selected_rows_functor.cu 28.4 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
2 3 4 5 6 7 8 9 10 11 12 13 14

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

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

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */

T
typhoonzero 已提交
15
#include <set>
16
#include <vector>
T
typhoonzero 已提交
17

Y
Yi Wang 已提交
18
#include "paddle/fluid/operators/math/selected_rows_functor.h"
19
#include "paddle/fluid/platform/bfloat16.h"
20
#include "paddle/fluid/platform/device/gpu/gpu_primitives.h"
C
chengduo 已提交
21
#include "paddle/fluid/platform/float16.h"
22
#include "paddle/phi/kernels/funcs/math_function.h"
23 24 25 26 27

namespace paddle {
namespace operators {
namespace math {
template <typename T>
Q
QI JUN 已提交
28 29
struct SelectedRowsAdd<platform::CUDADeviceContext, T> {
  void operator()(const platform::CUDADeviceContext& context,
30 31
                  const phi::SelectedRows& input1,
                  const phi::SelectedRows& input2, phi::SelectedRows* output) {
32
    auto in1_height = input1.height();
33 34 35 36 37 38
    PADDLE_ENFORCE_EQ(
        in1_height, input2.height(),
        platform::errors::InvalidArgument("The two inputs height must be equal."
                                          "But recieved first input height  = "
                                          "[%d], second input height = [%d]",
                                          in1_height, input2.height()));
39 40
    output->set_height(in1_height);

D
dzhwinter 已提交
41
    framework::Vector<int64_t> in1_rows(input1.rows());
42 43 44 45 46 47 48 49 50 51 52 53 54 55
    auto& in2_rows = input2.rows();
    std::vector<int64_t> out_rows;
    out_rows.reserve(in1_rows.size() + in2_rows.size());

    // concat rows
    out_rows.insert(out_rows.end(), in1_rows.begin(), in1_rows.end());
    out_rows.insert(out_rows.end(), in2_rows.begin(), in2_rows.end());
    output->set_rows(out_rows);

    auto* out_value = output->mutable_value();
    auto& in1_value = input1.value();
    auto& in2_value = input2.value();

    auto in1_row_numel = in1_value.numel() / in1_rows.size();
56 57 58 59 60 61 62 63 64 65 66 67
    PADDLE_ENFORCE_EQ(
        in1_row_numel, in2_value.numel() / in2_rows.size(),
        platform::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But recieved first input width = [%d], second input width = [%d]",
            in1_row_numel, in2_value.numel() / in2_rows.size()));
    PADDLE_ENFORCE_EQ(
        in1_row_numel, out_value->numel() / out_rows.size(),
        platform::errors::InvalidArgument(
            "The input and oupput width must be equal."
            "But recieved input width = [%d], output width = [%d]",
            in1_row_numel, out_value->numel() / out_rows.size()));
68 69 70 71 72

    auto* out_data = out_value->data<T>();
    auto* in1_data = in1_value.data<T>();

    auto in1_place = input1.place();
73 74 75
    PADDLE_ENFORCE_EQ(platform::is_gpu_place(in1_place), true,
                      platform::errors::InvalidArgument(
                          "The running enviroment is not on the GPU place."));
76
    auto in2_place = input2.place();
77 78 79
    PADDLE_ENFORCE_EQ(platform::is_gpu_place(in2_place), true,
                      platform::errors::InvalidArgument(
                          "The running enviroment is not on the GPU place."));
80
    auto out_place = context.GetPlace();
81 82 83
    PADDLE_ENFORCE_EQ(platform::is_gpu_place(out_place), true,
                      platform::errors::InvalidArgument(
                          "The running enviroment is not on the GPU place."));
84

85
    memory::Copy(out_place, out_data, in1_place, in1_data,
86
                 in1_value.numel() * sizeof(T), context.stream());
87 88

    auto* in2_data = in2_value.data<T>();
89
    memory::Copy(out_place, out_data + in1_value.numel(), in2_place, in2_data,
Q
QI JUN 已提交
90
                 in2_value.numel() * sizeof(T), context.stream());
91 92 93
  }
};

Q
QI JUN 已提交
94 95
template struct SelectedRowsAdd<platform::CUDADeviceContext, float>;
template struct SelectedRowsAdd<platform::CUDADeviceContext, double>;
96 97

namespace {
Q
QI JUN 已提交
98
template <typename T, int block_size>
99 100
__global__ void SelectedRowsAddTensorKernel(const T* selected_rows,
                                            const int64_t* rows, T* tensor_out,
Q
QI JUN 已提交
101
                                            int64_t row_numel) {
C
chengduo 已提交
102
  const int ty = blockIdx.x;
103 104 105 106 107 108 109 110 111
  int tid = threadIdx.x;

  selected_rows += ty * row_numel;
  tensor_out += rows[ty] * row_numel;

  for (int index = tid; index < row_numel; index += block_size) {
    // Since index in rows of SelectedRows can be duplicate, we can not use
    // tensor_out[index] += selected_rows[index]; Instead, we have to use
    // AtomicAdd to avoid concurrent write error.
Q
qijun 已提交
112
    paddle::platform::CudaAtomicAdd(tensor_out + index, selected_rows[index]);
113 114 115 116 117
  }
}
}  // namespace

template <typename T>
Q
QI JUN 已提交
118 119
struct SelectedRowsAddTensor<platform::CUDADeviceContext, T> {
  void operator()(const platform::CUDADeviceContext& context,
120
                  const phi::SelectedRows& input1,
121 122 123 124
                  const framework::Tensor& input2, framework::Tensor* output) {
    auto in1_height = input1.height();
    auto in2_dims = input2.dims();
    auto out_dims = output->dims();
125 126 127 128 129 130 131 132 133 134 135 136
    PADDLE_ENFORCE_EQ(
        in1_height, in2_dims[0],
        platform::errors::InvalidArgument(
            "The two inputs height must be equal."
            "But recieved first input height = [%d], first input height = [%d]",
            in1_height, in2_dims[0]));
    PADDLE_ENFORCE_EQ(
        in1_height, out_dims[0],
        platform::errors::InvalidArgument(
            "The input and output height must be equal."
            "But recieved input height = [%d], output height = [%d]",
            in1_height, out_dims[0]));
137 138

    auto& in1_value = input1.value();
139
    auto& in1_rows = input1.rows();
140 141

    int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
142 143 144 145 146 147 148 149 150 151 152 153
    PADDLE_ENFORCE_EQ(
        in1_row_numel, input2.numel() / in1_height,
        platform::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But recieved first input width = [%d], second input width = [%d]",
            in1_row_numel, input2.numel() / in1_height));
    PADDLE_ENFORCE_EQ(
        in1_row_numel, output->numel() / in1_height,
        platform::errors::InvalidArgument(
            "The input and output width must be equal."
            "But recieved input width = [%d], output width = [%d]",
            in1_row_numel, output->numel() / in1_height));
154 155 156 157 158

    auto* in1_data = in1_value.data<T>();
    auto* in2_data = input2.data<T>();
    auto* out_data = output->data<T>();

159
    phi::funcs::SetConstant<platform::CUDADeviceContext, T> functor;
C
chengduo 已提交
160
    functor(context, output, static_cast<T>(0));
161

Q
QI JUN 已提交
162
    const int block_size = 256;
163
    dim3 threads(block_size, 1);
C
chengduo 已提交
164
    dim3 grid(in1_rows.size(), 1);
165
    paddle::framework::MixVector<int64_t> mixv_in1_rows(&in1_rows);
Q
QI JUN 已提交
166 167
    SelectedRowsAddTensorKernel<
        T, block_size><<<grid, threads, 0, context.stream()>>>(
168
        in1_data, mixv_in1_rows.CUDAData(context.GetPlace()), out_data,
Y
Yu Yang 已提交
169
        in1_row_numel);
170 171 172

    auto out_eigen = framework::EigenVector<T>::Flatten(*output);
    auto in2_eigen = framework::EigenVector<T>::Flatten(input2);
Q
QI JUN 已提交
173
    out_eigen.device(*context.eigen_device()) = out_eigen + in2_eigen;
174 175 176
  }
};

P
phlrain 已提交
177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
template <typename T>
struct SelectedRowsAddTensor<phi::GPUContext, T> {
  void operator()(const phi::GPUContext& context,
                  const phi::SelectedRows& input1,
                  const framework::Tensor& input2, framework::Tensor* output) {
    auto in1_height = input1.height();
    auto in2_dims = input2.dims();
    auto out_dims = output->dims();
    PADDLE_ENFORCE_EQ(
        in1_height, in2_dims[0],
        platform::errors::InvalidArgument(
            "The two inputs height must be equal."
            "But recieved first input height = [%d], first input height = [%d]",
            in1_height, in2_dims[0]));
    PADDLE_ENFORCE_EQ(
        in1_height, out_dims[0],
        platform::errors::InvalidArgument(
            "The input and output height must be equal."
            "But recieved input height = [%d], output height = [%d]",
            in1_height, out_dims[0]));

    auto& in1_value = input1.value();
    auto& in1_rows = input1.rows();

    int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
    PADDLE_ENFORCE_EQ(
        in1_row_numel, input2.numel() / in1_height,
        platform::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But recieved first input width = [%d], second input width = [%d]",
            in1_row_numel, input2.numel() / in1_height));
    PADDLE_ENFORCE_EQ(
        in1_row_numel, output->numel() / in1_height,
        platform::errors::InvalidArgument(
            "The input and output width must be equal."
            "But recieved input width = [%d], output width = [%d]",
            in1_row_numel, output->numel() / in1_height));

    auto* in1_data = in1_value.data<T>();
    auto* in2_data = input2.data<T>();
    auto* out_data = output->data<T>();

    phi::funcs::SetConstant<phi::GPUContext, T> functor;
    functor(context, output, static_cast<T>(0));

    const int block_size = 256;
    dim3 threads(block_size, 1);
    dim3 grid(in1_rows.size(), 1);
    paddle::framework::MixVector<int64_t> mixv_in1_rows(&in1_rows);
    SelectedRowsAddTensorKernel<
        T, block_size><<<grid, threads, 0, context.stream()>>>(
        in1_data, mixv_in1_rows.CUDAData(context.GetPlace()), out_data,
        in1_row_numel);

    auto out_eigen = framework::EigenVector<T>::Flatten(*output);
    auto in2_eigen = framework::EigenVector<T>::Flatten(input2);
    out_eigen.device(*context.eigen_device()) = out_eigen + in2_eigen;
  }
};

Q
QI JUN 已提交
237 238
template struct SelectedRowsAddTensor<platform::CUDADeviceContext, float>;
template struct SelectedRowsAddTensor<platform::CUDADeviceContext, double>;
C
chengduo 已提交
239 240 241
template struct SelectedRowsAdd<platform::CUDADeviceContext, platform::float16>;
template struct SelectedRowsAddTensor<platform::CUDADeviceContext,
                                      platform::float16>;
Q
QI JUN 已提交
242

P
phlrain 已提交
243 244 245 246 247
template struct SelectedRowsAddTensor<phi::GPUContext, float>;
template struct SelectedRowsAddTensor<phi::GPUContext, double>;
template struct SelectedRowsAdd<phi::GPUContext, platform::float16>;
template struct SelectedRowsAddTensor<phi::GPUContext, platform::float16>;

Q
QI JUN 已提交
248
template <typename T>
Q
QI JUN 已提交
249 250
struct SelectedRowsAddTo<platform::CUDADeviceContext, T> {
  void operator()(const platform::CUDADeviceContext& context,
251 252
                  const phi::SelectedRows& input1, const int64_t input2_offset,
                  phi::SelectedRows* input2) {
Q
QI JUN 已提交
253
    auto in1_height = input1.height();
254 255 256 257 258 259
    PADDLE_ENFORCE_EQ(
        in1_height, input2->height(),
        platform::errors::InvalidArgument("The two inputs height must be equal."
                                          "But recieved first input height = "
                                          "[%d], second input height = [%d]",
                                          in1_height, input2->height()));
Q
QI JUN 已提交
260

261
    auto& in1_rows = input1.rows();
Q
QI JUN 已提交
262 263 264 265 266 267
    auto& in2_rows = *(input2->mutable_rows());

    auto& in1_value = input1.value();
    auto* in2_value = input2->mutable_value();

    // concat rows
268
    paddle::framework::MixVector<int64_t> mixv_in2_rows(&in2_rows);
Y
Fix CI  
Yu Yang 已提交
269
    if (in1_rows.size()) {
270
      mixv_in2_rows.Extend(in1_rows.begin(), in1_rows.end());
Y
Fix CI  
Yu Yang 已提交
271
    }
Q
QI JUN 已提交
272 273

    auto in1_place = input1.place();
274 275 276
    PADDLE_ENFORCE_EQ(platform::is_gpu_place(in1_place), true,
                      platform::errors::InvalidArgument(
                          "The running enviroment is not on the GPU place."));
Q
QI JUN 已提交
277
    auto in2_place = input2->place();
278 279 280
    PADDLE_ENFORCE_EQ(platform::is_gpu_place(in1_place), true,
                      platform::errors::InvalidArgument(
                          "The running enviroment is not on the GPU place."));
Q
QI JUN 已提交
281 282 283

    auto* in1_data = in1_value.data<T>();
    auto* in2_data = in2_value->data<T>();
284
    memory::Copy(in2_place, in2_data + input2_offset, in1_place, in1_data,
Q
QI JUN 已提交
285
                 in1_value.numel() * sizeof(T), context.stream());
Q
QI JUN 已提交
286 287 288
  }
};

Q
QI JUN 已提交
289 290 291 292
template struct SelectedRowsAddTo<platform::CUDADeviceContext, float>;
template struct SelectedRowsAddTo<platform::CUDADeviceContext, double>;
template struct SelectedRowsAddTo<platform::CUDADeviceContext, int>;
template struct SelectedRowsAddTo<platform::CUDADeviceContext, int64_t>;
C
chengduo 已提交
293 294
template struct SelectedRowsAddTo<platform::CUDADeviceContext,
                                  platform::float16>;
Q
QI JUN 已提交
295 296 297 298 299 300 301

namespace {
template <typename T, int block_size>
__global__ void SelectedRowsAddToTensorKernel(const T* selected_rows,
                                              const int64_t* rows,
                                              T* tensor_out,
                                              int64_t row_numel) {
C
chengduo 已提交
302
  const int ty = blockIdx.x;
Q
QI JUN 已提交
303 304 305 306 307 308 309 310 311 312 313 314 315 316
  int tid = threadIdx.x;

  selected_rows += ty * row_numel;
  tensor_out += rows[ty] * row_numel;

  for (int index = tid; index < row_numel; index += block_size) {
    // Since index in rows of SelectedRows can be duplicate, we have to use
    // Atomic Operation to avoid concurrent write error.
    paddle::platform::CudaAtomicAdd(tensor_out + index, selected_rows[index]);
  }
}
}  // namespace

template <typename T>
Q
QI JUN 已提交
317 318
struct SelectedRowsAddToTensor<platform::CUDADeviceContext, T> {
  void operator()(const platform::CUDADeviceContext& context,
319
                  const phi::SelectedRows& input1, framework::Tensor* input2) {
Q
QI JUN 已提交
320 321
    auto in1_height = input1.height();
    auto in2_dims = input2->dims();
322 323 324 325 326 327
    PADDLE_ENFORCE_EQ(
        in1_height, in2_dims[0],
        platform::errors::InvalidArgument("The two inputs height must be equal."
                                          "But recieved first input height = "
                                          "[%d], second input height = [%d]",
                                          in1_height, in2_dims[0]));
Q
QI JUN 已提交
328 329

    auto& in1_value = input1.value();
330
    auto& in1_rows = input1.rows();
Q
QI JUN 已提交
331 332

    int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
333 334 335 336 337 338
    PADDLE_ENFORCE_EQ(
        in1_row_numel, input2->numel() / in1_height,
        platform::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But recieved first input width = [%d], second input width = [%d]",
            in1_row_numel, input2->numel() / in1_height));
Q
QI JUN 已提交
339 340 341 342 343

    auto* in1_data = in1_value.data<T>();
    auto* in2_data = input2->data<T>();
    const int block_size = 256;
    dim3 threads(block_size, 1);
C
chengduo 已提交
344
    dim3 grid(in1_rows.size(), 1);
345
    paddle::framework::MixVector<int64_t> mixv_in1_rows(&in1_rows);
Q
QI JUN 已提交
346 347
    SelectedRowsAddToTensorKernel<
        T, block_size><<<grid, threads, 0, context.stream()>>>(
348
        in1_data, mixv_in1_rows.CUDAData(context.GetPlace()), in2_data,
Y
Yu Yang 已提交
349
        in1_row_numel);
Q
QI JUN 已提交
350 351 352
  }
};

P
phlrain 已提交
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
template <typename T>
struct SelectedRowsAddToTensor<phi::GPUContext, T> {
  void operator()(const phi::GPUContext& context,
                  const phi::SelectedRows& input1, framework::Tensor* input2) {
    auto in1_height = input1.height();
    auto in2_dims = input2->dims();
    PADDLE_ENFORCE_EQ(
        in1_height, in2_dims[0],
        platform::errors::InvalidArgument("The two inputs height must be equal."
                                          "But recieved first input height = "
                                          "[%d], second input height = [%d]",
                                          in1_height, in2_dims[0]));

    auto& in1_value = input1.value();
    auto& in1_rows = input1.rows();

    int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
    PADDLE_ENFORCE_EQ(
        in1_row_numel, input2->numel() / in1_height,
        platform::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But recieved first input width = [%d], second input width = [%d]",
            in1_row_numel, input2->numel() / in1_height));

    auto* in1_data = in1_value.data<T>();
    auto* in2_data = input2->data<T>();
    const int block_size = 256;
    dim3 threads(block_size, 1);
    dim3 grid(in1_rows.size(), 1);
    paddle::framework::MixVector<int64_t> mixv_in1_rows(&in1_rows);
    SelectedRowsAddToTensorKernel<
        T, block_size><<<grid, threads, 0, context.stream()>>>(
        in1_data, mixv_in1_rows.CUDAData(context.GetPlace()), in2_data,
        in1_row_numel);
  }
};

Q
QI JUN 已提交
390 391 392 393
template struct SelectedRowsAddToTensor<platform::CUDADeviceContext, float>;
template struct SelectedRowsAddToTensor<platform::CUDADeviceContext, double>;
template struct SelectedRowsAddToTensor<platform::CUDADeviceContext, int>;
template struct SelectedRowsAddToTensor<platform::CUDADeviceContext, int64_t>;
C
chengduo 已提交
394 395
template struct SelectedRowsAddToTensor<platform::CUDADeviceContext,
                                        platform::float16>;
P
phlrain 已提交
396 397 398 399 400
template struct SelectedRowsAddToTensor<phi::GPUContext, float>;
template struct SelectedRowsAddToTensor<phi::GPUContext, double>;
template struct SelectedRowsAddToTensor<phi::GPUContext, int>;
template struct SelectedRowsAddToTensor<phi::GPUContext, int64_t>;
template struct SelectedRowsAddToTensor<phi::GPUContext, platform::float16>;
T
typhoonzero 已提交
401 402 403 404 405 406 407

namespace scatter {

template <typename T, int block_size>
__global__ void MergeAddKernel(const T* input, const int64_t* input_rows,
                               T* out, const int64_t* out_rows,
                               size_t out_rows_size, int64_t row_numel) {
S
sneaxiy 已提交
408
  const int ty = blockIdx.x;
T
typhoonzero 已提交
409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
  int tid = threadIdx.x;
  __shared__ size_t out_idx;

  if (tid == 0) {
    for (size_t i = 0; i < out_rows_size; i++) {
      if (input_rows[ty] == out_rows[i]) {
        out_idx = i;
      }
    }
  }

  __syncthreads();

  input += ty * row_numel;
  out += out_idx * row_numel;
  for (int index = tid; index < row_numel; index += block_size) {
    paddle::platform::CudaAtomicAdd(out + index, input[index]);
  }
}

P
phlrain 已提交
429 430 431
template <typename DeviceContext, typename T>
struct MergeAddImpl {
  phi::SelectedRows operator()(const DeviceContext& context,
432 433 434
                               const phi::SelectedRows& input,
                               const bool sorted_result = false) {
    phi::SelectedRows out;
S
sneaxiy 已提交
435 436 437 438
    (*this)(context, input, &out);
    return out;
  }

P
phlrain 已提交
439 440
  void operator()(const DeviceContext& context, const phi::SelectedRows& input,
                  phi::SelectedRows* output, const bool sorted_result = false) {
D
dzhwinter 已提交
441
    framework::Vector<int64_t> input_rows(input.rows());
Q
Qiao Longfei 已提交
442 443 444 445
    if (input_rows.size() == 0) {
      return;
    }

446
    phi::SelectedRows& out = *output;
T
typhoonzero 已提交
447
    std::set<int64_t> row_set(input_rows.begin(), input_rows.end());
Q
Qiao Longfei 已提交
448 449
    std::vector<int64_t> merge_rows_cpu(row_set.begin(), row_set.end());
    framework::Vector<int64_t> merge_rows(merge_rows_cpu);
T
typhoonzero 已提交
450 451

    auto input_width = input.value().dims()[1];
T
wip  
typhoonzero 已提交
452 453 454 455

    out.set_rows(merge_rows);
    out.set_height(input.height());
    out.mutable_value()->mutable_data<T>(
456
        phi::make_ddim({static_cast<int64_t>(merge_rows.size()), input_width}),
T
typhoonzero 已提交
457 458
        context.GetPlace());

P
phlrain 已提交
459
    phi::funcs::SetConstant<DeviceContext, T> constant_functor;
C
chengduo 已提交
460
    constant_functor(context, out.mutable_value(), static_cast<T>(0));
T
typhoonzero 已提交
461

T
wip  
typhoonzero 已提交
462
    auto* out_data = out.mutable_value()->data<T>();
T
typhoonzero 已提交
463 464 465 466
    auto* input_data = input.value().data<T>();

    const int block_size = 256;
    dim3 threads(block_size, 1);
S
sneaxiy 已提交
467
    dim3 grid1(input_rows.size(), 1);
T
typhoonzero 已提交
468

469 470
    paddle::framework::MixVector<int64_t> mix_vector_input(&input_rows);
    paddle::framework::MixVector<int64_t> mix_vector_out(out.mutable_rows());
S
sneaxiy 已提交
471
    MergeAddKernel<T, 256><<<grid1, threads, 0, context.stream()>>>(
472 473 474 475
        input_data, mix_vector_input.CUDAData(context.GetPlace()), out_data,
        mix_vector_out.CUDAMutableData(context.GetPlace()), out.rows().size(),
        input_width);
    mix_vector_out.CopyToCPU();
T
typhoonzero 已提交
476
  }
477

P
phlrain 已提交
478
  void operator()(const DeviceContext& context,
479 480
                  const std::vector<const phi::SelectedRows*>& inputs,
                  phi::SelectedRows* output, const bool sorted_result = false) {
481
    if (inputs.size() == 0) {
M
minqiyang 已提交
482
      VLOG(3) << "no input! return";
483 484
      return;
    }
485
    const phi::SelectedRows* has_value_input = nullptr;
486
    for (auto* in : inputs) {
Q
Qiao Longfei 已提交
487
      if (in->rows().size() > 0) {
488 489 490 491 492
        has_value_input = in;
        break;
      }
    }
    if (has_value_input == nullptr) {
M
minqiyang 已提交
493
      VLOG(3) << "no input has value! just return" << std::endl;
494 495 496 497
      return;
    }
    auto input_width = has_value_input->value().dims()[1];
    auto input_height = has_value_input->height();
498
    phi::SelectedRows& out = *output;
499 500
    std::set<int64_t> merged_row_set;
    for (auto* input : inputs) {
Q
Qiao Longfei 已提交
501
      if (input->rows().size() == 0) {
502 503
        continue;
      }
504
      PADDLE_ENFORCE_EQ(input_width, input->value().dims()[1],
505 506 507
                        platform::errors::InvalidArgument(
                            "All input should have same "
                            "dimension except for the first one."));
508
      PADDLE_ENFORCE_EQ(input_height, input->height(),
509 510
                        platform::errors::InvalidArgument(
                            "All input should have same height."));
511 512
      merged_row_set.insert(input->rows().begin(), input->rows().end());
    }
Q
Qiao Longfei 已提交
513
    std::vector<int64_t> merge_rows_cpu(merged_row_set.begin(),
Q
format  
Qiao Longfei 已提交
514
                                        merged_row_set.end());
Q
Qiao Longfei 已提交
515
    framework::Vector<int64_t> merge_rows(merge_rows_cpu);
516 517 518 519

    out.set_rows(merge_rows);
    out.set_height(input_height);
    out.mutable_value()->mutable_data<T>(
520
        phi::make_ddim({static_cast<int64_t>(merge_rows.size()), input_width}),
521 522
        context.GetPlace());

P
phlrain 已提交
523
    phi::funcs::SetConstant<DeviceContext, T> constant_functor;
C
chengduo 已提交
524
    constant_functor(context, out.mutable_value(), static_cast<T>(0));
525 526 527 528 529 530 531

    auto* out_data = out.mutable_value()->data<T>();

    const int block_size = 256;
    dim3 threads(block_size, 1);

    for (auto* input : inputs) {
Q
Qiao Longfei 已提交
532
      if (input->rows().size() == 0) {
Q
Qiao Longfei 已提交
533 534
        continue;
      }
535 536
      auto* input_data = input->value().data<T>();
      auto& input_rows = input->rows();
537 538
      dim3 grid1(input_rows.size(), 1);

539 540
      paddle::framework::MixVector<int64_t> mix_vector_input(&input_rows);
      paddle::framework::MixVector<int64_t> mix_vector_out(out.mutable_rows());
541
      MergeAddKernel<T, 256><<<grid1, threads, 0, context.stream()>>>(
542 543 544 545
          input_data, mix_vector_input.CUDAData(context.GetPlace()), out_data,
          mix_vector_out.CUDAMutableData(context.GetPlace()), out.rows().size(),
          input_width);
      mix_vector_out.CopyToCPU();
546 547
    }
  }
T
typhoonzero 已提交
548 549
};

P
phlrain 已提交
550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612
template <typename T>
struct MergeAdd<platform::CUDADeviceContext, T> {
  // unary functor, merge by adding duplicated rows in
  // the input SelectedRows object.
  phi::SelectedRows operator()(const platform::CUDADeviceContext& context,
                               const phi::SelectedRows& input,
                               const bool sorted_result) {
    return MergeAddImpl<platform::CUDADeviceContext, T>()(context, input,
                                                          sorted_result);
  }

  void operator()(const platform::CUDADeviceContext& context,
                  const phi::SelectedRows& input, phi::SelectedRows* output,
                  const bool sorted_result) {
    MergeAddImpl<platform::CUDADeviceContext, T>()(context, input, output,
                                                   sorted_result);
  }

  void operator()(const platform::CUDADeviceContext& context,
                  const std::vector<const phi::SelectedRows*>& inputs,
                  phi::SelectedRows* output, const bool sorted_result) {
    MergeAddImpl<platform::CUDADeviceContext, T>()(context, inputs, output,
                                                   sorted_result);
  }
};

template <typename T>
struct MergeAdd<phi::GPUContext, T> {
  // unary functor, merge by adding duplicated rows in
  // the input SelectedRows object.
  phi::SelectedRows operator()(const phi::GPUContext& context,
                               const phi::SelectedRows& input,
                               const bool sorted_result) {
    return MergeAddImpl<phi::GPUContext, T>()(context, input, sorted_result);
  }

  void operator()(const phi::GPUContext& context,
                  const phi::SelectedRows& input, phi::SelectedRows* output,
                  const bool sorted_result) {
    MergeAddImpl<phi::GPUContext, T>()(context, input, output, sorted_result);
  }

  void operator()(const phi::GPUContext& context,
                  const std::vector<const phi::SelectedRows*>& inputs,
                  phi::SelectedRows* output, const bool sorted_result) {
    MergeAddImpl<phi::GPUContext, T>()(context, inputs, output, sorted_result);
  }
};

#define TEMPLATE_SPECIALIZED_FOR_MERGEADD(dtype)                    \
  template struct MergeAddImpl<platform::CUDADeviceContext, dtype>; \
  template struct MergeAddImpl<phi::GPUContext, dtype>;             \
  template struct MergeAdd<platform::CUDADeviceContext, dtype>;     \
  template struct MergeAdd<phi::GPUContext, dtype>;

TEMPLATE_SPECIALIZED_FOR_MERGEADD(float)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(double)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(int)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(int64_t)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(platform::float16)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(platform::bfloat16)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(platform::complex<float>)
TEMPLATE_SPECIALIZED_FOR_MERGEADD(platform::complex<double>)
T
wip  
typhoonzero 已提交
613 614 615 616 617

template <typename T, int block_size>
__global__ void UpdateToTensorKernel(const T* selected_rows,
                                     const int64_t* rows, const ScatterOps& op,
                                     T* tensor_out, int64_t row_numel) {
C
chengduo 已提交
618
  const int ty = blockIdx.x;
T
wip  
typhoonzero 已提交
619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663
  int tid = threadIdx.x;

  selected_rows += ty * row_numel;
  tensor_out += rows[ty] * row_numel;
  // FIXME(typhoonzero): use macro fix the below messy code.
  switch (op) {
    case ScatterOps::ASSIGN:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] = selected_rows[index];
      }
      break;
    case ScatterOps::ADD:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] += selected_rows[index];
      }
      break;
    case ScatterOps::SUB:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] -= selected_rows[index];
      }
      break;
    case ScatterOps::SUBBY:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] = selected_rows[index] - tensor_out[index];
      }
      break;
    case ScatterOps::MUL:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] *= selected_rows[index];
      }
      break;
    case ScatterOps::DIV:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] /= selected_rows[index];
      }
      break;
    case ScatterOps::DIVBY:
      for (int index = tid; index < row_numel; index += block_size) {
        tensor_out[index] = selected_rows[index] / tensor_out[index];
      }
      break;
  }
}

template <typename T>
T
typhoonzero 已提交
664 665
struct UpdateToTensor<platform::CUDADeviceContext, T> {
  void operator()(const platform::CUDADeviceContext& context,
666
                  const ScatterOps& op, const phi::SelectedRows& input1,
T
typhoonzero 已提交
667
                  framework::Tensor* input2) {
T
wip  
typhoonzero 已提交
668 669
    // NOTE: Use SelectedRowsAddToTensor for better performance
    //       no additional MergeAdd called.
T
typhoonzero 已提交
670 671
    MergeAdd<platform::CUDADeviceContext, T> merge_func;
    auto merged_in1 = merge_func(context, input1);
T
wip  
typhoonzero 已提交
672 673 674

    auto in1_height = merged_in1.height();
    auto in2_dims = input2->dims();
675 676 677 678 679 680
    PADDLE_ENFORCE_EQ(
        in1_height, in2_dims[0],
        platform::errors::InvalidArgument("The two inputs height must be equal."
                                          "But recieved first input height = "
                                          "[%d], second input height = [%d]",
                                          in1_height, in2_dims[0]));
T
wip  
typhoonzero 已提交
681 682 683 684 685

    auto& in1_value = merged_in1.value();
    auto& in1_rows = merged_in1.rows();

    int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
686 687 688 689 690 691
    PADDLE_ENFORCE_EQ(
        in1_row_numel, input2->numel() / in1_height,
        platform::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But recieved first input width = [%d], second input width = [%d]",
            in1_row_numel, input2->numel() / in1_height));
T
wip  
typhoonzero 已提交
692

T
typhoonzero 已提交
693 694
    auto* in1_data = in1_value.template data<T>();
    auto* in2_data = input2->data<T>();
T
wip  
typhoonzero 已提交
695

T
typhoonzero 已提交
696
    dim3 threads(platform::PADDLE_CUDA_NUM_THREADS, 1);
C
chengduo 已提交
697
    dim3 grid(in1_rows.size(), 1);
T
typhoonzero 已提交
698
    UpdateToTensorKernel<T, platform::PADDLE_CUDA_NUM_THREADS><<<
D
dzhwinter 已提交
699 700
        grid, threads, 0, context.stream()>>>(in1_data, in1_rows.cuda_data(),
                                              op, in2_data, in1_row_numel);
T
wip  
typhoonzero 已提交
701 702
  }
};
T
typhoonzero 已提交
703
}  // namespace scatter
704 705 706
}  // namespace math
}  // namespace operators
}  // namespace paddle