#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 >
static void	test_simple_padding (const Eigen::SyclDevice &sycl_device)

template<typename DataType , int DataLayout, typename IndexType >
static void	test_padded_expr (const Eigen::SyclDevice &sycl_device)

template<typename DataType , typename dev_Selector >
void	sycl_padding_test_per_device (dev_Selector s)

	EIGEN_DECLARE_TEST (cxx11_tensor_padding_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_padding_sycl )

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

References CALL_SUBTEST.

◆ sycl_padding_test_per_device()

template<typename DataType , typename dev_Selector >

void sycl_padding_test_per_device ( dev_Selector s )

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

References s.

◆ test_padded_expr()

template<typename DataType , int DataLayout, typename IndexType >

static void test_padded_expr ( const Eigen::SyclDevice & sycl_device )

static

                                                                  {
   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.setRandom();
  
   array<std::pair<IndexType, IndexType>, 4> paddings;
   paddings[0] = std::make_pair(0, 0);
   paddings[1] = std::make_pair(2, 1);
   paddings[2] = std::make_pair(3, 4);
   paddings[3] = std::make_pair(0, 0);
  
   Eigen::DSizes<IndexType, 2> reshape_dims;
   reshape_dims[0] = 12;
   reshape_dims[1] = 84;
  
   Tensor<DataType, 2, DataLayout, IndexType> result(reshape_dims);
  
   DataType* gpu_data1 = static_cast<DataType*>(sycl_device.allocate(tensor.size() * sizeof(DataType)));
   DataType* gpu_data2 = static_cast<DataType*>(sycl_device.allocate(result.size() * sizeof(DataType)));
   TensorMap<Tensor<DataType, 4, DataLayout, IndexType>> gpu1(gpu_data1, tensorRange);
   TensorMap<Tensor<DataType, 2, DataLayout, IndexType>> gpu2(gpu_data2, reshape_dims);
  
   sycl_device.memcpyHostToDevice(gpu_data1, tensor.data(), (tensor.size()) * sizeof(DataType));
   gpu2.device(sycl_device) = gpu1.pad(paddings).reshape(reshape_dims);
   sycl_device.memcpyDeviceToHost(result.data(), gpu_data2, (result.size()) * sizeof(DataType));
  
   for (IndexType i = 0; i < 2; ++i) {
     for (IndexType j = 0; j < 6; ++j) {
       for (IndexType k = 0; k < 12; ++k) {
         for (IndexType l = 0; l < 7; ++l) {
           const float result_value =
               DataLayout == ColMajor ? result(i + 2 * j, k + 12 * l) : result(j + 6 * i, l + 7 * k);
           if (j >= 2 && j < 5 && k >= 3 && k < 8) {
             VERIFY_IS_EQUAL(result_value, tensor(i, j - 2, k - 3, l));
           } else {
             VERIFY_IS_EQUAL(result_value, 0.0f);
           }
         }
       }
     }
   }
   sycl_device.deallocate(gpu_data1);
   sycl_device.deallocate(gpu_data2);
 }

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

◆ test_simple_padding()