generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

updated cmake file to allow specifying GPU archs in BIN(PTX) format

Browse Source
This commit is contained in:
Alexey Spizhevoy 2011-01-25 14:13:12 +00:00
parent 397a63539c
commit dbff16eb85
1 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
CMakeLists.txt
View File

@ -708,15 +708,15 @@ if(WITH_CUDA)
set(HAVE_CUDA 1)
message(STATUS "CUDA detected: " ${CUDA_VERSION})
set(CUDA_ARCH_GPU "1.3 2.0" CACHE STRING "Specify 'real' GPU architectures to build binaries for")
set(CUDA_ARCH_BIN "1.3 2.0" CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
set(CUDA_ARCH_PTX "1.1 1.3" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
# These variables are used in config templates
string(REGEX REPLACE "\\." "" ARCH_GPU_NO_POINTS "${CUDA_ARCH_GPU}")
string(REGEX REPLACE "\\." "" ARCH_GPU_NO_POINTS "${CUDA_ARCH_BIN}")
string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
# Ckeck if user specified 1.0 compute capability
string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_GPU} ${CUDA_ARCH_PTX}")
string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
if(NOT ${HAS_ARCH_10} STREQUAL "")
set(OPENCV_ARCH_GPU_OR_PTX_10 1)
endif()
@ -724,9 +724,13 @@ if(WITH_CUDA)
set(NVCC_FLAGS_EXTRA "")
# Tell nvcc to add binaries for the specified GPUs
string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_GPU_NO_POINTS}")
string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_GPU_NO_POINTS}")
foreach(ARCH IN LISTS ARCH_LIST)
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
if (ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
else()
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
endif()
endforeach()
# Tell nvcc to add PTX intermediate code for the specified architectures