2 message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
6 if(NOT APPLE 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: ${_generations} Generations are suppered.")
69 unset(CUDA_ARCH_BIN CACHE)
70 unset(CUDA_ARCH_PTX CACHE)
73 set(__cuda_arch_ptx "")
74 if(CUDA_GENERATION STREQUAL "Fermi")
75 set(__cuda_arch_bin "2.0")
76 elseif(CUDA_GENERATION STREQUAL "Kepler")
77 set(__cuda_arch_bin "3.0 3.5 3.7")
78 elseif(CUDA_GENERATION STREQUAL "Maxwell")
79 set(__cuda_arch_bin "5.0 5.2")
80 elseif(CUDA_GENERATION STREQUAL "Pascal")
81 set(__cuda_arch_bin "6.0 6.1")
82 elseif(CUDA_GENERATION STREQUAL "Volta")
83 set(__cuda_arch_bin "7.0")
84 elseif(CUDA_GENERATION STREQUAL "Auto")
85 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
86 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
87 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
88 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
89 if(NOT _nvcc_res EQUAL 0)
90 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
92 set(__cuda_arch_bin "${_nvcc_out}")
93 string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
97 if(NOT DEFINED __cuda_arch_bin)
99 set(__cuda_arch_bin "3.2")
100 set(__cuda_arch_ptx "")
102 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
103 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
104 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
105 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
106 if(NOT _nvcc_res EQUAL 0)
107 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
108 set(__cuda_arch_bin "5.3 6.2 7.0")
110 set(__cuda_arch_bin "${_nvcc_out}")
111 string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
113 set(__cuda_arch_ptx "")
115 if(${CUDA_VERSION} VERSION_LESS "9.0")
116 set(__cuda_arch_bin "2.0 3.0 3.5 3.7 5.0 5.2 6.0 6.1")
118 set(__cuda_arch_bin "3.0 3.5 3.7 5.0 5.2 6.0 6.1 7.0")
123 set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
124 set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
126 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
127 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
129 # Ckeck if user specified 1.0 compute capability: we don't support it
130 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
131 set(CUDA_ARCH_BIN_OR_PTX_10 0)
132 if(NOT ${HAS_ARCH_10} STREQUAL "")
133 set(CUDA_ARCH_BIN_OR_PTX_10 1)
136 # NVCC flags to be set
137 set(NVCC_FLAGS_EXTRA "")
139 # These vars will be passed into the templates
140 set(OPENCV_CUDA_ARCH_BIN "")
141 set(OPENCV_CUDA_ARCH_PTX "")
142 set(OPENCV_CUDA_ARCH_FEATURES "")
144 # Tell NVCC to add binaries for the specified GPUs
145 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
146 foreach(ARCH IN LISTS ARCH_LIST)
147 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
148 # User explicitly specified PTX for the concrete BIN
149 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
150 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
151 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
153 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
154 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
155 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
156 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
159 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -D_FORCE_INLINES)
161 # Tell NVCC to add PTX intermediate code for the specified architectures
162 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
163 foreach(ARCH IN LISTS ARCH_LIST)
164 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
165 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
166 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
169 # These vars will be processed in other scripts
170 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
171 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
174 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
177 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
179 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
182 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
185 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
187 macro(ocv_cuda_filter_options)
188 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
189 set(${var}_backup_in_cuda_compile_ "${${var}}")
191 # we remove /EHa as it generates warnings under windows
192 string(REPLACE "/EHa" "" ${var} "${${var}}")
194 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
195 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
197 # we remove -Wsign-promo as it generates warnings under linux
198 string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
200 # we remove -Wno-sign-promo as it generates warnings under linux
201 string(REPLACE "-Wno-sign-promo" "" ${var} "${${var}}")
203 # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
204 # but NVCC uses C compiler by default
205 string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")
207 # we remove -frtti because it's used for C++ compiler
208 # but NVCC uses C compiler by default
209 string(REPLACE "-frtti" "" ${var} "${${var}}")
211 string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
213 # cc1: warning: command line option '-Wsuggest-override' is valid for C++/ObjC++ but not for C
214 string(REPLACE "-Wsuggest-override" "" ${var} "${${var}}")
216 # issue: #11552 (from OpenCVCompilerOptions.cmake)
217 string(REGEX REPLACE "-Wimplicit-fallthrough(=[0-9]+)? " "" ${var} "${${var}}")
219 # removal of custom specified options
220 if(OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS)
221 foreach(__flag ${OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS})
222 string(REPLACE "${__flag}" "" ${var} "${${var}}")
228 macro(ocv_cuda_compile VAR)
229 ocv_cuda_filter_options()
231 if(BUILD_SHARED_LIBS)
232 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
236 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
239 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
242 if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
243 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xlinker --unresolved-symbols=ignore-in-shared-libs)
246 # disabled because of multiple warnings during building nvcc auto generated files
247 if(CV_GCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "4.6.0")
248 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
251 CUDA_COMPILE(${VAR} ${ARGN})
253 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
254 set(${var} "${${var}_backup_in_cuda_compile_}")
255 unset(${var}_backup_in_cuda_compile_)
259 unset(CUDA_ARCH_BIN CACHE)
260 unset(CUDA_ARCH_PTX CACHE)
264 set(CUDA_LIBS_PATH "")
265 foreach(p ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
266 get_filename_component(_tmp ${p} PATH)
267 list(APPEND CUDA_LIBS_PATH ${_tmp})
271 foreach(p ${CUDA_cublas_LIBRARY})
272 get_filename_component(_tmp ${p} PATH)
273 list(APPEND CUDA_LIBS_PATH ${_tmp})
278 foreach(p ${CUDA_cufft_LIBRARY})
279 get_filename_component(_tmp ${p} PATH)
280 list(APPEND CUDA_LIBS_PATH ${_tmp})
284 list(REMOVE_DUPLICATES CUDA_LIBS_PATH)
285 link_directories(${CUDA_LIBS_PATH})
287 set(CUDA_LIBRARIES_ABS ${CUDA_LIBRARIES})
288 ocv_convert_to_lib_name(CUDA_LIBRARIES ${CUDA_LIBRARIES})
289 set(CUDA_npp_LIBRARY_ABS ${CUDA_npp_LIBRARY})
290 ocv_convert_to_lib_name(CUDA_npp_LIBRARY ${CUDA_npp_LIBRARY})
292 set(CUDA_cublas_LIBRARY_ABS ${CUDA_cublas_LIBRARY})
293 ocv_convert_to_lib_name(CUDA_cublas_LIBRARY ${CUDA_cublas_LIBRARY})
297 set(CUDA_cufft_LIBRARY_ABS ${CUDA_cufft_LIBRARY})
298 ocv_convert_to_lib_name(CUDA_cufft_LIBRARY ${CUDA_cufft_LIBRARY})