#include <iostream>
#include <chrono>
#include <ctime>
#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_striding (const Eigen::SyclDevice &sycl_device)

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

template<typename Dev_selector >
void	tensorStridingPerDevice (Dev_selector &s)

	EIGEN_DECLARE_TEST (cxx11_tensor_striding_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_striding_sycl )

                                                {
   for (const auto& device : Eigen::get_sycl_supported_devices()) {
     CALL_SUBTEST(tensorStridingPerDevice(device));
   }
 }

References CALL_SUBTEST, and tensorStridingPerDevice().

◆ tensorStridingPerDevice()

template<typename Dev_selector >

void tensorStridingPerDevice ( Dev_selector & s )

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

References s.

Referenced by EIGEN_DECLARE_TEST().

◆ test_simple_striding()

template<typename DataType , int DataLayout, typename IndexType >

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

static

                                                                      {
   Eigen::array<IndexType, 4> tensor_dims = {{2, 3, 5, 7}};
   Eigen::array<IndexType, 4> stride_dims = {{1, 1, 3, 3}};
  
   Tensor<DataType, 4, DataLayout, IndexType> tensor(tensor_dims);
   Tensor<DataType, 4, DataLayout, IndexType> no_stride(tensor_dims);
   Tensor<DataType, 4, DataLayout, IndexType> stride(stride_dims);
  
   std::size_t tensor_bytes = tensor.size() * sizeof(DataType);
   std::size_t no_stride_bytes = no_stride.size() * sizeof(DataType);
   std::size_t stride_bytes = stride.size() * sizeof(DataType);
   DataType* d_tensor = static_cast<DataType*>(sycl_device.allocate(tensor_bytes));
   DataType* d_no_stride = static_cast<DataType*>(sycl_device.allocate(no_stride_bytes));
   DataType* d_stride = static_cast<DataType*>(sycl_device.allocate(stride_bytes));
  
   Eigen::TensorMap<Eigen::Tensor<DataType, 4, DataLayout, IndexType> > gpu_tensor(d_tensor, tensor_dims);
   Eigen::TensorMap<Eigen::Tensor<DataType, 4, DataLayout, IndexType> > gpu_no_stride(d_no_stride, tensor_dims);
   Eigen::TensorMap<Eigen::Tensor<DataType, 4, DataLayout, IndexType> > gpu_stride(d_stride, stride_dims);
  
   tensor.setRandom();
   array<IndexType, 4> strides;
   strides[0] = 1;
   strides[1] = 1;
   strides[2] = 1;
   strides[3] = 1;
   sycl_device.memcpyHostToDevice(d_tensor, tensor.data(), tensor_bytes);
   gpu_no_stride.device(sycl_device) = gpu_tensor.stride(strides);
   sycl_device.memcpyDeviceToHost(no_stride.data(), d_no_stride, no_stride_bytes);
  
   // no_stride = tensor.stride(strides);
  
   VERIFY_IS_EQUAL(no_stride.dimension(0), 2);
   VERIFY_IS_EQUAL(no_stride.dimension(1), 3);
   VERIFY_IS_EQUAL(no_stride.dimension(2), 5);
   VERIFY_IS_EQUAL(no_stride.dimension(3), 7);
  
   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;
   // Tensor<float, 4, DataLayout> stride;
   //   stride = tensor.stride(strides);
  
   gpu_stride.device(sycl_device) = gpu_tensor.stride(strides);
   sycl_device.memcpyDeviceToHost(stride.data(), d_stride, stride_bytes);
  
   VERIFY_IS_EQUAL(stride.dimension(0), 1);
   VERIFY_IS_EQUAL(stride.dimension(1), 1);
   VERIFY_IS_EQUAL(stride.dimension(2), 3);
   VERIFY_IS_EQUAL(stride.dimension(3), 3);
  
   for (IndexType i = 0; i < 1; ++i) {
     for (IndexType j = 0; j < 1; ++j) {
       for (IndexType k = 0; k < 3; ++k) {
         for (IndexType l = 0; l < 3; ++l) {
           VERIFY_IS_EQUAL(tensor(2 * i, 4 * j, 2 * k, 3 * l), stride(i, j, k, l));
         }
       }
     }
   }
  
   sycl_device.deallocate(d_tensor);
   sycl_device.deallocate(d_no_stride);
   sycl_device.deallocate(d_stride);
 }

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::TensorBase< Derived, AccessLevel >::stride(), Eigen::internal::strides(), and VERIFY_IS_EQUAL.

◆ test_striding_as_lvalue()