lod_tensor_test.cu 2.1 KB
Newer Older
1
//   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
D
dzhwinter 已提交
2 3 4 5 6
//
// 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
//
D
dzhwinter 已提交
7
//     http://www.apache.org/licenses/LICENSE-2.0
D
dzhwinter 已提交
8 9 10 11 12 13
//
// 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.
14 15 16

#include <cuda.h>
#include <cuda_runtime.h>
D
dzhwinter 已提交
17
#include <stdio.h>
18

Y
Yi Wang 已提交
19 20
#include "gtest/gtest.h"
#include "paddle/fluid/framework/lod_tensor.h"
21
#include "paddle/fluid/platform/init.h"
Y
Yi Wang 已提交
22
#include "paddle/fluid/platform/place.h"
23 24 25 26 27 28 29 30

__global__ void test(size_t* a, int size) {
  for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < size;
       i += blockDim.x * gridDim.x) {
    a[i] *= 2;
  }
}

D
dzhwinter 已提交
31
TEST(LoD, data) {
X
Xin Pan 已提交
32
  paddle::framework::InitDevices(true);
D
dzhwinter 已提交
33 34 35 36 37 38

  paddle::framework::LoD lod{{0, 1, 2}};
  lod.push_back({0, 2, 4, 5});
  lod.push_back(std::vector<size_t>({0, 1, 6, 8, 10, 11}));

  auto& v = lod[0];
Y
Yu Yang 已提交
39 40
  paddle::platform::CUDAPlace gpu(0);
  test<<<1, 1>>>(v.CUDAMutableData(gpu), v.size());
D
dzhwinter 已提交
41 42 43 44 45 46
  cudaDeviceSynchronize();
  for (size_t i = 0; i < v.size(); ++i) {
    EXPECT_EQ(v[i], i * 2);
  }
}

47
TEST(LoDTensor, LoDInGPU) {
X
Xin Pan 已提交
48
  paddle::framework::InitDevices(true);
D
dzhwinter 已提交
49

50
  paddle::framework::LoDTensor lod_tensor;
D
dzhwinter 已提交
51
  paddle::platform::CUDAPlace place(0);
52 53 54 55

  paddle::framework::LoD src_lod;
  src_lod.push_back(std::vector<size_t>{0, 2, 4, 6, 8, 10, 12, 14});

56 57
  lod_tensor.Resize({14, 16});
  lod_tensor.mutable_data<float>(place);
58 59

  lod_tensor.set_lod(src_lod);
60 61
  EXPECT_EQ(lod_tensor.lod_element(0, 2).first, 4UL);
  EXPECT_EQ(lod_tensor.lod_element(0, 4).first, 8UL);
62 63 64

  auto lod = lod_tensor.lod();

Y
Yu Yang 已提交
65
  test<<<1, 8>>>(lod[0].CUDAMutableData(place), lod[0].size());
66 67 68
  cudaDeviceSynchronize();

  for (size_t i = 0; i < src_lod[0].size(); ++i) {
69
    EXPECT_EQ(lod[0].data()[i], src_lod[0].data()[i] * 2);
70
  }
71
}