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)
30 find_cuda_helper_libs(nvcuvid)
34 message(STATUS "CUDA detected: " ${CUDA_VERSION})
36 set(CUDA_ARCH_BIN "1.1 1.2 1.3 2.0 2.1(2.0) 3.0" CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
37 set(CUDA_ARCH_PTX "2.0 3.0" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
39 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
40 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
42 # Ckeck if user specified 1.0 compute capability: we don't support it
43 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
44 set(CUDA_ARCH_BIN_OR_PTX_10 0)
45 if(NOT ${HAS_ARCH_10} STREQUAL "")
46 set(CUDA_ARCH_BIN_OR_PTX_10 1)
49 # NVCC flags to be set
50 set(NVCC_FLAGS_EXTRA "")
52 # These vars will be passed into the templates
53 set(OPENCV_CUDA_ARCH_BIN "")
54 set(OPENCV_CUDA_ARCH_PTX "")
55 set(OPENCV_CUDA_ARCH_FEATURES "")
57 # Tell NVCC to add binaries for the specified GPUs
58 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
59 foreach(ARCH IN LISTS ARCH_LIST)
60 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
61 # User explicitly specified PTX for the concrete BIN
62 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
63 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
64 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
66 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
67 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
68 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
69 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
73 # Tell NVCC to add PTX intermediate code for the specified architectures
74 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
75 foreach(ARCH IN LISTS ARCH_LIST)
76 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
77 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
78 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
81 if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "arm")
82 set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --target-cpu-architecture=ARM")
85 # These vars will be processed in other scripts
86 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
87 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
89 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
91 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
94 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
97 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
99 find_cuda_helper_libs(npp)
101 macro(ocv_cuda_compile VAR)
102 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
103 set(${var}_backup_in_cuda_compile_ "${${var}}")
105 # we reomove /EHa as it leasd warnings under windows
106 string(REPLACE "/EHa" "" ${var} "${${var}}")
108 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
109 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
112 if(BUILD_SHARED_LIBS)
113 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
117 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
120 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
123 # disabled because of multiple warnings during building nvcc auto generated files
124 if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
125 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
128 CUDA_COMPILE(${VAR} ${ARGN})
130 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
131 set(${var} "${${var}_backup_in_cuda_compile_}")
132 unset(${var}_backup_in_cuda_compile_)
136 unset(CUDA_ARCH_BIN CACHE)
137 unset(CUDA_ARCH_PTX CACHE)