parallel_executor.h 2.7 KB
Newer Older
Y
Yang Yang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/* Copyright (c) 2016 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. */

#pragma once

Y
yuyang18 已提交
17
#include <paddle/fluid/framework/details/build_strategy.h>
X
Xin Pan 已提交
18
#include <string>
Y
Yang Yang 已提交
19
#include <unordered_set>
X
Xin Pan 已提交
20
#include <vector>
Y
yuyang18 已提交
21
#include "paddle/fluid/framework/details/execution_strategy.h"
22
#include "paddle/fluid/framework/details/multi_devices_graph_builder.h"
Y
Yang Yang 已提交
23 24 25 26 27 28
#include "paddle/fluid/framework/executor.h"
#include "paddle/fluid/framework/op_info.h"
#include "paddle/fluid/framework/program_desc.h"
#include "paddle/fluid/framework/scope.h"
#include "paddle/fluid/framework/tensor.h"
#include "paddle/fluid/platform/device_context.h"
29

Y
Yang Yang 已提交
30 31 32
namespace paddle {
namespace framework {

Y
Yu Yang 已提交
33
class ParallelExecutorPrivate;
Y
Use mtx  
Yu Yang 已提交
34

Y
yuyang18 已提交
35
using details::BuildStrategy;
Y
yuyang18 已提交
36 37
using details::ExecutionStrategy;

Y
Yang Yang 已提交
38
class ParallelExecutor {
Y
Yu Yang 已提交
39 40
  DISABLE_COPY_AND_ASSIGN(ParallelExecutor);

Y
Yu Yang 已提交
41
 public:
Y
yuyang18 已提交
42 43 44 45 46 47
  explicit ParallelExecutor(const std::vector<platform::Place> &places,
                            const std::unordered_set<std::string> &params,
                            const std::unordered_set<std::string> &bcast_vars,
                            const ProgramDesc &main_program,
                            const std::string &loss_var_name, Scope *scope,
                            const std::vector<Scope *> &local_scopes,
Y
yuyang18 已提交
48
                            const ExecutionStrategy &exec_strategy,
49
                            const BuildStrategy &build_strategy,
T
typhoonzero 已提交
50
                            size_t num_trainers = 1, size_t trainer_id = 0);
Y
Yu Yang 已提交
51

52 53
  ~ParallelExecutor();

Y
yuyang18 已提交
54
  std::vector<Scope *> &GetLocalScopes();
55

Y
Yu Yang 已提交
56 57 58 59 60
  /**
   * Feed tensors to local scopes. The size of tensors should be equal to the
   * size of local scopes.
   */
  void FeedTensorsIntoLocalScopes(
Y
yuyang18 已提交
61
      const std::vector<std::unordered_map<std::string, LoDTensor>> &tensors);
Y
Yu Yang 已提交
62 63

  void FeedAndSplitTensorIntoLocalScopes(
Y
yuyang18 已提交
64
      const std::unordered_map<std::string, LoDTensor> &tensors);
Y
Yu Yang 已提交
65

Y
yuyang18 已提交
66 67
  void Run(const std::vector<std::string> &fetch_tensors,
           const std::string &fetched_var_name);
Y
Yang Yang 已提交
68

Y
yuyang18 已提交
69
  void BCastParamsToGPUs(const std::unordered_set<std::string> &vars) const;
T
typhoonzero 已提交
70

Y
Yang Yang 已提交
71
 private:
Y
yuyang18 已提交
72
  ParallelExecutorPrivate *member_;
73
  std::unique_ptr<details::SSAGraphBuilder> builder_;
Y
Yang Yang 已提交
74 75 76 77
};

}  // namespace framework
}  // namespace paddle