threaded_ssa_graph_executor.h 2.0 KB
Newer Older
Y
Yu Yang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
//   Copyright (c) 2018 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

X
Xin Pan 已提交
17 18 19 20 21 22
#include <deque>
#include <string>
#include <unordered_set>
#include <utility>
#include <vector>

Y
Yu Yang 已提交
23
#include <functional>
Y
Yu Yang 已提交
24
#include "ThreadPool.h"  // ThreadPool in thrird party
T
typhoonzero 已提交
25
#include "paddle/fluid/framework/blocking_queue.h"
Y
Yu Yang 已提交
26 27 28 29 30 31 32 33 34 35 36 37 38
#include "paddle/fluid/framework/details/ssa_graph_executor.h"

namespace paddle {
namespace framework {
class Scope;

namespace details {

class ThreadedSSAGraphExecutor : public SSAGraphExecutor {
 public:
  ThreadedSSAGraphExecutor(size_t num_threads, bool use_event,
                           const std::vector<Scope *> &local_scopes,
                           const std::vector<platform::Place> &places,
X
Xin Pan 已提交
39 40
                           std::unique_ptr<SSAGraph> &&graph,
                           bool allow_op_delay);
Y
Yu Yang 已提交
41 42 43 44 45 46 47 48

  // Run a SSAGraph by a thread pool
  // Use topological sort algorithm
  FeedFetchList Run(const std::vector<std::string> &fetch_tensors) override;

  ~ThreadedSSAGraphExecutor() {}

 private:
X
Xin Pan 已提交
49
  void RunOp(BlockingQueue<VarHandleBase *> *ready_var_q,
Y
Yu Yang 已提交
50
             details::OpHandleBase *op);
Y
Yu Yang 已提交
51 52 53 54 55 56 57 58

 private:
  std::unique_ptr<::ThreadPool> pool_;
  std::vector<Scope *> local_scopes_;
  std::vector<platform::Place> places_;
  platform::DeviceContextPool fetch_ctxs_;
  const bool use_event_;
  std::unique_ptr<platform::EnforceNotMet> exception_;
X
Xin Pan 已提交
59
  std::atomic<int> running_ops_;
X
Xin Pan 已提交
60
  bool allow_op_delay_;
Y
Yu Yang 已提交
61 62 63 64 65
};

}  // namespace details
}  // namespace framework
}  // namespace paddle