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

Removed the sycl include from Eigen/Core and moved it to Unsupported/Eigen/CXX11/Tensor; added TensorReduction for sycl (full reduction and partial reduction); added TensorReduction test case for sycl (full reduction and partial reduction); fixed the tile size on TensorSyclRun.h based on the device max work group size;

Browse Source
This commit is contained in:
Mehdi Goli
2016-11-04 18:18:19 +00:00
parent 0585b2965d
commit 0ebe3808ca
18 changed files with 663 additions and 249 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.

147
unsupported/test/cxx11_tensor_reduction_sycl.cpp Normal file
View File

@@ -0,0 +1,147 @@
// 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() {
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);
const int num_rows = 452;
const int num_cols = 765;
array<int, 2> tensorRange = {{num_rows, num_cols}};
Tensor<float, 2> in(tensorRange);
in.setRandom();
Tensor<float, 0> full_redux;
Tensor<float, 0> full_redux_g;
full_redux = in.sum();
float* out_data = (float*)sycl_device.allocate(sizeof(float));
TensorMap<Tensor<float, 2> > in_gpu(in.data(), tensorRange);
TensorMap<Tensor<float, 0> > full_redux_gpu(out_data);
full_redux_gpu.device(sycl_device) = in_gpu.sum();
sycl_device.deallocate(out_data);
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux_gpu(), full_redux());
}
static void test_first_dim_reductions_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);
int dim_x = 145;
int dim_y = 1;
int dim_z = 67;
array<int, 3> tensorRange = {{dim_x, dim_y, dim_z}};
Tensor<float, 3> in(tensorRange);
in.setRandom();
Eigen::array<int, 1> red_axis;
red_axis[0] = 0;
Tensor<float, 2> redux = in.sum(red_axis);
array<int, 2> reduced_tensorRange = {{dim_y, dim_z}};
Tensor<float, 2> redux_g(reduced_tensorRange);
TensorMap<Tensor<float, 3> > in_gpu(in.data(), tensorRange);
float* out_data = (float*)sycl_device.allocate(dim_y*dim_z*sizeof(float));
TensorMap<Tensor<float, 2> > redux_gpu(out_data, dim_y, dim_z );
redux_gpu.device(sycl_device) = in_gpu.sum(red_axis);
sycl_device.deallocate(out_data);
// Check that the CPU and GPU reductions return the same result.
for(int j=0; j<dim_y; j++ )
for(int k=0; k<dim_z; k++ )
VERIFY_IS_APPROX(redux_gpu(j,k), redux(j,k));
}
static void test_last_dim_reductions_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);
int dim_x = 567;
int dim_y = 1;
int dim_z = 47;
array<int, 3> tensorRange = {{dim_x, dim_y, dim_z}};
Tensor<float, 3> in(tensorRange);
in.setRandom();
Eigen::array<int, 1> red_axis;
red_axis[0] = 2;
Tensor<float, 2> redux = in.sum(red_axis);
array<int, 2> reduced_tensorRange = {{dim_x, dim_y}};
Tensor<float, 2> redux_g(reduced_tensorRange);
TensorMap<Tensor<float, 3> > in_gpu(in.data(), tensorRange);
float* out_data = (float*)sycl_device.allocate(dim_x*dim_y*sizeof(float));
TensorMap<Tensor<float, 2> > redux_gpu(out_data, dim_x, dim_y );
redux_gpu.device(sycl_device) = in_gpu.sum(red_axis);
sycl_device.deallocate(out_data);
// Check that the CPU and GPU reductions return the same result.
for(int j=0; j<dim_x; j++ )
for(int k=0; k<dim_y; k++ )
VERIFY_IS_APPROX(redux_gpu(j,k), redux(j,k));
}
void test_cxx11_tensor_reduction_sycl() {
CALL_SUBTEST((test_full_reductions_sycl()));
CALL_SUBTEST((test_first_dim_reductions_sycl()));
CALL_SUBTEST((test_last_dim_reductions_sycl()));
}