device_context.cc 8.9 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
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. */
Y
Yi Wang 已提交
11
#include "paddle/fluid/platform/device_context.h"
12

13
#include <set>
14
#include <string>
Y
Yu Yang 已提交
15
#include <unordered_set>
16 17
#include <vector>

F
fengjiayi 已提交
18
#include "paddle/fluid/memory/memory.h"
F
fengjiayi 已提交
19
#ifdef PADDLE_WITH_CUDA
F
fengjiayi 已提交
20
#include "paddle/fluid/framework/rw_lock.h"
F
fengjiayi 已提交
21 22
#endif

Q
qijun 已提交
23 24 25
namespace paddle {
namespace platform {

D
dzhwinter 已提交
26 27
DeviceContextPool* DeviceContextPool::pool = nullptr;

Y
Yu Yang 已提交
28
platform::DeviceContext* DeviceContextPool::Get(const platform::Place& place) {
D
dzhwinter 已提交
29 30 31 32 33 34
  auto it = device_contexts_.find(place);
  if (it == device_contexts_.end()) {
    PADDLE_THROW(
        "'Place' is not supported, Please re-compile with WITH_GPU "
        "option");
  }
Y
Yu Yang 已提交
35
  return it->second.get();
D
dzhwinter 已提交
36 37 38 39 40
}

DeviceContextPool::DeviceContextPool(
    const std::vector<platform::Place>& places) {
  PADDLE_ENFORCE_GT(places.size(), 0);
Y
Yu Yang 已提交
41
  using PtrType = std::unique_ptr<DeviceContext>;
42
  std::set<Place> set;
Y
Yu Yang 已提交
43 44 45 46 47 48
  for (auto& p : places) {
    set.insert(p);
  }

  for (auto& p : set) {
    if (platform::is_cpu_place(p)) {
49
#ifdef PADDLE_WITH_MKLDNN
Y
Yu Yang 已提交
50 51
      device_contexts_.emplace(
          p, PtrType(new MKLDNNDeviceContext(boost::get<CPUPlace>(p))));
52
#else
Y
Yu Yang 已提交
53 54
      device_contexts_.emplace(
          p, PtrType(new CPUDeviceContext(boost::get<CPUPlace>(p))));
55
#endif
Y
Yu Yang 已提交
56
    } else if (platform::is_gpu_place(p)) {
D
dzhwinter 已提交
57
#ifdef PADDLE_WITH_CUDA
Y
Yu Yang 已提交
58 59
      device_contexts_.emplace(
          p, PtrType(new CUDADeviceContext(boost::get<CUDAPlace>(p))));
D
dzhwinter 已提交
60 61
#else
      PADDLE_THROW(
D
dzhwinter 已提交
62
          "'CUDAPlace' is not supported, Please re-compile with WITH_GPU "
D
dzhwinter 已提交
63
          "option");
C
chengduoZH 已提交
64 65 66 67 68 69 70 71 72 73
#endif
    } else if (platform::is_cuda_pinned_place(p)) {
#ifdef PADDLE_WITH_CUDA
      device_contexts_.emplace(
          p,
          PtrType(new CUDAPinnedDeviceContext(boost::get<CUDAPinnedPlace>(p))));
#else
      PADDLE_THROW(
          "'CUDAPlace' is not supported, Please re-compile with WITH_GPU "
          "option");
D
dzhwinter 已提交
74 75 76 77 78
#endif
    }
  }
}

79 80 81 82
CPUDeviceContext::CPUDeviceContext() {
  eigen_device_.reset(new Eigen::DefaultDevice());
}

D
dzhwinter 已提交
83
CPUDeviceContext::CPUDeviceContext(CPUPlace place) : place_(place) {
84 85 86 87 88 89 90
  eigen_device_.reset(new Eigen::DefaultDevice());
}

Eigen::DefaultDevice* CPUDeviceContext::eigen_device() const {
  return eigen_device_.get();
}

D
dzhwinter 已提交
91
Place CPUDeviceContext::GetPlace() const { return place_; }
92

93
#ifdef PADDLE_WITH_CUDA
94

Q
init  
qijun 已提交
95 96 97 98 99 100 101
class EigenCudaStreamDevice : public Eigen::StreamInterface {
 public:
  EigenCudaStreamDevice() : scratch_(nullptr), semaphore_(nullptr) {
    Eigen::initializeDeviceProp();
  }
  ~EigenCudaStreamDevice() override {}

D
dzhwinter 已提交
102
  void Reinitialize(const cudaStream_t* cuda_stream, CUDAPlace place) {
Q
init  
qijun 已提交
103 104 105 106 107 108 109 110 111 112 113 114
    stream_ = cuda_stream;
    place_ = place;
    device_prop_ = &Eigen::m_deviceProperties[place.device];
  }

  const cudaStream_t& stream() const override { return *stream_; }

  const cudaDeviceProp& deviceProperties() const override {
    return *device_prop_;
  }

  void* allocate(size_t num_bytes) const override {
Q
qijun 已提交
115
    return paddle::memory::Alloc(place_, num_bytes);
Q
init  
qijun 已提交
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
  }

  void deallocate(void* buffer) const override {
    paddle::memory::Free(place_, buffer);
  }

  void* scratchpad() const override {
    if (scratch_ == NULL) {
      scratch_ = allocate(Eigen::kCudaScratchSize + sizeof(unsigned int));
    }
    return scratch_;
  }

  unsigned int* semaphore() const override {
    if (semaphore_ == NULL) {
      char* scratch =
          static_cast<char*>(scratchpad()) + Eigen::kCudaScratchSize;
      semaphore_ = reinterpret_cast<unsigned int*>(scratch);
      PADDLE_ENFORCE(
          cudaMemsetAsync(semaphore_, 0, sizeof(unsigned int), *stream_));
    }
    return semaphore_;
  }

 private:
D
dzhwinter 已提交
141
  CUDAPlace place_;
Q
init  
qijun 已提交
142 143
  const cudaStream_t* stream_;         // not owned;
  const cudaDeviceProp* device_prop_;  // not owned;
Q
qijun 已提交
144
  mutable void* scratch_;
Q
init  
qijun 已提交
145 146 147
  mutable unsigned int* semaphore_;
};

148 149 150
class CudnnHolder {
 public:
  CudnnHolder(const cudaStream_t* stream, const CUDAPlace& place)
F
fengjiayi 已提交
151
      : workspace_(nullptr), workspace_len_(0), stream_(stream), place_(place) {
152
    PADDLE_ENFORCE(dynload::cudnnCreate(&cudnn_handle_));
F
fengjiayi 已提交
153
    PADDLE_ENFORCE(dynload::cudnnSetStream(cudnn_handle_, *stream_));
154 155
  }

F
fengjiayi 已提交
156 157 158 159
  cudnnHandle_t cudnn_handle() const { return cudnn_handle_; }

  void RunFunc(const std::function<void(void*)>& cudnn_func,
               size_t required_workspace_len) {
F
fengjiayi 已提交
160
    std::lock_guard<std::mutex> lock(mtx_);
F
fengjiayi 已提交
161
    if (required_workspace_len > workspace_len_) {
F
fengjiayi 已提交
162
      ReallocateWorkspace(required_workspace_len);
163
    }
F
fengjiayi 已提交
164
    cudnn_func(workspace_);
165 166
  }

F
fengjiayi 已提交
167 168 169 170 171 172
  ~CudnnHolder() {
    PADDLE_ENFORCE(dynload::cudnnDestroy(cudnn_handle_));
    if (workspace_ != nullptr) {
      paddle::memory::Free(place_, workspace_);
    }
  }
173 174

 private:
F
fengjiayi 已提交
175
  void ReallocateWorkspace(size_t required_workspace_len) {
F
fengjiayi 已提交
176 177 178
    if (required_workspace_len <= workspace_len_) {
      return;
    }
F
fengjiayi 已提交
179
    void* new_workspace = paddle::memory::Alloc(place_, required_workspace_len);
F
fengjiayi 已提交
180 181 182 183 184 185
    if (workspace_ != nullptr) {
      // Maybe someone is using the current workspace
      PADDLE_ENFORCE(cudaStreamSynchronize(*stream_));
      paddle::memory::Free(place_, workspace_);
    }
    workspace_ = new_workspace;
F
fengjiayi 已提交
186
    workspace_len_ = required_workspace_len;
F
fengjiayi 已提交
187 188
  }

189 190 191 192 193 194
  cudnnHandle_t cudnn_handle_;
  void* workspace_;
  size_t workspace_len_;

  const cudaStream_t* stream_;  // not owned;
  const CUDAPlace place_;
F
fengjiayi 已提交
195

F
fengjiayi 已提交
196
  std::mutex mtx_;
197 198 199 200
};

CUDADeviceContext::CUDADeviceContext(CUDAPlace place)
    : place_(place), cudnn_holder_(nullptr) {
201
  SetDeviceId(place_.device);
K
Kexin Zhao 已提交
202
  compute_capability = GetCUDAComputeCapability(place_.device);
203 204
  multi_process = GetCUDAMultiProcessors(place_.device);
  max_threads_per_mp = GetCUDAMaxThreadsPerMultiProcessor(place_.device);
Q
init  
qijun 已提交
205 206 207
  PADDLE_ENFORCE(cudaStreamCreate(&stream_));
  eigen_stream_.reset(new EigenCudaStreamDevice());
  eigen_stream_->Reinitialize(&stream_, place);
208
  eigen_device_.reset(new Eigen::GpuDevice(eigen_stream_.get()));
209 210
  PADDLE_ENFORCE(dynload::cublasCreate(&cublas_handle_));
  PADDLE_ENFORCE(dynload::cublasSetStream(cublas_handle_, stream_));
D
dzhwinter 已提交
211
  if (dynload::HasCUDNN()) {
212
    cudnn_holder_.reset(new CudnnHolder(&stream_, place));
D
dzhwinter 已提交
213
  }
214 215 216 217
}

CUDADeviceContext::~CUDADeviceContext() {
  SetDeviceId(place_.device);
L
liaogang 已提交
218
  Wait();
219
  PADDLE_ENFORCE(dynload::cublasDestroy(cublas_handle_));
220 221
  eigen_stream_.reset();
  eigen_device_.reset();
Q
init  
qijun 已提交
222
  PADDLE_ENFORCE(cudaStreamDestroy(stream_));
223 224
}

L
liaogang 已提交
225
Place CUDADeviceContext::GetPlace() const { return place_; }
226

L
liaogang 已提交
227
void CUDADeviceContext::Wait() const {
Q
init  
qijun 已提交
228
  PADDLE_ENFORCE(cudaStreamSynchronize(stream_));
229 230 231
  PADDLE_ENFORCE(cudaGetLastError());
}

K
Kexin Zhao 已提交
232 233 234 235
int CUDADeviceContext::GetComputeCapability() const {
  return compute_capability;
}

236 237 238 239
int CUDADeviceContext::GetMaxPhysicalThreadCount() const {
  return multi_process * max_threads_per_mp;
}

240 241 242 243
Eigen::GpuDevice* CUDADeviceContext::eigen_device() const {
  return eigen_device_.get();
}

244
cublasHandle_t CUDADeviceContext::cublas_handle() const {
245 246 247
  return cublas_handle_;
}

248
cudnnHandle_t CUDADeviceContext::cudnn_handle() const {
F
fengjiayi 已提交
249
  return cudnn_holder_->cudnn_handle();
250 251
}

F
fengjiayi 已提交
252 253 254
void CUDADeviceContext::RunCudnnFuncWithWorkspace(
    const std::function<void(void*)>& cudnn_func, size_t workspace_len) const {
  cudnn_holder_->RunFunc(cudnn_func, workspace_len);
255
}
256

257
cudaStream_t CUDADeviceContext::stream() const { return stream_; }
Q
qijun 已提交
258

C
chengduoZH 已提交
259 260 261 262 263 264 265 266 267 268 269 270 271 272
CUDAPinnedDeviceContext::CUDAPinnedDeviceContext() {
  eigen_device_.reset(new Eigen::DefaultDevice());
}

CUDAPinnedDeviceContext::CUDAPinnedDeviceContext(CUDAPinnedPlace place)
    : place_(place) {
  eigen_device_.reset(new Eigen::DefaultDevice());
}

Eigen::DefaultDevice* CUDAPinnedDeviceContext::eigen_device() const {
  return eigen_device_.get();
}

Place CUDAPinnedDeviceContext::GetPlace() const { return place_; }
L
Luo Tao 已提交
273
#endif
Q
qijun 已提交
274

T
tensor-tang 已提交
275 276
#ifdef PADDLE_WITH_MKLDNN
MKLDNNDeviceContext::MKLDNNDeviceContext(CPUPlace place)
277 278
    : CPUDeviceContext(place), engine_(mkldnn::engine::cpu, 0), p_blobs_() {
  p_blobs_.reset(new std::unordered_map<std::string, std::shared_ptr<void>>());
T
tensor-tang 已提交
279 280
}

281 282 283 284
void MKLDNNDeviceContext::SetBlob(const std::string& name,
                                  std::shared_ptr<void> data) const {
  std::unordered_map<std::string, std::shared_ptr<void>>* p;
  p = p_blobs_.get();
T
tensor-tang 已提交
285

286
  auto it = p->find(name);
T
tensor-tang 已提交
287

288 289 290 291 292
  if (it == p->end()) {
    (*p)[name] = data;  // create new blob
  } else {
    it->second = data;  // set data to existing blob
  }
T
tensor-tang 已提交
293

294
  return;
T
tensor-tang 已提交
295 296
}

297 298 299 300
std::shared_ptr<void> MKLDNNDeviceContext::GetBlob(
    const std::string& name) const {
  std::unordered_map<std::string, std::shared_ptr<void>>* p;
  p = p_blobs_.get();
T
tensor-tang 已提交
301

302
  auto it = p->find(name);
T
tensor-tang 已提交
303

304 305
  if (it != p->end()) {
    return it->second;
T
tensor-tang 已提交
306
  }
307 308

  return nullptr;
T
tensor-tang 已提交
309 310 311 312
}

#endif

Q
qijun 已提交
313
}  // namespace platform
Q
qijun 已提交
314
}  // namespace paddle