1 if(${CMAKE_VERSION} VERSION_LESS "2.8.3")
2 message(STATUS "WITH_CUDA flag requires CMake 2.8.3 or newer. CUDA support is disabled.")
7 message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
11 if(CMAKE_COMPILER_IS_GNUCXX AND NOT APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
12 message(STATUS "CUDA compilation is disabled (due to Clang unsupported on your platform).")
16 find_package(CUDA 4.2 QUIET)
29 if(${CUDA_VERSION} VERSION_LESS "5.5")
30 find_cuda_helper_libs(npp)
32 find_cuda_helper_libs(nppc)
33 find_cuda_helper_libs(nppi)
34 find_cuda_helper_libs(npps)
35 set(CUDA_npp_LIBRARY ${CUDA_nppc_LIBRARY} ${CUDA_nppi_LIBRARY} ${CUDA_npps_LIBRARY})
39 find_cuda_helper_libs(nvcuvid)
43 message(STATUS "CUDA detected: " ${CUDA_VERSION})
45 set(_generations "Fermi" "Kepler")
46 if(NOT CMAKE_CROSSCOMPILING)
47 list(APPEND _generations "Auto")
49 set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
50 if( CMAKE_VERSION VERSION_GREATER "2.8" )
51 set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
55 if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
56 string(REPLACE ";" ", " _generations "${_generations}")
57 message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
59 unset(CUDA_ARCH_BIN CACHE)
60 unset(CUDA_ARCH_PTX CACHE)
63 set(__cuda_arch_ptx "")
64 if(CUDA_GENERATION STREQUAL "Fermi")
65 set(__cuda_arch_bin "2.0 2.1(2.0)")
66 elseif(CUDA_GENERATION STREQUAL "Kepler")
67 set(__cuda_arch_bin "3.0")
68 elseif(CUDA_GENERATION STREQUAL "Auto")
69 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
70 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
71 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
72 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
73 if(NOT _nvcc_res EQUAL 0)
74 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
76 set(__cuda_arch_bin "${_nvcc_out}")
80 if(NOT DEFINED __cuda_arch_bin)
81 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
82 set(__cuda_arch_ptx "2.0 3.0")
85 set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
86 set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
88 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
89 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
91 # Ckeck if user specified 1.0 compute capability: we don't support it
92 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
93 set(CUDA_ARCH_BIN_OR_PTX_10 0)
94 if(NOT ${HAS_ARCH_10} STREQUAL "")
95 set(CUDA_ARCH_BIN_OR_PTX_10 1)
98 # NVCC flags to be set
99 set(NVCC_FLAGS_EXTRA "")
101 # These vars will be passed into the templates
102 set(OPENCV_CUDA_ARCH_BIN "")
103 set(OPENCV_CUDA_ARCH_PTX "")
104 set(OPENCV_CUDA_ARCH_FEATURES "")
106 # Tell NVCC to add binaries for the specified GPUs
107 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
108 foreach(ARCH IN LISTS ARCH_LIST)
109 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
110 # User explicitly specified PTX for the concrete BIN
111 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
112 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
113 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
115 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
116 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
117 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
118 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
122 # Tell NVCC to add PTX intermediate code for the specified architectures
123 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
124 foreach(ARCH IN LISTS ARCH_LIST)
125 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
126 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
127 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
130 if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "arm")
131 set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --target-cpu-architecture=ARM")
134 # These vars will be processed in other scripts
135 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
136 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
138 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
140 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
143 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
146 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
148 macro(ocv_cuda_compile VAR)
149 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
150 set(${var}_backup_in_cuda_compile_ "${${var}}")
152 # we remove /EHa as it generates warnings under windows
153 string(REPLACE "/EHa" "" ${var} "${${var}}")
155 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
156 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
158 # we remove -Wsign-promo as it generates warnings under linux
159 string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
162 if(BUILD_SHARED_LIBS)
163 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
167 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
170 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
173 # disabled because of multiple warnings during building nvcc auto generated files
174 if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
175 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
178 CUDA_COMPILE(${VAR} ${ARGN})
180 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
181 set(${var} "${${var}_backup_in_cuda_compile_}")
182 unset(${var}_backup_in_cuda_compile_)
186 unset(CUDA_ARCH_BIN CACHE)
187 unset(CUDA_ARCH_PTX CACHE)