Add Cuda event and stream API (#32460)

* add cuda event and stream api

* add cuda event and stream api

* add get_current_stream api

* add get_current_stream api

* init streams

* modify get_current_stream

* modify get_cuttent_stream

* add synchronize func

* add current_stream doc and test file

* move get_current_stream into CUDA macro

* move CudaEvent into CUDA macro

* move _get_current_stream and _device_synchronize into cuda macro

* modify the macro of cuda stream and event

* add test case for synchronize

* add paddle.devices.cuda module

* event and stream support hip

* add doc for stream and event class

* move cuda stream and event into single pybind

* add cuda_streams_py.cc to cmakelist

* add _device_synchronize and _get_current_stream to core module

* add test case for cudastream and cudaevent

* move __all__ in streams.py

* fix test fail

* add cuda to devices __all__

* fix current_stream doc writing error

* move devices to device direction, and merge device.py into __init__.py

* add required:gpu to sample codes

* remove cuda direction from device/__init__.py

paddle/fluid/platform/event.h

@@ -22,6 +22,7 @@ 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 {
@@ -117,5 +118,98 @@ class MemEvent {
   std::string annotation_;
 };
 
+class CudaEvent {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+ public:
+  CudaEvent() {
+#ifdef PADDLE_WITH_HIP
+    hipEventCreateWithFlags(&event_, flags_);
+#else
+    cudaEventCreateWithFlags(&event_, flags_);
+#endif
+  }
+
+  CudaEvent(unsigned int flags) : flags_(flags) {
+#ifdef PADDLE_WITH_HIP
+    hipEventCreateWithFlags(&event_, flags_);
+#else
+    cudaEventCreateWithFlags(&event_, flags_);
+#endif
+  }
+
+  void Record(paddle::platform::stream::CUDAStream& stream) {
+#ifdef PADDLE_WITH_HIP
+    PADDLE_ENFORCE_CUDA_SUCCESS(hipEventRecord(event_, stream.raw_stream()));
+#else
+    PADDLE_ENFORCE_CUDA_SUCCESS(cudaEventRecord(event_, stream.raw_stream()));
+#endif
+  }
+
+  bool Query() {
+#ifdef PADDLE_WITH_HIP
+    gpuError_t err = hipEventQuery(event_);
+    if (err == hipSuccess) {
+      return true;
+    }
+    if (err == hipErrorNotReady) {
+      return false;
+    }
+#else
+    gpuError_t err = cudaEventQuery(event_);
+    if (err == cudaSuccess) {
+      return true;
+    }
+    if (err == cudaErrorNotReady) {
+      return false;
+    }
+#endif
+    PADDLE_ENFORCE_CUDA_SUCCESS(err);
+    return false;
+  }
+
+  void Synchronize() {
+#ifdef PADDLE_WITH_HIP
+    PADDLE_ENFORCE_CUDA_SUCCESS(hipEventSynchronize(event_));
+#else
+    PADDLE_ENFORCE_CUDA_SUCCESS(cudaEventSynchronize(event_));
+#endif
+  }
+  gpuEvent_t GetRawCudaEvent() { return event_; }
+
+ private:
+#ifdef PADDLE_WITH_HIP
+  unsigned int flags_ = hipEventDefault;
+#else
+  unsigned int flags_ = cudaEventDefault;
+#endif
+  gpuEvent_t event_;
+#endif
+};
+
+static unsigned int get_cuda_flags(bool enable_timing, bool blocking,
+                                   bool interprocess) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+
+#ifdef PADDLE_WITH_HIP
+  unsigned int flags =
+      (blocking ? hipEventBlockingSync : hipEventDefault) |
+      (enable_timing ? hipEventDefault : hipEventDisableTiming) |
+      (interprocess ? hipEventInterprocess : hipEventDefault);
+  return flags;
+#else
+  unsigned int flags =
+      (blocking ? cudaEventBlockingSync : cudaEventDefault) |
+      (enable_timing ? cudaEventDefault : cudaEventDisableTiming) |
+      (interprocess ? cudaEventInterprocess : cudaEventDefault);
+  return flags;
+#endif
+
+#else
+  PADDLE_THROW(platform::errors::Unavailable(
+      "Paddle is not compiled with CUDA. Cannot get the cuda event flags."));
+  return 0;
+#endif
+}
+
 }  // namespace platform
 }  // namespace paddle

paddle/fluid/platform/stream/cuda_stream.cc

@@ -14,6 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/platform/stream/cuda_stream.h"
 #include "paddle/fluid/platform/cuda_device_guard.h"
+#include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
 
 namespace paddle {
@@ -95,6 +96,28 @@ void CUDAStream::Wait() const {
   PADDLE_ENFORCE_CUDA_SUCCESS(e_sync);
 }
 
+CUDAStream* get_current_stream(int deviceId) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+  if (deviceId == -1) {
+    deviceId = platform::GetCurrentDeviceId();
+  }
+
+  auto& pool = platform::DeviceContextPool::Instance();
+
+  platform::Place device = CUDAPlace(deviceId);
+
+  auto stream = static_cast<platform::CUDADeviceContext*>(pool.Get(device))
+                    ->context()
+                    ->Stream()
+                    .get();
+  return stream;
+#else
+  PADDLE_THROW(platform::errors::Unavailable(
+      "Paddle is not compiled with CUDA. Cannot visit cuda current stream."));
+  return nullptr;
+#endif
+}
+
 }  // namespace stream
 }  // namespace platform
 }  // namespace paddle

paddle/fluid/platform/stream/cuda_stream.h

@@ -33,8 +33,9 @@ enum class Priority : uint8_t {
   kHigh = 0x1,
   kNormal = 0x2,
 };
-
+#endif
 class CUDAStream final {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  public:
   CUDAStream() = default;
   explicit CUDAStream(const Place& place,
@@ -93,6 +94,37 @@ class CUDAStream final {
 #endif
   void Destroy();
 
+  bool Query() const {
+#ifdef PADDLE_WITH_HIP
+    hipError_t err = hipStreamQuery(stream_);
+    if (err == hipSuccess) {
+      return true;
+    }
+    if (err == hipErrorNotReady) {
+      return false;
+    }
+#else
+    cudaError_t err = cudaStreamQuery(stream_);
+    if (err == cudaSuccess) {
+      return true;
+    }
+    if (err == cudaErrorNotReady) {
+      return false;
+    }
+#endif
+
+    PADDLE_ENFORCE_CUDA_SUCCESS(err);
+    return false;
+  }
+
+  void Synchronize() const {
+#ifdef PADDLE_WITH_HIP
+    PADDLE_ENFORCE_CUDA_SUCCESS(hipStreamSynchronize(stream_));
+#else
+    PADDLE_ENFORCE_CUDA_SUCCESS(cudaStreamSynchronize(stream_));
+#endif
+  }
+
  private:
   Place place_;
 #ifdef PADDLE_WITH_HIP
@@ -102,11 +134,11 @@ class CUDAStream final {
 #endif
   Priority priority_{Priority::kNormal};
   std::unique_ptr<StreamCallbackManager<gpuStream_t>> callback_manager_;
-
+#endif
   DISABLE_COPY_AND_ASSIGN(CUDAStream);
 };
 
-#endif
+CUDAStream* get_current_stream(int deviceId);
 
 }  // namespace stream
 }  // namespace platform

paddle/fluid/pybind/CMakeLists.txt

@@ -57,7 +57,8 @@ set(PYBIND_SRCS
   inference_api.cc
   compatible.cc
   io.cc
-  generator_py.cc)
+  generator_py.cc
+  cuda_streams_py.cc)
 
 if(WITH_ASCEND)
   set(PYBIND_DEPS ${PYBIND_DEPS} ascend_wrapper)

paddle/fluid/pybind/cuda_streams_py.cc

+// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#include <string>
+#include <vector>
+
+#include "paddle/fluid/platform/event.h"
+#include "paddle/fluid/platform/stream/cuda_stream.h"
+#include "paddle/fluid/pybind/cuda_streams_py.h"
+
+namespace py = pybind11;
+
+namespace paddle {
+namespace pybind {
+void BindCudaStream(py::module *m_ptr) {
+  auto &m = *m_ptr;
+
+  // Bind Methods
+  m.def("_get_current_stream",
+        [](int deviceId) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+          return paddle::platform::stream::get_current_stream(deviceId);
+#else
+          PADDLE_THROW(platform::errors::Unavailable(
+              "Paddle is not compiled with CUDA. Cannot visit cuda current "
+              "stream."));
+#endif
+        },
+        py::return_value_policy::reference);
+
+  m.def("_device_synchronize", [](int device_id) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+    if (device_id == -1) {
+      device_id = paddle::platform::GetCurrentDeviceId();
+    }
+
+    int curr_device_id = paddle::platform::GetCurrentDeviceId();
+    paddle::platform::SetDeviceId(device_id);
+#ifdef PADDLE_WITH_HIP
+    PADDLE_ENFORCE_CUDA_SUCCESS(hipDeviceSynchronize());
+#else
+    PADDLE_ENFORCE_CUDA_SUCCESS(cudaDeviceSynchronize());
+#endif
+    paddle::platform::SetDeviceId(curr_device_id);
+#else
+    PADDLE_THROW(platform::errors::Unavailable(
+        "Paddle is not compiled with CUDA. Cannot visit device synchronize."));
+#endif
+  });
+
+  py::class_<paddle::platform::stream::CUDAStream>(m, "CUDAStream", R"DOC(
+      The handle of the CUDA stream.
+
+      Parameters:
+        device(paddle.CUDAPlace()|int|None, optional): The device which wanted to allocate the stream. 
+        If device is None or negative integer, device will be the current device. 
+        If device is positive integer, it must less than the device count. Default: None. 
+
+        priority(int|None, optional): The priority of stream. The priority can be 1(high) or 2(normal).
+        If prioriyt is None, the priority is 2(normal). Default: None. 
+
+      Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+            s1 = paddle.device.cuda.Stream(paddle.CUDAPlace(0), 1)
+            s2 = paddle.device.cuda.Stream(0, 1)
+            s3 = paddle.device.cuda.Stream()
+
+  )DOC")
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+      .def("wait_event",
+           [](paddle::platform::stream::CUDAStream &self,
+              paddle::platform::CudaEvent &event) {
+             self.WaitEvent(event.GetRawCudaEvent());
+           },
+           R"DOC(
+      Makes all future work submitted to stream wait for all work captured in event.
+
+      Parameters:
+        event(CUDAEvent): The event to wait on.
+      
+      Examples:
+        .. code-block:: python
+
+          # required: gpu
+          import paddle
+          s = paddle.device.cuda.Stream(paddle.CUDAPlace(0), 1)
+          event = paddle.device.cuda.Event()
+          s.wait_event(event)
+
+           )DOC")
+      .def("wait_stream",
+           [](paddle::platform::stream::CUDAStream &self,
+              paddle::platform::stream::CUDAStream &stream) {
+             auto event = paddle::platform::CudaEvent();
+             event.Record(stream);
+
+             self.WaitEvent(event.GetRawCudaEvent());
+           },
+           R"DOC(
+      Synchronizes with the given stream.
+
+      Parameters:
+        stream(CUDAStream): The stream to synchronize with.
+      
+      Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+            s1 = paddle.device.cuda.Stream(paddle.CUDAPlace(0), 1)
+            s2 = paddle.device.cuda.Stream(0, 1)
+            s1.wait_stream(s2)
+
+           )DOC")
+      .def("query",
+           [](paddle::platform::stream::CUDAStream &self) {
+             return self.Query();
+           },
+           R"DOC(
+      Return the status whether if all operations in stream have completed.
+
+      Returns: A boolean value.
+
+      Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+            s = paddle.device.cuda.Stream(paddle.CUDAPlace(0), 1)
+            is_done = s.query()
+
+           )DOC")
+      .def("synchronize",
+           [](paddle::platform::stream::CUDAStream &self) {
+             self.Synchronize();
+           },
+           R"DOC(
+      Waits for stream tasks to complete.
+
+      Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+            s = paddle.device.cuda.Stream(paddle.CUDAPlace(0), 1)
+            s.synchronize()
+
+           )DOC")
+      .def("record_event",
+           [](paddle::platform::stream::CUDAStream &self,
+              paddle::platform::CudaEvent *event) {
+             if (event == nullptr) {
+               auto event_tmp = paddle::platform::CudaEvent();
+               event = &event_tmp;
+             }
+             event->Record(self);
+             return event;
+           },
+           R"DOC(
+      Record a CUDA event in the stream.
+
+      Parameters:
+          event(CUDAEvent, optional): The event to be record. If event is None, a new event is created.
+          Default: None.
+      
+      Returns:
+          The recored event.
+
+      Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+            s = paddle.device.cuda.Stream(paddle.CUDAPlace(0), 1)
+            event = s.record_event()
+
+           )DOC",
+           py::arg("event") = nullptr)
+#endif
+      .def("__init__",
+           [](paddle::platform::stream::CUDAStream &self,
+              platform::CUDAPlace *device, int priority) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+             if (priority != 1 && priority != 2) {
+               PADDLE_THROW(platform::errors::InvalidArgument(
+                   "Priority should be 1(high) or 2(normal) "));
+             }
+             auto prio = paddle::platform::stream::Priority(priority);
+
+             if (device == nullptr) {
+               int curr_device_id = platform::GetCurrentDeviceId();
+               auto device_tmp = platform::CUDAPlace(curr_device_id);
+               device = &device_tmp;
+             }
+
+             new (&self) paddle::platform::stream::CUDAStream(*device, prio);
+#else
+            PADDLE_THROW(platform::errors::Unavailable(
+        "Class CUDAStream can only be initialized on the GPU platform."));
+#endif
+           },
+           py::arg("device") = nullptr, py::arg("priority") = 2)
+      .def(
+          "__init__",
+          [](paddle::platform::stream::CUDAStream &self, int device,
+             int priority) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+            if (priority != 1 && priority != 2) {
+              PADDLE_THROW(platform::errors::InvalidArgument(
+                  "Priority should be 1(high) or 2(normal) "));
+            }
+            auto prio = paddle::platform::stream::Priority(priority);
+
+            int device_count = platform::GetCUDADeviceCount();
+            if (device < 0) {
+              device = platform::GetCurrentDeviceId();
+            }
+            if (device >= device_count) {
+              PADDLE_THROW(platform::errors::InvalidArgument(
+                  "The device id  must be inside [0, %d), but input device=%d.",
+                  device_count, device));
+            }
+
+            new (&self) paddle::platform::stream::CUDAStream(
+                platform::CUDAPlace(device), prio);
+#else
+            PADDLE_THROW(platform::errors::Unavailable(
+        "Class CUDAStream can only be initialized on the GPU platform."));
+#endif
+          },
+          py::arg("device") = -1, py::arg("priority") = 2)
+      .def("__init__", [](paddle::platform::stream::CUDAStream &self) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+        auto prio = paddle::platform::stream::Priority::kNormal;
+
+        int device_id = platform::GetCurrentDeviceId();
+
+        new (&self) paddle::platform::stream::CUDAStream(
+            platform::CUDAPlace(device_id), prio);
+#else
+            PADDLE_THROW(platform::errors::Unavailable(
+        "Class CUDAStream can only be initialized on the GPU platform."));
+#endif
+      });
+
+  py::class_<paddle::platform::CudaEvent>(m, "CUDAEvent", R"DOC(
+      The handle of the CUDA event.
+
+      Parameters:
+        enable_timing(bool, optional): Whether the event will measure time. Default: False.
+        blocking(bool, optional): Whether the wait() func will be blocking. Default: False;
+        interprocess(bool, optional): Whether the event can be shared between processes. Defalut: False.
+      
+      Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+            event = paddle.device.cuda.Event()
+
+  )DOC")
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+      .def("record",
+           [](paddle::platform::CudaEvent &self,
+              paddle::platform::stream::CUDAStream *stream) {
+             if (stream == nullptr) {
+               stream = paddle::platform::stream::get_current_stream(-1);
+             }
+             self.Record(*stream);
+           },
+           R"DOC(
+          Records the event in the given stream.
+
+          Parameters:
+            stream(CUDAStream, optional): The handle of CUDA stream. If None, the stream is the current stream. Default: None.
+          
+          Examples:
+            .. code-block:: python
+
+              # required: gpu
+              import paddle
+              event = paddle.device.cuda.Event()
+              event.record()
+    
+        )DOC",
+           py::arg("stream") = nullptr)
+      .def("query",
+           [](paddle::platform::CudaEvent &self) { return self.Query(); },
+           R"DOC(
+          Queries the event's status.
+
+          Returns: A boolean which indicates all work currently captured by the event has been completed.
+
+          Examples:
+            .. code-block:: python
+
+                # required: gpu
+                import paddle
+                event = paddle.device.cuda.Event()
+                is_done = event.query()
+
+           )DOC")
+      .def("synchronize",
+           [](paddle::platform::CudaEvent &self) { self.Synchronize(); }, R"DOC(
+            Waits for an event to complete.
+
+            Examples:
+              .. code-block:: python
+
+                # required: gpu
+                import paddle
+                event = paddle.device.cuda.Event()
+                event.synchronize()
+
+           )DOC")
+#endif
+      .def("__init__",
+           [](paddle::platform::CudaEvent &self, bool enable_timing,
+              bool blocking, bool interprocess) {
+#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
+             unsigned int flags = platform::get_cuda_flags(
+                 enable_timing, blocking, interprocess);
+             new (&self) paddle::platform::CudaEvent(flags);
+#else
+             PADDLE_THROW(platform::errors::Unavailable(
+                 "Class CUDAEvent can only be initialized on the GPU "
+                 "platform."));
+
+#endif
+           },
+           py::arg("enable_timing") = false, py::arg("blocking") = false,
+           py::arg("interprocess") = false);
+}
+
+}  // namespace pybind
+
+}  // namespace paddle

paddle/fluid/pybind/cuda_streams_py.h

+// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+//
+// 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.
+
+#pragma once
+
+#include "pybind11/pybind11.h"
+#include "pybind11/stl.h"
+
+namespace py = pybind11;
+
+namespace paddle {
+namespace pybind {
+
+void BindCudaStream(py::module* m);
+
+}  // namespace pybind
+}  // namespace paddle

paddle/fluid/pybind/pybind.cc

@@ -70,6 +70,7 @@ limitations under the License. */
 #include "paddle/fluid/platform/monitor.h"
 #include "paddle/fluid/platform/place.h"
 #include "paddle/fluid/platform/profiler.h"
+#include "paddle/fluid/pybind/cuda_streams_py.h"
 #include "paddle/fluid/pybind/io.h"
 #ifdef PADDLE_WITH_ASCEND
 #include "paddle/fluid/pybind/ascend_wrapper_py.h"
@@ -77,6 +78,7 @@ limitations under the License. */
 #include "paddle/fluid/pybind/box_helper_py.h"
 #include "paddle/fluid/pybind/compatible.h"
 #include "paddle/fluid/pybind/const_value.h"
+#include "paddle/fluid/pybind/cuda_streams_py.h"
 #include "paddle/fluid/pybind/data_set_py.h"
 #include "paddle/fluid/pybind/exception.h"
 #include "paddle/fluid/pybind/fleet_wrapper_py.h"
@@ -471,6 +473,8 @@ PYBIND11_MODULE(core_avx, m) {
 PYBIND11_MODULE(core_noavx, m) {
 #endif
 
+  BindCudaStream(&m);
+
   // Not used, just make sure cpu_info.cc is linked.
   paddle::platform::CpuTotalPhysicalMemory();
 

python/paddle/__init__.py

@@ -52,6 +52,7 @@ import paddle.metric  # noqa: F401
 import paddle.regularizer  # noqa: F401
 import paddle.incubate  # noqa: F401
 import paddle.autograd  # noqa: F401
+import paddle.device  # noqa: F401
 
 import paddle.jit  # noqa: F401
 import paddle.amp  # noqa: F401

python/paddle/device.py → python/paddle/device/__init__.py

-#   Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
-#
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# 
 # 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.
@@ -20,7 +20,7 @@ from paddle.fluid import framework
 from paddle.fluid.dygraph.parallel import ParallelEnv
 from paddle.fluid.framework import is_compiled_with_cuda  # noqa: F401
 from paddle.fluid.framework import is_compiled_with_rocm  # noqa: F401
-
+from . import cuda
 
 __all__ = [  # noqa
     'get_cudnn_version',

python/paddle/device/cuda/__init__.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# 
+# 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.
+
+from paddle.fluid import core
+
+from .streams import Stream  # noqa: F401
+from .streams import Event  # noqa: F401
+
+__all__ = [
+    'Stream',
+    'Event',
+    'current_stream',
+    'synchronize',
+]
+
+
+def current_stream(device=None):
+    '''
+    Return the current CUDA stream by the device.
+
+    Parameters:
+        device(paddle.CUDAPlace()|int, optional): The device or the ID of the device which want to get stream from. 
+        If device is None, the device is the current device. Default: None.
+    
+    Returns:
+        CUDAStream: the stream to the device.
+    
+    Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+
+            s1 = paddle.device.cuda.current_stream()
+
+            s2 = paddle.device.cuda.current_stream(0)
+
+            s3 = paddle.device.cuda.current_stream(paddle.CUDAPlace(0))
+
+    '''
+
+    device_id = -1
+
+    if device is not None:
+        if isinstance(device, int):
+            device_id = device
+        elif isinstance(device, core.CUDAPlace):
+            device_id = device.get_device_id()
+        else:
+            raise ValueError("device type must be int or paddle.CUDAPlace")
+
+    return core._get_current_stream(device_id)
+
+
+def synchronize(device=None):
+    '''
+    Wait for the compute on the given CUDA device to finish.
+
+    Parameters:
+        device(paddle.CUDAPlace()|int, optional): The device or the ID of the device.
+        If device is None, the device is the current device. Default: None.
+    
+    Examples:
+        .. code-block:: python
+
+            # required: gpu
+            import paddle
+
+            paddle.device.cuda.synchronize()
+            paddle.device.cuda.synchronize(0)
+            paddle.device.cuda.synchronize(paddle.CUDAPlace(0))
+
+    '''
+
+    device_id = -1
+
+    if device is not None:
+        if isinstance(device, int):
+            device_id = device
+        elif isinstance(device, core.CUDAPlace):
+            device_id = device.get_device_id()
+        else:
+            raise ValueError("device type must be int or paddle.CUDAPlace")
+
+    return core._device_synchronize(device_id)

python/paddle/device/cuda/streams.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# 
+# 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.
+
+from paddle.fluid.core import CUDAStream as Stream
+from paddle.fluid.core import CUDAEvent as Event

python/paddle/fluid/core.py

@@ -274,6 +274,8 @@ if avx_supported():
         from .core_avx import _cuda_synchronize
         from .core_avx import _is_compiled_with_heterps
         from .core_avx import _promote_types_if_complex_exists
+        from .core_avx import _device_synchronize
+        from .core_avx import _get_current_stream
         if sys.platform != 'win32':
             from .core_avx import _set_process_pids
             from .core_avx import _erase_process_pids
@@ -323,6 +325,8 @@ if load_noavx:
         from .core_noavx import _cuda_synchronize
         from .core_noavx import _is_compiled_with_heterps
         from .core_noavx import _promote_types_if_complex_exists
+        from .core_noavx import _device_synchronize
+        from .core_noavx import _get_current_stream
         if sys.platform != 'win32':
             from .core_noavx import _set_process_pids
             from .core_noavx import _erase_process_pids

python/paddle/fluid/tests/unittests/test_cuda_stream_event.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# 
+# 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.
+
+from paddle.device import cuda
+import paddle
+
+import unittest
+
+
+class TestCurrentStream(unittest.TestCase):
+    def test_current_stream(self):
+        if paddle.is_compiled_with_cuda():
+            s = cuda.current_stream()
+            self.assertTrue(isinstance(s, cuda.Stream))
+
+            s1 = cuda.current_stream(0)
+            self.assertTrue(isinstance(s1, cuda.Stream))
+
+            s2 = cuda.current_stream(paddle.CUDAPlace(0))
+            self.assertTrue(isinstance(s2, cuda.Stream))
+
+            self.assertEqual(s1, s2)
+
+            self.assertRaises(ValueError, cuda.current_stream, "gpu:0")
+
+
+class TestSynchronize(unittest.TestCase):
+    def test_synchronize(self):
+        if paddle.is_compiled_with_cuda():
+            self.assertIsNone(cuda.synchronize())
+            self.assertIsNone(cuda.synchronize(0))
+            self.assertIsNone(cuda.synchronize(paddle.CUDAPlace(0)))
+
+            self.assertRaises(ValueError, cuda.synchronize, "gpu:0")
+
+
+class TestCUDAStream(unittest.TestCase):
+    def test_cuda_stream(self):
+        if paddle.is_compiled_with_cuda():
+            s = paddle.device.cuda.Stream()
+            self.assertIsNotNone(s)
+
+    def test_cuda_stream_synchronize(self):
+        if paddle.is_compiled_with_cuda():
+            s = paddle.device.cuda.Stream()
+            e1 = paddle.device.cuda.Event(True, False, False)
+            e2 = paddle.device.cuda.Event(True, False, False)
+
+            e1.record(s)
+            e1.query()
+            tensor1 = paddle.to_tensor(paddle.rand([1000, 1000]))
+            tensor2 = paddle.matmul(tensor1, tensor1)
+            s.synchronize()
+            e2.record(s)
+            e2.synchronize()
+
+            self.assertTrue(s.query())
+
+    def test_cuda_stream_wait_event_and_record_event(self):
+        if paddle.is_compiled_with_cuda():
+            s1 = cuda.Stream(0)
+            tensor1 = paddle.to_tensor(paddle.rand([1000, 1000]))
+            tensor2 = paddle.matmul(tensor1, tensor1)
+            e1 = cuda.Event(False, False, False)
+            s1.record_event(e1)
+
+            s2 = cuda.Stream(0)
+            s2.wait_event(e1)
+            s2.synchronize()
+
+            self.assertTrue(e1.query() and s1.query() and s2.query())
+
+
+class TestCUDAEvent(unittest.TestCase):
+    def test_cuda_event(self):
+        if paddle.is_compiled_with_cuda():
+            e = paddle.device.cuda.Event(True, False, False)
+            self.assertIsNotNone(e)
+            s = paddle.device.cuda.current_stream()
+
+    def test_cuda_event_methods(self):
+        if paddle.is_compiled_with_cuda():
+            e = paddle.device.cuda.Event(True, False, False)
+            s = paddle.device.cuda.current_stream()
+            event_query_1 = e.query()
+            tensor1 = paddle.to_tensor(paddle.rand([1000, 1000]))
+            tensor2 = paddle.matmul(tensor1, tensor1)
+            s.record_event(e)
+            e.synchronize()
+            event_query_2 = e.query()
+
+            self.assertTrue(event_query_1)
+            self.assertTrue(event_query_2)
+
+
+if __name__ == "__main__":
+    unittest.main()

python/setup.py.in

@@ -224,6 +224,8 @@ packages=['paddle',
           'paddle.tensor',
           'paddle.onnx',
           'paddle.autograd',
+          'paddle.device',
+          'paddle.device.cuda',
           ]
 
 with open('@PADDLE_SOURCE_DIR@/python/requirements.txt') as f: