diff options
Diffstat (limited to 'Modules/FindCUDA.cmake')
-rw-r--r-- | Modules/FindCUDA.cmake | 64 |
1 files changed, 43 insertions, 21 deletions
diff --git a/Modules/FindCUDA.cmake b/Modules/FindCUDA.cmake index f152437..0a98bf5 100644 --- a/Modules/FindCUDA.cmake +++ b/Modules/FindCUDA.cmake @@ -650,9 +650,11 @@ endif() # Look for the SDK stuff. As of CUDA 3.0 NVSDKCUDA_ROOT has been replaced with # NVSDKCOMPUTE_ROOT with the old CUDA C contents moved into the C subdirectory find_path(CUDA_SDK_ROOT_DIR common/inc/cutil.h + HINTS "$ENV{NVSDKCOMPUTE_ROOT}/C" - "$ENV{NVSDKCUDA_ROOT}" + ENV NVSDKCUDA_ROOT "[HKEY_LOCAL_MACHINE\\SOFTWARE\\NVIDIA Corporation\\Installed Products\\NVIDIA SDK 10\\Compute;InstallDir]" + PATHS "/Developer/GPU\ Computing/C" ) @@ -898,14 +900,6 @@ endfunction() macro(CUDA_WRAP_SRCS cuda_target format generated_files) - if( ${format} MATCHES "PTX" ) - set( compile_to_ptx ON ) - elseif( ${format} MATCHES "OBJ") - set( compile_to_ptx OFF ) - else() - message( FATAL_ERROR "Invalid format flag passed to CUDA_WRAP_SRCS: '${format}'. Use OBJ or PTX.") - endif() - # Set up all the command line flags here, so that they can be overridden on a per target basis. set(nvcc_flags "") @@ -1002,12 +996,12 @@ macro(CUDA_WRAP_SRCS cuda_target format generated_files) # Only add the CMAKE_{C,CXX}_FLAGS if we are propagating host flags. We # always need to set the SHARED_FLAGS, though. if(CUDA_PROPAGATE_HOST_FLAGS) - set(CUDA_HOST_FLAGS "set(CMAKE_HOST_FLAGS ${CMAKE_${CUDA_C_OR_CXX}_FLAGS} ${CUDA_HOST_SHARED_FLAGS})") + set(_cuda_host_flags "set(CMAKE_HOST_FLAGS ${CMAKE_${CUDA_C_OR_CXX}_FLAGS} ${CUDA_HOST_SHARED_FLAGS})") else() - set(CUDA_HOST_FLAGS "set(CMAKE_HOST_FLAGS ${CUDA_HOST_SHARED_FLAGS})") + set(_cuda_host_flags "set(CMAKE_HOST_FLAGS ${CUDA_HOST_SHARED_FLAGS})") endif() - set(CUDA_NVCC_FLAGS_CONFIG "# Build specific configuration flags") + set(_cuda_nvcc_flags_config "# Build specific configuration flags") # Loop over all the configuration types to generate appropriate flags for run_nvcc.cmake foreach(config ${CUDA_configuration_types}) string(TOUPPER ${config} config_upper) @@ -1016,27 +1010,31 @@ macro(CUDA_WRAP_SRCS cuda_target format generated_files) if(CUDA_PROPAGATE_HOST_FLAGS) # nvcc chokes on -g3 in versions previous to 3.0, so replace it with -g - if(CMAKE_COMPILER_IS_GNUCC AND CUDA_VERSION VERSION_LESS "3.0") + set(_cuda_fix_g3 FALSE) + + if(CMAKE_COMPILER_IS_GNUCC) + if (CUDA_VERSION VERSION_LESS "3.0" OR + CUDA_VERSION VERSION_EQUAL "4.1" OR + CUDA_VERSION VERSION_EQUAL "4.2" + ) + set(_cuda_fix_g3 TRUE) + endif() + endif() + if(_cuda_fix_g3) string(REPLACE "-g3" "-g" _cuda_C_FLAGS "${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}}") else() set(_cuda_C_FLAGS "${CMAKE_${CUDA_C_OR_CXX}_FLAGS_${config_upper}}") endif() - set(CUDA_HOST_FLAGS "${CUDA_HOST_FLAGS}\nset(CMAKE_HOST_FLAGS_${config_upper} ${_cuda_C_FLAGS})") + set(_cuda_host_flags "${_cuda_host_flags}\nset(CMAKE_HOST_FLAGS_${config_upper} ${_cuda_C_FLAGS})") endif() # Note that if we ever want CUDA_NVCC_FLAGS_<CONFIG> to be string (instead of a list # like it is currently), we can remove the quotes around the # ${CUDA_NVCC_FLAGS_${config_upper}} variable like the CMAKE_HOST_FLAGS_<CONFIG> variable. - set(CUDA_NVCC_FLAGS_CONFIG "${CUDA_NVCC_FLAGS_CONFIG}\nset(CUDA_NVCC_FLAGS_${config_upper} ${CUDA_NVCC_FLAGS_${config_upper}} ;; ${CUDA_WRAP_OPTION_NVCC_FLAGS_${config_upper}})") + set(_cuda_nvcc_flags_config "${_cuda_nvcc_flags_config}\nset(CUDA_NVCC_FLAGS_${config_upper} ${CUDA_NVCC_FLAGS_${config_upper}} ;; ${CUDA_WRAP_OPTION_NVCC_FLAGS_${config_upper}})") endforeach() - if(compile_to_ptx) - # Don't use any of the host compilation flags for PTX targets. - set(CUDA_HOST_FLAGS) - set(CUDA_NVCC_FLAGS_CONFIG) - endif() - # Get the list of definitions from the directory property get_directory_property(CUDA_NVCC_DEFINITIONS COMPILE_DEFINITIONS) if(CUDA_NVCC_DEFINITIONS) @@ -1059,6 +1057,30 @@ macro(CUDA_WRAP_SRCS cuda_target format generated_files) get_source_file_property(_is_header ${file} HEADER_FILE_ONLY) if(${file} MATCHES ".*\\.cu$" AND NOT _is_header) + # Allow per source file overrides of the format. + get_source_file_property(_cuda_source_format ${file} CUDA_SOURCE_PROPERTY_FORMAT) + if(NOT _cuda_source_format) + set(_cuda_source_format ${format}) + endif() + + if( ${_cuda_source_format} MATCHES "PTX" ) + set( compile_to_ptx ON ) + elseif( ${_cuda_source_format} MATCHES "OBJ") + set( compile_to_ptx OFF ) + else() + message( FATAL_ERROR "Invalid format flag passed to CUDA_WRAP_SRCS for file '${file}': '${_cuda_source_format}'. Use OBJ or PTX.") + endif() + + + if(compile_to_ptx) + # Don't use any of the host compilation flags for PTX targets. + set(CUDA_HOST_FLAGS) + set(CUDA_NVCC_FLAGS_CONFIG) + else() + set(CUDA_HOST_FLAGS ${_cuda_host_flags}) + set(CUDA_NVCC_FLAGS_CONFIG ${_cuda_nvcc_flags_config}) + endif() + # Determine output directory cuda_compute_build_path("${file}" cuda_build_path) set(cuda_compile_intermediate_directory "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${cuda_target}.dir/${cuda_build_path}") |