if(${CMAKE_VERSION} VERSION_LESS "2.8.3")
  message(STATUS "WITH_CUDA flag requires CMake 2.8.3 or newer. CUDA support is disabled.")
  return()
endif()

if(WIN32 AND NOT MSVC)
  message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
  return()
endif()

if(CMAKE_COMPILER_IS_GNUCXX AND NOT APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
  message(STATUS "CUDA compilation is disabled (due to Clang unsupported on your platform).")
  return()
endif()

set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})

foreach(var INCLUDE LIBRARY PROGRAM)
  set(__old_frpm_${var} "${CMAKE_FIND_ROOT_PATH_MODE_${var}}")
endforeach()

set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY BOTH)
set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE NEVER)

find_package(CUDA 4.2 QUIET)

foreach(var INCLUDE LIBRARY PROGRAM)
  set(CMAKE_FIND_ROOT_PATH_MODE_${var} "${__old_frpm_${var}}")
endforeach()

list(REMOVE_AT CMAKE_MODULE_PATH 0)

if(CUDA_FOUND)
  set(HAVE_CUDA 1)

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

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

  if(WITH_NVCUVID)
    find_cuda_helper_libs(nvcuvid)
    set(HAVE_NVCUVID 1)
  endif()

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

  set(_generations "Fermi" "Kepler")
  if(NOT CMAKE_CROSSCOMPILING)
    list(APPEND _generations "Auto")
  endif()
  set(CUDA_GENERATION "" CACHE STRING "Build CUDA device code only for specific GPU architecture. Leave empty to build for all architectures.")
  if( CMAKE_VERSION VERSION_GREATER "2.8" )
    set_property( CACHE CUDA_GENERATION PROPERTY STRINGS "" ${_generations} )
  endif()

  if(CUDA_GENERATION)
    if(NOT ";${_generations};" MATCHES ";${CUDA_GENERATION};")
      string(REPLACE ";" ", " _generations "${_generations}")
      message(FATAL_ERROR "ERROR: ${_generations} Generations are suppered.")
    endif()
    unset(CUDA_ARCH_BIN CACHE)
    unset(CUDA_ARCH_PTX CACHE)
  endif()

  set(__cuda_arch_ptx "")
  if(CUDA_GENERATION STREQUAL "Fermi")
    set(__cuda_arch_bin "2.0 2.1(2.0)")
  elseif(CUDA_GENERATION STREQUAL "Kepler")
    if(${CUDA_VERSION} VERSION_LESS "5.0")
      set(__cuda_arch_bin "3.0")
    else()
      set(__cuda_arch_bin "3.0 3.5")
    endif()
  elseif(CUDA_GENERATION STREQUAL "Auto")
    execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
                     WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
                     RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
                     ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
    if(NOT _nvcc_res EQUAL 0)
      message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
    else()
      set(__cuda_arch_bin "${_nvcc_out}")
      string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
    endif()
  endif()

  if(NOT DEFINED __cuda_arch_bin)
    if(ANDROID)
      set(__cuda_arch_bin "3.2")
      set(__cuda_arch_ptx "")
    else()
      if(${CUDA_VERSION} VERSION_LESS "5.0")
        set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
      else()
        set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
      endif()
      set(__cuda_arch_ptx "3.0")
    endif()
  endif()

  set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
  set(CUDA_ARCH_PTX ${__cuda_arch_ptx} 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}")

  if(ANDROID)
    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-target-os-variant=Android")
  endif()

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

  OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)

  if(CUDA_FAST_MATH)
    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)
  endif()

  mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)

  macro(ocv_cuda_compile VAR)
    foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
      set(${var}_backup_in_cuda_compile_ "${${var}}")

      # we remove /EHa as it generates warnings under windows
      string(REPLACE "/EHa" "" ${var} "${${var}}")

      # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
      string(REPLACE "-ggdb3" "" ${var} "${${var}}")

      # we remove -Wsign-promo as it generates warnings under linux
      string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")

      # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
      # but NVCC uses C compiler by default
      string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")

      # we remove -frtti because it's used for C++ compiler
      # but NVCC uses C compiler by default
      string(REPLACE "-frtti" "" ${var} "${${var}}")
    endforeach()

    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()

    # disabled because of multiple warnings during building nvcc auto generated files
    if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
      ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
    endif()

    CUDA_COMPILE(${VAR} ${ARGN})

    foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
      set(${var} "${${var}_backup_in_cuda_compile_}")
      unset(${var}_backup_in_cuda_compile_)
    endforeach()
  endmacro()
else()
  unset(CUDA_ARCH_BIN CACHE)
  unset(CUDA_ARCH_PTX CACHE)
endif()