devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged in benoitsteiner/opencl (pull request PR-246)

Browse Source 
Improved support for OpenCL
This commit is contained in:
Benoit Steiner
2016-11-08 22:28:44 +00:00
parent 73985ead27 dcc14bee64
commit db3903498d
24 changed files with 879 additions and 469 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
unsupported/test/CMakeLists.txt
View File

@@ -1,6 +1,6 @@
# generate split test header file only if it does not yet exist
# in order to prevent a rebuild everytime cmake is configured
if(NOT EXISTS ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h)
if(NOT EXISTS ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h)
file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h "")
foreach(i RANGE 1 999)
file(APPEND ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h
@@ -16,11 +16,11 @@ endif()
set_property(GLOBAL PROPERTY EIGEN_CURRENT_SUBPROJECT "Unsupported")
add_custom_target(BuildUnsupported)
include_directories(../../test ../../unsupported ../../Eigen
include_directories(../../test ../../unsupported ../../Eigen
${CMAKE_CURRENT_BINARY_DIR}/../../test)
find_package (Threads)
find_package(GoogleHash)
if(GOOGLEHASH_FOUND)
add_definitions("-DEIGEN_GOOGLEHASH_SUPPORT")
@@ -134,7 +134,7 @@ ei_add_test(cxx11_tensor_roundings)
ei_add_test(cxx11_tensor_layout_swap)
ei_add_test(cxx11_tensor_io)
if("${CMAKE_SIZEOF_VOID_P}" EQUAL "8")
# This test requires __uint128_t which is only available on 64bit systems
# This test requires __uint128_t which is only available on 64bit systems
ei_add_test(cxx11_tensor_uint128)
endif()
endif()
@@ -145,6 +145,7 @@ if(EIGEN_TEST_CXX11)
ei_add_test_sycl(cxx11_tensor_forced_eval_sycl "-std=c++11")
ei_add_test_sycl(cxx11_tensor_broadcast_sycl "-std=c++11")
ei_add_test_sycl(cxx11_tensor_device_sycl "-std=c++11")
ei_add_test_sycl(cxx11_tensor_reduction_sycl "-std=c++11")
endif(EIGEN_TEST_SYCL)
# It should be safe to always run these tests as there is some fallback code for
# older compiler that don't support cxx11.

85
unsupported/test/cxx11_tensor_broadcast_sycl.cpp
View File

@@ -25,55 +25,50 @@ using Eigen::SyclDevice;
using Eigen::Tensor;
using Eigen::TensorMap;
// Types used in tests:
using TestTensor = Tensor<float, 3>;
using TestTensorMap = TensorMap<Tensor<float, 3>>;
static void test_broadcast_sycl(){
static void test_broadcast_sycl(const Eigen::SyclDevice &sycl_device){
cl::sycl::gpu_selector s;
cl::sycl::queue q(s, [=](cl::sycl::exception_list l) {
for (const auto& e : l) {
try {
std::rethrow_exception(e);
} catch (cl::sycl::exception e) {
std::cout << e.what() << std::endl;
// BROADCAST test:
array<int, 4> in_range = {{2, 3, 5, 7}};
array<int, 4> broadcasts = {{2, 3, 1, 4}};
array<int, 4> out_range; // = in_range * broadcasts
for (size_t i = 0; i < out_range.size(); ++i)
out_range[i] = in_range[i] * broadcasts[i];
Tensor<float, 4> input(in_range);
Tensor<float, 4> out(out_range);
for (size_t i = 0; i < in_range.size(); ++i)
VERIFY_IS_EQUAL(out.dimension(i), out_range[i]);
for (int i = 0; i < input.size(); ++i)
input(i) = static_cast<float>(i);
float * gpu_in_data = static_cast<float*>(sycl_device.allocate(input.dimensions().TotalSize()*sizeof(float)));
float * gpu_out_data = static_cast<float*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(float)));
TensorMap<Tensor<float, 4>> gpu_in(gpu_in_data, in_range);
TensorMap<Tensor<float, 4>> gpu_out(gpu_out_data, out_range);
sycl_device.memcpyHostToDevice(gpu_in_data, input.data(),(input.dimensions().TotalSize())*sizeof(float));
gpu_out.device(sycl_device) = gpu_in.broadcast(broadcasts);
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < 4; ++i) {
for (int j = 0; j < 9; ++j) {
for (int k = 0; k < 5; ++k) {
for (int l = 0; l < 28; ++l) {
VERIFY_IS_APPROX(input(i%2,j%3,k%5,l%7), out(i,j,k,l));
}
}
}
});
SyclDevice sycl_device(q);
// BROADCAST test:
array<int, 4> in_range = {{2, 3, 5, 7}};
array<int, in_range.size()> broadcasts = {{2, 3, 1, 4}};
array<int, in_range.size()> out_range; // = in_range * broadcasts
for (size_t i = 0; i < out_range.size(); ++i)
out_range[i] = in_range[i] * broadcasts[i];
Tensor<float, in_range.size()> input(in_range);
Tensor<float, out_range.size()> output(out_range);
for (int i = 0; i < input.size(); ++i)
input(i) = static_cast<float>(i);
TensorMap<decltype(input)> gpu_in(input.data(), in_range);
TensorMap<decltype(output)> gpu_out(output.data(), out_range);
gpu_out.device(sycl_device) = gpu_in.broadcast(broadcasts);
sycl_device.deallocate(output.data());
for (size_t i = 0; i < in_range.size(); ++i)
VERIFY_IS_EQUAL(output.dimension(i), out_range[i]);
for (int i = 0; i < 4; ++i) {
for (int j = 0; j < 9; ++j) {
for (int k = 0; k < 5; ++k) {
for (int l = 0; l < 28; ++l) {
VERIFY_IS_APPROX(input(i%2,j%3,k%5,l%7), output(i,j,k,l));
}
}
}
}
printf("Broadcast Test Passed\n");
}
printf("Broadcast Test Passed\n");
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
void test_cxx11_tensor_broadcast_sycl() {
CALL_SUBTEST(test_broadcast_sycl());
cl::sycl::gpu_selector s;
Eigen::SyclDevice sycl_device(s);
CALL_SUBTEST(test_broadcast_sycl(sycl_device));
}

20
unsupported/test/cxx11_tensor_device_sycl.cpp
View File

@@ -20,20 +20,12 @@
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
void test_device_sycl() {
cl::sycl::gpu_selector s;
cl::sycl::queue q(s, [=](cl::sycl::exception_list l) {
for (const auto& e : l) {
try {
std::rethrow_exception(e);
} catch (cl::sycl::exception e) {
std::cout << e.what() << std::endl;
}
}
});
Eigen::SyclDevice sycl_device(q);
printf("Helo from ComputeCpp: Device Exists\n");
void test_device_sycl(const Eigen::SyclDevice &sycl_device) {
std::cout <<"Helo from ComputeCpp: the requested device exists and the device name is : "
<< sycl_device.m_queue.get_device(). template get_info<cl::sycl::info::device::name>() <<std::endl;;
}
void test_cxx11_tensor_device_sycl() {
CALL_SUBTEST(test_device_sycl());
cl::sycl::gpu_selector s;
Eigen::SyclDevice sycl_device(s);
CALL_SUBTEST(test_device_sycl(sycl_device));
}

44
unsupported/test/cxx11_tensor_forced_eval_sycl.cpp
View File

@@ -22,18 +22,7 @@
using Eigen::Tensor;
void test_forced_eval_sycl() {
cl::sycl::gpu_selector s;
cl::sycl::queue q(s, [=](cl::sycl::exception_list l) {
for (const auto& e : l) {
try {
std::rethrow_exception(e);
} catch (cl::sycl::exception e) {
std::cout << e.what() << std::endl;
}
}
});
SyclDevice sycl_device(q);
void test_forced_eval_sycl(const Eigen::SyclDevice &sycl_device) {
int sizeDim1 = 100;
int sizeDim2 = 200;
@@ -43,17 +32,22 @@ void test_forced_eval_sycl() {
Eigen::Tensor<float, 3> in2(tensorRange);
Eigen::Tensor<float, 3> out(tensorRange);
float * gpu_in1_data = static_cast<float*>(sycl_device.allocate(in1.dimensions().TotalSize()*sizeof(float)));
float * gpu_in2_data = static_cast<float*>(sycl_device.allocate(in2.dimensions().TotalSize()*sizeof(float)));
float * gpu_out_data = static_cast<float*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(float)));
in1 = in1.random() + in1.constant(10.0f);
in2 = in2.random() + in2.constant(10.0f);
// creating TensorMap from tensor
Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in1(in1.data(), tensorRange);
Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in2(in2.data(), tensorRange);
Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_out(out.data(), tensorRange);
// creating TensorMap from tensor
Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in1(gpu_in1_data, tensorRange);
Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_in2(gpu_in2_data, tensorRange);
Eigen::TensorMap<Eigen::Tensor<float, 3>> gpu_out(gpu_out_data, tensorRange);
sycl_device.memcpyHostToDevice(gpu_in1_data, in1.data(),(in1.dimensions().TotalSize())*sizeof(float));
sycl_device.memcpyHostToDevice(gpu_in2_data, in2.data(),(in1.dimensions().TotalSize())*sizeof(float));
/// c=(a+b)*b
gpu_out.device(sycl_device) =(gpu_in1 + gpu_in2).eval() * gpu_in2;
sycl_device.deallocate(out.data());
gpu_out.device(sycl_device) =(gpu_in1 + gpu_in2).eval() * gpu_in2;
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -62,7 +56,15 @@ void test_forced_eval_sycl() {
}
}
}
printf("(a+b)*b Test Passed\n");
printf("(a+b)*b Test Passed\n");
sycl_device.deallocate(gpu_in1_data);
sycl_device.deallocate(gpu_in2_data);
sycl_device.deallocate(gpu_out_data);
}
void test_cxx11_tensor_forced_eval_sycl() { CALL_SUBTEST(test_forced_eval_sycl()); }
void test_cxx11_tensor_forced_eval_sycl() {
cl::sycl::gpu_selector s;
Eigen::SyclDevice sycl_device(s);
CALL_SUBTEST(test_forced_eval_sycl(sycl_device));
}

138
unsupported/test/cxx11_tensor_reduction_sycl.cpp Normal file
View File

@@ -0,0 +1,138 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2015
// Mehdi Goli Codeplay Software Ltd.
// Ralph Potter Codeplay Software Ltd.
// Luke Iwanski Codeplay Software Ltd.
// Contact: <eigen@codeplay.com>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#define EIGEN_TEST_NO_LONGDOUBLE
#define EIGEN_TEST_NO_COMPLEX
#define EIGEN_TEST_FUNC cxx11_tensor_reduction_sycl
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
#define EIGEN_USE_SYCL
#include "main.h"
#include <unsupported/Eigen/CXX11/Tensor>
static void test_full_reductions_sycl(const Eigen::SyclDevice& sycl_device) {
const int num_rows = 452;
const int num_cols = 765;
array<int, 2> tensorRange = {{num_rows, num_cols}};
Tensor<float, 2> in(tensorRange);
Tensor<float, 0> full_redux;
Tensor<float, 0> full_redux_gpu;
in.setRandom();
full_redux = in.sum();
float* gpu_in_data = static_cast<float*>(sycl_device.allocate(in.dimensions().TotalSize()*sizeof(float)));
float* gpu_out_data =(float*)sycl_device.allocate(sizeof(float));
TensorMap<Tensor<float, 2> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<float, 0> > out_gpu(gpu_out_data);
sycl_device.memcpyHostToDevice(gpu_in_data, in.data(),(in.dimensions().TotalSize())*sizeof(float));
out_gpu.device(sycl_device) = in_gpu.sum();
sycl_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_data, sizeof(float));
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux_gpu(), full_redux());
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
static void test_first_dim_reductions_sycl(const Eigen::SyclDevice& sycl_device) {
int dim_x = 145;
int dim_y = 1;
int dim_z = 67;
array<int, 3> tensorRange = {{dim_x, dim_y, dim_z}};
Eigen::array<int, 1> red_axis;
red_axis[0] = 0;
array<int, 2> reduced_tensorRange = {{dim_y, dim_z}};
Tensor<float, 3> in(tensorRange);
Tensor<float, 2> redux(reduced_tensorRange);
Tensor<float, 2> redux_gpu(reduced_tensorRange);
in.setRandom();
redux= in.sum(red_axis);
float* gpu_in_data = static_cast<float*>(sycl_device.allocate(in.dimensions().TotalSize()*sizeof(float)));
float* gpu_out_data = static_cast<float*>(sycl_device.allocate(redux_gpu.dimensions().TotalSize()*sizeof(float)));
TensorMap<Tensor<float, 3> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<float, 2> > out_gpu(gpu_out_data, reduced_tensorRange);
sycl_device.memcpyHostToDevice(gpu_in_data, in.data(),(in.dimensions().TotalSize())*sizeof(float));
out_gpu.device(sycl_device) = in_gpu.sum(red_axis);
sycl_device.memcpyDeviceToHost(redux_gpu.data(), gpu_out_data, redux_gpu.dimensions().TotalSize()*sizeof(float));
// Check that the CPU and GPU reductions return the same result.
for(int j=0; j<reduced_tensorRange[0]; j++ )
for(int k=0; k<reduced_tensorRange[1]; k++ )
VERIFY_IS_APPROX(redux_gpu(j,k), redux(j,k));
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
static void test_last_dim_reductions_sycl(const Eigen::SyclDevice &sycl_device) {
int dim_x = 567;
int dim_y = 1;
int dim_z = 47;
array<int, 3> tensorRange = {{dim_x, dim_y, dim_z}};
Eigen::array<int, 1> red_axis;
red_axis[0] = 2;
array<int, 2> reduced_tensorRange = {{dim_x, dim_y}};
Tensor<float, 3> in(tensorRange);
Tensor<float, 2> redux(reduced_tensorRange);
Tensor<float, 2> redux_gpu(reduced_tensorRange);
in.setRandom();
redux= in.sum(red_axis);
float* gpu_in_data = static_cast<float*>(sycl_device.allocate(in.dimensions().TotalSize()*sizeof(float)));
float* gpu_out_data = static_cast<float*>(sycl_device.allocate(redux_gpu.dimensions().TotalSize()*sizeof(float)));
TensorMap<Tensor<float, 3> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<float, 2> > out_gpu(gpu_out_data, reduced_tensorRange);
sycl_device.memcpyHostToDevice(gpu_in_data, in.data(),(in.dimensions().TotalSize())*sizeof(float));
out_gpu.device(sycl_device) = in_gpu.sum(red_axis);
sycl_device.memcpyDeviceToHost(redux_gpu.data(), gpu_out_data, redux_gpu.dimensions().TotalSize()*sizeof(float));
// Check that the CPU and GPU reductions return the same result.
for(int j=0; j<reduced_tensorRange[0]; j++ )
for(int k=0; k<reduced_tensorRange[1]; k++ )
VERIFY_IS_APPROX(redux_gpu(j,k), redux(j,k));
sycl_device.deallocate(gpu_in_data);
sycl_device.deallocate(gpu_out_data);
}
void test_cxx11_tensor_reduction_sycl() {
cl::sycl::gpu_selector s;
Eigen::SyclDevice sycl_device(s);
CALL_SUBTEST((test_full_reductions_sycl(sycl_device)));
CALL_SUBTEST((test_first_dim_reductions_sycl(sycl_device)));
CALL_SUBTEST((test_last_dim_reductions_sycl(sycl_device)));
}

67
unsupported/test/cxx11_tensor_sycl.cpp
View File

@@ -27,42 +27,33 @@ using Eigen::SyclDevice;
using Eigen::Tensor;
using Eigen::TensorMap;
// Types used in tests:
using TestTensor = Tensor<float, 3>;
using TestTensorMap = TensorMap<Tensor<float, 3>>;
void test_sycl_cpu() {
cl::sycl::gpu_selector s;
cl::sycl::queue q(s, [=](cl::sycl::exception_list l) {
for (const auto& e : l) {
try {
std::rethrow_exception(e);
} catch (cl::sycl::exception e) {
std::cout << e.what() << std::endl;
}
}
});
SyclDevice sycl_device(q);
void test_sycl_cpu(const Eigen::SyclDevice &sycl_device) {
int sizeDim1 = 100;
int sizeDim2 = 100;
int sizeDim3 = 100;
array<int, 3> tensorRange = {{sizeDim1, sizeDim2, sizeDim3}};
TestTensor in1(tensorRange);
TestTensor in2(tensorRange);
TestTensor in3(tensorRange);
TestTensor out(tensorRange);
in1 = in1.random();
Tensor<float, 3> in1(tensorRange);
Tensor<float, 3> in2(tensorRange);
Tensor<float, 3> in3(tensorRange);
Tensor<float, 3> out(tensorRange);
in2 = in2.random();
in3 = in3.random();
TestTensorMap gpu_in1(in1.data(), tensorRange);
TestTensorMap gpu_in2(in2.data(), tensorRange);
TestTensorMap gpu_in3(in3.data(), tensorRange);
TestTensorMap gpu_out(out.data(), tensorRange);
float * gpu_in1_data = static_cast<float*>(sycl_device.allocate(in1.dimensions().TotalSize()*sizeof(float)));
float * gpu_in2_data = static_cast<float*>(sycl_device.allocate(in2.dimensions().TotalSize()*sizeof(float)));
float * gpu_in3_data = static_cast<float*>(sycl_device.allocate(in3.dimensions().TotalSize()*sizeof(float)));
float * gpu_out_data = static_cast<float*>(sycl_device.allocate(out.dimensions().TotalSize()*sizeof(float)));
TensorMap<Tensor<float, 3>> gpu_in1(gpu_in1_data, tensorRange);
TensorMap<Tensor<float, 3>> gpu_in2(gpu_in2_data, tensorRange);
TensorMap<Tensor<float, 3>> gpu_in3(gpu_in3_data, tensorRange);
TensorMap<Tensor<float, 3>> gpu_out(gpu_out_data, tensorRange);
/// a=1.2f
gpu_in1.device(sycl_device) = gpu_in1.constant(1.2f);
sycl_device.deallocate(in1.data());
sycl_device.memcpyDeviceToHost(in1.data(), gpu_in1_data ,(in1.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -74,7 +65,7 @@ void test_sycl_cpu() {
/// a=b*1.2f
gpu_out.device(sycl_device) = gpu_in1 * 1.2f;
sycl_device.deallocate(out.data());
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data ,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -86,8 +77,9 @@ void test_sycl_cpu() {
printf("a=b*1.2f Test Passed\n");
/// c=a*b
sycl_device.memcpyHostToDevice(gpu_in2_data, in2.data(),(in2.dimensions().TotalSize())*sizeof(float));
gpu_out.device(sycl_device) = gpu_in1 * gpu_in2;
sycl_device.deallocate(out.data());
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -101,7 +93,7 @@ void test_sycl_cpu() {
/// c=a+b
gpu_out.device(sycl_device) = gpu_in1 + gpu_in2;
sycl_device.deallocate(out.data());
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -115,7 +107,7 @@ void test_sycl_cpu() {
/// c=a*a
gpu_out.device(sycl_device) = gpu_in1 * gpu_in1;
sycl_device.deallocate(out.data());
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -125,12 +117,11 @@ void test_sycl_cpu() {
}
}
}
printf("c= a*a Test Passed\n");
//a*3.14f + b*2.7f
gpu_out.device(sycl_device) = gpu_in1 * gpu_in1.constant(3.14f) + gpu_in2 * gpu_in2.constant(2.7f);
sycl_device.deallocate(out.data());
sycl_device.memcpyDeviceToHost(out.data(),gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -143,8 +134,9 @@ void test_sycl_cpu() {
printf("a*3.14f + b*2.7f Test Passed\n");
///d= (a>0.5? b:c)
sycl_device.memcpyHostToDevice(gpu_in3_data, in3.data(),(in3.dimensions().TotalSize())*sizeof(float));
gpu_out.device(sycl_device) =(gpu_in1 > gpu_in1.constant(0.5f)).select(gpu_in2, gpu_in3);
sycl_device.deallocate(out.data());
sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.dimensions().TotalSize())*sizeof(float));
for (int i = 0; i < sizeDim1; ++i) {
for (int j = 0; j < sizeDim2; ++j) {
for (int k = 0; k < sizeDim3; ++k) {
@@ -155,8 +147,13 @@ void test_sycl_cpu() {
}
}
printf("d= (a>0.5? b:c) Test Passed\n");
sycl_device.deallocate(gpu_in1_data);
sycl_device.deallocate(gpu_in2_data);
sycl_device.deallocate(gpu_in3_data);
sycl_device.deallocate(gpu_out_data);
}
void test_cxx11_tensor_sycl() {
CALL_SUBTEST(test_sycl_cpu());
cl::sycl::gpu_selector s;
Eigen::SyclDevice sycl_device(s);
CALL_SUBTEST(test_sycl_cpu(sycl_device));
}