device_context.cc 11.0 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
#include <set>
13
#include <string>
Y
Yu Yang 已提交
14
#include <unordered_set>
15 16
#include <vector>

Y
Yi Wang 已提交
17
#include "paddle/fluid/memory/memory.h"
18 19
#ifdef PADDLE_WITH_CUDA
#include "paddle/fluid/framework/rw_lock.h"
S
sneaxiy 已提交
20
#include "paddle/fluid/platform/cuda_device_guard.h"
21
#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 {
S
sneaxiy 已提交
123 124 125
    if (UNLIKELY(num_bytes == 0)) {
      return nullptr;
    }
Y
Yu Yang 已提交
126 127
    auto buf = paddle::memory::Alloc(place_, num_bytes,
                                     memory::Allocator::kScratchpad);
128
    void* retv = buf->ptr();
S
sneaxiy 已提交
129 130 131 132
    {
      std::lock_guard<std::mutex> lock(mtx_);
      allocations_.emplace(retv, std::move(buf));
    }
133
    return retv;
Q
init  
qijun 已提交
134 135
  }

S
sneaxiy 已提交
136 137 138 139 140 141
  void deallocate(void* buffer) const override {
    if (LIKELY(buffer)) {
      std::lock_guard<std::mutex> lock(mtx_);
      allocations_.erase(buffer);
    }
  }
Q
init  
qijun 已提交
142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161

  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
162
  CUDAPlace place_;
Q
init  
qijun 已提交
163 164
  const cudaStream_t* stream_;         // not owned;
  const cudaDeviceProp* device_prop_;  // not owned;
Q
qijun 已提交
165
  mutable void* scratch_;
Q
init  
qijun 已提交
166
  mutable unsigned int* semaphore_;
S
sneaxiy 已提交
167
  mutable std::mutex mtx_;  // to protect allocations_
Y
Yu Yang 已提交
168
  mutable std::unordered_map<void*, memory::AllocationPtr> allocations_;
Q
init  
qijun 已提交
169 170
};

S
sneaxiy 已提交
171
CudnnHolder::CudnnHolder(const cudaStream_t* stream, const CUDAPlace& place)
Y
Yu Yang 已提交
172
    : workspace_(nullptr), stream_(stream), place_(place) {
S
sneaxiy 已提交
173 174 175
  PADDLE_ENFORCE(dynload::cudnnCreate(&cudnn_handle_));
  PADDLE_ENFORCE(dynload::cudnnSetStream(cudnn_handle_, *stream_));
}
176

S
sneaxiy 已提交
177 178
CudnnHolder::~CudnnHolder() {
  PADDLE_ENFORCE(dynload::cudnnDestroy(cudnn_handle_));
S
sneaxiy 已提交
179
}
180

S
sneaxiy 已提交
181
void CudnnHolder::ReallocateWorkspace(size_t required_workspace_len) {
Y
Yu Yang 已提交
182
  if (required_workspace_len <= WorkspaceSize()) {
S
sneaxiy 已提交
183
    return;
Y
Yu Yang 已提交
184
  }
S
sneaxiy 已提交
185 186 187
  if (workspace_ != nullptr) {
    // Maybe someone is using the current workspace
    PADDLE_ENFORCE(cudaStreamSynchronize(*stream_));
Y
Yu Yang 已提交
188
    workspace_.reset();
189
  }
Y
Yu Yang 已提交
190 191
  workspace_ = paddle::memory::Alloc(place_, required_workspace_len,
                                     paddle::memory::Allocator::kScratchpad);
S
sneaxiy 已提交
192
}
193 194 195

CUDADeviceContext::CUDADeviceContext(CUDAPlace place)
    : place_(place), cudnn_holder_(nullptr) {
Y
Yu Yang 已提交
196
  CUDADeviceGuard guard(place_.device);
C
chengduo 已提交
197 198 199
  compute_capability_ = GetCUDAComputeCapability(place_.device);
  multi_process_ = GetCUDAMultiProcessors(place_.device);
  max_threads_per_mp_ = GetCUDAMaxThreadsPerMultiProcessor(place_.device);
Q
init  
qijun 已提交
200 201 202
  PADDLE_ENFORCE(cudaStreamCreate(&stream_));
  eigen_stream_.reset(new EigenCudaStreamDevice());
  eigen_stream_->Reinitialize(&stream_, place);
203
  eigen_device_.reset(new Eigen::GpuDevice(eigen_stream_.get()));
204 205
  PADDLE_ENFORCE(dynload::cublasCreate(&cublas_handle_));
  PADDLE_ENFORCE(dynload::cublasSetStream(cublas_handle_, stream_));
D
dzhwinter 已提交
206
  if (dynload::HasCUDNN()) {
207
    cudnn_holder_.reset(new CudnnHolder(&stream_, place));
D
dzhwinter 已提交
208
  }
S
sneaxiy 已提交
209

C
chengduo 已提交
210 211 212
  driver_version_ = GetCUDADriverVersion(place_.device);
  runtime_version_ = GetCUDARuntimeVersion(place_.device);

213 214 215 216 217 218
  LOG_FIRST_N(WARNING, 1) << "Please NOTE: 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;
219 220 221 222
  size_t cudnn_dso_ver = dynload::cudnnGetVersion();
  LOG_FIRST_N(WARNING, 1) << "device: " << place_.device
                          << ", cuDNN Version: " << cudnn_dso_ver / 1000 << "."
                          << (cudnn_dso_ver % 100) / 10 << ".";
S
sneaxiy 已提交
223
  callback_manager_.reset(new StreamCallbackManager(stream_));
224 225 226 227
}

CUDADeviceContext::~CUDADeviceContext() {
  SetDeviceId(place_.device);
L
liaogang 已提交
228
  Wait();
S
sneaxiy 已提交
229
  WaitStreamCallback();
230
  PADDLE_ENFORCE(dynload::cublasDestroy(cublas_handle_));
231 232
  eigen_stream_.reset();
  eigen_device_.reset();
Q
init  
qijun 已提交
233
  PADDLE_ENFORCE(cudaStreamDestroy(stream_));
234 235
}

L
liaogang 已提交
236
Place CUDADeviceContext::GetPlace() const { return place_; }
237

L
liaogang 已提交
238
void CUDADeviceContext::Wait() const {
Q
init  
qijun 已提交
239
  PADDLE_ENFORCE(cudaStreamSynchronize(stream_));
240 241 242
  PADDLE_ENFORCE(cudaGetLastError());
}

K
Kexin Zhao 已提交
243
int CUDADeviceContext::GetComputeCapability() const {
C
chengduo 已提交
244
  return compute_capability_;
K
Kexin Zhao 已提交
245 246
}

247
int CUDADeviceContext::GetMaxPhysicalThreadCount() const {
C
chengduo 已提交
248
  return multi_process_ * max_threads_per_mp_;
249 250
}

251 252 253 254
Eigen::GpuDevice* CUDADeviceContext::eigen_device() const {
  return eigen_device_.get();
}

255
cublasHandle_t CUDADeviceContext::cublas_handle() const {
256 257 258
  return cublas_handle_;
}

259 260 261 262
cudnnHandle_t CUDADeviceContext::cudnn_handle() const {
  return cudnn_holder_->cudnn_handle();
}

S
sneaxiy 已提交
263 264
CudnnWorkspaceHandle CUDADeviceContext::cudnn_workspace_handle() const {
  return CudnnWorkspaceHandle(cudnn_holder_.get());
265
}
266

267
cudaStream_t CUDADeviceContext::stream() const { return stream_; }
Q
qijun 已提交
268

C
chengduoZH 已提交
269 270 271 272 273 274 275 276 277 278 279 280 281 282
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
283
#endif
Q
qijun 已提交
284

T
tensor-tang 已提交
285 286
#ifdef PADDLE_WITH_MKLDNN
MKLDNNDeviceContext::MKLDNNDeviceContext(CPUPlace place)
287 288 289
    : 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
290 291
}

S
Sylwester Fraczek 已提交
292 293 294 295 296 297 298 299
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; }

300 301
void MKLDNNDeviceContext::SetBlob(const std::string& name,
                                  std::shared_ptr<void> data) const {
302 303 304 305
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;

  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
306

307
  std::lock_guard<std::mutex> lock(*p_mutex_.get());
T
tensor-tang 已提交
308

309 310 311 312 313 314 315
  // 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;
316
  } else {
317
    pBlob = map_it->second;
318
  }
T
tensor-tang 已提交
319

320 321 322 323 324 325 326 327 328 329
  // 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
330
  return;
T
tensor-tang 已提交
331 332
}

333 334
std::shared_ptr<void> MKLDNNDeviceContext::GetBlob(
    const std::string& name) const {
335 336
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;
T
tensor-tang 已提交
337

338
  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
339

340 341 342 343 344 345 346 347 348 349 350
  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;
351

352 353
  // lock will be automatically released when out of scope
  return key_it->second;
T
tensor-tang 已提交
354 355 356 357
}

#endif

Q
qijun 已提交
358
}  // namespace platform
Q
qijun 已提交
359
}  // namespace paddle