device_context.cc 10.7 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;

28 29 30 31 32 33 34 35
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; }

Y
Yu Yang 已提交
36
platform::DeviceContext* DeviceContextPool::Get(const platform::Place& place) {
D
dzhwinter 已提交
37 38 39 40 41 42
  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
43
  return it->second.get();
D
dzhwinter 已提交
44 45
}

C
chengduozh 已提交
46 47 48 49 50 51 52 53 54 55
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
56 57 58
DeviceContextPool::DeviceContextPool(
    const std::vector<platform::Place>& places) {
  PADDLE_ENFORCE_GT(places.size(), 0);
Y
Yu Yang 已提交
59
  using PtrType = std::unique_ptr<DeviceContext>;
60
  std::set<Place> set;
Y
Yu Yang 已提交
61 62 63 64 65 66
  for (auto& p : places) {
    set.insert(p);
  }

  for (auto& p : set) {
    if (platform::is_cpu_place(p)) {
67
#ifdef PADDLE_WITH_MKLDNN
Y
Yu Yang 已提交
68 69
      device_contexts_.emplace(
          p, PtrType(new MKLDNNDeviceContext(boost::get<CPUPlace>(p))));
70
#else
Y
Yu Yang 已提交
71 72
      device_contexts_.emplace(
          p, PtrType(new CPUDeviceContext(boost::get<CPUPlace>(p))));
73
#endif
Y
Yu Yang 已提交
74
    } else if (platform::is_gpu_place(p)) {
D
dzhwinter 已提交
75
#ifdef PADDLE_WITH_CUDA
Y
Yu Yang 已提交
76 77
      device_contexts_.emplace(
          p, PtrType(new CUDADeviceContext(boost::get<CUDAPlace>(p))));
D
dzhwinter 已提交
78 79
#else
      PADDLE_THROW(
D
dzhwinter 已提交
80
          "'CUDAPlace' is not supported, Please re-compile with WITH_GPU "
D
dzhwinter 已提交
81
          "option");
C
chengduoZH 已提交
82 83 84 85 86 87 88 89 90 91
#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
92 93 94 95 96
#endif
    }
  }
}

97 98 99 100
CPUDeviceContext::CPUDeviceContext() {
  eigen_device_.reset(new Eigen::DefaultDevice());
}

D
dzhwinter 已提交
101
CPUDeviceContext::CPUDeviceContext(CPUPlace place) : place_(place) {
102 103 104 105 106 107 108
  eigen_device_.reset(new Eigen::DefaultDevice());
}

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

D
dzhwinter 已提交
109
Place CPUDeviceContext::GetPlace() const { return place_; }
110

111
#ifdef PADDLE_WITH_CUDA
112

Q
init  
qijun 已提交
113 114 115 116 117 118 119
class EigenCudaStreamDevice : public Eigen::StreamInterface {
 public:
  EigenCudaStreamDevice() : scratch_(nullptr), semaphore_(nullptr) {
    Eigen::initializeDeviceProp();
  }
  ~EigenCudaStreamDevice() override {}

D
dzhwinter 已提交
120
  void Reinitialize(const cudaStream_t* cuda_stream, CUDAPlace place) {
Q
init  
qijun 已提交
121 122 123 124 125 126 127 128 129 130 131 132
    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
133
    return paddle::memory::Alloc(place_, num_bytes);
Q
init  
qijun 已提交
134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158
  }

  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
159
  CUDAPlace place_;
Q
init  
qijun 已提交
160 161
  const cudaStream_t* stream_;         // not owned;
  const cudaDeviceProp* device_prop_;  // not owned;
Q
qijun 已提交
162
  mutable void* scratch_;
Q
init  
qijun 已提交
163 164 165
  mutable unsigned int* semaphore_;
};

166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
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
202
    workspace_ = paddle::memory::Alloc(place_, required_workspace_len);
203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
    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) {
218
  SetDeviceId(place_.device);
C
chengduo 已提交
219 220 221
  compute_capability_ = GetCUDAComputeCapability(place_.device);
  multi_process_ = GetCUDAMultiProcessors(place_.device);
  max_threads_per_mp_ = GetCUDAMaxThreadsPerMultiProcessor(place_.device);
Q
init  
qijun 已提交
222 223 224
  PADDLE_ENFORCE(cudaStreamCreate(&stream_));
  eigen_stream_.reset(new EigenCudaStreamDevice());
  eigen_stream_->Reinitialize(&stream_, place);
225
  eigen_device_.reset(new Eigen::GpuDevice(eigen_stream_.get()));
226 227
  PADDLE_ENFORCE(dynload::cublasCreate(&cublas_handle_));
  PADDLE_ENFORCE(dynload::cublasSetStream(cublas_handle_, stream_));
D
dzhwinter 已提交
228
  if (dynload::HasCUDNN()) {
229
    cudnn_holder_.reset(new CudnnHolder(&stream_, place));
D
dzhwinter 已提交
230
  }
S
sneaxiy 已提交
231

C
chengduo 已提交
232 233 234 235 236 237 238 239 240 241
  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
242
  callback_manager_.reset(new StreamCallbackManager(stream_));
243 244 245 246
}

CUDADeviceContext::~CUDADeviceContext() {
  SetDeviceId(place_.device);
L
liaogang 已提交
247
  Wait();
S
sneaxiy 已提交
248
  WaitStreamCallback();
249
  PADDLE_ENFORCE(dynload::cublasDestroy(cublas_handle_));
250 251
  eigen_stream_.reset();
  eigen_device_.reset();
Q
init  
qijun 已提交
252
  PADDLE_ENFORCE(cudaStreamDestroy(stream_));
253 254
}

L
liaogang 已提交
255
Place CUDADeviceContext::GetPlace() const { return place_; }
256

L
liaogang 已提交
257
void CUDADeviceContext::Wait() const {
Q
init  
qijun 已提交
258
  PADDLE_ENFORCE(cudaStreamSynchronize(stream_));
259 260 261
  PADDLE_ENFORCE(cudaGetLastError());
}

K
Kexin Zhao 已提交
262
int CUDADeviceContext::GetComputeCapability() const {
C
chengduo 已提交
263
  return compute_capability_;
K
Kexin Zhao 已提交
264 265
}

266
int CUDADeviceContext::GetMaxPhysicalThreadCount() const {
C
chengduo 已提交
267
  return multi_process_ * max_threads_per_mp_;
268 269
}

270 271 272 273
Eigen::GpuDevice* CUDADeviceContext::eigen_device() const {
  return eigen_device_.get();
}

274
cublasHandle_t CUDADeviceContext::cublas_handle() const {
275 276 277
  return cublas_handle_;
}

278 279 280 281 282 283 284 285
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);
}
286

287
cudaStream_t CUDADeviceContext::stream() const { return stream_; }
Q
qijun 已提交
288

C
chengduoZH 已提交
289 290 291 292 293 294 295 296 297 298 299 300 301 302
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
303
#endif
Q
qijun 已提交
304

T
tensor-tang 已提交
305 306
#ifdef PADDLE_WITH_MKLDNN
MKLDNNDeviceContext::MKLDNNDeviceContext(CPUPlace place)
307 308 309
    : 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
310 311
}

312 313
void MKLDNNDeviceContext::SetBlob(const std::string& name,
                                  std::shared_ptr<void> data) const {
314 315 316 317
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;

  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
318

319
  std::lock_guard<std::mutex> lock(*p_mutex_.get());
T
tensor-tang 已提交
320

321 322 323 324 325 326 327
  // 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;
328
  } else {
329
    pBlob = map_it->second;
330
  }
T
tensor-tang 已提交
331

332 333 334 335 336 337 338 339 340 341
  // 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
342
  return;
T
tensor-tang 已提交
343 344
}

345 346
std::shared_ptr<void> MKLDNNDeviceContext::GetBlob(
    const std::string& name) const {
347 348
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;
T
tensor-tang 已提交
349

350
  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
351

352 353 354 355 356 357 358 359 360 361 362
  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;
363

364 365
  // lock will be automatically released when out of scope
  return key_it->second;
T
tensor-tang 已提交
366 367 368 369
}

#endif

Q
qijun 已提交
370
}  // namespace platform
Q
qijun 已提交
371
}  // namespace paddle