device_context.cc 10.8 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");
  }
35
  return it->second.get().get();
D
dzhwinter 已提交
36 37
}

38 39 40 41 42 43 44 45 46 47 48
template <typename DevCtx, typename PlaceType>
inline void EmplaceDeviceContext(
    std::map<Place, std::shared_future<std::unique_ptr<DeviceContext>>>*
        map_ptr,
    platform::Place p) {
  using PtrType = std::unique_ptr<DeviceContext>;
  map_ptr->emplace(p, std::async(std::launch::deferred, [=] {
                     // lazy evaluation. i.e., only create device context at
                     // first `Get`
                     return PtrType(new DevCtx(boost::get<PlaceType>(p)));
                   }));
C
chengduozh 已提交
49 50
}

D
dzhwinter 已提交
51 52 53
DeviceContextPool::DeviceContextPool(
    const std::vector<platform::Place>& places) {
  PADDLE_ENFORCE_GT(places.size(), 0);
54
  std::set<Place> set;
Y
Yu Yang 已提交
55 56 57 58 59 60
  for (auto& p : places) {
    set.insert(p);
  }

  for (auto& p : set) {
    if (platform::is_cpu_place(p)) {
61
#ifdef PADDLE_WITH_MKLDNN
62
      EmplaceDeviceContext<MKLDNNDeviceContext, CPUPlace>(&device_contexts_, p);
63
#else
64
      EmplaceDeviceContext<CPUDeviceContext, CPUPlace>(&device_contexts_, p);
65
#endif
Y
Yu Yang 已提交
66
    } else if (platform::is_gpu_place(p)) {
D
dzhwinter 已提交
67
#ifdef PADDLE_WITH_CUDA
68
      EmplaceDeviceContext<CUDADeviceContext, CUDAPlace>(&device_contexts_, p);
D
dzhwinter 已提交
69 70
#else
      PADDLE_THROW(
D
dzhwinter 已提交
71
          "'CUDAPlace' is not supported, Please re-compile with WITH_GPU "
D
dzhwinter 已提交
72
          "option");
C
chengduoZH 已提交
73 74 75
#endif
    } else if (platform::is_cuda_pinned_place(p)) {
#ifdef PADDLE_WITH_CUDA
76 77
      EmplaceDeviceContext<CUDAPinnedDeviceContext, CUDAPinnedPlace>(
          &device_contexts_, p);
C
chengduoZH 已提交
78 79 80 81
#else
      PADDLE_THROW(
          "'CUDAPlace' is not supported, Please re-compile with WITH_GPU "
          "option");
D
dzhwinter 已提交
82 83 84 85 86
#endif
    }
  }
}

87 88 89 90
CPUDeviceContext::CPUDeviceContext() {
  eigen_device_.reset(new Eigen::DefaultDevice());
}

D
dzhwinter 已提交
91
CPUDeviceContext::CPUDeviceContext(CPUPlace place) : place_(place) {
92 93 94 95 96 97 98
  eigen_device_.reset(new Eigen::DefaultDevice());
}

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

D
dzhwinter 已提交
99
Place CPUDeviceContext::GetPlace() const { return place_; }
100

101
#ifdef PADDLE_WITH_CUDA
102

Q
init  
qijun 已提交
103 104 105 106 107 108 109
class EigenCudaStreamDevice : public Eigen::StreamInterface {
 public:
  EigenCudaStreamDevice() : scratch_(nullptr), semaphore_(nullptr) {
    Eigen::initializeDeviceProp();
  }
  ~EigenCudaStreamDevice() override {}

D
dzhwinter 已提交
110
  void Reinitialize(const cudaStream_t* cuda_stream, CUDAPlace place) {
Q
init  
qijun 已提交
111 112 113 114 115 116 117 118 119 120 121 122
    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
123
    return paddle::memory::Alloc(place_, num_bytes);
Q
init  
qijun 已提交
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148
  }

  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
149
  CUDAPlace place_;
Q
init  
qijun 已提交
150 151
  const cudaStream_t* stream_;         // not owned;
  const cudaDeviceProp* device_prop_;  // not owned;
Q
qijun 已提交
152
  mutable void* scratch_;
Q
init  
qijun 已提交
153 154 155
  mutable unsigned int* semaphore_;
};

156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191
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
192
    workspace_ = paddle::memory::Alloc(place_, required_workspace_len);
193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
    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) {
208
  SetDeviceId(place_.device);
C
chengduo 已提交
209 210 211
  compute_capability_ = GetCUDAComputeCapability(place_.device);
  multi_process_ = GetCUDAMultiProcessors(place_.device);
  max_threads_per_mp_ = GetCUDAMaxThreadsPerMultiProcessor(place_.device);
Q
init  
qijun 已提交
212 213 214
  PADDLE_ENFORCE(cudaStreamCreate(&stream_));
  eigen_stream_.reset(new EigenCudaStreamDevice());
  eigen_stream_->Reinitialize(&stream_, place);
215
  eigen_device_.reset(new Eigen::GpuDevice(eigen_stream_.get()));
216 217
  PADDLE_ENFORCE(dynload::cublasCreate(&cublas_handle_));
  PADDLE_ENFORCE(dynload::cublasSetStream(cublas_handle_, stream_));
D
dzhwinter 已提交
218
  if (dynload::HasCUDNN()) {
219
    cudnn_holder_.reset(new CudnnHolder(&stream_, place));
D
dzhwinter 已提交
220
  }
S
sneaxiy 已提交
221

C
chengduo 已提交
222 223 224 225 226 227 228 229 230 231
  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;

S
sneaxiy 已提交
232
  callback_manager_.reset(new StreamCallbackManager(stream_));
233 234 235 236
}

CUDADeviceContext::~CUDADeviceContext() {
  SetDeviceId(place_.device);
L
liaogang 已提交
237
  Wait();
S
sneaxiy 已提交
238
  WaitStreamCallback();
239
  PADDLE_ENFORCE(dynload::cublasDestroy(cublas_handle_));
240 241
  eigen_stream_.reset();
  eigen_device_.reset();
Q
init  
qijun 已提交
242
  PADDLE_ENFORCE(cudaStreamDestroy(stream_));
243 244
}

L
liaogang 已提交
245
Place CUDADeviceContext::GetPlace() const { return place_; }
246

L
liaogang 已提交
247
void CUDADeviceContext::Wait() const {
Q
init  
qijun 已提交
248
  PADDLE_ENFORCE(cudaStreamSynchronize(stream_));
249 250 251
  PADDLE_ENFORCE(cudaGetLastError());
}

K
Kexin Zhao 已提交
252
int CUDADeviceContext::GetComputeCapability() const {
C
chengduo 已提交
253
  return compute_capability_;
K
Kexin Zhao 已提交
254 255
}

256
int CUDADeviceContext::GetMaxPhysicalThreadCount() const {
C
chengduo 已提交
257
  return multi_process_ * max_threads_per_mp_;
258 259
}

260 261 262 263
Eigen::GpuDevice* CUDADeviceContext::eigen_device() const {
  return eigen_device_.get();
}

264
cublasHandle_t CUDADeviceContext::cublas_handle() const {
265 266 267
  return cublas_handle_;
}

268 269 270 271 272 273 274 275
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);
}
276

277
cudaStream_t CUDADeviceContext::stream() const { return stream_; }
Q
qijun 已提交
278

C
chengduoZH 已提交
279 280 281 282 283 284 285 286 287 288 289 290 291 292
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
293
#endif
Q
qijun 已提交
294

T
tensor-tang 已提交
295 296
#ifdef PADDLE_WITH_MKLDNN
MKLDNNDeviceContext::MKLDNNDeviceContext(CPUPlace place)
297 298 299
    : 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
300 301
}

S
Sylwester Fraczek 已提交
302 303 304 305 306 307 308 309
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; }

310 311
void MKLDNNDeviceContext::SetBlob(const std::string& name,
                                  std::shared_ptr<void> data) const {
312 313 314 315
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;

  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
316

317
  std::lock_guard<std::mutex> lock(*p_mutex_.get());
T
tensor-tang 已提交
318

319 320 321 322 323 324 325
  // 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;
326
  } else {
327
    pBlob = map_it->second;
328
  }
T
tensor-tang 已提交
329

330 331 332 333 334 335 336 337 338 339
  // Find Key in found (or newly created) KeyBlob
  auto key_it = pBlob->find(name);

  if (key_it == pBlob->end()) {
    (*pBlob)[name] = data;  // create new blob
  } else {
    key_it->second = data;  // set data to existing blob
  }

  // lock will be automatically released when out of scope
340
  return;
T
tensor-tang 已提交
341 342
}

343 344
std::shared_ptr<void> MKLDNNDeviceContext::GetBlob(
    const std::string& name) const {
345 346
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;
T
tensor-tang 已提交
347

348
  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
349

350 351 352 353 354 355 356 357 358 359 360
  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;
361

362 363
  // lock will be automatically released when out of scope
  return key_it->second;
T
tensor-tang 已提交
364 365 366 367
}

#endif

Q
qijun 已提交
368
}  // namespace platform
Q
qijun 已提交
369
}  // namespace paddle