device_context.cc 10.9 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>

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

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
}

C
chengduozh 已提交
38 39 40 41 42 43 44 45 46 47
const std::vector<const DeviceContext*>
DeviceContextPool::GetAllDeviceContexts() const {
  std::vector<const DeviceContext*> all_device_ctx;
  all_device_ctx.reserve(device_contexts_.size());
  for (auto& dev_ctx : device_contexts_) {
    all_device_ctx.emplace_back(dev_ctx.second.get());
  }
  return all_device_ctx;
}

D
dzhwinter 已提交
48 49 50
DeviceContextPool::DeviceContextPool(
    const std::vector<platform::Place>& places) {
  PADDLE_ENFORCE_GT(places.size(), 0);
Y
Yu Yang 已提交
51
  using PtrType = std::unique_ptr<DeviceContext>;
52
  std::set<Place> set;
Y
Yu Yang 已提交
53 54 55
  for (auto& p : places) {
    set.insert(p);
  }
D
dzhwinter 已提交
56
  VLOG(3) << "pool start";
Y
Yu Yang 已提交
57 58
  for (auto& p : set) {
    if (platform::is_cpu_place(p)) {
59
#ifdef PADDLE_WITH_MKLDNN
Y
Yu Yang 已提交
60 61
      device_contexts_.emplace(
          p, PtrType(new MKLDNNDeviceContext(boost::get<CPUPlace>(p))));
62
#else
D
dzhwinter 已提交
63
      VLOG(3) << "cpu context start";
Y
Yu Yang 已提交
64 65
      device_contexts_.emplace(
          p, PtrType(new CPUDeviceContext(boost::get<CPUPlace>(p))));
66
#endif
Y
Yu Yang 已提交
67
    } else if (platform::is_gpu_place(p)) {
D
dzhwinter 已提交
68
#ifdef PADDLE_WITH_CUDA
D
dzhwinter 已提交
69
      VLOG(3) << "gpu context start";
Y
Yu Yang 已提交
70 71
      device_contexts_.emplace(
          p, PtrType(new CUDADeviceContext(boost::get<CUDAPlace>(p))));
D
dzhwinter 已提交
72 73
#else
      PADDLE_THROW(
D
dzhwinter 已提交
74
          "'CUDAPlace' is not supported, Please re-compile with WITH_GPU "
D
dzhwinter 已提交
75
          "option");
C
chengduoZH 已提交
76 77 78
#endif
    } else if (platform::is_cuda_pinned_place(p)) {
#ifdef PADDLE_WITH_CUDA
D
dzhwinter 已提交
79
      VLOG(3) << "gpu pin start";
C
chengduoZH 已提交
80 81 82 83 84 85 86
      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
87 88
#endif
    }
D
dzhwinter 已提交
89
    VLOG(3) << "pool finish";
D
dzhwinter 已提交
90 91 92
  }
}

93 94 95 96
CPUDeviceContext::CPUDeviceContext() {
  eigen_device_.reset(new Eigen::DefaultDevice());
}

D
dzhwinter 已提交
97
CPUDeviceContext::CPUDeviceContext(CPUPlace place) : place_(place) {
98 99 100 101 102 103 104
  eigen_device_.reset(new Eigen::DefaultDevice());
}

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

D
dzhwinter 已提交
105
Place CPUDeviceContext::GetPlace() const { return place_; }
106

107
#ifdef PADDLE_WITH_CUDA
108

Q
init  
qijun 已提交
109 110 111 112 113 114 115
class EigenCudaStreamDevice : public Eigen::StreamInterface {
 public:
  EigenCudaStreamDevice() : scratch_(nullptr), semaphore_(nullptr) {
    Eigen::initializeDeviceProp();
  }
  ~EigenCudaStreamDevice() override {}

D
dzhwinter 已提交
116
  void Reinitialize(const cudaStream_t* cuda_stream, CUDAPlace place) {
Q
init  
qijun 已提交
117 118 119 120 121 122 123 124 125 126 127 128
    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
129
    return paddle::memory::Alloc(place_, num_bytes);
Q
init  
qijun 已提交
130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
  }

  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
155
  CUDAPlace place_;
Q
init  
qijun 已提交
156 157
  const cudaStream_t* stream_;         // not owned;
  const cudaDeviceProp* device_prop_;  // not owned;
Q
qijun 已提交
158
  mutable void* scratch_;
Q
init  
qijun 已提交
159 160 161
  mutable unsigned int* semaphore_;
};

162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
class CudnnHolder {
 public:
  CudnnHolder(const cudaStream_t* stream, const CUDAPlace& place)
      : workspace_(nullptr), workspace_len_(0), stream_(stream), place_(place) {
    PADDLE_ENFORCE(dynload::cudnnCreate(&cudnn_handle_));
    PADDLE_ENFORCE(dynload::cudnnSetStream(cudnn_handle_, *stream_));
  }

  cudnnHandle_t cudnn_handle() const { return cudnn_handle_; }

  void RunFunc(const std::function<void(void*)>& cudnn_func,
               size_t required_workspace_len) {
    std::lock_guard<std::mutex> lock(mtx_);
    if (required_workspace_len > workspace_len_) {
      ReallocateWorkspace(required_workspace_len);
    }
    cudnn_func(workspace_);
  }

  ~CudnnHolder() {
    PADDLE_ENFORCE(dynload::cudnnDestroy(cudnn_handle_));
    if (workspace_ != nullptr) {
      paddle::memory::Free(place_, workspace_);
    }
  }

 private:
  void ReallocateWorkspace(size_t required_workspace_len) {
    if (required_workspace_len <= workspace_len_) {
      return;
    }
    if (workspace_ != nullptr) {
      // Maybe someone is using the current workspace
      PADDLE_ENFORCE(cudaStreamSynchronize(*stream_));
      paddle::memory::Free(place_, workspace_);
    }
F
fengjiayi 已提交
198
    workspace_ = paddle::memory::Alloc(place_, required_workspace_len);
199 200 201 202 203 204 205 206 207 208 209 210 211 212 213
    workspace_len_ = required_workspace_len;
  }

  cudnnHandle_t cudnn_handle_;
  void* workspace_;
  size_t workspace_len_;

  const cudaStream_t* stream_;  // not owned;
  const CUDAPlace place_;

  std::mutex mtx_;
};

CUDADeviceContext::CUDADeviceContext(CUDAPlace place)
    : place_(place), cudnn_holder_(nullptr) {
214
  SetDeviceId(place_.device);
C
chengduo 已提交
215 216 217
  compute_capability_ = GetCUDAComputeCapability(place_.device);
  multi_process_ = GetCUDAMultiProcessors(place_.device);
  max_threads_per_mp_ = GetCUDAMaxThreadsPerMultiProcessor(place_.device);
Q
init  
qijun 已提交
218 219 220
  PADDLE_ENFORCE(cudaStreamCreate(&stream_));
  eigen_stream_.reset(new EigenCudaStreamDevice());
  eigen_stream_->Reinitialize(&stream_, place);
221
  eigen_device_.reset(new Eigen::GpuDevice(eigen_stream_.get()));
D
dzhwinter 已提交
222 223
  PADDLE_ENFORCE(dynload::cublasCreate(&cublas_handle_));
  PADDLE_ENFORCE(dynload::cublasSetStream(cublas_handle_, stream_));
D
dzhwinter 已提交
224
  if (dynload::HasCUDNN()) {
225
    cudnn_holder_.reset(new CudnnHolder(&stream_, place));
D
dzhwinter 已提交
226
  }
S
sneaxiy 已提交
227

C
chengduo 已提交
228 229 230 231 232 233 234 235 236
  driver_version_ = GetCUDADriverVersion(place_.device);
  runtime_version_ = GetCUDARuntimeVersion(place_.device);

  LOG(INFO) << "device: " << place_.device
            << ", CUDA Capability: " << compute_capability_
            << ", Driver Version: " << driver_version_ / 1000 << "."
            << (driver_version_ % 100) / 10
            << ", Runtime Version: " << runtime_version_ / 1000 << "."
            << (runtime_version_ % 100) / 10;
D
dzhwinter 已提交
237

D
dzhwinter 已提交
238
#ifndef _WIN32
S
sneaxiy 已提交
239
  callback_manager_.reset(new StreamCallbackManager(stream_));
D
dzhwinter 已提交
240
#endif  // NOT WIN32
241 242 243 244
}

CUDADeviceContext::~CUDADeviceContext() {
  SetDeviceId(place_.device);
L
liaogang 已提交
245
  Wait();
S
sneaxiy 已提交
246
  WaitStreamCallback();
247
  PADDLE_ENFORCE(dynload::cublasDestroy(cublas_handle_));
248 249
  eigen_stream_.reset();
  eigen_device_.reset();
Q
init  
qijun 已提交
250
  PADDLE_ENFORCE(cudaStreamDestroy(stream_));
251 252
}

L
liaogang 已提交
253
Place CUDADeviceContext::GetPlace() const { return place_; }
254

L
liaogang 已提交
255
void CUDADeviceContext::Wait() const {
Q
init  
qijun 已提交
256
  PADDLE_ENFORCE(cudaStreamSynchronize(stream_));
257 258 259
  PADDLE_ENFORCE(cudaGetLastError());
}

K
Kexin Zhao 已提交
260
int CUDADeviceContext::GetComputeCapability() const {
C
chengduo 已提交
261
  return compute_capability_;
K
Kexin Zhao 已提交
262 263
}

264
int CUDADeviceContext::GetMaxPhysicalThreadCount() const {
C
chengduo 已提交
265
  return multi_process_ * max_threads_per_mp_;
266 267
}

268 269 270 271
Eigen::GpuDevice* CUDADeviceContext::eigen_device() const {
  return eigen_device_.get();
}

272
cublasHandle_t CUDADeviceContext::cublas_handle() const {
273 274 275
  return cublas_handle_;
}

276 277 278 279 280 281 282 283
cudnnHandle_t CUDADeviceContext::cudnn_handle() const {
  return cudnn_holder_->cudnn_handle();
}

void CUDADeviceContext::RunCudnnFuncWithWorkspace(
    const std::function<void(void*)>& cudnn_func, size_t workspace_len) const {
  cudnn_holder_->RunFunc(cudnn_func, workspace_len);
}
284

285
cudaStream_t CUDADeviceContext::stream() const { return stream_; }
Q
qijun 已提交
286

C
chengduoZH 已提交
287 288 289 290 291 292 293 294 295 296 297 298 299 300
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
301
#endif
Q
qijun 已提交
302

T
tensor-tang 已提交
303 304
#ifdef PADDLE_WITH_MKLDNN
MKLDNNDeviceContext::MKLDNNDeviceContext(CPUPlace place)
305 306 307
    : CPUDeviceContext(place), engine_(mkldnn::engine::cpu, 0), p_blobmap_() {
  p_blobmap_.reset(new BlobMap());
  p_mutex_.reset(new std::mutex());
T
tensor-tang 已提交
308 309
}

S
Sylwester Fraczek 已提交
310 311 312 313 314 315 316 317
namespace {
// Current thread's id.
thread_local int cur_thread_id = 0;
}

void set_cur_thread_id(int tid) { cur_thread_id = tid; }
int get_cur_thread_id(void) { return cur_thread_id; }

318 319
void MKLDNNDeviceContext::SetBlob(const std::string& name,
                                  std::shared_ptr<void> data) const {
320 321 322 323
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;

  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
324

325
  std::lock_guard<std::mutex> lock(*p_mutex_.get());
T
tensor-tang 已提交
326

327 328 329 330 331 332 333
  // Find KeyBlob for current thread
  auto map_it = pMap->find(tid);

  if (map_it == pMap->end()) {
    // 1st time to set blob in current thread
    pBlob = std::shared_ptr<KeyBlob>(new KeyBlob());
    (*pMap)[tid] = pBlob;
334
  } else {
335
    pBlob = map_it->second;
336
  }
T
tensor-tang 已提交
337

338 339
  // Find Key in found (or newly created) KeyBlob
  auto key_it = pBlob->find(name);
T
tensor-tang 已提交
340

341 342
  if (key_it == pBlob->end()) {
    (*pBlob)[name] = data;  // create new blob
343
  } else {
344
    key_it->second = data;  // set data to existing blob
345
  }
T
tensor-tang 已提交
346

347
  // lock will be automatically released when out of scope
348
  return;
T
tensor-tang 已提交
349 350
}

351 352
std::shared_ptr<void> MKLDNNDeviceContext::GetBlob(
    const std::string& name) const {
353 354
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;
T
tensor-tang 已提交
355

356
  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
357

358 359 360 361 362 363 364 365 366 367 368
  std::lock_guard<std::mutex> lock(*p_mutex_.get());

  // Find KeyBlob for current thread firstly
  auto map_it = pMap->find(tid);
  if (map_it == pMap->end()) return nullptr;
  pBlob = map_it->second;

  // Find Blob via name
  auto key_it = pBlob->find(name);

  if (key_it == pBlob->end()) return nullptr;
369

370 371
  // lock will be automatically released when out of scope
  return key_it->second;
T
tensor-tang 已提交
372 373 374 375
}

#endif

Q
qijun 已提交
376
}  // namespace platform
Q
qijun 已提交
377
}  // namespace paddle