cxx11_tensor_inflation_sycl.cpp File Reference

#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>

Macros
#define	EIGEN_TEST_NO_LONGDOUBLE
#define	EIGEN_TEST_NO_COMPLEX
#define	EIGEN_DEFAULT_DENSE_INDEX_TYPE   int64_t
#define	EIGEN_USE_SYCL

Functions
template<typename DataType , int DataLayout, typename IndexType >
void	test_simple_inflation_sycl (const Eigen::SyclDevice &sycl_device)
template<typename DataType , typename dev_Selector >
void	sycl_inflation_test_per_device (dev_Selector s)
	EIGEN_DECLARE_TEST (cxx11_tensor_inflation_sycl)

Macro Definition Documentation

EIGEN_DEFAULT_DENSE_INDEX_TYPE

#define EIGEN_DEFAULT_DENSE_INDEX_TYPE   int64_t

EIGEN_TEST_NO_COMPLEX

#define EIGEN_TEST_NO_COMPLEX

EIGEN_TEST_NO_LONGDOUBLE

#define EIGEN_TEST_NO_LONGDOUBLE

EIGEN_USE_SYCL

#define EIGEN_USE_SYCL

Function Documentation

EIGEN_DECLARE_TEST()

EIGEN_DECLARE_TEST

(

cxx11_tensor_inflation_sycl

)

                                                {
  for (const auto& device : Eigen::get_sycl_supported_devices()) {
    CALL_SUBTEST(sycl_inflation_test_per_device<half>(device));
    CALL_SUBTEST(sycl_inflation_test_per_device<float>(device));
  }
}

References CALL_SUBTEST.

sycl_inflation_test_per_device()

template<typename DataType , typename dev_Selector >

void sycl_inflation_test_per_device

(

dev_Selector

s

)

                                                    {
  QueueInterface queueInterface(s);
  auto sycl_device = Eigen::SyclDevice(&queueInterface);
  test_simple_inflation_sycl<DataType, RowMajor, int64_t>(sycl_device);
  test_simple_inflation_sycl<DataType, ColMajor, int64_t>(sycl_device);
}

References s.

test_simple_inflation_sycl()

template<typename DataType , int DataLayout, typename IndexType >

void test_simple_inflation_sycl

(

const Eigen::SyclDevice &

sycl_device

)

                                                                    {
  IndexType sizeDim1 = 2;
  IndexType sizeDim2 = 3;
  IndexType sizeDim3 = 5;
  IndexType sizeDim4 = 7;
  array<IndexType, 4> tensorRange = {{sizeDim1, sizeDim2, sizeDim3, sizeDim4}};
  Tensor<DataType, 4, DataLayout, IndexType> tensor(tensorRange);
  Tensor<DataType, 4, DataLayout, IndexType> no_stride(tensorRange);
  tensor.setRandom();
 
  array<IndexType, 4> strides;
  strides[0] = 1;
  strides[1] = 1;
  strides[2] = 1;
  strides[3] = 1;
 
  const size_t tensorBuffSize = tensor.size() * sizeof(DataType);
  DataType* gpu_data_tensor = static_cast<DataType*>(sycl_device.allocate(tensorBuffSize));
  DataType* gpu_data_no_stride = static_cast<DataType*>(sycl_device.allocate(tensorBuffSize));
 
  TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu_tensor(gpu_data_tensor, tensorRange);
  TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu_no_stride(gpu_data_no_stride, tensorRange);
 
  sycl_device.memcpyHostToDevice(gpu_data_tensor, tensor.data(), tensorBuffSize);
  gpu_no_stride.device(sycl_device) = gpu_tensor.inflate(strides);
  sycl_device.memcpyDeviceToHost(no_stride.data(), gpu_data_no_stride, tensorBuffSize);
 
  VERIFY_IS_EQUAL(no_stride.dimension(0), sizeDim1);
  VERIFY_IS_EQUAL(no_stride.dimension(1), sizeDim2);
  VERIFY_IS_EQUAL(no_stride.dimension(2), sizeDim3);
  VERIFY_IS_EQUAL(no_stride.dimension(3), sizeDim4);
 
  for (IndexType i = 0; i < 2; ++i) {
    for (IndexType j = 0; j < 3; ++j) {
      for (IndexType k = 0; k < 5; ++k) {
        for (IndexType l = 0; l < 7; ++l) {
          VERIFY_IS_EQUAL(tensor(i, j, k, l), no_stride(i, j, k, l));
        }
      }
    }
  }
 
  strides[0] = 2;
  strides[1] = 4;
  strides[2] = 2;
  strides[3] = 3;
 
  IndexType inflatedSizeDim1 = 3;
  IndexType inflatedSizeDim2 = 9;
  IndexType inflatedSizeDim3 = 9;
  IndexType inflatedSizeDim4 = 19;
  array<IndexType, 4> inflatedTensorRange = {{inflatedSizeDim1, inflatedSizeDim2, inflatedSizeDim3, inflatedSizeDim4}};
 
  Tensor<DataType, 4, DataLayout, IndexType> inflated(inflatedTensorRange);
 
  const size_t inflatedTensorBuffSize = inflated.size() * sizeof(DataType);
  DataType* gpu_data_inflated = static_cast<DataType*>(sycl_device.allocate(inflatedTensorBuffSize));
  TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu_inflated(gpu_data_inflated, inflatedTensorRange);
  gpu_inflated.device(sycl_device) = gpu_tensor.inflate(strides);
  sycl_device.memcpyDeviceToHost(inflated.data(), gpu_data_inflated, inflatedTensorBuffSize);
 
  VERIFY_IS_EQUAL(inflated.dimension(0), inflatedSizeDim1);
  VERIFY_IS_EQUAL(inflated.dimension(1), inflatedSizeDim2);
  VERIFY_IS_EQUAL(inflated.dimension(2), inflatedSizeDim3);
  VERIFY_IS_EQUAL(inflated.dimension(3), inflatedSizeDim4);
 
  for (IndexType i = 0; i < inflatedSizeDim1; ++i) {
    for (IndexType j = 0; j < inflatedSizeDim2; ++j) {
      for (IndexType k = 0; k < inflatedSizeDim3; ++k) {
        for (IndexType l = 0; l < inflatedSizeDim4; ++l) {
          if (i % strides[0] == 0 && j % strides[1] == 0 && k % strides[2] == 0 && l % strides[3] == 0) {
            VERIFY_IS_EQUAL(inflated(i, j, k, l),
                            tensor(i / strides[0], j / strides[1], k / strides[2], l / strides[3]));
          } else {
            VERIFY_IS_EQUAL(0, inflated(i, j, k, l));
          }
        }
      }
    }
  }
  sycl_device.deallocate(gpu_data_tensor);
  sycl_device.deallocate(gpu_data_no_stride);
  sycl_device.deallocate(gpu_data_inflated);
}

References Eigen::Tensor< Scalar_, NumIndices_, Options_, IndexType_ >::data(), Eigen::TensorBase< Derived, AccessLevel >::device(), Eigen::Tensor< Scalar_, NumIndices_, Options_, IndexType_ >::dimension(), i, j, k, Eigen::TensorBase< Derived, AccessLevel >::setRandom(), Eigen::Tensor< Scalar_, NumIndices_, Options_, IndexType_ >::size(), Eigen::internal::strides(), and VERIFY_IS_EQUAL.