test_matmul_api.cc 5.2 KB
Newer Older
Z
zyfncg 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
/* 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 <gtest/gtest.h>
#include <memory>

#include "paddle/pten/hapi/include/linalg.h"

#include "paddle/pten/core/dense_tensor.h"
#include "paddle/pten/core/kernel_registry.h"
#include "paddle/pten/hapi/lib/utils/allocator.h"
#include "paddle/pten/kernels/cuda/utils.h"

PT_DECLARE_MODULE(LinalgCPU);

#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
PT_DECLARE_MODULE(LinalgCUDA);
#endif

namespace framework = paddle::framework;
using DDim = paddle::framework::DDim;

TEST(API, matmul_cpu) {
  // 1. create tensor
  const auto alloc = std::make_shared<paddle::experimental::DefaultAllocator>(
      paddle::platform::CPUPlace());
  auto dense_x = std::make_shared<pten::DenseTensor>(
      alloc,
      pten::DenseTensorMeta(pten::DataType::FLOAT32,
                            framework::make_ddim({3, 3}),
                            pten::DataLayout::NCHW));

  auto* dense_x_data = dense_x->mutable_data<float>();

  auto dense_y = std::make_shared<pten::DenseTensor>(
      alloc,
      pten::DenseTensorMeta(pten::DataType::FLOAT32,
                            framework::make_ddim({3, 3}),
                            pten::DataLayout::NCHW));
  auto* dense_y_data = dense_y->mutable_data<float>();

  for (size_t i = 0; i < 9; ++i) {
    dense_x_data[i] = 1.0;
    dense_y_data[i] = 2.0;
  }
  std::vector<float> sum(9, 6.0);

  paddle::experimental::Tensor x(dense_x);
  paddle::experimental::Tensor y(dense_y);

  // 2. test API
  auto out = paddle::experimental::matmul(x, y, false, false);

  // 3. check result
  ASSERT_EQ(out.shape().size(), 2);
  ASSERT_EQ(out.shape()[0], 3);
  ASSERT_EQ(out.shape()[1], 3);
  ASSERT_EQ(out.numel(), 9);
  ASSERT_EQ(out.type(), pten::DataType::FLOAT32);
  ASSERT_EQ(out.layout(), pten::DataLayout::NCHW);
  ASSERT_EQ(out.initialized(), true);

  auto dense_out = std::dynamic_pointer_cast<pten::DenseTensor>(out.impl());

  for (size_t i = 0; i < 9; i++) {
    ASSERT_NEAR(sum[i], dense_out->data<float>()[i], 1e-6f);
  }
}

#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
TEST(API, matmul_cuda) {
  // Prepare CPU Dense Tensor
  const auto alloc_cpu =
      std::make_shared<paddle::experimental::DefaultAllocator>(
          paddle::platform::CPUPlace());
  auto ref_x = std::make_shared<pten::DenseTensor>(
      alloc_cpu,
      pten::DenseTensorMeta(pten::DataType::FLOAT32,
                            framework::make_ddim({3, 3}),
                            pten::DataLayout::NCHW));

  auto* ref_x_data = ref_x->mutable_data<float>();

  auto ref_y = std::make_shared<pten::DenseTensor>(
      alloc_cpu,
      pten::DenseTensorMeta(pten::DataType::FLOAT32,
                            framework::make_ddim({3, 3}),
                            pten::DataLayout::NCHW));
  auto* ref_y_data = ref_y->mutable_data<float>();

  for (size_t i = 0; i < 9; ++i) {
    ref_x_data[i] = 1.0;
    ref_y_data[i] = 2.0;
  }
  std::vector<float> sum(9, 6.0);

  // 1. create tensor
  const auto alloc_cuda =
      std::make_shared<paddle::experimental::DefaultAllocator>(
          paddle::platform::CUDAPlace());
  auto dense_x = std::make_shared<pten::DenseTensor>(
      alloc_cuda,
      pten::DenseTensorMeta(pten::DataType::FLOAT32,
                            framework::make_ddim({3, 3}),
                            pten::DataLayout::NCHW));

  auto dense_y = std::make_shared<pten::DenseTensor>(
      alloc_cuda,
      pten::DenseTensorMeta(pten::DataType::FLOAT32,
                            framework::make_ddim({3, 3}),
                            pten::DataLayout::NCHW));

  auto& pool = paddle::platform::DeviceContextPool::Instance();
  auto place = paddle::platform::CUDAPlace();
  auto* dev_ctx = pool.GetByPlace(place);

  pten::Copy(*dev_ctx, *ref_x.get(), dense_x.get());
  pten::Copy(*dev_ctx, *ref_y.get(), dense_y.get());

  paddle::experimental::Tensor x(dense_x);
  paddle::experimental::Tensor y(dense_y);

  // 2. test API
  auto out = paddle::experimental::matmul(x, y, false, false);

  // 3. check result
  ASSERT_EQ(out.shape().size(), 2);
  ASSERT_EQ(out.shape()[0], 3);
  ASSERT_EQ(out.shape()[1], 3);
  ASSERT_EQ(out.numel(), 9);
  ASSERT_EQ(out.type(), pten::DataType::FLOAT32);
  ASSERT_EQ(out.layout(), pten::DataLayout::NCHW);
  ASSERT_EQ(out.initialized(), true);

  auto dense_out = std::dynamic_pointer_cast<pten::DenseTensor>(out.impl());

  auto ref_out = std::make_shared<pten::DenseTensor>(
      alloc_cpu,
      pten::DenseTensorMeta(
          pten::DataType::FLOAT32, out.shape(), pten::DataLayout::NCHW));

  pten::Copy(*dev_ctx, *dense_out.get(), ref_out.get());

  for (size_t i = 0; i < 9; i++) {
    ASSERT_NEAR(sum[i], ref_out->data<float>()[i], 1e-6f);
  }
}

#endif