executor.h 4.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Q
qijun 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

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

L
Liu Yiqun 已提交
17 18 19
#include <map>
#include <string>
#include <vector>
20
#include "paddle/fluid/framework/garbage_collector.h"
Y
Yi Wang 已提交
21 22 23 24 25
#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"
Q
qijun 已提交
26 27 28

namespace paddle {
namespace framework {
Q
Qiao Longfei 已提交
29

S
sneaxiy 已提交
30 31 32 33 34 35
template <typename T>
std::unordered_map<std::string, T> GetNonPersistableReferenceCount(
    const ProgramDesc& prog, size_t block_id) {
  auto& block = prog.Block(block_id);
  std::unordered_map<std::string, T> ref_cnts;

S
sneaxiy 已提交
36 37 38 39 40 41 42 43 44
  auto update_ref_cnts = [&](OpDesc* op_desc, const VariableNameMap& name_map) {
    for (auto& name_pair : name_map) {
      for (auto& name : name_pair.second) {
        auto* var_desc = block.FindVar(name);
        if (var_desc == nullptr || var_desc->Persistable()) continue;
        auto type = var_desc->Proto()->type().type();
        if (type != proto::VarType::LOD_TENSOR &&
            type != proto::VarType::SELECTED_ROWS) {
          continue;
S
sneaxiy 已提交
45 46
        }

S
sneaxiy 已提交
47 48 49 50 51
        auto it = ref_cnts.find(name);
        if (it != ref_cnts.end()) {
          ++it->second;
        } else {
          ref_cnts[name] = 1;
S
sneaxiy 已提交
52 53 54
        }
      }
    }
S
sneaxiy 已提交
55 56 57 58 59
  };

  for (auto op_desc : block.AllOps()) {
    update_ref_cnts(op_desc, op_desc->Inputs());
    update_ref_cnts(op_desc, op_desc->Outputs());
S
sneaxiy 已提交
60 61 62 63
  }
  return ref_cnts;
}

Q
Qiao Longfei 已提交
64 65 66 67
struct ExecutorPrepareContext {
  ExecutorPrepareContext(const framework::ProgramDesc& prog, size_t block_id);
  ~ExecutorPrepareContext();

S
sneaxiy 已提交
68 69
  void ResetReferenceCount() { cur_ref_cnts_ = ref_cnts_; }

Q
Qiao Longfei 已提交
70 71 72
  const framework::ProgramDesc& prog_;
  size_t block_id_;
  std::vector<std::unique_ptr<OperatorBase>> ops_;
S
sneaxiy 已提交
73 74

  std::unordered_map<std::string, int> ref_cnts_;
S
sneaxiy 已提交
75
  std::unordered_map<std::string, int> cur_ref_cnts_;
Q
Qiao Longfei 已提交
76 77
};

Q
qijun 已提交
78 79
class Executor {
 public:
D
dzhwinter 已提交
80 81
  // TODO(dzhwinter) : Do not rely on this function, it will be removed
  explicit Executor(const platform::DeviceContext& device)
D
dzhwinter 已提交
82
      : Executor(device.GetPlace()) {}
D
dzhwinter 已提交
83

D
dzhwinter 已提交
84
  explicit Executor(const platform::Place& place);
Y
Yang Yang 已提交
85

W
Wu Yi 已提交
86
  /*
Y
Yancey1989 已提交
87 88
   * Close this Executor.
   * Calling this method will send complete messages to all pserver instances.
W
Wu Yi 已提交
89
   */
Y
Yancey1989 已提交
90
  void Close();
W
Wu Yi 已提交
91

Y
Yang Yang 已提交
92 93 94 95 96 97 98
  /* @Brief
   * Runtime evaluation of the given ProgramDesc under certain Scope
   *
   * @param
   *  ProgramDesc
   *  Scope
   */
Y
Yu Yang 已提交
99 100
  void Run(const ProgramDesc& prog, Scope* scope, int block_id,
           bool create_local_scope = true, bool create_vars = true);
Q
qijun 已提交
101

X
fix  
Xin Pan 已提交
102
  // This API is very slow.
103
  void Run(const ProgramDesc& program, Scope* scope,
104 105
           std::map<std::string, const LoDTensor*>* feed_targets,
           std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
106
           bool create_local_scope = true, bool create_vars = true,
107
           const std::string& feed_holder_name = "feed",
108
           const std::string& fetch_holder_name = "fetch");
109

Q
Qiao Longfei 已提交
110 111
  static std::unique_ptr<ExecutorPrepareContext> Prepare(
      const ProgramDesc& program, int block_id);
Y
Yu Yang 已提交
112

T
typhoonzero 已提交
113 114 115
  static std::vector<std::shared_ptr<ExecutorPrepareContext>> Prepare(
      const ProgramDesc& program, const std::vector<int>& block_ids);

L
Liu Yiqun 已提交
116
  void CreateVariables(const ProgramDesc& pdesc, Scope* scope, int block_id);
117

Y
Yu Yang 已提交
118 119
  void RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
                          bool create_local_scope = true,
Q
qiaolongfei 已提交
120
                          bool create_vars = true, bool keep_kids = false);
Y
Yu Yang 已提交
121

X
fix  
Xin Pan 已提交
122
  // This API is very slow.
123
  void RunPreparedContext(ExecutorPrepareContext* ctx, Scope* scope,
124 125
                          std::map<std::string, const LoDTensor*>* feed_targets,
                          std::map<std::string, LoDTensor*>* fetch_targets,
W
Wu Yi 已提交
126
                          bool create_local_scope = true,
L
Liu Yiqun 已提交
127
                          bool create_vars = true,
128
                          const std::string& feed_holder_name = "feed",
L
Liu Yiqun 已提交
129
                          const std::string& fetch_holder_name = "fetch");
130

131 132
  void EnableMKLDNN(const ProgramDesc& program);

Q
qijun 已提交
133
 private:
D
dzhwinter 已提交
134
  const platform::Place place_;
Q
qijun 已提交
135
};
Q
qijun 已提交
136 137 138

}  // namespace framework
}  // namespace paddle