multi_trainer.cc 2.8 KB
Newer Older
X
xiexionghang 已提交
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 26
/* 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. */

#include <string>
#include <vector>
#include "paddle/fluid/framework/data_feed_factory.h"
#include "paddle/fluid/framework/device_worker_factory.h"
#include "paddle/fluid/framework/trainer.h"

namespace paddle {
namespace framework {

void MultiTrainer::Initialize(const TrainerDesc& trainer_desc,
                              Dataset* dataset) {
  thread_num_ = trainer_desc.thread_num();
27 28 29 30 31
  for (int i = 0; i < trainer_desc.downpour_param().stat_var_names_size();
       i++) {
    need_merge_var_names_.push_back(
        trainer_desc.downpour_param().stat_var_names(i));
  }
X
xiexionghang 已提交
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
  SetDataset(dataset);
  // get filelist from trainer_desc here
  const std::vector<paddle::framework::DataFeed*> readers =
      dataset->GetReaders();
  VLOG(3) << "readers num: " << readers.size();
  // change thread num to readers num
  thread_num_ = readers.size();
  VLOG(3) << "worker thread num: " << thread_num_;
  workers_.resize(thread_num_);
  for (int i = 0; i < thread_num_; ++i) {
    workers_[i] = DeviceWorkerFactory::CreateDeviceWorker(
        trainer_desc.device_worker_name());
    workers_[i]->Initialize(trainer_desc);
    workers_[i]->SetDeviceIndex(i);
    workers_[i]->SetDataFeed(readers[i]);
  }

  // set debug here
  SetDebug(trainer_desc.debug());
}

// call only after all resources are set in current trainer
void MultiTrainer::InitTrainerEnv(const ProgramDesc& main_program,
                                  const platform::Place& place) {
  for (int i = 0; i < thread_num_; ++i) {
    workers_[i]->SetPlace(place);
58
    workers_[i]->SetReaderPlace(place);
X
xiexionghang 已提交
59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
    workers_[i]->SetRootScope(root_scope_);
    workers_[i]->CreateDeviceResource(main_program);  // Program
    workers_[i]->BindingDataFeedMemory();
  }
}

void MultiTrainer::Run() {
  VLOG(3) << "Going to run";
  for (int thidx = 0; thidx < thread_num_; ++thidx) {
    if (!debug_) {
      threads_.push_back(
          std::thread(&DeviceWorker::TrainFiles, workers_[thidx].get()));
    } else {
      threads_.push_back(std::thread(&DeviceWorker::TrainFilesWithProfiler,
                                     workers_[thidx].get()));
    }
  }
}

void MultiTrainer::Finalize() {
  for (auto& th : threads_) {
    th.join();
  }
  root_scope_->DropKids();
}

}  // end namespace framework
}  // end namespace paddle