PaddlePaddle · zhiqiu · Jul 29, 2022 · Jul 19, 2022 · Jul 21, 2022 · Jul 26, 2022
@@ -18,24 +18,12 @@ if(WITH_NCCL OR WITH_RCCL)
   cc_library(
     processgroup_nccl
     SRCS ProcessGroupNCCL.cc NCCLTools.cc Common.cc
-    DEPS place
-         cuda_stream
-         enforce
-         collective_helper
-         device_context
-         phi_api
-         eager_api)
+    DEPS place enforce collective_helper device_context phi_api eager_api)
   if(WITH_DISTRIBUTE AND WITH_PSCORE)
     cc_library(
       processgroup_heter
       SRCS ProcessGroupHeter.cc NCCLTools.cc Common.cc
-      DEPS place
-           cuda_stream
-           enforce
-           collective_helper
-           device_context
-           phi_api
-           eager_api)
+      DEPS place enforce collective_helper device_context phi_api eager_api)
   endif()
 endif()
 

@@ -33,7 +33,6 @@
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/gen_comm_id_helper.h"
 #include "paddle/fluid/platform/place.h"
-#include "paddle/fluid/platform/stream/cuda_stream.h"
 
 #if defined(PADDLE_WITH_NCCL) || defined(PADDLE_WITH_RCCL)
 #include "paddle/fluid/distributed/collective/NCCLTools.h"

@@ -28,7 +28,6 @@
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/gen_comm_id_helper.h"
 #include "paddle/fluid/platform/place.h"
-#include "paddle/fluid/platform/stream/cuda_stream.h"
 
 #if defined(PADDLE_WITH_NCCL) || defined(PADDLE_WITH_RCCL)
 #include "paddle/fluid/distributed/collective/NCCLTools.h"
@@ -46,7 +45,6 @@ namespace paddle {
 namespace distributed {
 
 using Place = paddle::platform::Place;
-using CUDAStream = platform::stream::CUDAStream;
 using CUDADeviceContext = paddle::platform::CUDADeviceContext;
 
 class ProcessGroupNCCL : public ProcessGroup {

diff --git a/paddle/fluid/framework/garbage_collector.h b/paddle/fluid/framework/garbage_collector.h
@@ -25,6 +25,7 @@
 #ifdef PADDLE_WITH_MLU
 #include "paddle/fluid/platform/device/mlu/device_context.h"
 #endif
+#include "paddle/fluid/platform/stream_callback_manager.h"
 
 namespace paddle {
 namespace framework {

@@ -30,7 +30,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/place.h"
 #include "paddle/phi/core/ddim.h"
 #include "paddle/phi/core/dense_tensor.h"
-#include "paddle/phi/core/stream.h"
 
 namespace paddle {
 

diff --git a/paddle/fluid/memory/allocation/stream_safe_cuda_allocator.cc b/paddle/fluid/memory/allocation/stream_safe_cuda_allocator.cc
@@ -195,7 +195,7 @@ phi::Allocation* StreamSafeCUDAAllocator::AllocateImpl(size_t size) {
       default_stream_,
       this);
   VLOG(8) << "Allocate " << allocation->size() << " bytes at address "
-          << allocation->ptr();
+          << allocation->ptr() << "  , stream: " << default_stream_;
   return allocation;
 }
 

diff --git a/paddle/fluid/operators/decode_jpeg_op.cu b/paddle/fluid/operators/decode_jpeg_op.cu
@@ -19,7 +19,6 @@
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/platform/dynload/nvjpeg.h"
 #include "paddle/fluid/platform/enforce.h"
-#include "paddle/fluid/platform/stream/cuda_stream.h"
 
 namespace paddle {
 namespace operators {
@@ -69,7 +68,7 @@ class GPUDecodeJpegKernel : public framework::OpKernel<T> {
     nvjpegStatus_t info_status =
         platform::dynload::nvjpegGetImageInfo(nvjpeg_handle,
                                               x_data,
-                                              (size_t)x->numel(),
+                                              static_cast<size_t>(x->numel()),
                                               &components,
                                               &subsampling,
                                               widths,

diff --git a/paddle/fluid/platform/CMakeLists.txt b/paddle/fluid/platform/CMakeLists.txt
@@ -113,7 +113,6 @@ endif()
 
 add_subdirectory(device)
 add_subdirectory(dynload)
-add_subdirectory(stream)
 
 cc_library(
   cpu_helper
@@ -130,7 +129,7 @@ if(WITH_DGC)
 endif()
 
 if(WITH_GPU OR WITH_ROCM)
-  set(GPU_CTX_DEPS dynload_cuda dynamic_loader cuda_stream)
+  set(GPU_CTX_DEPS dynload_cuda dynamic_loader)
 endif()
 
 if(WITH_IPU)

diff --git a/paddle/fluid/platform/device_context.cc b/paddle/fluid/platform/device_context.cc
@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/place.h"
 #include "paddle/fluid/platform/profiler.h"
 #include "paddle/fluid/platform/profiler/event_tracing.h"
-#include "paddle/fluid/platform/stream/cuda_stream.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/allocator.h"
 
@@ -534,24 +533,11 @@ void CudnnWorkspaceHandle::ReallocWorkspace(size_t required_workspace_bytes) {
   allocation_ = memory::Alloc(device_context_, required_workspace_bytes);
 }
 
-CUDADeviceContext::CUDADeviceContext(CUDAPlace place) : phi::GPUContext(place) {
-  phi::GPUContext::PartialInitWithoutAllocator();
-  cuda_stream_.reset(new stream::CUDAStream(phi::GPUContext::stream(), place));
-}
+CUDADeviceContext::CUDADeviceContext(CUDAPlace place)
+    : phi::GPUContext(place) {}
 
 CUDADeviceContext::~CUDADeviceContext() = default;
 
-stream::CUDAStream* CUDADeviceContext::GetCudaStream() const {
-  return cuda_stream_.get();
-}
-
-stream::CUDAStream* CUDADeviceContext::SetCudaStream(
-    stream::CUDAStream* new_stream_ptr) {
-  auto* old_stream_ptr = cuda_stream_.release();
-  cuda_stream_.reset(new_stream_ptr);
-  return old_stream_ptr;
-}
-
 CUDAPinnedDeviceContext::CUDAPinnedDeviceContext() {
   eigen_device_.reset(new Eigen::DefaultDevice());
 }

diff --git a/paddle/fluid/platform/device_context.h b/paddle/fluid/platform/device_context.h
@@ -67,9 +67,6 @@ limitations under the License. */
 #include "glog/logging.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/place.h"
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-#include "paddle/fluid/platform/stream/cuda_stream.h"
-#endif
 #ifdef PADDLE_WITH_ASCEND_CL
 #include "paddle/fluid/platform/device/npu/enforce_npu.h"
 #include "paddle/fluid/platform/device/npu/npu_stream.h"
@@ -279,16 +276,8 @@ class CUDADeviceContext : public phi::GPUContext {
   explicit CUDADeviceContext(CUDAPlace place);
   virtual ~CUDADeviceContext();
 
-  // NOTE: Just for compatibility with the past, please delete if there is an
-  // elegant way.
-  stream::CUDAStream* GetCudaStream() const;
-  stream::CUDAStream* SetCudaStream(stream::CUDAStream*);
-
  private:
-  // NOTE: Just for compatibility with the past, please delete if there is an
-  // elegant way.
-  std::unique_ptr<stream::CUDAStream> cuda_stream_;
-
+  int place_holder_;  // TO BE REMOVED
   DISABLE_COPY_AND_ASSIGN(CUDADeviceContext);
 };
 

diff --git a/paddle/fluid/platform/event.h b/paddle/fluid/platform/event.h
@@ -25,7 +25,6 @@ limitations under the License. */
 #include <hip/hip_runtime.h>
 #endif
 #include "paddle/fluid/platform/place.h"
-#include "paddle/fluid/platform/stream/cuda_stream.h"
 
 namespace paddle {
 namespace platform {

diff --git a/paddle/fluid/platform/stream/CMakeLists.txt b/paddle/fluid/platform/stream/CMakeLists.txt
diff --git a/paddle/fluid/platform/stream/cuda_stream.cc b/paddle/fluid/platform/stream/cuda_stream.cc