ProcessGroupStream.cc 12.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "paddle/fluid/distributed/collective/ProcessGroupStream.h"

namespace paddle {
namespace distributed {

ProcessGroupStream::ProcessGroupStream(int rank,
                                       int size,
                                       const platform::Place& place,
                                       int gid)
    : ProcessGroup(rank, size, place, gid) {}

26
const phi::DeviceContext& ProcessGroupStream::GetDeviceContext(
27 28 29 30 31 32
    const Place& place, bool use_calc_stream) const {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support get device_context.", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllGather(
33 34
    phi::DenseTensor* out_tensor,
    const phi::DenseTensor& in_tensor,
35
    bool sync_op) {
36 37
  return AllGather(out_tensor,
                   in_tensor,
38 39 40 41 42
                   sync_op,
                   /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllGather(
43 44
    phi::DenseTensor* out_tensor,
    const phi::DenseTensor& in_tensor,
45 46 47 48 49 50
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do all_gather", GetBackendName()));
}

51
std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllReduce(
52 53 54
    phi::DenseTensor* out_tensor,
    const phi::DenseTensor& in_tensor,
    const AllreduceOptions& opts,
55
    bool sync_op) {
56 57 58
  return AllReduce(out_tensor,
                   in_tensor,
                   opts,
59 60 61 62 63
                   sync_op,
                   /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllReduce(
64 65 66
    phi::DenseTensor* out_tensor,
    const phi::DenseTensor& in_tensor,
    const AllreduceOptions& opts,
67 68 69
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
70
      "ProcessGroup%s does not support do all_reduce", GetBackendName()));
71 72
}

73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Broadcast(
    phi::DenseTensor* out_tensor,
    const phi::DenseTensor& in_tensor,
    const BroadcastOptions& opts,
    bool sync_op) {
  return Broadcast(out_tensor,
                   in_tensor,
                   opts,
                   sync_op,
                   /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Broadcast(
    phi::DenseTensor* out_tensor,
    const phi::DenseTensor& in_tensor,
    const BroadcastOptions& opts,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do broadcast", GetBackendName()));
}

95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Recv(
    phi::DenseTensor* tensor, int src_rank, bool sync_op) {
  return Recv(tensor,
              src_rank,
              sync_op,
              /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Recv(
    phi::DenseTensor* tensor,
    int src_rank,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do recv", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::RecvPartial(
    phi::DenseTensor* tensor,
    int src_rank,
    int64_t offset,
    int64_t length,
    bool sync_op) {
  return RecvPartial(tensor,
                     src_rank,
                     offset,
                     length,
                     sync_op,
                     /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::RecvPartial(
    phi::DenseTensor* tensor,
    int src_rank,
    int64_t offset,
    int64_t length,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do recv_partial", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Send(
    phi::DenseTensor* tensor, int dst_rank, bool sync_op) {
  return Send(tensor,
              dst_rank,
              sync_op,
              /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Send(
    phi::DenseTensor*, int dst_rank, bool sync_op, bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do send", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::SendPartial(
    phi::DenseTensor* tensor,
    int dst_rank,
    int64_t offset,
    int64_t length,
    bool sync_op) {
  return SendPartial(tensor,
                     dst_rank,
                     offset,
                     length,
                     sync_op,
                     /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::SendPartial(
    phi::DenseTensor* tensor,
    int dst_rank,
    int64_t offset,
    int64_t length,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do send partial", GetBackendName()));
}

176
// TODO(sunyilun): methods below will be removed later
177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllToAll(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    bool sync_op) {
  return AllToAll(in_tensors,
                  out_tensors,
                  sync_op,
                  /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllToAll(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do alltoall", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllToAllSingle(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    std::vector<int64_t>& in_sizes,
    std::vector<int64_t>& out_sizes,
    bool sync_op) {
  return AllToAllSingle(in_tensors,
                        out_tensors,
                        in_sizes,
                        out_sizes,
                        sync_op,
                        /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllToAllSingle(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    std::vector<int64_t>& in_sizes,
    std::vector<int64_t>& out_sizes,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do alltoall_single", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Reduce(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    const ReduceOptions& opts,
    bool sync_op) {
  return Reduce(in_tensors,
                out_tensors,
                opts,
                sync_op,
                /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Reduce(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    const ReduceOptions& opts,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do reduce", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::ReduceScatter(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    const ReduceScatterOptions& opts,
    bool sync_op) {
  return ReduceScatter(in_tensors,
                       out_tensors,
                       opts,
                       sync_op,
                       /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::ReduceScatter(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    const ReduceScatterOptions& opts,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do reduce_scatter", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Scatter(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    const ScatterOptions& opts,
    bool sync_op) {
  return Scatter(in_tensors,
                 out_tensors,
                 opts,
                 sync_op,
                 /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Scatter(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
    const ScatterOptions& opts,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do scatter", GetBackendName()));
}

287 288 289
std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Send_Partial(
    phi::DenseTensor& tensors,
    int dst_rank,
290 291
    int64_t offset,
    int64_t length,
292 293 294 295 296 297 298 299 300 301 302 303
    bool sync_op) {
  return Send_Partial(tensors,
                      dst_rank,
                      offset,
                      length,
                      sync_op,
                      /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Send_Partial(
    phi::DenseTensor& tensors,
    int dst_rank,
304 305
    int64_t offset,
    int64_t length,
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
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do send_partial", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Recv(
    std::vector<phi::DenseTensor>& tensors, int src_rank, bool sync_op) {
  return Recv(tensors,
              src_rank,
              sync_op,
              /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Recv(
    std::vector<phi::DenseTensor>& tensors,
    int src_rank,
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do recv", GetBackendName()));
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Recv_Partial(
    phi::DenseTensor& tensors,
    int src_rank,
332 333
    int64_t offset,
    int64_t length,
334 335 336 337 338 339 340 341 342 343 344 345
    bool sync_op) {
  return Recv_Partial(tensors,
                      src_rank,
                      offset,
                      length,
                      sync_op,
                      /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::Recv_Partial(
    phi::DenseTensor& tensors,
    int src_rank,
346 347
    int64_t offset,
    int64_t length,
348 349 350 351 352 353
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do recv_partial", GetBackendName()));
}

354 355 356
std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllGather_Partial(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
357 358
    int64_t offset,
    int64_t length,
359 360 361 362 363 364 365 366 367 368 369 370
    bool sync_op) {
  return AllGather_Partial(in_tensors,
                           out_tensors,
                           offset,
                           length,
                           sync_op,
                           /*use_calc_stream*/ false);
}

std::shared_ptr<ProcessGroup::Task> ProcessGroupStream::AllGather_Partial(
    std::vector<phi::DenseTensor>& in_tensors,
    std::vector<phi::DenseTensor>& out_tensors,
371 372
    int64_t offset,
    int64_t length,
373 374 375 376 377 378
    bool sync_op,
    bool use_calc_stream) {
  PADDLE_THROW(platform::errors::InvalidArgument(
      "ProcessGroup%s does not support do recv_partial", GetBackendName()));
}

379 380
}  // namespace distributed
}  // namespace paddle