mirror of
https://gitlab.com/libeigen/eigen.git
synced 2026-04-10 11:34:33 +08:00
Partial OpenCL support via SYCL compatible with ComputeCpp CE.
This commit is contained in:
Luke Iwanski
@@ -138,6 +138,13 @@ endif()
|
||||
endif()
|
||||
|
||||
if(EIGEN_TEST_CXX11)
|
||||
if(EIGEN_TEST_SYCL)
|
||||
ei_add_test_sycl(cxx11_tensor_sycl "-std=c++11")
|
||||
ei_add_test_sycl(cxx11_tensor_sycl_forced_eval "-std=c++11")
|
||||
ei_add_test_sycl(cxx11_tensor_sycl_broadcast "-std=c++11")
|
||||
ei_add_test_sycl(cxx11_tensor_sycl_device "-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.
|
||||
set(CMAKE_CXX_STANDARD 11)
|
||||
|
||||
157
unsupported/test/cxx11_tensor_sycl.cpp
Normal file
157
unsupported/test/cxx11_tensor_sycl.cpp
Normal file
@@ -0,0 +1,157 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.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_sycl
|
||||
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
|
||||
#define EIGEN_USE_SYCL
|
||||
|
||||
#include "main.h"
|
||||
#include <unsupported/Eigen/CXX11/Tensor>
|
||||
|
||||
using Eigen::array;
|
||||
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);
|
||||
|
||||
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();
|
||||
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);
|
||||
|
||||
/// a=1.2f
|
||||
gpu_in1.device(sycl_device) = gpu_in1.constant(1.2f);
|
||||
sycl_device.deallocate(in1.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(in1(i,j,k), 1.2f);
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("a=1.2f Test passed\n");
|
||||
|
||||
/// a=b*1.2f
|
||||
gpu_out.device(sycl_device) = gpu_in1 * 1.2f;
|
||||
sycl_device.deallocate(out.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i,j,k),
|
||||
in1(i,j,k) * 1.2f);
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("a=b*1.2f Test Passed\n");
|
||||
|
||||
/// c=a*b
|
||||
gpu_out.device(sycl_device) = gpu_in1 * gpu_in2;
|
||||
sycl_device.deallocate(out.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i,j,k),
|
||||
in1(i,j,k) *
|
||||
in2(i,j,k));
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("c=a*b Test Passed\n");
|
||||
|
||||
/// c=a+b
|
||||
gpu_out.device(sycl_device) = gpu_in1 + gpu_in2;
|
||||
sycl_device.deallocate(out.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i,j,k),
|
||||
in1(i,j,k) +
|
||||
in2(i,j,k));
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("c=a+b Test Passed\n");
|
||||
|
||||
/// c=a*a
|
||||
gpu_out.device(sycl_device) = gpu_in1 * gpu_in1;
|
||||
sycl_device.deallocate(out.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i,j,k),
|
||||
in1(i,j,k) *
|
||||
in1(i,j,k));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
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());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i,j,k),
|
||||
in1(i,j,k) * 3.14f
|
||||
+ in2(i,j,k) * 2.7f);
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("a*3.14f + b*2.7f Test Passed\n");
|
||||
|
||||
///d= (a>0.5? b:c)
|
||||
gpu_out.device(sycl_device) =(gpu_in1 > gpu_in1.constant(0.5f)).select(gpu_in2, gpu_in3);
|
||||
sycl_device.deallocate(out.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i, j, k), (in1(i, j, k) > 0.5f)
|
||||
? in2(i, j, k)
|
||||
: in3(i, j, k));
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("d= (a>0.5? b:c) Test Passed\n");
|
||||
|
||||
}
|
||||
void test_cxx11_tensor_sycl() {
|
||||
CALL_SUBTEST(test_sycl_cpu());
|
||||
}
|
||||
76
unsupported/test/cxx11_tensor_sycl_broadcast.cpp
Normal file
76
unsupported/test/cxx11_tensor_sycl_broadcast.cpp
Normal file
@@ -0,0 +1,76 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.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_sycl_broadcast
|
||||
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
|
||||
#define EIGEN_USE_SYCL
|
||||
|
||||
#include "main.h"
|
||||
#include <unsupported/Eigen/CXX11/Tensor>
|
||||
|
||||
using Eigen::array;
|
||||
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_sycl_broadcast(){
|
||||
|
||||
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);
|
||||
// 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");
|
||||
}
|
||||
|
||||
void test_cxx11_tensor_sycl_broadcast() {
|
||||
CALL_SUBTEST(test_sycl_broadcast());
|
||||
}
|
||||
37
unsupported/test/cxx11_tensor_sycl_device.cpp
Normal file
37
unsupported/test/cxx11_tensor_sycl_device.cpp
Normal file
@@ -0,0 +1,37 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.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_sycl_device
|
||||
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
|
||||
#define EIGEN_USE_SYCL
|
||||
|
||||
#include "main.h"
|
||||
#include <unsupported/Eigen/CXX11/Tensor>
|
||||
|
||||
|
||||
void test_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;
|
||||
}
|
||||
}
|
||||
});
|
||||
SyclDevice sycl_device(q);
|
||||
printf("Helo from ComputeCpp: Device Exists\n");
|
||||
}
|
||||
void test_cxx11_tensor_sycl_device() {
|
||||
CALL_SUBTEST(test_sycl_device());
|
||||
}
|
||||
64
unsupported/test/cxx11_tensor_sycl_forced_eval.cpp
Normal file
64
unsupported/test/cxx11_tensor_sycl_forced_eval.cpp
Normal file
@@ -0,0 +1,64 @@
|
||||
// This file is part of Eigen, a lightweight C++ template library
|
||||
// for linear algebra.
|
||||
//
|
||||
// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.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_sycl_forced_eval
|
||||
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
|
||||
#define EIGEN_USE_SYCL
|
||||
|
||||
#include "main.h"
|
||||
#include <unsupported/Eigen/CXX11/Tensor>
|
||||
|
||||
using Eigen::Tensor;
|
||||
|
||||
void test_sycl_gpu() {
|
||||
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);
|
||||
|
||||
int sizeDim1 = 100;
|
||||
int sizeDim2 = 200;
|
||||
int sizeDim3 = 200;
|
||||
Eigen::array<int, 3> tensorRange = {{sizeDim1, sizeDim2, sizeDim3}};
|
||||
Eigen::Tensor<float, 3> in1(tensorRange);
|
||||
Eigen::Tensor<float, 3> in2(tensorRange);
|
||||
Eigen::Tensor<float, 3> out(tensorRange);
|
||||
|
||||
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);
|
||||
|
||||
/// c=(a+b)*b
|
||||
gpu_out.device(sycl_device) =(gpu_in1 + gpu_in2).eval() * gpu_in2;
|
||||
sycl_device.deallocate(out.data());
|
||||
for (int i = 0; i < sizeDim1; ++i) {
|
||||
for (int j = 0; j < sizeDim2; ++j) {
|
||||
for (int k = 0; k < sizeDim3; ++k) {
|
||||
VERIFY_IS_APPROX(out(i, j, k),
|
||||
(in1(i, j, k) + in2(i, j, k)) * in2(i, j, k));
|
||||
}
|
||||
}
|
||||
}
|
||||
printf("(a+b)*b Test Passed\n");
|
||||
}
|
||||
|
||||
void test_cxx11_tensor_sycl_forced_eval() { CALL_SUBTEST(test_sycl_gpu()); }
|
||||
Reference in New Issue
Block a user