2 message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
6 if(NOT UNIX AND CV_CLANG)
7 message(STATUS "CUDA compilation is disabled (due to Clang unsupported on your platform).")
12 if(((NOT CMAKE_VERSION VERSION_LESS "3.9.0") # requires https://gitlab.kitware.com/cmake/cmake/merge_requests/663
13 OR OPENCV_CUDA_FORCE_EXTERNAL_CMAKE_MODULE)
14 AND NOT OPENCV_CUDA_FORCE_BUILTIN_CMAKE_MODULE)
15 ocv_update(CUDA_LINK_LIBRARIES_KEYWORD "LINK_PRIVATE")
16 find_host_package(CUDA "${MIN_VER_CUDA}" QUIET)
18 # Use OpenCV's patched "FindCUDA" module
19 set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
22 set(CUDA_TARGET_OS_VARIANT "Android")
24 find_host_package(CUDA "${MIN_VER_CUDA}" QUIET)
26 list(REMOVE_AT CMAKE_MODULE_PATH 0)
41 find_cuda_helper_libs(nvcuvid)
43 find_cuda_helper_libs(nvcuvenc)
45 if(CUDA_nvcuvid_LIBRARY)
48 if(CUDA_nvcuvenc_LIBRARY)
53 message(STATUS "CUDA detected: " ${CUDA_VERSION})
55 set(_generations "Fermi" "Kepler" "Maxwell" "Pascal" "Volta")
56 if(NOT CMAKE_CROSSCOMPILING)
57 list(APPEND _generations "Auto")
59 set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
60 if( CMAKE_VERSION VERSION_GREATER "2.8" )
61 set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
65 if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
66 string(REPLACE ";" ", " _generations "${_generations}")
67 message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
69 unset(CUDA_ARCH_BIN CACHE)
70 unset(CUDA_ARCH_PTX CACHE)
73 set(__cuda_arch_ptx "")
74 if(CUDA_GENERATION STREQUAL "Fermi")
75 set(__cuda_arch_bin "2.0")
76 elseif(CUDA_GENERATION STREQUAL "Kepler")
77 set(__cuda_arch_bin "3.0 3.5 3.7")
78 elseif(CUDA_GENERATION STREQUAL "Maxwell")
79 set(__cuda_arch_bin "5.0 5.2")
80 elseif(CUDA_GENERATION STREQUAL "Pascal")
81 set(__cuda_arch_bin "6.0 6.1")
82 elseif(CUDA_GENERATION STREQUAL "Volta")
83 set(__cuda_arch_bin "7.0")
84 elseif(CUDA_GENERATION STREQUAL "Turing")
85 set(__cuda_arch_bin "7.5")
86 elseif(CUDA_GENERATION STREQUAL "Auto")
87 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
88 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
89 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
90 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
91 if(NOT _nvcc_res EQUAL 0)
92 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
94 set(__cuda_arch_bin "${_nvcc_out}")
95 string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
99 if(NOT DEFINED __cuda_arch_bin)
101 set(__cuda_arch_bin "3.2")
102 set(__cuda_arch_ptx "")
104 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
105 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
106 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
107 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
108 if(NOT _nvcc_res EQUAL 0)
109 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
110 set(__cuda_arch_bin "5.3 6.2 7.0 7.5")
112 set(__cuda_arch_bin "${_nvcc_out}")
113 string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
115 set(__cuda_arch_ptx "")
117 if(CUDA_VERSION VERSION_LESS "9.0")
118 set(__cuda_arch_bin "2.0 3.0 3.5 3.7 5.0 5.2 6.0 6.1")
119 elseif(CUDA_VERSION VERSION_LESS "10.0")
120 set(__cuda_arch_bin "3.0 3.5 3.7 5.0 5.2 6.0 6.1 7.0")
122 set(__cuda_arch_bin "3.0 3.5 3.7 5.0 5.2 6.0 6.1 7.0 7.5")
127 set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
128 set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
130 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
131 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
133 # Ckeck if user specified 1.0 compute capability: we don't support it
134 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
135 set(CUDA_ARCH_BIN_OR_PTX_10 0)
136 if(NOT ${HAS_ARCH_10} STREQUAL "")
137 set(CUDA_ARCH_BIN_OR_PTX_10 1)
140 # NVCC flags to be set
141 set(NVCC_FLAGS_EXTRA "")
143 # These vars will be passed into the templates
144 set(OPENCV_CUDA_ARCH_BIN "")
145 set(OPENCV_CUDA_ARCH_PTX "")
146 set(OPENCV_CUDA_ARCH_FEATURES "")
148 # Tell NVCC to add binaries for the specified GPUs
149 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
150 foreach(ARCH IN LISTS ARCH_LIST)
151 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
152 # User explicitly specified PTX for the concrete BIN
153 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
154 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
155 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
157 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
158 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
159 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
160 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
163 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -D_FORCE_INLINES)
165 # Tell NVCC to add PTX intermediate code for the specified architectures
166 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
167 foreach(ARCH IN LISTS ARCH_LIST)
168 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
169 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
170 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
173 # These vars will be processed in other scripts
174 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
175 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
178 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
181 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
183 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
186 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
189 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
191 macro(ocv_cuda_filter_options)
192 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
193 set(${var}_backup_in_cuda_compile_ "${${var}}")
196 # we remove -Winconsistent-missing-override and -Qunused-arguments
197 # just in case we are compiling CUDA with gcc but OpenCV with clang
198 string(REPLACE "-Winconsistent-missing-override" "" ${var} "${${var}}")
199 string(REPLACE "-Qunused-arguments" "" ${var} "${${var}}")
202 # we remove /EHa as it generates warnings under windows
203 string(REPLACE "/EHa" "" ${var} "${${var}}")
205 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
206 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
208 # we remove -Wsign-promo as it generates warnings under linux
209 string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
211 # we remove -Wno-sign-promo as it generates warnings under linux
212 string(REPLACE "-Wno-sign-promo" "" ${var} "${${var}}")
214 # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
215 # but NVCC uses C compiler by default
216 string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")
218 # we remove -frtti because it's used for C++ compiler
219 # but NVCC uses C compiler by default
220 string(REPLACE "-frtti" "" ${var} "${${var}}")
222 string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
224 # cc1: warning: command line option '-Wsuggest-override' is valid for C++/ObjC++ but not for C
225 string(REPLACE "-Wsuggest-override" "" ${var} "${${var}}")
227 # issue: #11552 (from OpenCVCompilerOptions.cmake)
228 string(REGEX REPLACE "-Wimplicit-fallthrough(=[0-9]+)? " "" ${var} "${${var}}")
230 # removal of custom specified options
231 if(OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS)
232 foreach(__flag ${OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS})
233 string(REPLACE "${__flag}" "" ${var} "${${var}}")
239 macro(ocv_cuda_compile VAR)
240 ocv_cuda_filter_options()
242 if(BUILD_SHARED_LIBS)
243 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
247 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
250 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
253 if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
254 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xlinker --unresolved-symbols=ignore-in-shared-libs)
257 # disabled because of multiple warnings during building nvcc auto generated files
258 if(CV_GCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "4.6.0")
259 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
262 CUDA_COMPILE(${VAR} ${ARGN})
264 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
265 set(${var} "${${var}_backup_in_cuda_compile_}")
266 unset(${var}_backup_in_cuda_compile_)
270 unset(CUDA_ARCH_BIN CACHE)
271 unset(CUDA_ARCH_PTX CACHE)
275 set(CUDA_LIBS_PATH "")
276 foreach(p ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
277 get_filename_component(_tmp ${p} PATH)
278 list(APPEND CUDA_LIBS_PATH ${_tmp})
282 foreach(p ${CUDA_cublas_LIBRARY})
283 get_filename_component(_tmp ${p} PATH)
284 list(APPEND CUDA_LIBS_PATH ${_tmp})
289 foreach(p ${CUDA_cufft_LIBRARY})
290 get_filename_component(_tmp ${p} PATH)
291 list(APPEND CUDA_LIBS_PATH ${_tmp})
295 list(REMOVE_DUPLICATES CUDA_LIBS_PATH)
296 link_directories(${CUDA_LIBS_PATH})
298 set(CUDA_LIBRARIES_ABS ${CUDA_LIBRARIES})
299 ocv_convert_to_lib_name(CUDA_LIBRARIES ${CUDA_LIBRARIES})
300 set(CUDA_npp_LIBRARY_ABS ${CUDA_npp_LIBRARY})
301 ocv_convert_to_lib_name(CUDA_npp_LIBRARY ${CUDA_npp_LIBRARY})
303 set(CUDA_cublas_LIBRARY_ABS ${CUDA_cublas_LIBRARY})
304 ocv_convert_to_lib_name(CUDA_cublas_LIBRARY ${CUDA_cublas_LIBRARY})
308 set(CUDA_cufft_LIBRARY_ABS ${CUDA_cufft_LIBRARY})
309 ocv_convert_to_lib_name(CUDA_cufft_LIBRARY ${CUDA_cufft_LIBRARY})