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

Fix failing tests on arm/ppc

Browse Source
This commit is contained in:
Antonio Sanchez
2025-02-28 13:15:33 -08:00
parent ea37d9e73e
commit c53002f5fb
5 changed files with 53 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
unsupported/test/CMakeLists.txt
View File

@@ -215,18 +215,6 @@ if(CUDA_FOUND AND EIGEN_TEST_CUDA)
message(STATUS "Flags used to compile cuda code: " ${CMAKE_CXX_FLAGS})
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
set(CUDA_NVCC_FLAGS "-ccbin ${CMAKE_C_COMPILER}" CACHE STRING "nvcc flags" FORCE)
endif()
if(EIGEN_TEST_CUDA_CLANG)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 --cuda-gpu-arch=sm_${EIGEN_CUDA_COMPUTE_ARCH}")
endif()
set(EIGEN_CUDA_RELAXED_CONSTEXPR "--expt-relaxed-constexpr")
if (${CUDA_VERSION} STREQUAL "7.0")
set(EIGEN_CUDA_RELAXED_CONSTEXPR "--relaxed-constexpr")
endif()
if( (NOT EIGEN_TEST_CXX11) OR (CMAKE_VERSION VERSION_LESS 3.3))
set(EIGEN_CUDA_CXX11_FLAG "-std=c++11")
else()
@@ -234,8 +222,39 @@ if(CUDA_FOUND AND EIGEN_TEST_CUDA)
set(EIGEN_CUDA_CXX11_FLAG "")
endif()
set(CUDA_NVCC_FLAGS "${EIGEN_CUDA_CXX11_FLAG} ${EIGEN_CUDA_RELAXED_CONSTEXPR} -arch compute_${EIGEN_CUDA_COMPUTE_ARCH} -Xcudafe \"--display_error_number\" ${CUDA_NVCC_FLAGS}")
cuda_include_directories("${CMAKE_CURRENT_BINARY_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}/include")
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
set(CUDA_NVCC_FLAGS "-ccbin ${CMAKE_C_COMPILER}" CACHE STRING "nvcc flags" FORCE)
endif()
if(EIGEN_TEST_CUDA_CLANG)
string(APPEND CMAKE_CXX_FLAGS " --cuda-path=${CUDA_TOOLKIT_ROOT_DIR} ${EIGEN_CUDA_CXX11_FLAG}")
foreach(ARCH IN LISTS EIGEN_CUDA_COMPUTE_ARCH)
string(APPEND CMAKE_CXX_FLAGS " --cuda-gpu-arch=sm_${ARCH}")
endforeach()
else()
set(CUDA_PROPAGATE_HOST_FLAGS OFF)
set(NVCC_ARCH_FLAGS)
# Define an -arch=sm_<arch>, otherwise if GPU does not exactly match one of
# those in the arch list for -gencode, the kernels will fail to run with
# cudaErrorNoKernelImageForDevice
# This can happen with newer cards (e.g. sm_75) and compiling with older
# versions of nvcc (e.g. 9.2) that do not support their specific arch.
list(LENGTH EIGEN_CUDA_COMPUTE_ARCH EIGEN_CUDA_COMPUTE_ARCH_SIZE)
if(EIGEN_CUDA_COMPUTE_ARCH_SIZE)
list(GET EIGEN_CUDA_COMPUTE_ARCH 0 EIGEN_CUDA_COMPUTE_DEFAULT)
set(NVCC_ARCH_FLAGS " -arch=sm_${EIGEN_CUDA_COMPUTE_DEFAULT}")
endif()
foreach(ARCH IN LISTS EIGEN_CUDA_COMPUTE_ARCH)
string(APPEND NVCC_ARCH_FLAGS " -gencode arch=compute_${ARCH},code=sm_${ARCH}")
endforeach()
set(CUDA_NVCC_FLAGS "--expt-relaxed-constexpr -Xcudafe \"--display_error_number\" ${NVCC_ARCH_FLAGS} ${CUDA_NVCC_FLAGS} ${EIGEN_CUDA_CXX11_FLAG}")
cuda_include_directories("${CMAKE_CURRENT_BINARY_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}/include")
endif()
set(EIGEN_CUDA_RELAXED_CONSTEXPR "--expt-relaxed-constexpr")
if (${CUDA_VERSION} STREQUAL "7.0")
set(EIGEN_CUDA_RELAXED_CONSTEXPR "--relaxed-constexpr")
endif()
set(EIGEN_ADD_TEST_FILENAME_EXTENSION "cu")
ei_add_test(cxx11_tensor_complex_cuda)
@@ -246,18 +265,13 @@ if(CUDA_FOUND AND EIGEN_TEST_CUDA)
ei_add_test(cxx11_tensor_scan_cuda)
# Contractions require arch 3.0 or higher
if (${EIGEN_CUDA_COMPUTE_ARCH} GREATER 29)
ei_add_test(cxx11_tensor_device)
ei_add_test(cxx11_tensor_cuda)
ei_add_test(cxx11_tensor_contract_cuda)
ei_add_test(cxx11_tensor_of_float16_cuda)
endif()
ei_add_test(cxx11_tensor_device)
ei_add_test(cxx11_tensor_cuda)
ei_add_test(cxx11_tensor_contract_cuda)
ei_add_test(cxx11_tensor_of_float16_cuda)
# The random number generation code requires arch 3.5 or greater.
if (${EIGEN_CUDA_COMPUTE_ARCH} GREATER 34)
ei_add_test(cxx11_tensor_random_cuda)
endif()
ei_add_test(cxx11_tensor_random_cuda)
unset(EIGEN_ADD_TEST_FILENAME_EXTENSION)
endif()