommonGradBroadcastCUDAKernelHeight<<<grid_size, block_size, 0,
                                                  stream>>>(
            x_data, y_data, out_data, dout_data, h, w, dy_op, dy_data, kh, kw,
            is_y);
      }
    } else {
      if (w < 16 || h < 16) {
        int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, h);
        int grid_size = w;
        CommonGradBroadcast1CUDAKernelHeight<<<grid_size, block_size, 0,
                                               stream>>>(
            x_data, y_data, out_data, dout_data, h, w, dx_op, dx_data, kh, kw,
            is_y);
      } else {
        dim3 block_size = dim3(BLOCK_X, BLOCK_Y);
        int grid_size = (w + BLOCK_X - 1) / BLOCK_X;
        FastCommonGradBroadcastCUDAKernelHeight<<<grid_size, block_size, 0,
                                                  stream>>>(
            x_data, y_data, out_data, dout_data, h, w, dx_op, dx_data, kh, kw,
            is_y);
      }
    }
  };

  auto FastBroadCastHeightCUDAF = [&](const std::vector<int> &broadcast_pos,
                                      bool x_large) {
    int h =
        std::accumulate(out_dims_array, out_dims_array + broadcast_pos.size(),
                        1, std::multiplies<int>());
    int w =
        std::accumulate(out_dims_array + broadcast_pos.size(),
                        out_dims_array + max_dim, 1, std::multiplies<int>());

    VLOG(3) << "FastBroadCastHeightCUDAF w:" << w << " h:" << h;

    if (w < 16 || h < 16) {
      int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, h);
      int grid_size = w;
      ElemwiseGradBroadcast1CUDAKernel<<<grid_size, block_size, 0, stream>>>(
          x_data, y_data, out_data, dout_data, h, w, x_large, dx_op, dy_op,
          dx_data, dy_data);
    } else {
      dim3 block_size = dim3(BLOCK_X, BLOCK_Y);
      int grid_size = (w + BLOCK_X - 1) / BLOCK_X;
      FastElemwiseGradBroadcast1CUDAKernel<<<grid_size, block_size, 0,
                                             stream>>>(
          x_data, y_data, out_data, dout_data, h, w, x_large, dx_op, dy_op,
          dx_data, dy_data);
    }
  };

  auto FastBroadCastAllCUDAF = [&](const std::vector<int> &broadcast_pos,
                                   int max_dim, bool is_x_large) {
    int axis = broadcast_pos[0];
    int pre = std::accumulate(out_dims_array, out_dims_array + axis, 1,
                              std::multiplies<int>());
1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
    int mid = 1;
    int post = 1;

    if (broadcast_pos.size() == 1) {
      mid = out_dims_array[axis];
      post =
          std::accumulate(out_dims_array + axis + 1, out_dims_array + max_dim,
                          1, std::multiplies<int>());
    } else {
      mid = std::accumulate(out_dims_array + axis,
                            out_dims_array + broadcast_pos.back() + 1, 1,
                            std::multiplies<int>());
      post =
          std::accumulate(out_dims_array + broadcast_pos.back() + 1,
                          out_dims_array + max_dim, 1, std::multiplies<int>());
    }
1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114

    VLOG(3) << "FastBroadCastAllCUDAF pre:" << pre << " mid:" << mid
            << " post:" << post;

    int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, mid);
    int grid_size = pre * post;

    FastCommonGradBroadcastAllCUDAKernel<<<grid_size, block_size, 0, stream>>>(
        x_data, y_data, out_data, dout_data, pre, mid, post, is_x_large, dx_op,
        dy_op, dx_data, dy_data);
  };

1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
  auto FastBroadCastOneCUDAF = [&](const std::vector<int> &broadcast_pos,
                                   int max_dim, bool is_x) {
    int axis = broadcast_pos[0];
    int pre = std::accumulate(out_dims_array, out_dims_array + axis, 1,
                              std::multiplies<int>());
    int mid = out_dims_array[axis];
    int post =
        std::accumulate(out_dims_array + axis + 1, out_dims_array + max_dim, 1,
                        std::multiplies<int>());

    int k_pre;
    int k_mid;
    int k_post;

    if (is_x) {
      k_pre = std::accumulate(y_dims_array, y_dims_array + axis, 1,
                              std::multiplies<int>());
      k_mid = y_dims_array[axis];
      k_post = std::accumulate(y_dims_array + axis + 1, y_dims_array + max_dim,
                               1, std::multiplies<int>());
      int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, mid);
      int grid_size = pre * post;
      // we need to calc y offset with blockid, so do x_pre/y_pre to get left
      // size.
      if (k_pre != pre) k_pre = pre / k_pre;

      FastCommonGradBroadcastOneCUDAKernel<<<grid_size, block_size, 0,
                                             stream>>>(
          x_data, y_data, out_data, dout_data, pre, mid, post, k_pre, k_mid,
          k_post, true, dx_op, dx_data);
    } else {
      k_pre = std::accumulate(x_dims_array, x_dims_array + axis, 1,
                              std::multiplies<int>());
      k_mid = x_dims_array[axis];
      k_post = std::accumulate(x_dims_array + axis + 1, x_dims_array + max_dim,
                               1, std::multiplies<int>());
      int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, mid);
      int grid_size = pre * post;
      if (k_pre != pre) k_pre = pre / k_pre;

      FastCommonGradBroadcastOneCUDAKernel<<<grid_size, block_size, 0,
                                             stream>>>(
          x_data, y_data, out_data, dout_data, pre, mid, post, k_pre, k_mid,
          k_post, false, dy_op, dy_data);
    }
    VLOG(3) << "FastBroadCastOneCUDAF pre:" << pre << " mid:" << mid
            << " post:" << post;
  };

1164 1165 1166 1167
  // do fast elementwise if: 1. only one input need to do broadcast, we can
  // fallback
  // to old fast path.
  // 2. if both x and y need broadcast, then do it one by one.
1168
  bool fast_broadcast = false;
1169 1170 1171 1172 1173 1174
  if (x_broadcast_pos.empty() && !y_broadcast_pos.empty()) {
    can_split_y = SplitDims(y_broadcast_pos, max_dim);
    if (can_split_y) {
      // only y need to do broadcast on h
      if (y_broadcast_pos[0] == 0) {
        FastBroadCastHeightCUDAF(y_broadcast_pos, true);
1175
        fast_broadcast = true;
1176
      }
1177 1178 1179
    } else if (y_broadcast_pos.size() == 1 ||
               CheckContiguousDims(y_broadcast_pos)) {  // for only one dim and
                                                        // contiguous broadcast.
1180 1181
      // If cannot split,  which means input has 3 parts
      FastBroadCastAllCUDAF(y_broadcast_pos, max_dim, true);
1182
      fast_broadcast = true;
1183 1184 1185 1186 1187 1188 1189
    }
  } else if (y_broadcast_pos.empty() && !x_broadcast_pos.empty()) {
    // only x need broadcast
    can_split_x = SplitDims(x_broadcast_pos, max_dim);
    if (can_split_x) {
      if (x_broadcast_pos[0] == 0) {
        FastBroadCastHeightCUDAF(x_broadcast_pos, false);
1190
        fast_broadcast = true;
1191
      }
1192 1193
    } else if (x_broadcast_pos.size() == 1 ||
               CheckContiguousDims(x_broadcast_pos)) {
1194
      FastBroadCastAllCUDAF(x_broadcast_pos, max_dim, false);
1195
      fast_broadcast = true;
1196 1197 1198 1199
    }
  } else if (!x_broadcast_pos.empty() && !y_broadcast_pos.empty()) {
    // do x and y broadcast each.
    can_split_y = SplitDims(y_broadcast_pos, max_dim);
1200 1201
    bool fast_broadcast_x = false;
    bool fast_broadcast_y = false;
1202 1203 1204 1205
    if (can_split_y) {
      // begin at start.
      if (y_broadcast_pos[0] == 0) {
        FastCommonCUDAF(y_broadcast_pos, true);
1206
        fast_broadcast_y = true;
1207
      }
1208 1209 1210
    } else if (y_broadcast_pos.size() == 1) {
      FastBroadCastOneCUDAF(y_broadcast_pos, max_dim, false);
      can_split_y = true;
1211
      fast_broadcast_y = true;
1212 1213 1214 1215 1216
    }
    can_split_x = SplitDims(x_broadcast_pos, max_dim);
    if (can_split_x) {
      if (x_broadcast_pos[0] == 0) {
        FastCommonCUDAF(x_broadcast_pos, false);
1217
        fast_broadcast_x = true;
1218
      }
1219 1220 1221
    } else if (x_broadcast_pos.size() == 1) {
      FastBroadCastOneCUDAF(x_broadcast_pos, max_dim, true);
      can_split_x = true;
1222
      fast_broadcast_x = true;
1223 1224 1225 1226
    }
    VLOG(3) << "CommonBroadcast can_split_y:" << can_split_y
            << " can_split_x:" << can_split_x;
    // if both x and y into fast path then return
1227 1228 1229 1230
    if (fast_broadcast_x && fast_broadcast_y) {
      fast_broadcast = true;
    }
    if (can_split_y && can_split_x && fast_broadcast) return;
1231
  }
1232

1233
  // Should remove memory copy, use reg instead.
1234 1235 1236
  if (fast_broadcast) {
    return;
  }
1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266
  int x_blocks = 0;
  int x_threads = 0;
  ComputeBroadcastKernelSize(x_dims_array, out_dims_array, &x_blocks,
                             &x_threads, max_dim);
  int y_blocks = 0;
  int y_threads = 0;
  ComputeBroadcastKernelSize(y_dims_array, out_dims_array, &y_blocks,
                             &y_threads, max_dim);

  auto x_strides_array_tmp = memory::Alloc(ctx, bytes);
  int *x_strides_array_gpu =
      reinterpret_cast<int *>(x_strides_array_tmp->ptr());
  memory::Copy(gplace, x_strides_array_gpu, cplace, x_strides_array.data(),
               bytes, ctx.stream());

  auto y_strides_array_tmp = memory::Alloc(ctx, bytes);
  int *y_strides_array_gpu =
      reinterpret_cast<int *>(y_strides_array_tmp->ptr());
  memory::Copy(gplace, y_strides_array_gpu, cplace, y_strides_array.data(),
               bytes, ctx.stream());

  auto out_dims_array_tmp = memory::Alloc(ctx, bytes);
  int *out_dims_array_gpu = reinterpret_cast<int *>(out_dims_array_tmp->ptr());
  memory::Copy(gplace, out_dims_array_gpu, cplace, out_dims_array, bytes,
               ctx.stream());

  const int out_size = std::accumulate(out_dims_array, out_dims_array + max_dim,
                                       1, std::multiplies<int>());
  int x_block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, x_threads);
  int y_block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, y_threads);
1267
  if (dx) {
1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283
    auto x_strides_order_tmp = memory::Alloc(ctx, bytes);
    int *x_strides_order_gpu =
        reinterpret_cast<int *>(x_strides_order_tmp->ptr());
    memory::Copy(gplace, x_strides_order_gpu, cplace, x_strides_order.data(),
                 bytes, ctx.stream());

    auto x_dims_order_tmp = memory::Alloc(ctx, bytes);
    int *x_dims_order_gpu = reinterpret_cast<int *>(x_dims_order_tmp->ptr());
    memory::Copy(gplace, x_dims_order_gpu, cplace, x_dims_order.data(), bytes,
                 ctx.stream());
    CommonGradBroadcastCUDAKernel<
        T, DX_OP><<<x_blocks, x_block_size, 0, ctx.stream()>>>(
        x_strides_array_gpu, y_strides_array_gpu, out_dims_array_gpu,
        x_strides_order_gpu, x_dims_order_gpu, x_data, y_data, out_data,
        dout_data, dx_data, out_size, max_dim, x_threads, dx_op);
  }
1284
  if (dy) {
1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302
    auto y_strides_order_tmp = memory::Alloc(ctx, bytes);
    int *y_strides_order_gpu =
        reinterpret_cast<int *>(y_strides_order_tmp->ptr());
    memory::Copy(gplace, y_strides_order_gpu, cplace, y_strides_order.data(),
                 bytes, ctx.stream());

    auto y_dims_order_tmp = memory::Alloc(ctx, bytes);
    int *y_dims_order_gpu = reinterpret_cast<int *>(y_dims_order_tmp->ptr());
    memory::Copy(gplace, y_dims_order_gpu, cplace, y_dims_order.data(), bytes,
                 ctx.stream());
    CommonGradBroadcastCUDAKernel<
        T, DY_OP><<<y_blocks, y_block_size, 0, ctx.stream()>>>(
        x_strides_array_gpu, y_strides_array_gpu, out_dims_array_gpu,
        y_strides_order_gpu, y_dims_order_gpu, x_data, y_data, out_data,
        dout_data, dy_data, out_size, max_dim, y_threads, dy_op);
  }
}

1303
#endif  // __NVCC__ or __HIPCC__
1304

1305
inline framework::DDim trim_trailing_singular_dims(
1306
    const framework::DDim &dims) {
1307
  // Remove trailing dimensions of size 1 for y
1308
  auto actual_dims_size = dims.size();
1309
  for (; actual_dims_size != 0; --actual_dims_size) {
1310
    if (dims[actual_dims_size - 1] != 1) break;
1311
  }
1312
  if (actual_dims_size == dims.size()) return dims;
1313 1314 1315 1316
  std::vector<int> trim_dims;
  trim_dims.resize(actual_dims_size);
  for (int i = 0; i < actual_dims_size; ++i) {
    trim_dims[i] = dims[i];
1317
  }
1318 1319 1320
  if (trim_dims.size() == 0) {
    return framework::DDim(framework::make_dim());
  }
1321 1322
  framework::DDim actual_dims = framework::make_ddim(trim_dims);
  return actual_dims;
1323 1324
}

Q
QI JUN 已提交
1325
template <typename T, typename DeviceContext>
C
chengduoZH 已提交
1326
class RowwiseTransformIterator;
1327

Q
QI JUN 已提交
1328
template <typename T, typename DeviceContext>
C
chengduoZH 已提交
1329
class MidWiseTransformIterator;
C
chengduoZH 已提交
1330

D
dzhwinter 已提交
1331
// NOTE(dzhwinter): ptrdiff_t in iterator is deperecated in c++17
C
chengduoZH 已提交
1332
template <typename T>
D
dzhwinter 已提交
1333 1334 1335
class RowwiseTransformIterator<T, platform::CPUDeviceContext>
    : public std::iterator<std::random_access_iterator_tag, T, std::ptrdiff_t,
                           T *, T &> {
C
chengduoZH 已提交
1336
 public:
1337
  RowwiseTransformIterator(const T *ptr, int n) : ptr_(ptr), i_(0), n_(n) {}
C
chengduoZH 已提交
1338

1339
  RowwiseTransformIterator<T, platform::CPUDeviceContext> &operator++() {
C
chengduoZH 已提交
1340
    ++i_;
C
chengduoZH 已提交
1341 1342 1343
    if (UNLIKELY(i_ == n_)) {
      i_ = 0;
    }
C
chengduoZH 已提交
1344 1345 1346
    return *this;
  }

P
peizhilin 已提交
1347
  RowwiseTransformIterator<T, platform::CPUDeviceContext> &operator+(int n) {
P
peizhilin 已提交
1348
    while (n-- > 0) {
P
peizhilin 已提交
1349 1350 1351 1352 1353 1354 1355 1356 1357
      ++i_;
      if (UNLIKELY(i_ == n_)) {
        i_ = 0;
      }
    }

    return *this;
  }

1358 1359
  bool operator==(const RowwiseTransformIterator<T, platform::CPUDeviceContext>
                      &rhs) const {
C
chengduoZH 已提交
1360
    return (ptr_ + i_) == &(*rhs);
C
chengduoZH 已提交
1361 1362
  }

1363 1364
  bool operator!=(const RowwiseTransformIterator<T, platform::CPUDeviceContext>
                      &rhs) const {
C
chengduoZH 已提交
1365
    return (ptr_ + i_) != &(*rhs);
C
chengduoZH 已提交
1366 1367
  }

1368
  const T &operator*() { return ptr_[i_]; }
C
chengduoZH 已提交
1369

C
chengduoZH 已提交
1370
 private:
1371
  const T *ptr_;
C
chengduoZH 已提交
1372
  int i_;
C
chengduoZH 已提交
1373
  int64_t n_;
C
chengduoZH 已提交
1374 1375 1376
};

template <typename T>
D
dzhwinter 已提交
1377 1378 1379
class MidWiseTransformIterator<T, platform::CPUDeviceContext>
    : public std::iterator<std::random_access_iterator_tag, T, std::ptrdiff_t,
                           T *, T &> {
C
chengduoZH 已提交
1380
 public:
1381
  MidWiseTransformIterator(const T *ptr, int n, int post)
C
chengduoZH 已提交
1382 1383
      : ptr_(ptr), i_(0), j_(0), n_(n), post_(post) {}

1384
  MidWiseTransformIterator<T, platform::CPUDeviceContext> &operator++() {
C
chengduoZH 已提交
1385
    ++j_;
C
chengduoZH 已提交
1386 1387
    if (UNLIKELY(j_ == post_)) {
      ++i_;
C
refine  
chengduoZH 已提交
1388
      j_ = 0;
C
chengduoZH 已提交
1389 1390 1391
      if (UNLIKELY(i_ == n_)) {
        i_ = 0;
      }
C
chengduoZH 已提交
1392
    }
C
chengduoZH 已提交
1393 1394 1395
    return *this;
  }

P
peizhilin 已提交
1396
  MidWiseTransformIterator<T, platform::CPUDeviceContext> &operator+(int n) {
P
peizhilin 已提交
1397
    while (n-- > 0) {
P
peizhilin 已提交
1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
      ++j_;
      if (UNLIKELY(j_ == post_)) {
        ++i_;
        j_ = 0;
        if (UNLIKELY(i_ == n_)) {
          i_ = 0;
        }
      }
    }
    return *this;
  }

1410 1411
  bool operator==(const MidWiseTransformIterator<T, platform::CPUDeviceContext>
                      &rhs) const {
C
chengduoZH 已提交
1412
    return (ptr_ + i_) == &(*rhs);
C
chengduoZH 已提交
1413 1414
  }

1415 1416
  bool operator!=(const MidWiseTransformIterator<T, platform::CPUDeviceContext>
                      &rhs) const {
C
chengduoZH 已提交
1417
    return (ptr_ + i_) != &(*rhs);
C
chengduoZH 已提交
1418 1419
  }

1420
  const T &operator*() { return ptr_[i_]; }
C
chengduoZH 已提交
1421

C
chengduoZH 已提交
1422
 private:
1423
  const T *ptr_;
C
refine  
chengduoZH 已提交
1424
  int64_t i_;
C
chengduoZH 已提交
1425 1426
  int64_t j_;
  int64_t n_;
C
refine  
chengduoZH 已提交
1427
  int64_t post_;
C
chengduoZH 已提交
1428 1429
};

1430
#if defined(__NVCC__) || defined(__HIPCC__)
C
chengduoZH 已提交
1431
template <typename T>
Q
QI JUN 已提交
1432
class RowwiseTransformIterator<T, platform::CUDADeviceContext>
C
chengduoZH 已提交
1433
    : public thrust::iterator_adaptor<
1434
          RowwiseTransformIterator<T, platform::CUDADeviceContext>, const T *> {
C
chengduoZH 已提交
1435 1436
 public:
  typedef thrust::iterator_adaptor<
1437
      RowwiseTransformIterator<T, platform::CUDADeviceContext>, const T *>
C
chengduoZH 已提交
1438
      super_t;
1439
  HOSTDEVICE RowwiseTransformIterator(const T *x, int n)
1440
      : super_t(x), begin_(x), n_(n) {}
C
chengduoZH 已提交
1441 1442 1443 1444
  friend class thrust::iterator_core_access;

 private:
  unsigned int n_;
1445
  const T *begin_;
C
chengduoZH 已提交
1446
  HOSTDEVICE typename super_t::reference dereference() const {
C
chengduoZH 已提交
1447 1448 1449 1450 1451
    return *(begin_ + (this->base() - begin_) % n_);
  }
};

template <typename T>
Q
QI JUN 已提交
1452
class MidWiseTransformIterator<T, platform::CUDADeviceContext>
C
chengduoZH 已提交
1453
    : public thrust::iterator_adaptor<
1454
          MidWiseTransformIterator<T, platform::CUDADeviceContext>, const T *> {
C
chengduoZH 已提交
1455 1456
 public:
  typedef thrust::iterator_adaptor<
1457
      MidWiseTransformIterator<T, platform::CUDADeviceContext>, const T *>
C
chengduoZH 已提交
1458
      super_t;
1459
  HOSTDEVICE MidWiseTransformIterator(const T *x, int n, int post)
1460
      : super_t(x), begin_(x), n_(n), post_(post) {}
C
chengduoZH 已提交
1461 1462 1463 1464 1465
  friend class thrust::iterator_core_access;

 private:
  unsigned int post_;
  unsigned int n_;
1466
  const T *begin_;
C
chengduoZH 已提交
1467
  HOSTDEVICE typename super_t::reference dereference() const {
C
chengduoZH 已提交
1468 1469 1470 1471 1472
    return *(begin_ + (((this->base() - begin_) / post_) % n_));
  }
};
#endif

1473 1474
template <typename Functor, typename T, typename DeviceContext,
          typename OutType = T>
C
chengduoZH 已提交
1475 1476
class TransformFunctor {
 public:
1477
  TransformFunctor(const framework::Tensor *x, const framework::Tensor *y,
1478 1479
                   framework::Tensor *z, const DeviceContext &ctx, Functor func,
                   const bool is_xsize_larger = true)
C
chengduoZH 已提交
1480 1481
      : x_(x->data<T>()),
        y_(y->data<T>()),
1482
        z_(z->mutable_data<OutType>(ctx.GetPlace())),
C
chengduoZH 已提交
1483 1484
        nx_(x->numel()),
        ctx_(ctx),
1485 1486 1487 1488 1489 1490
        func_(func),
        is_xsize_larger_(is_xsize_larger) {
    if (is_xsize_larger_ == false) {
      nx_ = y->numel();
    }
  }
C
chengduoZH 已提交
1491 1492

  inline void Run() const {
Q
QI JUN 已提交
1493
    platform::Transform<DeviceContext> trans;
C
chengduoZH 已提交
1494
    trans(ctx_, x_, x_ + nx_, y_, z_, func_);
C
chengduoZH 已提交
1495 1496 1497
  }

  inline void RunRowWise(int n, int pre) const {
Q
QI JUN 已提交
1498
    platform::Transform<DeviceContext> trans;
1499 1500 1501 1502 1503 1504 1505
    if (is_xsize_larger_) {
      trans(ctx_, x_, x_ + nx_,
            RowwiseTransformIterator<T, DeviceContext>(y_, n), z_, func_);
    } else {
      trans(ctx_, y_, y_ + nx_,
            RowwiseTransformIterator<T, DeviceContext>(x_, n), z_, func_);
    }
C
chengduoZH 已提交
1506 1507 1508
  }

  inline void RunMidWise(int n, int pre, int post) const {
Q
QI JUN 已提交
1509
    platform::Transform<DeviceContext> trans;
1510 1511 1512 1513 1514 1515
    if (is_xsize_larger_) {
      trans(ctx_, x_, x_ + nx_,
            MidWiseTransformIterator<T, DeviceContext>(y_, n, post), z_, func_);
    } else {
      trans(ctx_, y_, y_ + nx_,
            MidWiseTransformIterator<T, DeviceContext>(x_, n, post), z_, func_);
1516 1517 1518
    }
  }

C
chengduoZH 已提交
1519
 private:
1520 1521 1522
  const T *x_;
  const T *y_;
  OutType *z_;
C
chengduoZH 已提交
1523
  int64_t nx_;
1524
  const DeviceContext &ctx_;
C
chengduoZH 已提交
1525
  Functor func_;
1526
  bool is_xsize_larger_;
C
chengduoZH 已提交
1527 1528
};

Y
Yu Yang 已提交
1529 1530
template <typename T, typename DX_OP, typename DY_OP>
struct ElemwiseGradNoBroadcast {
1531 1532 1533 1534
  const T *x_;
  const T *y_;
  const T *out_;
  const T *dout_;
Y
Yu Yang 已提交
1535 1536 1537 1538 1539 1540

  HOSTDEVICE void operator()(size_t i) {
    if (dx_ != nullptr) {
      dx_[i] = dx_op_(x_[i], y_[i], out_[i], dout_[i]);
    }
    if (dy_ != nullptr) {
C
chengduoZH 已提交
1541
      dy_[i] = dy_op_(x_[i], y_[i], out_[i], dout_[i]);
Y
Yu Yang 已提交
1542 1543 1544 1545 1546
    }
  }

  DX_OP dx_op_;
  DY_OP dy_op_;
1547 1548
  T *dx_;
  T *dy_;
Y
Yu Yang 已提交
1549 1550 1551
};

template <typename T, typename DX_OP, typename DY_OP>
1552
static void ElemwiseGradBroadcast1CPU(const T *x, const T *y, const T *out,
1553 1554
                                      const T *dout, int h, int w,
                                      bool is_xsize_larger, DX_OP dx_op,
1555
                                      DY_OP dy_op, T *dx, T *dy) {
1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571
  if (is_xsize_larger) {
    for (int i = 0; i < h; ++i) {
      for (int j = 0; j < w; ++j) {
        int x_offset = i * w + j;
        if (dx != nullptr) {
          dx[x_offset] =
              dx_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
        }
        if (dy != nullptr) {
          T tmp = dy_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
          if (i == 0) {
            dy[j] = tmp;
          } else {
            dy[j] += tmp;
          }
        }
Y
Yu Yang 已提交
1572
      }
1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588
    }
  } else {  // x.dims < y.dims, broadcast for x.
    for (int i = 0; i < h; ++i) {
      for (int j = 0; j < w; ++j) {
        int y_offset = i * w + j;
        if (dy != nullptr) {
          dy[y_offset] =
              dy_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
        }
        if (dx != nullptr) {
          T tmp = dx_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
          if (i == 0) {
            dx[j] = tmp;
          } else {
            dx[j] += tmp;
          }
Y
Yu Yang 已提交
1589 1590 1591 1592 1593
        }
      }
    }
  }
}
1594

1595
#if defined(__NVCC__) || defined(__HIPCC__)
1596

Y
Yu Yang 已提交
1597
template <typename T, typename DX_OP, typename DY_OP>
1598
static void ElemwiseGradBroadcast1CUDA(gpuStream_t stream, const T *x,
1599
                                       const T *y, const T *out, const T *dout,
1600 1601
                                       int h, int w, bool is_xsize_larger,
                                       DX_OP dx_op, DY_OP dy_op, T *dx, T *dy) {
1602 1603 1604 1605 1606 1607
  // For small case use 1D block
  constexpr int half_walf = 16;
  if (w < half_walf || h < half_walf) {
    int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, h);
    int gird_size = w;
    ElemwiseGradBroadcast1CUDAKernel<<<gird_size, block_size, 0, stream>>>(
1608
        x, y, out, dout, h, w, is_xsize_larger, dx_op, dy_op, dx, dy);
1609 1610 1611 1612 1613
  } else {
    // suppose perfoemance improves with h increased.
    dim3 block_size = dim3(BLOCK_X, BLOCK_Y);
    int grid_size = (w + BLOCK_X - 1) / BLOCK_X;
    FastElemwiseGradBroadcast1CUDAKernel<<<grid_size, block_size, 0, stream>>>(
1614
        x, y, out, dout, h, w, is_xsize_larger, dx_op, dy_op, dx, dy);
1615
  }
Y
Yu Yang 已提交
1616 1617 1618 1619 1620
}

#endif

template <typename T, typename DX_OP, typename DY_OP>
1621 1622
static void ElemwiseGradBroadcast2CPU(const T *x, const T *y, const T *out,
                                      const T *dout, int pre, int n, int post,
1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641
                                      bool is_xsize_larger, DX_OP dx_op,
                                      DY_OP dy_op, T *dx, T *dy) {
  if (is_xsize_larger) {
    for (int i = 0; i < pre; ++i) {
      for (int j = 0; j < n; ++j) {
        for (int k = 0; k < post; ++k) {
          int x_offset = i * n * post + j * post + k;
          if (dx != nullptr) {
            dx[x_offset] =
                dx_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
          }
          if (dy != nullptr) {
            T tmp = dy_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
            if (i == 0 && k == 0) {
              dy[j] = tmp;
            } else {
              dy[j] += tmp;
            }
          }
Y
Yu Yang 已提交
1642
        }
1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660
      }
    }
  } else {  // x.dims < y.dims, broadcast for x.
    for (int i = 0; i < pre; ++i) {
      for (int j = 0; j < n; ++j) {
        for (int k = 0; k < post; ++k) {
          int y_offset = i * n * post + j * post + k;
          if (dy != nullptr) {
            dy[y_offset] =
                dy_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
          }
          if (dx != nullptr) {
            T tmp = dx_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
            if (i == 0 && k == 0) {
              dx[j] = tmp;
            } else {
              dx[j] += tmp;
            }
Y
Yu Yang 已提交
1661 1662 1663 1664 1665 1666 1667
          }
        }
      }
    }
  }
}

1668
#if defined(__NVCC__) || defined(__HIPCC__)
Y
Yu Yang 已提交
1669 1670
template <typename T, typename DX_OP, typename DY_OP>
static __global__ void ElemwiseGradBroadcast2CUDAKernel(
1671
    const T *x, const T *y, const T *out, const T *dout, int pre, int n,
1672
    int post, bool is_xsize_larger, DX_OP dx_op, DY_OP dy_op, T *dx, T *dy) {
Y
Yu Yang 已提交
1673 1674 1675
  int tid = threadIdx.x;
  int j = blockIdx.x;

C
chengduo 已提交
1676
  T val(0);
Y
Yu Yang 已提交
1677 1678
  int ttid = tid;

1679 1680 1681 1682 1683
  if (is_xsize_larger) {
    while (true) {
      int i = ttid / post;
      int k = ttid % post;
      if (i >= pre) break;
Y
Yu Yang 已提交
1684

1685
      int x_offset = i * n * post + j * post + k;
Y
Yu Yang 已提交
1686

1687 1688 1689 1690 1691 1692 1693 1694 1695
      if (dx != nullptr) {
        dx[x_offset] = dx_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
      }

      if (dy != nullptr) {
        val += dy_op(x[x_offset], y[j], out[x_offset], dout[x_offset]);
      }

      ttid += ELEMWISE_MAX_BLOCK_DIM;
Y
Yu Yang 已提交
1696 1697
    }

1698 1699 1700 1701 1702 1703 1704
    if (dy) {
      int h = pre * post;
      h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dy[j] = val;
      }
Y
Yu Yang 已提交
1705
    }
1706 1707 1708 1709 1710
  } else {  // x.dims < y.dims, broadcast for x.
    while (true) {
      int i = ttid / post;
      int k = ttid % post;
      if (i >= pre) break;
Y
Yu Yang 已提交
1711

1712
      int y_offset = i * n * post + j * post + k;
Y
Yu Yang 已提交
1713

1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731
      if (dy != nullptr) {
        dy[y_offset] = dy_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
      }

      if (dx != nullptr) {
        val += dx_op(x[j], y[y_offset], out[y_offset], dout[y_offset]);
      }

      ttid += ELEMWISE_MAX_BLOCK_DIM;
    }

    if (dx) {
      int h = pre * post;
      h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dx[j] = val;
      }
Y
Yu Yang 已提交
1732 1733 1734 1735 1736
    }
  }
}

template <typename T, typename DX_OP, typename DY_OP>
1737
static void ElemwiseGradBroadcast2CUDA(gpuStream_t stream, const T *x,
1738
                                       const T *y, const T *out, const T *dout,
1739 1740
                                       int pre, int n, int post,
                                       bool is_xsize_larger, DX_OP dx_op,
1741
                                       DY_OP dy_op, T *dx, T *dy) {
Y
Yu Yang 已提交
1742 1743
  int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, pre * post);
  int gird_size = n;
C
chengduoZH 已提交
1744
  ElemwiseGradBroadcast2CUDAKernel<<<gird_size, block_size, 0, stream>>>(
1745
      x, y, out, dout, pre, n, post, is_xsize_larger, dx_op, dy_op, dx, dy);
Y
Yu Yang 已提交
1746 1747 1748 1749
}

#endif

1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766
template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP>
void CommonElementwiseBroadcastBackward(
    const framework::ExecutionContext &ctx, const framework::DDim &x_dims,
    const framework::DDim &y_dims, const framework::Tensor &x,
    const framework::Tensor &y, const framework::Tensor &out,
    const framework::Tensor &dout, int axis, framework::Tensor *dx,
    framework::Tensor *dy, DX_OP dx_op, DY_OP dy_op) {
  int max_dim = std::max(x_dims.size(), y_dims.size());
  axis = (axis == -1 ? std::abs(x_dims.size() - y_dims.size()) : axis);
  std::vector<int> x_dims_array(max_dim);
  std::vector<int> y_dims_array(max_dim);
  std::vector<int> out_dims_array(max_dim);
  GetBroadcastDimsArrays(x_dims, y_dims, x_dims_array.data(),
                         y_dims_array.data(), out_dims_array.data(), max_dim,
                         axis);
  // for inplace strategy. memset will make dx and dout clear and get wrong
  // result.
1767
  if (dx && dx->IsSharedBufferWith(dout)) {
1768 1769
    dx->clear();
    dx->mutable_data<T>(x_dims, ctx.GetPlace());
1770 1771
  }

1772 1773 1774 1775
  VLOG(3) << "CommonElementwiseBroadcastBackward xdims:"
          << framework::make_ddim(x_dims_array)
          << " ydim:" << framework::make_ddim(y_dims_array);

1776
  if (platform::is_gpu_place(ctx.GetPlace())) {
1777
#if defined(__NVCC__) || defined(__HIPCC__)
1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789
    CommonGradBroadcastCUDA<T, DX_OP, DY_OP>(
        x, y, out, dout, dx, dy, x_dims_array.data(), y_dims_array.data(),
        out_dims_array.data(), max_dim,
        ctx.template device_context<platform::CUDADeviceContext>(), dx_op,
        dy_op);
#endif
  } else {
    CommonGradBroadcastCPU<T, DX_OP, DY_OP>(
        x, y, out, dout, dx, dy, x_dims_array.data(), y_dims_array.data(),
        out_dims_array.data(), max_dim,
        ctx.template device_context<platform::CPUDeviceContext>(), dx_op,
        dy_op);
1790 1791 1792
  }
}

1793 1794
template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP>
void ElemwiseGradComputeNoBroadcast(
1795 1796 1797 1798 1799
    const framework::ExecutionContext &ctx, const framework::DDim &x_dim,
    const framework::DDim &y_dim, const framework::Tensor &x,
    const framework::Tensor &y, const framework::Tensor &out,
    const framework::Tensor &dout, int axis, framework::Tensor *dx,
    framework::Tensor *dy, DX_OP dx_op, DY_OP dy_op) {
1800
  size_t N = static_cast<size_t>(framework::product(x_dim));
D
dzhwinter 已提交
1801
#if !defined(_WIN32)
1802 1803
  platform::ForRange<DeviceContext> for_range(
      ctx.template device_context<DeviceContext>(), N);
D
dzhwinter 已提交
1804 1805 1806 1807
#else
  platform::ForRange<DeviceContext> for_range(
      ctx.device_context<DeviceContext>(), N);
#endif  // !_WIN32
1808 1809 1810 1811 1812 1813 1814 1815
  for_range(ElemwiseGradNoBroadcast<T, DX_OP, DY_OP>{
      x.data<T>(), y.data<T>(), out.data<T>(), dout.data<T>(), dx_op, dy_op,
      dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
      dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace())});
}

template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP>
void ElemwiseGradComputeWithBroadcast(
1816 1817
    const framework::ExecutionContext &ctx, const framework::DDim &x_dims,
    const framework::DDim &y_dims, const framework::Tensor &x,
1818 1819 1820
    const framework::Tensor &y, const framework::Tensor &out,
    const framework::Tensor &dout, int axis, framework::Tensor *dx,
    framework::Tensor *dy, DX_OP dx_op, DY_OP dy_op) {
1821
  bool is_xsize_larger = true;
1822

1823 1824 1825 1826 1827
  int max_dim = x_dims.size();
  if (x_dims.size() < y_dims.size()) {
    is_xsize_larger = false;
    max_dim = y_dims.size();
  }
1828

1829
  axis = (axis == -1 ? std::abs(x_dims.size() - y_dims.size()) : axis);
1830 1831 1832 1833 1834 1835 1836 1837 1838
  PADDLE_ENFORCE_GE(
      axis, 0,
      platform::errors::InvalidArgument(
          "Axis should be great than or equal to 0, but received axis is %d.",
          axis));
  PADDLE_ENFORCE_LT(axis, max_dim,
                    platform::errors::InvalidArgument(
                        "Axis should be less than %d, but received axis is %d.",
                        max_dim, axis));
1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858

  int pre, n, post, is_run_common_broadcast, axis_trim = 0;
  if (is_xsize_larger) {
    auto y_dims_trimed = trim_trailing_singular_dims(y_dims);
    axis_trim = (y_dims_trimed.size() == 0) ? x_dims.size() : axis;
    get_mid_dims(x_dims, y_dims_trimed, axis_trim, &pre, &n, &post,
                 &is_run_common_broadcast);
  } else {
    auto x_dims_trimed = trim_trailing_singular_dims(x_dims);
    axis_trim = (x_dims_trimed.size() == 0) ? y_dims.size() : axis;
    get_mid_dims(y_dims, x_dims_trimed, axis_trim, &pre, &n, &post,
                 &is_run_common_broadcast);
  }
  // special case for common backward implementation.
  if (is_run_common_broadcast) {
    CommonElementwiseBroadcastBackward<DeviceContext, T, DX_OP, DY_OP>(
        ctx, x_dims, y_dims, x, y, out, dout, axis, dx, dy, dx_op, dy_op);
    return;
  }
  if (post == 1) {
1859
    if (platform::is_gpu_place(ctx.GetPlace())) {
1860
#if defined(__NVCC__) || defined(__HIPCC__)
1861 1862
      ElemwiseGradBroadcast1CUDA(
          ctx.template device_context<DeviceContext>().stream(), x.data<T>(),
1863 1864
          y.data<T>(), out.data<T>(), dout.data<T>(), pre, n, is_xsize_larger,
          dx_op, dy_op,
1865 1866 1867 1868 1869
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()));
#endif
    } else {
      ElemwiseGradBroadcast1CPU(
1870
          x.data<T>(), y.data<T>(), out.data<T>(), dout.data<T>(), pre, n,
1871
          is_xsize_larger, dx_op, dy_op,
1872
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
1873 1874 1875 1876
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()));
    }
  } else {
    if (platform::is_gpu_place(ctx.GetPlace())) {
1877
#if defined(__NVCC__) || defined(__HIPCC__)
1878 1879
      ElemwiseGradBroadcast2CUDA(
          ctx.template device_context<DeviceContext>().stream(), x.data<T>(),
1880 1881 1882
          y.data<T>(), out.data<T>(), dout.data<T>(), pre, n, post,
          is_xsize_larger, dx_op, dy_op,
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
1883 1884 1885 1886 1887
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()));
#endif
    } else {
      ElemwiseGradBroadcast2CPU(
          x.data<T>(), y.data<T>(), out.data<T>(), dout.data<T>(), pre, n, post,
1888
          is_xsize_larger, dx_op, dy_op,
1889 1890 1891 1892 1893 1894
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()));
    }
  }
}

1895 1896 1897 1898 1899 1900 1901 1902 1903
template <typename Functor, typename DeviceContext, typename T,
          typename OutType = T>
void CommonElementwiseBroadcastForward(
    const framework::ExecutionContext &ctx, const framework::Tensor *x,
    const framework::Tensor *y, framework::Tensor *z,
    const framework::DDim &x_dims, const framework::DDim &y_dims, Functor func,
    int axis, const bool is_xsize_larger = true) {
  int max_dim = std::max(x_dims.size(), y_dims.size());
  axis = (axis == -1 ? std::abs(x_dims.size() - y_dims.size()) : axis);
1904 1905 1906 1907 1908 1909 1910 1911 1912
  PADDLE_ENFORCE_GE(
      axis, 0,
      platform::errors::InvalidArgument(
          "Axis should be great than or equal to 0, but received axis is %d.",
          axis));
  PADDLE_ENFORCE_LT(axis, max_dim,
                    platform::errors::InvalidArgument(
                        "Axis should be less than %d, but received axis is %d.",
                        max_dim, axis));
1913 1914 1915 1916 1917 1918 1919
  std::vector<int> x_dims_array(max_dim);
  std::vector<int> y_dims_array(max_dim);
  std::vector<int> out_dims_array(max_dim);
  GetBroadcastDimsArrays(x_dims, y_dims, x_dims_array.data(),
                         y_dims_array.data(), out_dims_array.data(), max_dim,
                         axis);

1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934
  if (platform::is_gpu_place(ctx.GetPlace())) {
#if defined(__NVCC__) || defined(__HIPCC__)
    CommonForwardBroadcastCUDA<Functor, T, OutType>(
        x, y, z, x_dims_array.data(), y_dims_array.data(),
        out_dims_array.data(), max_dim,
        ctx.template device_context<platform::CUDADeviceContext>(), func,
        is_xsize_larger);
#endif
  } else {
    CommonForwardBroadcastCPU<Functor, T, OutType>(
        x, y, z, x_dims_array.data(), y_dims_array.data(),
        out_dims_array.data(), max_dim,
        ctx.template device_context<platform::CPUDeviceContext>(), func,
        is_xsize_larger);
  }
1935 1936
}

Y
Yu Yang 已提交
1937
template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP>
1938 1939 1940 1941 1942
void ElemwiseGradCompute(const framework::ExecutionContext &ctx,
                         const framework::Tensor &x, const framework::Tensor &y,
                         const framework::Tensor &out,
                         const framework::Tensor &dout, int axis,
                         framework::Tensor *dx, framework::Tensor *dy,
Y
Yu Yang 已提交
1943
                         DX_OP dx_op, DY_OP dy_op) {
1944 1945
  const framework::DDim &x_dim = x.dims();
  const framework::DDim &y_dim = y.dims();
Y
Yu Yang 已提交
1946
  if (x.dims() == y.dims()) {
1947 1948
    ElemwiseGradComputeNoBroadcast<DeviceContext, T, DX_OP, DY_OP>(
        ctx, x_dim, y_dim, x, y, out, dout, axis, dx, dy, dx_op, dy_op);
1949
  } else {
1950 1951 1952 1953 1954 1955 1956 1957 1958 1959
    ElemwiseGradComputeWithBroadcast<DeviceContext, T, DX_OP, DY_OP>(
        ctx, x_dim, y_dim, x, y, out, dout, axis, dx, dy, dx_op, dy_op);
  }
}

// NOTE(dzhwinter): Only used in elementwise_add, elementwise_sub.
// explicit gradient can cut off X, Y, Out from gradient op
// In elementwise_add, elementwise_sub, we use dout as fake X, Y, Out to reuse
// elementwise code.
template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP>
1960 1961 1962 1963 1964 1965
void ElemwiseExplicitGradCompute(const framework::ExecutionContext &ctx,
                                 const framework::Tensor &x,
                                 const framework::Tensor &y,
                                 const framework::Tensor &out,
                                 const framework::Tensor &dout, int axis,
                                 framework::Tensor *dx, framework::Tensor *dy,
1966
                                 DX_OP dx_op, DY_OP dy_op) {
1967 1968 1969
  const framework::DDim &x_dim = x.dims();
  const framework::DDim &y_dim = y.dims();
  if (x.dims() == y.dims()) {
1970
    ElemwiseGradComputeNoBroadcast<DeviceContext, T, DX_OP, DY_OP>(
1971
        ctx, x_dim, y_dim, dout, dout, out, dout, axis, dx, dy, dx_op, dy_op);
1972
  } else {
1973 1974
    ElemwiseGradComputeWithBroadcast<DeviceContext, T, DX_OP, DY_OP>(
        ctx, x_dim, y_dim, dout, dout, out, dout, axis, dx, dy, dx_op, dy_op);
1975 1976
  }
}
F
fengjiayi 已提交
1977

1978 1979
template <typename Functor, typename DeviceContext, typename T,
          typename OutType = T>
1980 1981 1982 1983
void ElementwiseComputeEx(const framework::ExecutionContext &ctx,
                          const framework::Tensor *x,
                          const framework::Tensor *y, int axis, Functor func,
                          framework::Tensor *z) {
F
fengjiayi 已提交
1984
  auto x_dims = x->dims();
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995
  auto y_dims = y->dims();
  bool is_xsize_larger = true;
  int max_dim = x_dims.size();
  if (x_dims.size() < y_dims.size()) {
    is_xsize_larger = false;
    max_dim = y_dims.size();
  }
  TransformFunctor<Functor, T, DeviceContext, OutType> functor(
      x, y, z, ctx.template device_context<DeviceContext>(), func,
      is_xsize_larger);
  if (x_dims == y_dims) {
F
fengjiayi 已提交
1996 1997 1998 1999
    functor.Run();
    return;
  }

2000
  axis = (axis == -1 ? std::abs(x_dims.size() - y_dims.size()) : axis);
2001 2002 2003 2004 2005 2006 2007 2008 2009
  PADDLE_ENFORCE_GE(
      axis, 0,
      platform::errors::InvalidArgument(
          "Axis should be great than or equal to 0, but received axis is %d.",
          axis));
  PADDLE_ENFORCE_LT(axis, max_dim,
                    platform::errors::InvalidArgument(
                        "Axis should be less than %d, but received axis is %d.",
                        max_dim, axis));
2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028

  int pre, n, post, is_run_common_broadcast, axis_trim = 0;
  if (is_xsize_larger) {
    auto y_dims_trimed = trim_trailing_singular_dims(y_dims);
    axis_trim = (y_dims_trimed.size() == 0) ? x_dims.size() : axis;
    get_mid_dims(x_dims, y_dims_trimed, axis_trim, &pre, &n, &post,
                 &is_run_common_broadcast);
  } else {
    auto x_dims_trimed = trim_trailing_singular_dims(x_dims);
    axis_trim = (x_dims_trimed.size() == 0) ? y_dims.size() : axis;
    get_mid_dims(y_dims, x_dims_trimed, axis_trim, &pre, &n, &post,
                 &is_run_common_broadcast);
  }
  // special case for common implementation.
  // case 1: x=[2,3,1,5], y=[2,1,4,1]
  // case 2: x=[2,3,4], y=[1,1,4]
  if (is_run_common_broadcast == 1) {
    CommonElementwiseBroadcastForward<Functor, DeviceContext, T, OutType>(
        ctx, x, y, z, x_dims, y_dims, func, axis, is_xsize_larger);
2029 2030
    return;
  }
2031

2032 2033 2034 2035 2036 2037 2038 2039 2040
  if (platform::is_gpu_place(ctx.GetPlace())) {
#if defined(__NVCC__) || defined(__HIPCC__)
    ComputeElementwiseCUDA<Functor, T, OutType>(
        x, y, z, pre, n, post,
        ctx.template device_context<platform::CUDADeviceContext>(), func,
        is_xsize_larger);
#endif
    return;
  }
F
fengjiayi 已提交
2041 2042 2043 2044 2045 2046 2047 2048 2049
  if (post == 1) {
    functor.RunRowWise(n, pre);
    return;
  } else {
    functor.RunMidWise(n, pre, post);
    return;
  }
}

2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156
// FusedElemwiseAndAct
// --- forward
template <typename T, typename CompoundFunctor, bool KeepIntermediateOut>
struct FusedElemwiseAndActNoBroadcast {
  HOSTDEVICE void operator()(size_t i) {
    T y_val = y_[i];
    T x_val = x_[i];
    if (KeepIntermediateOut) {
      T intermeidiate_out = compound_functor_.GetIntermediateOut(x_val, y_val);
      intermediate_out_[i] = intermeidiate_out;
      out_[i] =
          compound_functor_.GetOutUseIntermediateOut(x_val, intermeidiate_out);
    } else {
      out_[i] = compound_functor_.GetOut(x_val, y_val);
    }
  }

  const T *x_;
  const T *y_;
  CompoundFunctor compound_functor_;
  T *out_;
  T *intermediate_out_;
};

// FusedElemwiseAndActBroadcast1:
// In this case, X and Y can be reshaped to a matrix.
// For example shape(X) = (2, 3, 4, 5), shape(Y) = (4, 5) and axis = -1 or 2,
// X can be reshaped to (6, 20) and Y can be reshaped to (1, 20)
template <typename T, typename CompoundFunctor, bool BcastY,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
static void FusedElemwiseAndActBroadcast1CPU(const T *x, const T *y,
                                             CompoundFunctor compound_functor,
                                             int h, int w, T *out,
                                             T *intermediate_out) {
  for (int i = 0; i < h; ++i) {
    for (int j = 0; j < w; ++j) {
      int offset = i * w + j;

      T y_val = BcastY ? y[j] : y[offset];
      T x_val = BcastY ? x[offset] : x[j];
      int64_t intermediate_out_offset;
      if (KeepIntermediateOut) {
        T intermeidiate_out = compound_functor.GetIntermediateOut(x_val, y_val);

        if (SameShapeOfIntermediateOutAndOut) {
          // for the case of f1(f2(x, y))
          intermediate_out_offset = offset;
        } else if (BcastY) {
          intermediate_out_offset = j;
        } else {
          intermediate_out_offset = offset;
        }

        intermediate_out[intermediate_out_offset] = intermeidiate_out;
        out[offset] =
            compound_functor.GetOutUseIntermediateOut(x_val, intermeidiate_out);
      } else {
        out[offset] = compound_functor.GetOut(x_val, y_val);
      }
    }
  }
}

// FusedElemwiseAndActBroadcast2
// In this case, X and Y can be reshaped to a matrix.
// For example shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4) and axis = 1,
// X can be reshaped to (2, 12, 5) and Y can be reshaped to (1, 12, 1)
// pre = 2, n = 12, post = 5
template <typename T, typename CompoundFunctor, bool BcastY,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
static void FusedElemwiseAndActBroadcast2CPU(const T *x, const T *y, int pre,
                                             int n, int post,
                                             CompoundFunctor compound_functor,
                                             T *out, T *intermediate_out) {
  for (int i = 0; i < pre; ++i) {
    for (int j = 0; j < n; ++j) {
      for (int k = 0; k < post; ++k) {
        int offset = i * n * post + j * post + k;

        T y_val = BcastY ? y[j] : y[offset];
        T x_val = BcastY ? x[offset] : x[j];
        int64_t intermediate_out_offset;

        if (KeepIntermediateOut) {
          T intermeidiate_out =
              compound_functor.GetIntermediateOut(x_val, y_val);

          if (SameShapeOfIntermediateOutAndOut) {
            // for the case of f1(f2(x, y))
            intermediate_out_offset = offset;
          } else if (BcastY) {
            intermediate_out_offset = j;
          } else {
            intermediate_out_offset = offset;
          }

          intermediate_out[intermediate_out_offset] = intermeidiate_out;
          out[offset] = compound_functor.GetOutUseIntermediateOut(
              x_val, intermeidiate_out);
        } else {
          out[offset] = compound_functor.GetOut(x_val, y_val);
        }
      }
    }
  }
}

2157
#if defined(__NVCC__) || defined(__HIPCC__)
2158 2159 2160 2161 2162
template <typename T, typename CompoundFunctor, bool BcastY,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
static __global__ void FusedElemwiseAndActBroadcast1CUDAKernel(
    const T *x, const T *y, int h, int w, CompoundFunctor compound_functor,
    T *out, T *intermediate_out) {
2163 2164
  int i = blockIdx.x;
  int j = threadIdx.x;
2165

2166
  while (j < w) {
2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191
    int offset = i * w + j;

    T y_val = BcastY ? y[j] : y[offset];
    T x_val = BcastY ? x[offset] : x[j];
    int64_t intermediate_out_offset;

    if (KeepIntermediateOut) {
      T intermeidiate_out = compound_functor.GetIntermediateOut(x_val, y_val);

      if (SameShapeOfIntermediateOutAndOut) {
        // for the case of f1(f2(x, y))
        intermediate_out_offset = offset;
      } else if (BcastY) {
        intermediate_out_offset = j;
      } else {
        intermediate_out_offset = offset;
      }

      intermediate_out[intermediate_out_offset] = intermeidiate_out;
      out[offset] =
          compound_functor.GetOutUseIntermediateOut(x_val, intermeidiate_out);
    } else {
      out[offset] = compound_functor.GetOut(x_val, y_val);
    }

2192
    j += ELEMWISE_MAX_BLOCK_DIM;
2193 2194 2195 2196 2197
  }
}

template <typename T, typename CompoundFunctor, bool BcastY,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
2198
static void FusedElemwiseAndActBroadcast1CUDA(gpuStream_t stream, const T *x,
2199 2200 2201 2202
                                              const T *y,
                                              CompoundFunctor compound_functor,
                                              int h, int w, T *out,
                                              T *intermediate_out) {
2203 2204
  int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, w);
  int gird_size = h;
2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254
  FusedElemwiseAndActBroadcast1CUDAKernel<
      T, CompoundFunctor, BcastY, KeepIntermediateOut,
      SameShapeOfIntermediateOutAndOut><<<gird_size, block_size, 0, stream>>>(
      x, y, h, w, compound_functor, out, intermediate_out);
}

template <typename T, typename CompoundFunctor, bool BcastY,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
static __global__ void FusedElemwiseAndActBroadcast2CUDAKernel(
    const T *x, const T *y, CompoundFunctor compound_functor, int pre, int n,
    int post, T *out, T *intermediate_out) {
  int tid = threadIdx.x;
  int j = blockIdx.x;

  while (true) {
    int i = tid / post;
    int k = tid % post;
    if (i >= pre) break;

    int offset = i * n * post + j * post + k;

    T y_val = BcastY ? y[j] : y[offset];
    T x_val = BcastY ? x[offset] : x[j];
    int64_t intermediate_out_offset;

    if (KeepIntermediateOut) {
      T intermeidiate_out = compound_functor.GetIntermediateOut(x_val, y_val);

      if (SameShapeOfIntermediateOutAndOut) {
        // for the case of f1(f2(x, y))
        intermediate_out_offset = offset;
      } else if (BcastY) {
        intermediate_out_offset = j;
      } else {
        intermediate_out_offset = offset;
      }

      intermediate_out[intermediate_out_offset] = intermeidiate_out;
      out[offset] =
          compound_functor.GetOutUseIntermediateOut(x_val, intermeidiate_out);
    } else {
      out[offset] = compound_functor.GetOut(x_val, y_val);
    }

    tid += ELEMWISE_MAX_BLOCK_DIM;
  }
}

template <typename T, typename CompoundFunctor, bool BcastY,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
2255
static void FusedElemwiseAndActBroadcast2CUDA(gpuStream_t stream, const T *x,
2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303
                                              const T *y, int pre, int n,
                                              int post,
                                              CompoundFunctor compound_functor,
                                              T *out, T *intermediate_out) {
  int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, pre * post);
  int gird_size = n;

  FusedElemwiseAndActBroadcast2CUDAKernel<
      T, CompoundFunctor, BcastY, KeepIntermediateOut,
      SameShapeOfIntermediateOutAndOut><<<gird_size, block_size, 0, stream>>>(
      x, y, compound_functor, pre, n, post, out, intermediate_out);
}

#endif

template <typename DeviceContext, typename T, typename CompoundFunctor,
          bool KeepIntermediateOut>
void FusedElemwiseAndActComputeNoBroadcast(
    const framework::ExecutionContext &ctx, const framework::DDim &x_dim,
    const framework::Tensor &x, const framework::Tensor &y,
    CompoundFunctor compound_functor, framework::Tensor *out,
    framework::Tensor *intermediate_out) {
  size_t N = static_cast<size_t>(framework::product(x_dim));

  platform::ForRange<DeviceContext> for_range(
      ctx.template device_context<DeviceContext>(), N);

  for_range(
      FusedElemwiseAndActNoBroadcast<T, CompoundFunctor, KeepIntermediateOut>{
          x.data<T>(), y.data<T>(), compound_functor,
          out->mutable_data<T>(ctx.GetPlace()),
          intermediate_out == nullptr
              ? nullptr
              : intermediate_out->mutable_data<T>(ctx.GetPlace())});
}

template <typename DeviceContext, typename T, typename CompoundFunctor,
          bool BcastY, bool KeepIntermediateOut,
          bool SameShapeOfIntermediateOutAndOut>
void FusedElemwiseAndActComputeWithBroadcast(
    const framework::ExecutionContext &ctx, const framework::DDim &x_dim,
    const framework::DDim &y_dim_untrimed, const framework::Tensor &x,
    const framework::Tensor &y, CompoundFunctor compound_functor, int axis,
    framework::Tensor *out, framework::Tensor *intermediate_out) {
  axis = (axis == -1 ? x_dim.size() - y_dim_untrimed.size() : axis);
  auto y_dim = trim_trailing_singular_dims(y_dim_untrimed);
  axis = (y_dim.size() == 0) ? x_dim.size() : axis;

2304 2305
  int pre, n, post, is_run_common_broadcast;
  get_mid_dims(x_dim, y_dim, axis, &pre, &n, &post, &is_run_common_broadcast);
2306 2307 2308 2309
  if (post == 1) {
    int h = pre;
    int w = n;
    if (platform::is_gpu_place(ctx.GetPlace())) {
2310
#if defined(__NVCC__) || defined(__HIPCC__)
2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332
      FusedElemwiseAndActBroadcast1CUDA<T, CompoundFunctor, BcastY,
                                        KeepIntermediateOut,
                                        SameShapeOfIntermediateOutAndOut>(
          ctx.template device_context<DeviceContext>().stream(), x.data<T>(),
          y.data<T>(), compound_functor, h, w,
          out->mutable_data<T>(ctx.GetPlace()),
          intermediate_out == nullptr
              ? nullptr
              : intermediate_out->mutable_data<T>(ctx.GetPlace()));
#endif
    } else {
      FusedElemwiseAndActBroadcast1CPU<T, CompoundFunctor, BcastY,
                                       KeepIntermediateOut,
                                       SameShapeOfIntermediateOutAndOut>(
          x.data<T>(), y.data<T>(), compound_functor, h, w,
          out->mutable_data<T>(ctx.GetPlace()),
          intermediate_out == nullptr
              ? nullptr
              : intermediate_out->mutable_data<T>(ctx.GetPlace()));
    }
  } else {
    if (platform::is_gpu_place(ctx.GetPlace())) {
2333
#if defined(__NVCC__) || defined(__HIPCC__)
2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357
      FusedElemwiseAndActBroadcast2CUDA<T, CompoundFunctor, BcastY,
                                        KeepIntermediateOut,
                                        SameShapeOfIntermediateOutAndOut>(
          ctx.template device_context<DeviceContext>().stream(), x.data<T>(),
          y.data<T>(), pre, n, post, compound_functor,
          out->mutable_data<T>(ctx.GetPlace()),
          intermediate_out == nullptr
              ? nullptr
              : intermediate_out->mutable_data<T>(ctx.GetPlace()));
#endif
    } else {
      FusedElemwiseAndActBroadcast2CPU<T, CompoundFunctor, BcastY,
                                       KeepIntermediateOut,
                                       SameShapeOfIntermediateOutAndOut>(
          x.data<T>(), y.data<T>(), pre, n, post, compound_functor,
          out->mutable_data<T>(ctx.GetPlace()),
          intermediate_out == nullptr
              ? nullptr
              : intermediate_out->mutable_data<T>(ctx.GetPlace()));
    }
  }
}

// --- backward
C
chengduo 已提交
2358 2359
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut>
2360 2361
struct FusedElemwiseAndActGradNoBroadcast {
  HOSTDEVICE void operator()(size_t i) {
2362 2363 2364
    T zero = static_cast<T>(0);
    T x_val = (x_ == nullptr) ? zero : x_[i];
    T y_val = (y_ == nullptr) ? zero : y_[i];
2365 2366 2367 2368 2369
    T out_val = out_[i];
    T dout_val = dout_[i];
    T intermediate_out_val = UseIntermediateOut
                                 ? intermediate_out_[i]
                                 : dx_op_.GetIntermediateOut(x_val, y_val);
2370
    if (dx_ != nullptr) {
2371 2372
      dx_[i] = dx_op_.UseIntermediateOut(x_val, y_val, intermediate_out_val,
                                         out_val, dout_val);
2373 2374
    }
    if (dy_ != nullptr) {
2375 2376
      dy_[i] = dy_op_.UseIntermediateOut(x_val, y_val, intermediate_out_val,
                                         out_val, dout_val);
C
chengduo 已提交
2377 2378
    }
    if (dintermediate_ != nullptr) {
2379 2380
      dintermediate_[i] = dintermediate_op_.UseIntermediateOut(
          x_val, intermediate_out_val, out_val, dout_val);
2381 2382 2383 2384 2385 2386 2387 2388 2389 2390
    }
  }

  const T *x_;
  const T *y_;
  const T *intermediate_out_;
  const T *out_;
  const T *dout_;
  DX_OP dx_op_;
  DY_OP dy_op_;
C
chengduo 已提交
2391
  DIntermediate_OP dintermediate_op_;
2392 2393
  T *dx_;
  T *dy_;
C
chengduo 已提交
2394
  T *dintermediate_;
2395 2396 2397
};

template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP,
C
chengduo 已提交
2398
          typename DIntermediate_OP, bool UseIntermediateOut>
2399 2400 2401 2402 2403
void FusedElemwiseAndActGradComputeNoBroadcast(
    const framework::ExecutionContext &ctx, const framework::DDim &x_dim,
    const framework::DDim &y_dim, const framework::Tensor *x,
    const framework::Tensor *y, const framework::Tensor *intermediate_out,
    const framework::Tensor *out, const framework::Tensor *dout, int axis,
C
chengduo 已提交
2404 2405 2406
    framework::Tensor *dx, framework::Tensor *dy,
    framework::Tensor *dintermediate, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op) {
2407 2408 2409
  size_t N = static_cast<size_t>(framework::product(x_dim));
  platform::ForRange<DeviceContext> for_range(
      ctx.template device_context<DeviceContext>(), N);
2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
  const T *x_data = nullptr;
  const T *y_data = nullptr;
  if (x->IsInitialized()) x_data = x->data<T>();
  if (y->IsInitialized()) y_data = y->data<T>();

  for_range(FusedElemwiseAndActGradNoBroadcast<
            T, DX_OP, DY_OP, DIntermediate_OP, UseIntermediateOut>{
      x_data, y_data, intermediate_out ? intermediate_out->data<T>() : nullptr,
      out->data<T>(), dout->data<T>(), dx_op, dy_op, dintermediate_op,
      dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
      dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()),
      dintermediate == nullptr ? nullptr : dintermediate->mutable_data<T>(
                                               ctx.GetPlace())});
2423 2424
}

C
chengduo 已提交
2425 2426 2427 2428 2429 2430 2431
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut, bool BcastY,
          bool SameShapeOfIntermediateOutAndOut>
static void FusedElemwiseAndActGradBroadcast1CPU(
    const T *x, const T *y, const T *intermediate_out, const T *out,
    const T *dout, int h, int w, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op, T *dx, T *dy, T *d_intermediate) {
2432
  int64_t tmp_out_idx, x_idx, y_idx;
2433
  T zero = static_cast<T>(0);
2434 2435 2436 2437 2438 2439 2440
  for (int i = 0; i < h; ++i) {
    for (int j = 0; j < w; ++j) {
      int offset = i * w + j;

      tmp_out_idx = BcastY ? j : offset;
      y_idx = BcastY ? j : offset;
      x_idx = BcastY ? offset : j;
2441 2442
      T x_val = (x == nullptr) ? zero : x[x_idx];
      T y_val = (y == nullptr) ? zero : y[y_idx];
2443 2444 2445 2446 2447 2448 2449

      if (SameShapeOfIntermediateOutAndOut) {
        tmp_out_idx = offset;
      }

      if (dx != nullptr) {
        T tmp = UseIntermediateOut
2450
                    ? dx_op.UseIntermediateOut(x_val, y_val,
C
chengduo 已提交
2451 2452
                                               intermediate_out[tmp_out_idx],
                                               out[offset], dout[offset])
2453
                    : dx_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466

        if (BcastY) {
          dx[x_idx] = tmp;
        } else {
          if (i == 0) {
            dx[x_idx] = tmp;
          } else {
            dx[x_idx] += tmp;
          }
        }
      }
      if (dy != nullptr) {
        T tmp = UseIntermediateOut
2467
                    ? dy_op.UseIntermediateOut(x_val, y_val,
C
chengduo 已提交
2468 2469
                                               intermediate_out[tmp_out_idx],
                                               out[offset], dout[offset])
2470
                    : dy_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2471 2472 2473 2474 2475 2476 2477 2478 2479 2480
        if (BcastY) {
          if (i == 0) {
            dy[y_idx] = tmp;
          } else {
            dy[y_idx] += tmp;
          }
        } else {
          dy[y_idx] = tmp;
        }
      }
C
chengduo 已提交
2481 2482 2483
      if (d_intermediate != nullptr) {
        T tmp = UseIntermediateOut
                    ? dintermediate_op.UseIntermediateOut(
2484
                          x_val, intermediate_out[tmp_out_idx], out[offset],
C
chengduo 已提交
2485
                          dout[offset])
2486 2487
                    : dintermediate_op.Recompute(x_val, y_val, out[offset],
                                                 dout[i]);
C
chengduo 已提交
2488 2489 2490 2491 2492 2493 2494 2495 2496 2497
        if (SameShapeOfIntermediateOutAndOut) {
          d_intermediate[tmp_out_idx] = tmp;
        } else {
          if (i == 0) {
            d_intermediate[tmp_out_idx] = tmp;
          } else {
            d_intermediate[tmp_out_idx] += tmp;
          }
        }
      }
2498 2499 2500 2501
    }
  }
}

C
chengduo 已提交
2502 2503 2504 2505 2506 2507 2508
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut, bool BcastY,
          bool SameShapeOfIntermediateOutAndOut>
static void FusedElemwiseAndActGradBroadcast2CPU(
    const T *x, const T *y, const T *intermediate_out, const T *out,
    const T *dout, int pre, int n, int post, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op, T *dx, T *dy, T *d_intermediate) {
2509
  int64_t tmp_out_idx, x_idx, y_idx;
2510
  T zero = static_cast<T>(0);
2511 2512 2513 2514 2515 2516 2517 2518 2519
  for (int i = 0; i < pre; ++i) {
    for (int j = 0; j < n; ++j) {
      for (int k = 0; k < post; ++k) {
        int offset = i * n * post + j * post + k;

        tmp_out_idx = BcastY ? j : offset;
        y_idx = BcastY ? j : offset;
        x_idx = BcastY ? offset : j;

2520 2521 2522
        T x_val = (x == nullptr) ? zero : x[x_idx];
        T y_val = (y == nullptr) ? zero : y[y_idx];

2523 2524 2525 2526 2527
        if (SameShapeOfIntermediateOutAndOut) {
          tmp_out_idx = offset;
        }

        if (dx != nullptr) {
2528 2529 2530 2531 2532 2533
          T tmp =
              UseIntermediateOut
                  ? dx_op.UseIntermediateOut(x_val, y_val,
                                             intermediate_out[tmp_out_idx],
                                             out[offset], dout[offset])
                  : dx_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545

          if (BcastY) {
            dx[x_idx] = tmp;
          } else {
            if (i == 0 && k == 0) {
              dx[x_idx] = tmp;
            } else {
              dx[x_idx] += tmp;
            }
          }
        }
        if (dy != nullptr) {
2546 2547 2548 2549 2550 2551
          T tmp =
              UseIntermediateOut
                  ? dy_op.UseIntermediateOut(x_val, y_val,
                                             intermediate_out[tmp_out_idx],
                                             out[offset], dout[offset])
                  : dy_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2552 2553 2554 2555 2556 2557 2558 2559 2560 2561
          if (BcastY) {
            if (i == 0 && k == 0) {
              dy[y_idx] = tmp;
            } else {
              dy[y_idx] += tmp;
            }
          } else {
            dy[y_idx] = tmp;
          }
        }
C
chengduo 已提交
2562 2563 2564
        if (d_intermediate != nullptr) {
          T tmp = UseIntermediateOut
                      ? dintermediate_op.UseIntermediateOut(
2565 2566 2567 2568
                            x_val, intermediate_out[tmp_out_idx], out[offset],
                            dout[offset])
                      : dintermediate_op.Recompute(x_val, y_val, out[offset],
                                                   dout[i]);
C
chengduo 已提交
2569 2570 2571 2572 2573 2574 2575 2576 2577 2578
          if (SameShapeOfIntermediateOutAndOut) {
            d_intermediate[tmp_out_idx] = tmp;
          } else {
            if (i == 0) {
              d_intermediate[tmp_out_idx] = tmp;
            } else {
              d_intermediate[tmp_out_idx] += tmp;
            }
          }
        }
2579 2580 2581 2582 2583
      }
    }
  }
}

2584
#if defined(__NVCC__) || defined(__HIPCC__)
C
chengduo 已提交
2585 2586 2587
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut, bool BcastY,
          bool SameShapeOfIntermediateOutAndOut>
2588 2589
static __global__ void FusedElemwiseAndActGradBroadcast1CUDAKernel(
    const T *x, const T *y, const T *intermediate_out, const T *out,
C
chengduo 已提交
2590 2591
    const T *dout, int h, int w, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op, T *dx, T *dy, T *d_intermediate) {
2592 2593 2594 2595 2596 2597
  __shared__ T sdata[BLOCK_Y][BLOCK_X];
  size_t idx = threadIdx.x + BLOCK_X * blockIdx.x;
  size_t width_stride = gridDim.x * BLOCK_X;

  size_t full_w = ROUNDUP(w, BLOCK_X);

2598
  T zero = static_cast<T>(0);
2599

2600 2601 2602 2603 2604
  for (size_t j = idx; j < full_w; j += width_stride) {
    T val(0), inter_val(0);
    if (j < w) {
      for (size_t i = threadIdx.y; i < h; i += BLOCK_Y) {
        size_t offset = i * w + j;
2605

2606 2607 2608 2609 2610
        size_t tmp_out_idx = BcastY ? j : offset;
        size_t y_idx = BcastY ? j : offset;
        size_t x_idx = BcastY ? offset : j;
        T x_val = (x == nullptr) ? zero : x[x_idx];
        T y_val = (y == nullptr) ? zero : y[y_idx];
2611

2612 2613 2614
        if (SameShapeOfIntermediateOutAndOut) {
          tmp_out_idx = offset;
        }
2615

2616 2617 2618
        if (dx != nullptr) {
          T tmp =
              UseIntermediateOut
2619 2620 2621 2622
                  ? dx_op.UseIntermediateOut(x_val, y_val,
                                             intermediate_out[tmp_out_idx],
                                             out[offset], dout[offset])
                  : dx_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2623

2624 2625 2626 2627 2628 2629 2630 2631 2632
          if (BcastY) {
            dx[x_idx] = tmp;
          } else {
            val += tmp;
          }
        }
        if (dy != nullptr) {
          T tmp =
              UseIntermediateOut
2633 2634 2635 2636
                  ? dy_op.UseIntermediateOut(x_val, y_val,
                                             intermediate_out[tmp_out_idx],
                                             out[offset], dout[offset])
                  : dy_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655
          if (BcastY) {
            val += tmp;
          } else {
            dy[y_idx] = tmp;
          }
        }
        if (d_intermediate != nullptr) {
          T tmp = UseIntermediateOut
                      ? dintermediate_op.UseIntermediateOut(
                            y[y_idx], intermediate_out[tmp_out_idx],
                            out[offset], dout[offset])
                      : dintermediate_op.Recompute(x_val, y_val, out[offset],
                                                   dout[offset]);
          if (SameShapeOfIntermediateOutAndOut) {
            d_intermediate[tmp_out_idx] = tmp;
          } else {
            inter_val += tmp;
          }
        }
C
chengduo 已提交
2656 2657
      }
    }
2658

2659 2660 2661 2662 2663 2664 2665 2666 2667
    // transpose, for ReduceSum with wrap
    sdata[threadIdx.y][threadIdx.x] = val;
    __syncthreads();
    val = sdata[threadIdx.x][threadIdx.y];
#pragma unroll
    for (int i = BLOCK_X >> 1; i > 0; i >>= 1) {
      // reduce sum with wrap
      val += platform::CudaShuffleXorSync(0xFFFFFFFF, val, i);
    }
2668

2669 2670 2671 2672
    size_t idx_j = j + threadIdx.y;
    if (BcastY) {
      if (dy) {
        if (threadIdx.x == 0 && (idx_j < w)) dy[idx_j] = val;
2673
      }
2674 2675 2676
    } else {
      if (dx) {
        if (threadIdx.x == 0 && (idx_j < w)) dx[idx_j] = val;
2677 2678
      }
    }
2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690

    if (!SameShapeOfIntermediateOutAndOut) {
      if (d_intermediate) {
        sdata[threadIdx.y][threadIdx.x] = inter_val;
        __syncthreads();
        inter_val = sdata[threadIdx.x][threadIdx.y];
#pragma unroll
        for (int i = BLOCK_X >> 1; i > 0; i >>= 1) {
          // reduce sum with wrap
          inter_val += platform::CudaShuffleXorSync(0xFFFFFFFF, inter_val, i);
        }
        if (threadIdx.x == 0 && (idx_j < w)) d_intermediate[idx_j] = inter_val;
C
chengduo 已提交
2691 2692
      }
    }
2693
  }  // end for
2694 2695
}

C
chengduo 已提交
2696 2697 2698 2699
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut, bool BcastY,
          bool SameShapeOfIntermediateOutAndOut>
static void FusedElemwiseAndActGradBroadcast1CUDA(
2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711
    const framework::ExecutionContext &ctx, const T *x, const T *y,
    const T *intermediate_out, const T *out, const T *dout, int h, int w,
    DX_OP dx_op, DY_OP dy_op, DIntermediate_OP dintermediate_op, T *dx, T *dy,
    T *d_intermediate) {
  gpuStream_t stream = ctx.cuda_device_context().stream();

  dim3 blocks(BLOCK_X, BLOCK_Y);
  int max_gpu_threads = ctx.cuda_device_context().GetMaxPhysicalThreadCount();
  int max_blocks = std::max(max_gpu_threads / (BLOCK_X * BLOCK_Y), 1);
  int theory_block = (w + BLOCK_X - 1) / BLOCK_X;
  dim3 grids(std::min(theory_block, max_blocks));

2712
  FusedElemwiseAndActGradBroadcast1CUDAKernel<
C
chengduo 已提交
2713
      T, DX_OP, DY_OP, DIntermediate_OP, UseIntermediateOut, BcastY,
2714
      SameShapeOfIntermediateOutAndOut><<<grids, blocks, 0, stream>>>(
C
chengduo 已提交
2715 2716
      x, y, intermediate_out, out, dout, h, w, dx_op, dy_op, dintermediate_op,
      dx, dy, d_intermediate);
2717 2718
}

C
chengduo 已提交
2719 2720 2721
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut, bool BcastY,
          bool SameShapeOfIntermediateOutAndOut>
2722 2723
static __global__ void FusedElemwiseAndActGradBroadcast2CUDAKernel(
    const T *x, const T *y, const T *intermediate_out, const T *out,
C
chengduo 已提交
2724 2725
    const T *dout, int pre, int n, int post, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op, T *dx, T *dy, T *d_intermediate) {
2726 2727 2728
  int tid = threadIdx.x;
  int j = blockIdx.x;

C
chengduo 已提交
2729
  T val(0), inter_val(0);
2730 2731
  int ttid = tid;
  int64_t tmp_out_idx, x_idx, y_idx;
2732
  T zero = static_cast<T>(0);
2733 2734 2735 2736 2737 2738 2739 2740 2741 2742
  while (true) {
    int i = ttid / post;
    int k = ttid % post;
    if (i >= pre) break;

    int offset = i * n * post + j * post + k;

    tmp_out_idx = BcastY ? j : offset;
    y_idx = BcastY ? j : offset;
    x_idx = BcastY ? offset : j;
2743 2744
    T x_val = (x == nullptr) ? zero : x[x_idx];
    T y_val = (y == nullptr) ? zero : y[y_idx];
2745 2746 2747 2748 2749 2750

    if (SameShapeOfIntermediateOutAndOut) {
      tmp_out_idx = offset;
    }

    if (dx != nullptr) {
2751 2752 2753 2754 2755
      T tmp = UseIntermediateOut
                  ? dx_op.UseIntermediateOut(x_val, y_val,
                                             intermediate_out[tmp_out_idx],
                                             out[offset], dout[offset])
                  : dx_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2756 2757 2758 2759 2760 2761 2762 2763

      if (BcastY) {
        dx[x_idx] = tmp;
      } else {
        val += tmp;
      }
    }
    if (dy != nullptr) {
2764 2765 2766 2767 2768
      T tmp = UseIntermediateOut
                  ? dy_op.UseIntermediateOut(x_val, y_val,
                                             intermediate_out[tmp_out_idx],
                                             out[offset], dout[offset])
                  : dy_op.Recompute(x_val, y_val, out[offset], dout[offset]);
2769 2770 2771 2772 2773 2774
      if (BcastY) {
        val += tmp;
      } else {
        dy[y_idx] = tmp;
      }
    }
C
chengduo 已提交
2775 2776 2777
    if (d_intermediate != nullptr) {
      T tmp = UseIntermediateOut
                  ? dintermediate_op.UseIntermediateOut(
2778
                        y_val, intermediate_out[tmp_out_idx], out[offset],
C
chengduo 已提交
2779
                        dout[offset])
2780
                  : dintermediate_op.Recompute(x_val, y_val, out[offset],
C
chengduo 已提交
2781 2782 2783 2784 2785 2786 2787
                                               dout[offset]);
      if (SameShapeOfIntermediateOutAndOut) {
        d_intermediate[tmp_out_idx] = tmp;
      } else {
        inter_val += tmp;
      }
    }
2788 2789 2790
    ttid += ELEMWISE_MAX_BLOCK_DIM;
  }

C
chengduo 已提交
2791 2792
  int h = pre * post;
  h = h > ELEMWISE_MAX_BLOCK_DIM ? ELEMWISE_MAX_BLOCK_DIM : h;
2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807
  if (BcastY) {
    if (dy) {
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dy[j] = val;
      }
    }
  } else {
    if (dx) {
      val = paddle::platform::reduceSum(val, tid, h);
      if (threadIdx.x == 0) {
        dx[j] = val;
      }
    }
  }
C
chengduo 已提交
2808 2809 2810 2811 2812 2813 2814 2815
  if (!SameShapeOfIntermediateOutAndOut) {
    if (d_intermediate) {
      inter_val = paddle::platform::reduceSum(inter_val, tid, h);
      if (threadIdx.x == 0) {
        d_intermediate[j] = inter_val;
      }
    }
  }
2816 2817
}

C
chengduo 已提交
2818 2819 2820
template <typename T, typename DX_OP, typename DY_OP, typename DIntermediate_OP,
          bool UseIntermediateOut, bool BcastY,
          bool SameShapeOfIntermediateOutAndOut>
2821
static void FusedElemwiseAndActGradBroadcast2CUDA(
2822
    gpuStream_t stream, const T *x, const T *y, const T *intermediate_out,
2823
    const T *out, const T *dout, int pre, int n, int post, DX_OP dx_op,
C
chengduo 已提交
2824 2825
    DY_OP dy_op, DIntermediate_OP dintermediate_op, T *dx, T *dy,
    T *dintermediate) {
2826 2827 2828
  int block_size = std::min(ELEMWISE_MAX_BLOCK_DIM, pre * post);
  int gird_size = n;
  FusedElemwiseAndActGradBroadcast2CUDAKernel<
C
chengduo 已提交
2829
      T, DX_OP, DY_OP, DIntermediate_OP, UseIntermediateOut, BcastY,
2830
      SameShapeOfIntermediateOutAndOut><<<gird_size, block_size, 0, stream>>>(
C
chengduo 已提交
2831 2832
      x, y, intermediate_out, out, dout, pre, n, post, dx_op, dy_op,
      dintermediate_op, dx, dy, dintermediate);
2833 2834 2835 2836
}
#endif

template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP,
C
chengduo 已提交
2837
          typename DIntermediate_OP, bool UseIntermediateOut, bool BcastY,
2838 2839 2840 2841 2842 2843
          bool SameShapeOfIntermediateOutAndOut>
void FusedElemwiseAndActGradComputeWithBroadcast(
    const framework::ExecutionContext &ctx, const framework::DDim &x_dim,
    const framework::DDim &y_dim_untrimed, const framework::Tensor *x,
    const framework::Tensor *y, const framework::Tensor *intermediate_out,
    const framework::Tensor *out, const framework::Tensor *dout, int axis,
C
chengduo 已提交
2844 2845 2846
    framework::Tensor *dx, framework::Tensor *dy,
    framework::Tensor *dintermediate, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op) {
2847 2848 2849 2850
  axis = (axis == -1 ? x_dim.size() - y_dim_untrimed.size() : axis);
  auto y_dim = trim_trailing_singular_dims(y_dim_untrimed);
  axis = (y_dim.size() == 0) ? x_dim.size() : axis;

2851 2852
  int pre, n, post, is_run_common_broadcast;
  get_mid_dims(x_dim, y_dim, axis, &pre, &n, &post, &is_run_common_broadcast);
2853 2854 2855 2856
  const T *x_data = nullptr;
  const T *y_data = nullptr;
  if (x->IsInitialized()) x_data = x->data<T>();
  if (y->IsInitialized()) y_data = y->data<T>();
2857 2858 2859
  if (post == 1) {
    int h = pre;
    int w = n;
2860

2861
    if (platform::is_gpu_place(ctx.GetPlace())) {
2862
#if defined(__NVCC__) || defined(__HIPCC__)
C
chengduo 已提交
2863 2864
      FusedElemwiseAndActGradBroadcast1CUDA<T, DX_OP, DY_OP, DIntermediate_OP,
                                            UseIntermediateOut, BcastY,
2865
                                            SameShapeOfIntermediateOutAndOut>(
2866
          ctx, x_data, y_data,
2867
          intermediate_out == nullptr ? nullptr : intermediate_out->data<T>(),
C
chengduo 已提交
2868
          out->data<T>(), dout->data<T>(), h, w, dx_op, dy_op, dintermediate_op,
2869
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
C
chengduo 已提交
2870 2871 2872
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()),
          dintermediate == nullptr ? nullptr : dintermediate->mutable_data<T>(
                                                   ctx.GetPlace()));
2873 2874
#endif
    } else {
C
chengduo 已提交
2875 2876
      FusedElemwiseAndActGradBroadcast1CPU<T, DX_OP, DY_OP, DIntermediate_OP,
                                           UseIntermediateOut, BcastY,
2877
                                           SameShapeOfIntermediateOutAndOut>(
2878
          x_data, y_data,
2879
          intermediate_out == nullptr ? nullptr : intermediate_out->data<T>(),
C
chengduo 已提交
2880
          out->data<T>(), dout->data<T>(), h, w, dx_op, dy_op, dintermediate_op,
2881
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
C
chengduo 已提交
2882 2883 2884
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()),
          dintermediate == nullptr ? nullptr : dintermediate->mutable_data<T>(
                                                   ctx.GetPlace()));
2885 2886 2887
    }
  } else {
    if (platform::is_gpu_place(ctx.GetPlace())) {
2888
#if defined(__NVCC__) || defined(__HIPCC__)
C
chengduo 已提交
2889 2890
      FusedElemwiseAndActGradBroadcast2CUDA<T, DX_OP, DY_OP, DIntermediate_OP,
                                            UseIntermediateOut, BcastY,
2891
                                            SameShapeOfIntermediateOutAndOut>(
2892
          ctx.template device_context<DeviceContext>().stream(), x_data, y_data,
2893 2894
          intermediate_out == nullptr ? nullptr : intermediate_out->data<T>(),
          out->data<T>(), dout->data<T>(), pre, n, post, dx_op, dy_op,
C
chengduo 已提交
2895
          dintermediate_op,
2896
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
C
chengduo 已提交
2897 2898 2899
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()),
          dintermediate == nullptr ? nullptr : dintermediate->mutable_data<T>(
                                                   ctx.GetPlace()));
2900 2901
#endif
    } else {
C
chengduo 已提交
2902 2903
      FusedElemwiseAndActGradBroadcast2CPU<T, DX_OP, DY_OP, DIntermediate_OP,
                                           UseIntermediateOut, BcastY,
2904
                                           SameShapeOfIntermediateOutAndOut>(
2905
          x_data, y_data,
2906 2907
          intermediate_out == nullptr ? nullptr : intermediate_out->data<T>(),
          out->data<T>(), dout->data<T>(), pre, n, post, dx_op, dy_op,
C
chengduo 已提交
2908
          dintermediate_op,
2909
          dx == nullptr ? nullptr : dx->mutable_data<T>(ctx.GetPlace()),
C
chengduo 已提交
2910 2911 2912
          dy == nullptr ? nullptr : dy->mutable_data<T>(ctx.GetPlace()),
          dintermediate == nullptr ? nullptr : dintermediate->mutable_data<T>(
                                                   ctx.GetPlace()));
2913 2914 2915 2916 2917
    }
  }
}

template <typename DeviceContext, typename T, typename DX_OP, typename DY_OP,
C
chengduo 已提交
2918 2919
          typename DIntermediate_OP, bool UseIntermediateOut,
          bool SameShapeOfIntermediateOutAndOut>
2920 2921 2922 2923
void FusedElemwiseAndActGradComputeEx(
    const framework::ExecutionContext &ctx, const framework::Tensor *x,
    const framework::Tensor *y, const framework::Tensor *out,
    const framework::Tensor *intermediate_out, const framework::Tensor *dout,
C
chengduo 已提交
2924 2925 2926
    int axis, framework::Tensor *dx, framework::Tensor *dy,
    framework::Tensor *dintermediate, DX_OP dx_op, DY_OP dy_op,
    DIntermediate_OP dintermediate_op) {
2927 2928 2929
  const framework::DDim &x_dim = x->dims();
  const framework::DDim &y_dim = y->dims();
  if (UseIntermediateOut) {
2930 2931 2932
    PADDLE_ENFORCE_NOT_NULL(
        intermediate_out,
        platform::errors::InvalidArgument("Intermediate out is null pointer."));
2933 2934
  }
  if (x_dim == y_dim) {
C
chengduo 已提交
2935 2936
    FusedElemwiseAndActGradComputeNoBroadcast<
        DeviceContext, T, DX_OP, DY_OP, DIntermediate_OP, UseIntermediateOut>(
2937
        ctx, x_dim, y_dim, x, y, intermediate_out, out, dout, axis, dx, dy,
C
chengduo 已提交
2938
        dintermediate, dx_op, dy_op, dintermediate_op);
2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953
  } else {  // Y is a scalar
    bool bcast_y = x_dim.size() >= y_dim.size();
    if (x_dim.size() == y_dim.size()) {
      for (int i = 0; i < x_dim.size(); ++i) {
        if (x_dim[i] < y_dim[i]) {
          bcast_y = false;
          break;
        }
      }
    }

    // z = f1(x, f2(y))
    // z = f1(f2(x, y))
    if (bcast_y) {  // Y should be broadcast.
      FusedElemwiseAndActGradComputeWithBroadcast<
C
chengduo 已提交
2954 2955 2956 2957
          DeviceContext, T, DX_OP, DY_OP, DIntermediate_OP, UseIntermediateOut,
          true /*BcastY*/, SameShapeOfIntermediateOutAndOut>(
          ctx, x_dim, y_dim, x, y, intermediate_out, out, dout, axis, dx, dy,
          dintermediate, dx_op, dy_op, dintermediate_op);
2958 2959
    } else {
      FusedElemwiseAndActGradComputeWithBroadcast<
C
chengduo 已提交
2960 2961 2962 2963
          DeviceContext, T, DX_OP, DY_OP, DIntermediate_OP, UseIntermediateOut,
          false /*BcastY*/, SameShapeOfIntermediateOutAndOut>(
          ctx, y_dim, x_dim, x, y, intermediate_out, out, dout, axis, dx, dy,
          dintermediate, dx_op, dy_op, dintermediate_op);
2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976
    }
  }
}

template <typename DeviceContext, typename T, typename CompoundFunctor,
          bool KeepIntermediateOut, bool SameShapeOfIntermediateOutAndOut>
void FusedElemwiseAndActComputeEx(const framework::ExecutionContext &ctx,
                                  const framework::Tensor &x,
                                  const framework::Tensor &y, int axis,
                                  CompoundFunctor compound_functor,
                                  framework::Tensor *out,
                                  framework::Tensor *intermediate_out) {
  if (KeepIntermediateOut) {
2977 2978 2979 2980 2981
    PADDLE_ENFORCE_NOT_NULL(
        intermediate_out,
        platform::errors::InvalidArgument(
            "The save_intermediate_out is opened, intermediate "
            "out is null pointer."));
2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992
  }

  const framework::DDim &x_dim = x.dims();
  const framework::DDim &y_dim = y.dims();
  if (x.dims() == y.dims()) {
    FusedElemwiseAndActComputeNoBroadcast<DeviceContext, T, CompoundFunctor,
                                          KeepIntermediateOut>(
        ctx, x_dim, x, y, compound_functor, out, intermediate_out);
  } else {
    // Whether the shape of Y is a continuous subsequence of X,
    // For more information please refer to the op's introduction.
2993
    bool bcast_y = x.numel() >= y.numel();
2994 2995 2996 2997
    // z = f1(x, f2(y))
    // z = f1(f2(x, y))
    if (bcast_y) {  // Y should be broadcast.
      // In this case,
2998 2999
      // for 'f2(y)', the shape of intermediate_out should be equal to the
      // shape
3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010
      // of Y.
      // for 'f2(x, y)', the shape of intermediate_out should be equal to the
      // shape of Out.
      // the shape of Out should be equal to the shape of X.
      FusedElemwiseAndActComputeWithBroadcast<
          DeviceContext, T, CompoundFunctor, true /*BcastY*/,
          KeepIntermediateOut, SameShapeOfIntermediateOutAndOut>(
          ctx, x_dim /*OutShape*/, y_dim, x, y, compound_functor, axis, out,
          intermediate_out);
    } else {
      // In this case,
3011 3012
      // for 'f2(y)', the shape of intermediate_out should be equal to the
      // shape
3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024
      // of Out.
      // for 'f2(x, y)', the shape of intermediate_out should be equal to the
      // shape of Out.
      // the shape of Out should be equal to the shape of Y.
      FusedElemwiseAndActComputeWithBroadcast<
          DeviceContext, T, CompoundFunctor, false /*BcastY*/,
          KeepIntermediateOut, SameShapeOfIntermediateOutAndOut>(
          ctx, y_dim /*OutShape*/, x_dim, x, y, compound_functor, axis, out,
          intermediate_out);
    }
  }
}
3025 3026 3027 3028 3029 3030 3031 3032

template <typename DeviceContext, typename T>
static inline void GetDoubleGradSafeTensor(
    const framework::ExecutionContext &ctx, const framework::Tensor *x,
    const framework::Tensor *ddx, framework::Tensor *ddx_safe) {
  if (ddx) {
    *ddx_safe = *ddx;
  } else {
3033 3034
    auto &dev_ctx = ctx.template device_context<DeviceContext>();
    *ddx_safe = ctx.AllocateTmpTensor<T, DeviceContext>(x->dims(), dev_ctx);
3035 3036 3037 3038 3039 3040
    math::SetConstant<DeviceContext, T> set_zero;
    set_zero(ctx.template device_context<DeviceContext>(), ddx_safe,
             static_cast<T>(0));
  }
}

3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060
// for broadcast backwards
static inline std::vector<int> GetReduceDim(const framework::DDim &in,
                                            const framework::DDim &out,
                                            int axis) {
  axis =
      (axis == -1 ? std::abs(static_cast<int>(out.size() - in.size())) : axis);
  std::vector<int> dims;
  for (int i = 0; i < axis; ++i) {
    dims.push_back(i);
  }
  for (int i = 0; i < in.size(); ++i) {
    if (out[i + axis] != in[i]) {
      dims.push_back(i + axis);
    }
  }
  for (int i = axis + in.size(); i < out.size(); ++i) {
    dims.push_back(i);
  }
  return dims;
}
3061 3062
}  // namespace operators
}  // namespace paddle