1 if(${CMAKE_VERSION} VERSION_LESS "2.8.3")
2 message(STATUS "WITH_CUDA flag requires CMake 2.8.3 or newer. CUDA support is disabled.")
7 message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
11 if(CMAKE_COMPILER_IS_GNUCXX AND NOT APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
12 message(STATUS "CUDA compilation is disabled (due to Clang unsupported on your platform).")
16 set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
18 foreach(var INCLUDE LIBRARY PROGRAM)
19 set(__old_frpm_${var} "${CMAKE_FIND_ROOT_PATH_MODE_${var}}")
22 set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
23 set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY BOTH)
24 set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE NEVER)
26 find_package(CUDA 4.2 QUIET)
28 foreach(var INCLUDE LIBRARY PROGRAM)
29 set(CMAKE_FIND_ROOT_PATH_MODE_${var} "${__old_frpm_${var}}")
32 list(REMOVE_AT CMAKE_MODULE_PATH 0)
46 find_cuda_helper_libs(nvcuvid)
50 message(STATUS "CUDA detected: " ${CUDA_VERSION})
52 set(_generations "Fermi" "Kepler")
53 if(NOT CMAKE_CROSSCOMPILING)
54 list(APPEND _generations "Auto")
56 set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
57 if( CMAKE_VERSION VERSION_GREATER "2.8" )
58 set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
62 if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
63 string(REPLACE ";" ", " _generations "${_generations}")
64 message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
66 unset(CUDA_ARCH_BIN CACHE)
67 unset(CUDA_ARCH_PTX CACHE)
70 set(__cuda_arch_ptx "")
71 if(CUDA_GENERATION STREQUAL "Fermi")
72 set(__cuda_arch_bin "2.0 2.1(2.0)")
73 elseif(CUDA_GENERATION STREQUAL "Kepler")
74 if(${CUDA_VERSION} VERSION_LESS "5.0")
75 set(__cuda_arch_bin "3.0")
77 set(__cuda_arch_bin "3.0 3.5")
79 elseif(CUDA_GENERATION STREQUAL "Auto")
80 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
81 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
82 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
83 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
84 if(NOT _nvcc_res EQUAL 0)
85 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
87 set(__cuda_arch_bin "${_nvcc_out}")
88 string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
92 if(NOT DEFINED __cuda_arch_bin)
94 set(__cuda_arch_bin "3.2")
95 set(__cuda_arch_ptx "")
97 if(${CUDA_VERSION} VERSION_LESS "5.0")
98 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
100 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
102 set(__cuda_arch_ptx "3.0")
106 set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
107 set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
109 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
110 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
112 # Ckeck if user specified 1.0 compute capability: we don't support it
113 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
114 set(CUDA_ARCH_BIN_OR_PTX_10 0)
115 if(NOT ${HAS_ARCH_10} STREQUAL "")
116 set(CUDA_ARCH_BIN_OR_PTX_10 1)
119 # NVCC flags to be set
120 set(NVCC_FLAGS_EXTRA "")
122 # These vars will be passed into the templates
123 set(OPENCV_CUDA_ARCH_BIN "")
124 set(OPENCV_CUDA_ARCH_PTX "")
125 set(OPENCV_CUDA_ARCH_FEATURES "")
127 # Tell NVCC to add binaries for the specified GPUs
128 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
129 foreach(ARCH IN LISTS ARCH_LIST)
130 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
131 # User explicitly specified PTX for the concrete BIN
132 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
133 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
134 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
136 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
137 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
138 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
139 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
143 # Tell NVCC to add PTX intermediate code for the specified architectures
144 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
145 foreach(ARCH IN LISTS ARCH_LIST)
146 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
147 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
148 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
151 # These vars will be processed in other scripts
152 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
153 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
156 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
157 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-target-os-variant=Android")
160 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
162 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
165 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
168 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
170 macro(ocv_cuda_compile VAR)
171 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
172 set(${var}_backup_in_cuda_compile_ "${${var}}")
174 # we remove /EHa as it generates warnings under windows
175 string(REPLACE "/EHa" "" ${var} "${${var}}")
177 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
178 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
180 # we remove -Wsign-promo as it generates warnings under linux
181 string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
183 # we remove -Wno-sign-promo as it generates warnings under linux
184 string(REPLACE "-Wno-sign-promo" "" ${var} "${${var}}")
186 # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
187 # but NVCC uses C compiler by default
188 string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")
190 # we remove -frtti because it's used for C++ compiler
191 # but NVCC uses C compiler by default
192 string(REPLACE "-frtti" "" ${var} "${${var}}")
195 if(BUILD_SHARED_LIBS)
196 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
200 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
203 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
206 # disabled because of multiple warnings during building nvcc auto generated files
207 if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
208 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
211 CUDA_COMPILE(${VAR} ${ARGN})
213 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
214 set(${var} "${${var}_backup_in_cuda_compile_}")
215 unset(${var}_backup_in_cuda_compile_)
219 unset(CUDA_ARCH_BIN CACHE)
220 unset(CUDA_ARCH_PTX CACHE)
224 set(CUDA_LIBS_PATH "")
225 foreach(p ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
226 get_filename_component(_tmp ${p} PATH)
227 list(APPEND CUDA_LIBS_PATH ${_tmp})
231 foreach(p ${CUDA_cublas_LIBRARY})
232 get_filename_component(_tmp ${p} PATH)
233 list(APPEND CUDA_LIBS_PATH ${_tmp})
238 foreach(p ${CUDA_cufft_LIBRARY})
239 get_filename_component(_tmp ${p} PATH)
240 list(APPEND CUDA_LIBS_PATH ${_tmp})
244 list(REMOVE_DUPLICATES CUDA_LIBS_PATH)
245 link_directories(${CUDA_LIBS_PATH})
247 set(CUDA_LIBRARIES_ABS ${CUDA_LIBRARIES})
248 ocv_convert_to_lib_name(CUDA_LIBRARIES ${CUDA_LIBRARIES})
249 set(CUDA_npp_LIBRARY_ABS ${CUDA_npp_LIBRARY})
250 ocv_convert_to_lib_name(CUDA_npp_LIBRARY ${CUDA_npp_LIBRARY})
252 set(CUDA_cublas_LIBRARY_ABS ${CUDA_cublas_LIBRARY})
253 ocv_convert_to_lib_name(CUDA_cublas_LIBRARY ${CUDA_cublas_LIBRARY})
257 set(CUDA_cufft_LIBRARY_ABS ${CUDA_cufft_LIBRARY})
258 ocv_convert_to_lib_name(CUDA_cufft_LIBRARY ${CUDA_cufft_LIBRARY})