OpenCVDetectCUDA.cmake

find_package(CUDA 4.1)

if(CUDA_FOUND)
  set(HAVE_CUDA 1)

  if(WITH_CUFFT)
    set(HAVE_CUFFT 1)
  endif()

  if(WITH_CUBLAS)
    set(HAVE_CUBLAS 1)
  endif()

  message(STATUS "CUDA detected: " ${CUDA_VERSION})

  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")
  set(CUDA_ARCH_PTX "2.0" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")

  string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
  string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")

  # Ckeck if user specified 1.0 compute capability: we don't support it
  string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
  set(CUDA_ARCH_BIN_OR_PTX_10 0)
  if(NOT ${HAS_ARCH_10} STREQUAL "")
    set(CUDA_ARCH_BIN_OR_PTX_10 1)
  endif()

  # NVCC flags to be set
  set(NVCC_FLAGS_EXTRA "")

  # These vars will be passed into the templates
  set(OPENCV_CUDA_ARCH_BIN "")
  set(OPENCV_CUDA_ARCH_PTX "")
  set(OPENCV_CUDA_ARCH_FEATURES "")

  # Tell NVCC to add binaries for the specified GPUs
  string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
  foreach(ARCH IN LISTS ARCH_LIST)
    if(ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
      # User explicitly specified PTX for the concrete BIN
      set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
      set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
      set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
    else()
      # User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
      set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
      set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
      set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
    endif()
  endforeach()

  # Tell NVCC to add PTX intermediate code for the specified architectures
  string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
    foreach(ARCH IN LISTS ARCH_LIST)
      set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
      set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
      set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
    endforeach()

  # These vars will be processed in other scripts
  set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
  set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")

  message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")

  unset(CUDA_npp_LIBRARY CACHE)
  find_cuda_helper_libs(npp)

  macro(OCV_CUDA_COMPILE VAR)
    if (BUILD_SHARED_LIBS)
      set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -DCVAPI_EXPORTS)
    endif()

    if(UNIX OR APPLE)
      set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fPIC)
    endif()
    if(APPLE)
      set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
    endif()

    # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1) 
    set(CMAKE_CXX_FLAGS_DEBUG_ ${CMAKE_CXX_FLAGS_DEBUG}) 
    string(REPLACE "-ggdb3" "" CMAKE_CXX_FLAGS_DEBUG ${CMAKE_CXX_FLAGS_DEBUG})
    CUDA_COMPILE(${VAR} ${ARGN})
    set(CMAKE_CXX_DEBUG_FLAGS ${CMAKE_CXX_FLAGS_DEBUG_})
  endmacro()
else()
  unset(CUDA_ARCH_BIN CACHE)
  unset(CUDA_ARCH_PTX CACHE)
endif()