2 message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
6 if(CMAKE_COMPILER_IS_GNUCXX AND NOT APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
7 message(STATUS "CUDA compilation is disabled (due to Clang unsupported on your platform).")
11 set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
13 foreach(var INCLUDE LIBRARY PROGRAM)
14 set(__old_frpm_${var} "${CMAKE_FIND_ROOT_PATH_MODE_${var}}")
17 set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
18 set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY BOTH)
19 set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE NEVER)
21 find_package(CUDA "${MIN_VER_CUDA}" QUIET)
23 foreach(var INCLUDE LIBRARY PROGRAM)
24 set(CMAKE_FIND_ROOT_PATH_MODE_${var} "${__old_frpm_${var}}")
27 list(REMOVE_AT CMAKE_MODULE_PATH 0)
41 find_cuda_helper_libs(nvcuvid)
43 find_cuda_helper_libs(nvcuvenc)
48 message(STATUS "CUDA detected: " ${CUDA_VERSION})
50 set(_generations "Fermi" "Kepler")
51 if(NOT CMAKE_CROSSCOMPILING)
52 list(APPEND _generations "Auto")
54 set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
55 if( CMAKE_VERSION VERSION_GREATER "2.8" )
56 set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
60 if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
61 string(REPLACE ";" ", " _generations "${_generations}")
62 message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
64 unset(CUDA_ARCH_BIN CACHE)
65 unset(CUDA_ARCH_PTX CACHE)
68 set(__cuda_arch_ptx "")
69 if(CUDA_GENERATION STREQUAL "Fermi")
70 set(__cuda_arch_bin "2.0 2.1(2.0)")
71 elseif(CUDA_GENERATION STREQUAL "Kepler")
72 if(${CUDA_VERSION} VERSION_LESS "5.0")
73 set(__cuda_arch_bin "3.0")
75 set(__cuda_arch_bin "3.0 3.5")
77 elseif(CUDA_GENERATION STREQUAL "Auto")
78 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
79 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
80 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
81 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
82 if(NOT _nvcc_res EQUAL 0)
83 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
85 set(__cuda_arch_bin "${_nvcc_out}")
86 string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
90 if(NOT DEFINED __cuda_arch_bin)
92 set(__cuda_arch_bin "3.2")
93 set(__cuda_arch_ptx "")
95 if(${CUDA_VERSION} VERSION_LESS "5.0")
96 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
98 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
100 set(__cuda_arch_ptx "3.0")
104 set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
105 set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
107 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
108 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
110 # Ckeck if user specified 1.0 compute capability: we don't support it
111 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
112 set(CUDA_ARCH_BIN_OR_PTX_10 0)
113 if(NOT ${HAS_ARCH_10} STREQUAL "")
114 set(CUDA_ARCH_BIN_OR_PTX_10 1)
117 # NVCC flags to be set
118 set(NVCC_FLAGS_EXTRA "")
120 # These vars will be passed into the templates
121 set(OPENCV_CUDA_ARCH_BIN "")
122 set(OPENCV_CUDA_ARCH_PTX "")
123 set(OPENCV_CUDA_ARCH_FEATURES "")
125 # Tell NVCC to add binaries for the specified GPUs
126 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
127 foreach(ARCH IN LISTS ARCH_LIST)
128 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
129 # User explicitly specified PTX for the concrete BIN
130 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
131 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
132 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
134 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
135 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
136 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
137 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
141 # Tell NVCC to add PTX intermediate code for the specified architectures
142 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
143 foreach(ARCH IN LISTS ARCH_LIST)
144 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
145 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
146 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
149 # These vars will be processed in other scripts
150 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
151 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
154 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
155 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-target-os-variant=Android")
158 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
160 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
163 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
166 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
168 macro(ocv_cuda_compile VAR)
169 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
170 set(${var}_backup_in_cuda_compile_ "${${var}}")
172 # we remove /EHa as it generates warnings under windows
173 string(REPLACE "/EHa" "" ${var} "${${var}}")
175 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
176 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
178 # we remove -Wsign-promo as it generates warnings under linux
179 string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
181 # we remove -fvisibility-inlines-hidden because it's used for C++ compiler
182 # but NVCC uses C compiler by default
183 string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
185 # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
186 # but NVCC uses C compiler by default
187 string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")
189 # we remove -frtti because it's used for C++ compiler
190 # but NVCC uses C compiler by default
191 string(REPLACE "-frtti" "" ${var} "${${var}}")
194 if(BUILD_SHARED_LIBS)
195 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
199 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
202 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
205 # disabled because of multiple warnings during building nvcc auto generated files
206 if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
207 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
210 CUDA_COMPILE(${VAR} ${ARGN})
212 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
213 set(${var} "${${var}_backup_in_cuda_compile_}")
214 unset(${var}_backup_in_cuda_compile_)
218 unset(CUDA_ARCH_BIN CACHE)
219 unset(CUDA_ARCH_PTX CACHE)