heter_resource.h 2.1 KB
Newer Older
T
Thunderbrook 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
/* Copyright (c) 2020 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
#include <cstddef>
#include <map>
#include <memory>
#include <vector>
#include "paddle/fluid/platform/cuda_device_guard.h"
#include "paddle/fluid/platform/enforce.h"

T
Thunderbrook 已提交
23
#ifdef PADDLE_WITH_HETERPS
T
Thunderbrook 已提交
24 25 26 27 28 29

namespace paddle {
namespace framework {

class GPUResource {
 public:
30
  GPUResource(std::vector<int>& device_id, int index);
T
Thunderbrook 已提交
31 32 33 34 35 36
  virtual ~GPUResource();
  GPUResource(const GPUResource&) = delete;
  GPUResource& operator=(const GPUResource&) = delete;

  int dev_id() const { return dev_id_; }
  int index() const { return index_; }
37 38 39
  gpuStream_t local_stream(int num) { return local_streams_[num]; }
  gpuStream_t remote_stream() { return remote_stream_; }
  gpuStream_t comm_stream(int num) { return comm_streams_[num]; }
T
Thunderbrook 已提交
40 41 42

  int dev_id_;
  int index_;
43
  std::vector<int> dev_ids_;
44 45 46
  gpuStream_t remote_stream_;
  std::vector<gpuStream_t> local_streams_;
  std::vector<gpuStream_t> comm_streams_;
T
Thunderbrook 已提交
47 48 49 50 51 52 53 54 55 56 57 58
};

class HeterPsResource {
 public:
  HeterPsResource(const std::vector<int>& dev_ids);
  HeterPsResource(const HeterPsResource&) = delete;
  HeterPsResource& operator=(const HeterPsResource&) = delete;
  virtual ~HeterPsResource() {}
  void enable_p2p();
  int total_gpu();
  int get_index_by_devid(int devid);
  int dev_id(int num);
59 60 61
  gpuStream_t local_stream(int gpu_num, int stream_num);
  gpuStream_t remote_stream(int gpu_num);
  gpuStream_t comm_stream(int gpu_num, int stream_num);
T
Thunderbrook 已提交
62 63 64 65 66 67 68 69 70

  std::vector<std::shared_ptr<GPUResource>> resources_;
  std::vector<int> dev_ids_;
  std::map<int, int> devid_2_index_;
};

}  // end namespace framework
}  // end namespace paddle
#endif