device_event_test.cc 2.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
// 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 "paddle/fluid/platform/device_event.h"
#include "glog/logging.h"
#include "gtest/gtest.h"

using ::paddle::platform::kCUDA;
using ::paddle::platform::kCPU;
21 22 23 24 25 26

using paddle::platform::DeviceEvent;
using paddle::platform::DeviceContextPool;

#ifdef PADDLE_WITH_CUDA
#include <cuda_runtime.h>
27 28 29 30 31 32 33 34 35 36 37 38

TEST(DeviceEvent, CUDA) {
  VLOG(1) << "In Test";
  using paddle::platform::CUDAPlace;

  auto& pool = DeviceContextPool::Instance();
  auto place = CUDAPlace(0);
  auto* context =
      static_cast<paddle::platform::CUDADeviceContext*>(pool.Get(place));

  ASSERT_NE(context, nullptr);
  // case 1. test for event_creator
39
  DeviceEvent event(place);
40 41
  ASSERT_NE(event.GetEvent().get(), nullptr);
  // case 2. test for event_recorder
42
  event.Record(context);
43 44 45 46 47 48 49 50 51 52 53 54 55 56
  bool status = event.Query();
  ASSERT_EQ(status, false);
  // case 3. test for event_finisher
  event.Finish();
  status = event.Query();
  ASSERT_EQ(status, true);

  // case 4. test for event_waiter
  float *src_fp32, *dst_fp32;
  int size = 1000000 * sizeof(float);
  cudaMallocHost(reinterpret_cast<void**>(&src_fp32), size);
  cudaMalloc(reinterpret_cast<void**>(&dst_fp32), size);
  cudaMemcpyAsync(dst_fp32, src_fp32, size, cudaMemcpyHostToDevice,
                  context->stream());
57
  event.Record(context);  // step 1. record it
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
  status = event.Query();
  ASSERT_EQ(status, false);

  event.Wait(kCUDA, context);  // step 2. add streamWaitEvent
  status = event.Query();
  ASSERT_EQ(status, false);  // async

  event.Wait(kCPU, context);  // step 3. EventSynchornize
  status = event.Query();
  ASSERT_EQ(status, true);  // sync

  // release resource
  cudaFree(dst_fp32);
  cudaFreeHost(src_fp32);
}
#endif
74 75 76 77 78 79 80 81 82 83 84 85 86

TEST(DeviceEvent, CPU) {
  using paddle::platform::CPUPlace;
  auto place = CPUPlace();
  DeviceEvent event(place);
  auto& pool = DeviceContextPool::Instance();
  auto* context = pool.Get(place);

  // TODO(Aurelius84): All DeviceContext should has Record/Wait
  event.Record(context);
  event.SetFininshed();
  bool status = event.Query();
  ASSERT_EQ(status, true);
87 88 89 90 91 92 93 94

  // test for Record again
  event.Record(context);
  status = event.Query();
  ASSERT_EQ(status, false);  // SCHEDULED

  event.Reset();
  ASSERT_EQ(status, false);  // INITIALIZED
95
}