Function TEST

test/cpp/fluid/framework/tensor_util_test.cc:24–124

namespace framework {

TEST(TensorCopy, Tensor) {
  phi::DenseTensor src_tensor;
  phi::DenseTensor dst_tensor;
  phi::CPUContext cpu_ctx((phi::CPUPlace()));

  int* src_ptr =
      src_tensor.mutable_data<int>(common::make_ddim({3, 3}), phi::CPUPlace());

  std::array<int, 9> arr = {1, 2, 3, 4, 5, 6, 7, 8, 9};
  memcpy(src_ptr, arr.data(), 9 * sizeof(int));
  src_tensor.set_layout(DataLayout::kAnyLayout);

  auto cpu_place = new phi::CPUPlace();
  TensorCopy(src_tensor, *cpu_place, &dst_tensor);

  const int* dst_ptr = dst_tensor.data<int>();
  EXPECT_NE(src_ptr, dst_ptr);
  for (size_t i = 0; i < 9; ++i) {
    EXPECT_EQ(src_ptr[i], dst_ptr[i]);
  }

  TensorCopy(dst_tensor, *cpu_place, &dst_tensor);
  for (size_t i = 0; i < 9; ++i) {
    EXPECT_EQ(src_ptr[i], dst_ptr[i]);
  }

  EXPECT_TRUE(dst_tensor.layout() == src_tensor.layout());

  phi::DenseTensor slice_tensor = src_tensor.Slice(1, 2);
  TensorCopy(slice_tensor, *cpu_place, &dst_tensor);
  const int* slice_ptr = slice_tensor.data<int>();
  dst_ptr = dst_tensor.data<int>();
  EXPECT_NE(dst_ptr, slice_ptr);
  for (size_t i = 0; i < 3; ++i) {
    EXPECT_EQ(dst_ptr[i], slice_ptr[i]);
  }
  EXPECT_TRUE(dst_tensor.layout() == src_tensor.layout());

#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  {
    phi::DenseTensor src_tensor;
    phi::DenseTensor gpu_tensor;
    phi::DenseTensor dst_tensor;

    int* src_ptr = src_tensor.mutable_data<int>(common::make_ddim({3, 3}),
                                                phi::CPUPlace());

    std::array<int, 9> arr = {1, 2, 3, 4, 5, 6, 7, 8, 9};
    memcpy(src_ptr, arr.data(), 9 * sizeof(int));

    // CPU phi::DenseTensor to GPU phi::DenseTensor
    auto gpu_place = new phi::GPUPlace(0);
    phi::GPUContext gpu_ctx(*gpu_place);
    gpu_ctx.SetAllocator(paddle::memory::allocation::AllocatorFacade::Instance()
                             .GetAllocator(*gpu_place, gpu_ctx.stream())
                             .get());
    gpu_ctx.PartialInitWithAllocator();
    TensorCopy(src_tensor, *gpu_place, gpu_ctx, &gpu_tensor);