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
#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 {
Y
Yu Yang 已提交
123 124
    auto buf = paddle::memory::Alloc(place_, num_bytes,
                                     memory::Allocator::kScratchpad);
125 126 127
    void* retv = buf->ptr();
    allocations_[buf->ptr()] = std::move(buf);
    return retv;
Q
init  
qijun 已提交
128 129
  }

S
sneaxiy 已提交
130
  void deallocate(void* buffer) const override { allocations_.erase(buffer); }
Q
init  
qijun 已提交
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150

  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
151
  CUDAPlace place_;
Q
init  
qijun 已提交
152 153
  const cudaStream_t* stream_;         // not owned;
  const cudaDeviceProp* device_prop_;  // not owned;
Q
qijun 已提交
154
  mutable void* scratch_;
Q
init  
qijun 已提交
155
  mutable unsigned int* semaphore_;
Y
Yu Yang 已提交
156
  mutable std::unordered_map<void*, memory::AllocationPtr> allocations_;
Q
init  
qijun 已提交
157 158
};

S
sneaxiy 已提交
159
CudnnHolder::CudnnHolder(const cudaStream_t* stream, const CUDAPlace& place)
Y
Yu Yang 已提交
160
    : workspace_(nullptr), stream_(stream), place_(place) {
S
sneaxiy 已提交
161 162 163
  PADDLE_ENFORCE(dynload::cudnnCreate(&cudnn_handle_));
  PADDLE_ENFORCE(dynload::cudnnSetStream(cudnn_handle_, *stream_));
}
164

S
sneaxiy 已提交
165 166
CudnnHolder::~CudnnHolder() {
  PADDLE_ENFORCE(dynload::cudnnDestroy(cudnn_handle_));
S
sneaxiy 已提交
167
}
168

S
sneaxiy 已提交
169
void CudnnHolder::ReallocateWorkspace(size_t required_workspace_len) {
Y
Yu Yang 已提交
170
  if (required_workspace_len <= WorkspaceSize()) {
S
sneaxiy 已提交
171
    return;
Y
Yu Yang 已提交
172
  }
S
sneaxiy 已提交
173 174 175
  if (workspace_ != nullptr) {
    // Maybe someone is using the current workspace
    PADDLE_ENFORCE(cudaStreamSynchronize(*stream_));
Y
Yu Yang 已提交
176
    workspace_.reset();
177
  }
Y
Yu Yang 已提交
178 179
  workspace_ = paddle::memory::Alloc(place_, required_workspace_len,
                                     paddle::memory::Allocator::kScratchpad);
S
sneaxiy 已提交
180
}
181 182 183

CUDADeviceContext::CUDADeviceContext(CUDAPlace place)
    : place_(place), cudnn_holder_(nullptr) {
Y
Yu Yang 已提交
184
  CUDADeviceGuard guard(place_.device);
C
chengduo 已提交
185 186 187
  compute_capability_ = GetCUDAComputeCapability(place_.device);
  multi_process_ = GetCUDAMultiProcessors(place_.device);
  max_threads_per_mp_ = GetCUDAMaxThreadsPerMultiProcessor(place_.device);
Q
init  
qijun 已提交
188 189 190
  PADDLE_ENFORCE(cudaStreamCreate(&stream_));
  eigen_stream_.reset(new EigenCudaStreamDevice());
  eigen_stream_->Reinitialize(&stream_, place);
191
  eigen_device_.reset(new Eigen::GpuDevice(eigen_stream_.get()));
192 193
  PADDLE_ENFORCE(dynload::cublasCreate(&cublas_handle_));
  PADDLE_ENFORCE(dynload::cublasSetStream(cublas_handle_, stream_));
D
dzhwinter 已提交
194
  if (dynload::HasCUDNN()) {
195
    cudnn_holder_.reset(new CudnnHolder(&stream_, place));
D
dzhwinter 已提交
196
  }
S
sneaxiy 已提交
197

C
chengduo 已提交
198 199 200
  driver_version_ = GetCUDADriverVersion(place_.device);
  runtime_version_ = GetCUDARuntimeVersion(place_.device);

201 202 203 204 205 206
  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;
207 208 209 210
  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
211
  callback_manager_.reset(new StreamCallbackManager(stream_));
212 213 214 215
}

CUDADeviceContext::~CUDADeviceContext() {
  SetDeviceId(place_.device);
L
liaogang 已提交
216
  Wait();
S
sneaxiy 已提交
217
  WaitStreamCallback();
218
  PADDLE_ENFORCE(dynload::cublasDestroy(cublas_handle_));
219 220
  eigen_stream_.reset();
  eigen_device_.reset();
Q
init  
qijun 已提交
221
  PADDLE_ENFORCE(cudaStreamDestroy(stream_));
222 223
}

L
liaogang 已提交
224
Place CUDADeviceContext::GetPlace() const { return place_; }
225

L
liaogang 已提交
226
void CUDADeviceContext::Wait() const {
Q
init  
qijun 已提交
227
  PADDLE_ENFORCE(cudaStreamSynchronize(stream_));
228 229 230
  PADDLE_ENFORCE(cudaGetLastError());
}

K
Kexin Zhao 已提交
231
int CUDADeviceContext::GetComputeCapability() const {
C
chengduo 已提交
232
  return compute_capability_;
K
Kexin Zhao 已提交
233 234
}

235
int CUDADeviceContext::GetMaxPhysicalThreadCount() const {
C
chengduo 已提交
236
  return multi_process_ * max_threads_per_mp_;
237 238
}

239 240 241 242
Eigen::GpuDevice* CUDADeviceContext::eigen_device() const {
  return eigen_device_.get();
}

243
cublasHandle_t CUDADeviceContext::cublas_handle() const {
244 245 246
  return cublas_handle_;
}

247 248 249 250
cudnnHandle_t CUDADeviceContext::cudnn_handle() const {
  return cudnn_holder_->cudnn_handle();
}

S
sneaxiy 已提交
251 252
CudnnWorkspaceHandle CUDADeviceContext::cudnn_workspace_handle() const {
  return CudnnWorkspaceHandle(cudnn_holder_.get());
253
}
254

255
cudaStream_t CUDADeviceContext::stream() const { return stream_; }
Q
qijun 已提交
256

C
chengduoZH 已提交
257 258 259 260 261 262 263 264 265 266 267 268 269 270
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
271
#endif
Q
qijun 已提交
272

T
tensor-tang 已提交
273 274
#ifdef PADDLE_WITH_MKLDNN
MKLDNNDeviceContext::MKLDNNDeviceContext(CPUPlace place)
275 276 277
    : 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
278 279
}

S
Sylwester Fraczek 已提交
280 281 282 283 284 285 286 287
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; }

288 289
void MKLDNNDeviceContext::SetBlob(const std::string& name,
                                  std::shared_ptr<void> data) const {
290 291 292 293
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;

  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
294

295
  std::lock_guard<std::mutex> lock(*p_mutex_.get());
T
tensor-tang 已提交
296

297 298 299 300 301 302 303
  // 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;
304
  } else {
305
    pBlob = map_it->second;
306
  }
T
tensor-tang 已提交
307

308 309 310 311 312 313 314 315 316 317
  // 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
318
  return;
T
tensor-tang 已提交
319 320
}

321 322
std::shared_ptr<void> MKLDNNDeviceContext::GetBlob(
    const std::string& name) const {
323 324
  BlobMap* pMap = p_blobmap_.get();
  std::shared_ptr<KeyBlob> pBlob = nullptr;
T
tensor-tang 已提交
325

326
  int tid = platform::get_cur_thread_id();
T
tensor-tang 已提交
327

328 329 330 331 332 333 334 335 336 337 338
  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;
339

340 341
  // lock will be automatically released when out of scope
  return key_it->second;
T
tensor-tang 已提交
342 343 344 345
}

#endif

Q
qijun 已提交
346
}  // namespace platform
Q
qijun 已提交
347
}  // namespace paddle