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)
33 if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "arm")
34 unset(CUDA_TOOLKIT_INCLUDE CACHE)
35 unset(CUDA_CUDART_LIBRARY CACHE)
36 unset(CUDA_cublas_LIBRARY CACHE)
37 unset(CUDA_cufft_LIBRARY CACHE)
38 unset(CUDA_npp_LIBRARY CACHE)
41 set(cuda_arm_path "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabi")
43 set(cuda_arm_path "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf")
46 set(CUDA_TOOLKIT_INCLUDE "${cuda_arm_path}/include" CACHE PATH "include path")
47 set(CUDA_INCLUDE_DIRS ${CUDA_TOOLKIT_INCLUDE})
49 set(cuda_arm_library_path "${cuda_arm_path}/lib")
51 set(CUDA_CUDART_LIBRARY "${cuda_arm_library_path}/libcudart.so" CACHE FILEPATH "cudart library")
52 set(CUDA_LIBRARIES ${CUDA_CUDART_LIBRARY})
53 set(CUDA_cublas_LIBRARY "${cuda_arm_library_path}/libcublas.so" CACHE FILEPATH "cublas library")
54 set(CUDA_cufft_LIBRARY "${cuda_arm_library_path}/libcufft.so" CACHE FILEPATH "cufft library")
55 set(CUDA_nppc_LIBRARY "${cuda_arm_library_path}/libnppc.so" CACHE FILEPATH "nppc library")
56 set(CUDA_nppi_LIBRARY "${cuda_arm_library_path}/libnppi.so" CACHE FILEPATH "nppi library")
57 set(CUDA_npps_LIBRARY "${cuda_arm_library_path}/libnpps.so" CACHE FILEPATH "npps library")
58 set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}" CACHE STRING "npp library")
60 unset(CUDA_npp_LIBRARY CACHE)
62 find_cuda_helper_libs(nppc)
63 find_cuda_helper_libs(nppi)
64 find_cuda_helper_libs(npps)
66 set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}" CACHE STRING "npp library")
71 find_cuda_helper_libs(nvcuvid)
75 message(STATUS "CUDA detected: " ${CUDA_VERSION})
77 set(_generations "Fermi" "Kepler")
78 if(NOT CMAKE_CROSSCOMPILING)
79 list(APPEND _generations "Auto")
81 set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
82 if( CMAKE_VERSION VERSION_GREATER "2.8" )
83 set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
87 if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
88 string(REPLACE ";" ", " _generations "${_generations}")
89 message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
91 unset(CUDA_ARCH_BIN CACHE)
92 unset(CUDA_ARCH_PTX CACHE)
95 set(__cuda_arch_ptx "")
96 if(CUDA_GENERATION STREQUAL "Fermi")
97 set(__cuda_arch_bin "2.0 2.1(2.0)")
98 elseif(CUDA_GENERATION STREQUAL "Kepler")
99 if(${CUDA_VERSION} VERSION_LESS "5.0")
100 set(__cuda_arch_bin "3.0")
102 set(__cuda_arch_bin "3.0 3.5")
104 elseif(CUDA_GENERATION STREQUAL "Auto")
105 execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
106 WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
107 RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
108 ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
109 if(NOT _nvcc_res EQUAL 0)
110 message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
112 set(__cuda_arch_bin "${_nvcc_out}")
116 if(NOT DEFINED __cuda_arch_bin)
117 if(${CUDA_VERSION} VERSION_LESS "5.0")
118 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
120 set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
122 set(__cuda_arch_ptx "3.0")
125 set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
126 set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
128 string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
129 string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
131 # Ckeck if user specified 1.0 compute capability: we don't support it
132 string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
133 set(CUDA_ARCH_BIN_OR_PTX_10 0)
134 if(NOT ${HAS_ARCH_10} STREQUAL "")
135 set(CUDA_ARCH_BIN_OR_PTX_10 1)
138 # NVCC flags to be set
139 set(NVCC_FLAGS_EXTRA "")
141 # These vars will be passed into the templates
142 set(OPENCV_CUDA_ARCH_BIN "")
143 set(OPENCV_CUDA_ARCH_PTX "")
144 set(OPENCV_CUDA_ARCH_FEATURES "")
146 # Tell NVCC to add binaries for the specified GPUs
147 string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
148 foreach(ARCH IN LISTS ARCH_LIST)
149 if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
150 # User explicitly specified PTX for the concrete BIN
151 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
152 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
153 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
155 # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
156 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
157 set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
158 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
162 # Tell NVCC to add PTX intermediate code for the specified architectures
163 string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
164 foreach(ARCH IN LISTS ARCH_LIST)
165 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
166 set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
167 set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
170 if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "arm")
171 set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --target-cpu-architecture=ARM")
174 # These vars will be processed in other scripts
175 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
176 set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
178 message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
180 OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
183 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
186 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
188 macro(ocv_cuda_compile VAR)
189 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
190 set(${var}_backup_in_cuda_compile_ "${${var}}")
192 # we remove /EHa as it generates warnings under windows
193 string(REPLACE "/EHa" "" ${var} "${${var}}")
195 # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
196 string(REPLACE "-ggdb3" "" ${var} "${${var}}")
198 # we remove -Wsign-promo as it generates warnings under linux
199 string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
202 if(BUILD_SHARED_LIBS)
203 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
207 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
210 set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
213 # disabled because of multiple warnings during building nvcc auto generated files
214 if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
215 ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
218 CUDA_COMPILE(${VAR} ${ARGN})
220 foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
221 set(${var} "${${var}_backup_in_cuda_compile_}")
222 unset(${var}_backup_in_cuda_compile_)
226 unset(CUDA_ARCH_BIN CACHE)
227 unset(CUDA_ARCH_PTX CACHE)