1if(WIN32 AND NOT MSVC)
2  message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
3  return()
4endif()
5
6if(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).")
8  return()
9endif()
10
11set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
12
13if(ANDROID)
14  set(CUDA_TARGET_OS_VARIANT "Android")
15endif()
16find_host_package(CUDA "${MIN_VER_CUDA}" QUIET)
17
18list(REMOVE_AT CMAKE_MODULE_PATH 0)
19
20if(CUDA_FOUND)
21  set(HAVE_CUDA 1)
22
23  if(WITH_CUFFT)
24    set(HAVE_CUFFT 1)
25  endif()
26
27  if(WITH_CUBLAS)
28    set(HAVE_CUBLAS 1)
29  endif()
30
31  if(WITH_NVCUVID)
32    find_cuda_helper_libs(nvcuvid)
33    if(WIN32)
34      find_cuda_helper_libs(nvcuvenc)
35    endif()
36    set(HAVE_NVCUVID 1)
37  endif()
38
39  message(STATUS "CUDA detected: " ${CUDA_VERSION})
40
41  set(_generations "Fermi" "Kepler")
42  if(NOT CMAKE_CROSSCOMPILING)
43    list(APPEND _generations "Auto")
44  endif()
45  set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
46  if( CMAKE_VERSION VERSION_GREATER "2.8" )
47    set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
48  endif()
49
50  if(CUDA_GENERATION)
51    if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
52      string(REPLACE ";" ", " _generations "${_generations}")
53      message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
54    endif()
55    unset(CUDA_ARCH_BIN CACHE)
56    unset(CUDA_ARCH_PTX CACHE)
57  endif()
58
59  set(__cuda_arch_ptx "")
60  if(CUDA_GENERATION STREQUAL "Fermi")
61    set(__cuda_arch_bin "2.0 2.1(2.0)")
62  elseif(CUDA_GENERATION STREQUAL "Kepler")
63    if(${CUDA_VERSION} VERSION_LESS "5.0")
64      set(__cuda_arch_bin "3.0")
65    else()
66      set(__cuda_arch_bin "3.0 3.5")
67    endif()
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.")
75    else()
76      set(__cuda_arch_bin "${_nvcc_out}")
77      string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
78    endif()
79  endif()
80
81  if(NOT DEFINED __cuda_arch_bin)
82    if(ANDROID)
83      if(ARM)
84        set(__cuda_arch_bin "3.2")
85        set(__cuda_arch_ptx "")
86      elseif(AARCH64)
87        set(__cuda_arch_bin "5.3")
88        set(__cuda_arch_ptx "")
89      endif()
90    else()
91      if(${CUDA_VERSION} VERSION_LESS "5.0")
92        set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
93      elseif(${CUDA_VERSION} VERSION_GREATER "6.5")
94        set(__cuda_arch_bin "2.0 2.1(2.0) 3.0 3.5")
95      else()
96        set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
97      endif()
98      set(__cuda_arch_ptx "3.0")
99    endif()
100  endif()
101
102  set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
103  set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
104
105  string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
106  string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
107
108  # Ckeck if user specified 1.0 compute capability: we don't support it
109  string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
110  set(CUDA_ARCH_BIN_OR_PTX_10 0)
111  if(NOT ${HAS_ARCH_10} STREQUAL "")
112    set(CUDA_ARCH_BIN_OR_PTX_10 1)
113  endif()
114
115  # NVCC flags to be set
116  set(NVCC_FLAGS_EXTRA "")
117
118  # These vars will be passed into the templates
119  set(OPENCV_CUDA_ARCH_BIN "")
120  set(OPENCV_CUDA_ARCH_PTX "")
121  set(OPENCV_CUDA_ARCH_FEATURES "")
122
123  # Tell NVCC to add binaries for the specified GPUs
124  string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
125  foreach(ARCH IN LISTS ARCH_LIST)
126    if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
127      # User explicitly specified PTX for the concrete BIN
128      set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
129      set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
130      set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
131    else()
132      # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
133      set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
134      set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
135      set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
136    endif()
137  endforeach()
138
139  # Tell NVCC to add PTX intermediate code for the specified architectures
140  string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
141  foreach(ARCH IN LISTS ARCH_LIST)
142    set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
143    set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
144    set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
145  endforeach()
146
147  # These vars will be processed in other scripts
148  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
149  set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
150
151  if(ANDROID)
152    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
153  endif()
154
155  message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
156
157  OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
158
159  if(CUDA_FAST_MATH)
160    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
161  endif()
162
163  mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
164
165  macro(ocv_cuda_compile VAR)
166    foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
167      set(${var}_backup_in_cuda_compile_ "${${var}}")
168
169      # we remove /EHa as it generates warnings under windows
170      string(REPLACE "/EHa" "" ${var} "${${var}}")
171
172      # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
173      string(REPLACE "-ggdb3" "" ${var} "${${var}}")
174
175      # we remove -Wsign-promo as it generates warnings under linux
176      string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
177
178      # we remove -Wno-sign-promo as it generates warnings under linux
179      string(REPLACE "-Wno-sign-promo" "" ${var} "${${var}}")
180
181      # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
182      # but NVCC uses C compiler by default
183      string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")
184
185      # we remove -frtti because it's used for C++ compiler
186      # but NVCC uses C compiler by default
187      string(REPLACE "-frtti" "" ${var} "${${var}}")
188
189      string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
190    endforeach()
191
192    if(BUILD_SHARED_LIBS)
193      set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
194    endif()
195
196    if(UNIX OR APPLE)
197      set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
198    endif()
199    if(APPLE)
200      set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
201    endif()
202
203    # disabled because of multiple warnings during building nvcc auto generated files
204    if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
205      ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
206    endif()
207
208    CUDA_COMPILE(${VAR} ${ARGN})
209
210    foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
211      set(${var} "${${var}_backup_in_cuda_compile_}")
212      unset(${var}_backup_in_cuda_compile_)
213    endforeach()
214  endmacro()
215else()
216  unset(CUDA_ARCH_BIN CACHE)
217  unset(CUDA_ARCH_PTX CACHE)
218endif()
219
220if(HAVE_CUDA)
221  set(CUDA_LIBS_PATH "")
222  foreach(p ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
223    get_filename_component(_tmp ${p} PATH)
224    list(APPEND CUDA_LIBS_PATH ${_tmp})
225  endforeach()
226
227  if(HAVE_CUBLAS)
228    foreach(p ${CUDA_cublas_LIBRARY})
229      get_filename_component(_tmp ${p} PATH)
230      list(APPEND CUDA_LIBS_PATH ${_tmp})
231    endforeach()
232  endif()
233
234  if(HAVE_CUFFT)
235    foreach(p ${CUDA_cufft_LIBRARY})
236      get_filename_component(_tmp ${p} PATH)
237      list(APPEND CUDA_LIBS_PATH ${_tmp})
238    endforeach()
239  endif()
240
241  list(REMOVE_DUPLICATES CUDA_LIBS_PATH)
242  link_directories(${CUDA_LIBS_PATH})
243
244  set(CUDA_LIBRARIES_ABS ${CUDA_LIBRARIES})
245  ocv_convert_to_lib_name(CUDA_LIBRARIES ${CUDA_LIBRARIES})
246  set(CUDA_npp_LIBRARY_ABS ${CUDA_npp_LIBRARY})
247  ocv_convert_to_lib_name(CUDA_npp_LIBRARY ${CUDA_npp_LIBRARY})
248  if(HAVE_CUBLAS)
249    set(CUDA_cublas_LIBRARY_ABS ${CUDA_cublas_LIBRARY})
250    ocv_convert_to_lib_name(CUDA_cublas_LIBRARY ${CUDA_cublas_LIBRARY})
251  endif()
252
253  if(HAVE_CUFFT)
254    set(CUDA_cufft_LIBRARY_ABS ${CUDA_cufft_LIBRARY})
255    ocv_convert_to_lib_name(CUDA_cufft_LIBRARY ${CUDA_cufft_LIBRARY})
256  endif()
257endif()
258