cxx11_tensor_device_sycl.cpp File Reference

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

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_device_memory (const Eigen::SyclDevice &sycl_device)
template<typename DataType , int DataLayout, typename IndexType >
void	test_device_exceptions (const Eigen::SyclDevice &sycl_device)
template<typename DataType >
void	sycl_device_test_per_device (const cl::sycl::device &d)
	EIGEN_DECLARE_TEST (cxx11_tensor_device_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_device_sycl

)

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

References CALL_SUBTEST, and sycl_device_test_per_device().

sycl_device_test_per_device()

template<typename DataType >

void sycl_device_test_per_device

(

const cl::sycl::device &

d

)

this test throw an exception. enable it if you want to see the exception

this test throw an exception. enable it if you want to see the exception

                                                        {
  std::cout << "Running on " << d.template get_info<cl::sycl::info::device::name>() << std::endl;
  QueueInterface queueInterface(d);
  auto sycl_device = Eigen::SyclDevice(&queueInterface);
  test_device_memory<DataType, RowMajor, int64_t>(sycl_device);
  test_device_memory<DataType, ColMajor, int64_t>(sycl_device);
  // test_device_exceptions<DataType, RowMajor>(sycl_device);
  // test_device_exceptions<DataType, ColMajor>(sycl_device);
}

Referenced by EIGEN_DECLARE_TEST().

test_device_exceptions()

template<typename DataType , int DataLayout, typename IndexType >

void test_device_exceptions

(

const Eigen::SyclDevice &

sycl_device

)

                                                                {
  VERIFY(sycl_device.ok());
  IndexType sizeDim1 = 100;
  array<IndexType, 1> tensorDims = {{sizeDim1}};
  DataType* gpu_data = static_cast<DataType*>(sycl_device.allocate(sizeDim1 * sizeof(DataType)));
  sycl_device.memset(gpu_data, 1, sizeDim1 * sizeof(DataType));
 
  TensorMap<Tensor<DataType, 1, DataLayout, IndexType>> in(gpu_data, tensorDims);
  TensorMap<Tensor<DataType, 1, DataLayout, IndexType>> out(gpu_data, tensorDims);
  out.device(sycl_device) = in / in.constant(0);
 
  sycl_device.synchronize();
  VERIFY(!sycl_device.ok());
  sycl_device.deallocate(gpu_data);
}

References out(), and VERIFY.

test_device_memory()

template<typename DataType , int DataLayout, typename IndexType >

void test_device_memory

(

const Eigen::SyclDevice &

sycl_device

)

                                                            {
  IndexType sizeDim1 = 100;
  array<IndexType, 1> tensorRange = {{sizeDim1}};
  Tensor<DataType, 1, DataLayout, IndexType> in(tensorRange);
  Tensor<DataType, 1, DataLayout, IndexType> in1(tensorRange);
  DataType* gpu_in_data = static_cast<DataType*>(sycl_device.allocate(in.size() * sizeof(DataType)));
 
  // memset
  memset(in1.data(), 1, in1.size() * sizeof(DataType));
  sycl_device.memset(gpu_in_data, 1, in.size() * sizeof(DataType));
  sycl_device.memcpyDeviceToHost(in.data(), gpu_in_data, in.size() * sizeof(DataType));
  for (IndexType i = 0; i < in.size(); i++) {
    VERIFY_IS_EQUAL(in(i), in1(i));
  }
 
  // fill
  DataType value = DataType(7);
  std::fill_n(in1.data(), in1.size(), value);
  sycl_device.fill(gpu_in_data, gpu_in_data + in.size(), value);
  sycl_device.memcpyDeviceToHost(in.data(), gpu_in_data, in.size() * sizeof(DataType));
  for (IndexType i = 0; i < in.size(); i++) {
    VERIFY_IS_EQUAL(in(i), in1(i));
  }
 
  sycl_device.deallocate(gpu_in_data);
}

References Eigen::Tensor< Scalar_, NumIndices_, Options_, IndexType_ >::data(), i, Eigen::Tensor< Scalar_, NumIndices_, Options_, IndexType_ >::size(), Eigen::value, and VERIFY_IS_EQUAL.