mirror of
https://github.com/opencv/opencv.git
synced 2024-11-27 12:40:05 +08:00
build: first class cuda support
This commit is contained in:
cudawarped
Alexander Smorkalov
parent
8c22b4b54b
commit
7d681cf80d
@ -73,6 +73,10 @@ if(POLICY CMP0077)
|
||||
cmake_policy(SET CMP0077 NEW) # CMake 3.13+: option() honors normal variables.
|
||||
endif()
|
||||
|
||||
if(POLICY CMP0091)
|
||||
cmake_policy(SET CMP0091 NEW) # CMake 3.15+: leave MSVC runtime selection out of default CMAKE_<LANG>_FLAGS_<CONFIG> flags
|
||||
endif()
|
||||
|
||||
if(POLICY CMP0146)
|
||||
cmake_policy(SET CMP0146 OLD) # CMake 3.27+: use CMake FindCUDA if available.
|
||||
endif()
|
||||
@ -549,6 +553,9 @@ OCV_OPTION(OPENCV_ENABLE_MEMALIGN "Enable posix_memalign or memalign usage"
|
||||
OCV_OPTION(OPENCV_DISABLE_FILESYSTEM_SUPPORT "Disable filesystem support" OFF)
|
||||
OCV_OPTION(OPENCV_DISABLE_THREAD_SUPPORT "Build the library without multi-threaded code." OFF)
|
||||
OCV_OPTION(OPENCV_SEMIHOSTING "Build the library for semihosting target (Arm). See https://developer.arm.com/documentation/100863/latest." OFF)
|
||||
OCV_OPTION(ENABLE_CUDA_FIRST_CLASS_LANGUAGE "Enable CUDA as a first class language, if enabled dependant projects will need to use CMake >= 3.18" (WITH_CUDA AND NOT CMAKE_VERSION VERSION_LESS 3.18)
|
||||
VISIBLE_IF (WITH_CUDA AND NOT CMAKE_VERSION VERSION_LESS 3.18)
|
||||
VERIFY HAVE_CUDA)
|
||||
|
||||
OCV_OPTION(ENABLE_PYLINT "Add target with Pylint checks" (BUILD_DOCS OR BUILD_EXAMPLES) IF (NOT CMAKE_CROSSCOMPILING AND NOT APPLE_FRAMEWORK) )
|
||||
OCV_OPTION(ENABLE_FLAKE8 "Add target with Python flake8 checker" (BUILD_DOCS OR BUILD_EXAMPLES) IF (NOT CMAKE_CROSSCOMPILING AND NOT APPLE_FRAMEWORK) )
|
||||
@ -657,6 +664,51 @@ include(cmake/OpenCVCompilerOptions.cmake)
|
||||
|
||||
ocv_cmake_hook(POST_COMPILER_OPTIONS)
|
||||
|
||||
# --- CUDA Support ---
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
|
||||
if(CMAKE_VERSION VERSION_LESS 3.18)
|
||||
message(WARNING "CUDA: First class language only supported for CMake versions >= 3.18, falling back to FindCUDA!")
|
||||
set(ENABLE_CUDA_FIRST_CLASS_LANGUAGE OFF CACHE BOOL "Enable CUDA as a first class language, if enabled dependant projects will need to use CMake >= 3.18" FORCE)
|
||||
else()
|
||||
|
||||
# Check CUDA_PATH if supplied
|
||||
if(UNIX AND CUDA_PATH AND NOT ENV{CUDA_PATH})
|
||||
set(ENV{CUDA_PATH} ${CUDA_PATH})
|
||||
elseif(WIN32 AND CUDA_PATH)
|
||||
set(ENV{PATH} "${CUDA_PATH}\\bin\;$ENV{PATH}")
|
||||
endif()
|
||||
include(CheckLanguage)
|
||||
check_language(CUDA)
|
||||
|
||||
# Fallback to checking default locations
|
||||
if(NOT CMAKE_CUDA_COMPILER)
|
||||
# Checking windows default search location isn't possible because the CUDA Toolkit is installed to C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/vXX.X
|
||||
if(WIN32)
|
||||
if(CMAKE_GENERATOR MATCHES "Visual Studio")
|
||||
message(STATUS "CUDA: Not detected, when using stand alone installations with the Visual Studio generator the path to the CUDA toolkit should be manually specified with -Tcuda=. e.g. -Tcuda=\"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/vXX.X\"")
|
||||
else()
|
||||
message(STATUS "CUDA: Not detected, for stand alone installations the path to the CUDA toolkit should be manually specified with -DCUDA_PATH=. e.g. -DCUDA_PATH=\"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/vXX.X\"")
|
||||
endif()
|
||||
elseif(UNIX)
|
||||
message(STATUS "CUDA: Not detected, make sure you have performed the mandatory Post-installation actions described in the CUDA installation guide.\n For stand alone installations you can set the CUDA_PATH environmental or CMake variable. e.g. export CUDA_PATH=/usr/local/cuda-XX.X or -DCUDA_PATH=/usr/local/cuda-XX.X.")
|
||||
message(STATUS "CUDA: Falling back to searching for the CUDA compiler in its default location (/usr/local/cuda)")
|
||||
set(CUDA_PATH "/usr/local/cuda" CACHE INTERNAL "")
|
||||
set(ENV{CUDA_PATH} ${CUDA_PATH})
|
||||
unset(CMAKE_CUDA_COMPILER CACHE)
|
||||
unset(CMAKE_CUDA_COMPILER)
|
||||
check_language(CUDA)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
cmake_policy(SET CMP0092 NEW) # CMake 3.15+: leave warning flags out of default CMAKE_<LANG>_FLAGS flags.
|
||||
if(CMAKE_CUDA_COMPILER)
|
||||
enable_language(CUDA)
|
||||
elseif(UNIX)
|
||||
message(WARNING "CUDA: Not detected! If you are not using the default host compiler (g++) then you need to specify both CMAKE_CUDA_HOST_COMPILER and CMAKE_CUDA_COMPILER. e.g. -DCMAKE_CUDA_HOST_COMPILER=/usr/bin/clang++ -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc.")
|
||||
endif()
|
||||
endif()
|
||||
endif()
|
||||
|
||||
# ----------------------------------------------------------------------------
|
||||
# CHECK FOR SYSTEM LIBRARIES, OPTIONS, ETC..
|
||||
# ----------------------------------------------------------------------------
|
||||
|
||||
@ -33,34 +33,44 @@ endif()
|
||||
# Ignore warning: This object file does not define any previously undefined public symbols, ...
|
||||
set(CMAKE_STATIC_LINKER_FLAGS "${CMAKE_STATIC_LINKER_FLAGS} /IGNORE:4221")
|
||||
|
||||
if(POLICY CMP0091)
|
||||
cmake_policy(GET CMP0091 MSVC_RUNTIME_SET_BY_ABSTRACTION)
|
||||
endif()
|
||||
|
||||
if(NOT BUILD_SHARED_LIBS AND BUILD_WITH_STATIC_CRT)
|
||||
foreach(flag_var
|
||||
CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
|
||||
CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO
|
||||
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
|
||||
CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
|
||||
if(${flag_var} MATCHES "/MD")
|
||||
string(REGEX REPLACE "/MD" "/MT" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
if(${flag_var} MATCHES "/MDd")
|
||||
string(REGEX REPLACE "/MDd" "/MTd" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
endforeach(flag_var)
|
||||
if(MSVC_RUNTIME_SET_BY_ABSTRACTION STREQUAL "NEW")
|
||||
set(CMAKE_MSVC_RUNTIME_LIBRARY "MultiThreaded$<$<CONFIG:Debug>:Debug>")
|
||||
else()
|
||||
foreach(flag_var
|
||||
CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
|
||||
CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO
|
||||
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
|
||||
CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
|
||||
if(${flag_var} MATCHES "/MD")
|
||||
string(REGEX REPLACE "/MD" "/MT" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
if(${flag_var} MATCHES "/MDd")
|
||||
string(REGEX REPLACE "/MDd" "/MTd" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
endforeach(flag_var)
|
||||
endif()
|
||||
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} /NODEFAULTLIB:atlthunk.lib")
|
||||
set(CMAKE_EXE_LINKER_FLAGS_DEBUG "${CMAKE_EXE_LINKER_FLAGS_DEBUG} /NODEFAULTLIB:libcmt.lib /NODEFAULTLIB:libcpmt.lib /NODEFAULTLIB:msvcrt.lib")
|
||||
set(CMAKE_EXE_LINKER_FLAGS_RELEASE "${CMAKE_EXE_LINKER_FLAGS_RELEASE} /NODEFAULTLIB:libcmtd.lib /NODEFAULTLIB:libcpmtd.lib /NODEFAULTLIB:msvcrtd.lib")
|
||||
else()
|
||||
foreach(flag_var
|
||||
CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
|
||||
CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO
|
||||
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
|
||||
CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
|
||||
if(${flag_var} MATCHES "/MT")
|
||||
string(REGEX REPLACE "/MT" "/MD" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
if(${flag_var} MATCHES "/MTd")
|
||||
string(REGEX REPLACE "/MTd" "/MDd" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
endforeach(flag_var)
|
||||
if(NOT MSVC_RUNTIME_SET_BY_ABSTRACTION STREQUAL "NEW")
|
||||
foreach(flag_var
|
||||
CMAKE_C_FLAGS CMAKE_C_FLAGS_DEBUG CMAKE_C_FLAGS_RELEASE
|
||||
CMAKE_C_FLAGS_MINSIZEREL CMAKE_C_FLAGS_RELWITHDEBINFO
|
||||
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
|
||||
CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
|
||||
if(${flag_var} MATCHES "/MT")
|
||||
string(REGEX REPLACE "/MT" "/MD" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
if(${flag_var} MATCHES "/MTd")
|
||||
string(REGEX REPLACE "/MTd" "/MDd" ${flag_var} "${${flag_var}}")
|
||||
endif()
|
||||
endforeach(flag_var)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
@ -1,10 +1,10 @@
|
||||
if((WIN32 AND NOT MSVC) OR OPENCV_CMAKE_FORCE_CUDA)
|
||||
message(STATUS "CUDA compilation is disabled (due to only Visual Studio compiler supported on your platform).")
|
||||
message(STATUS "CUDA: Compilation is disabled (due to only Visual Studio compiler supported on your platform).")
|
||||
return()
|
||||
endif()
|
||||
|
||||
if((NOT UNIX AND CV_CLANG) OR OPENCV_CMAKE_FORCE_CUDA)
|
||||
message(STATUS "CUDA compilation is disabled (due to Clang unsupported on your platform).")
|
||||
message(STATUS "CUDA: Compilation is disabled (due to Clang unsupported on your platform).")
|
||||
return()
|
||||
endif()
|
||||
|
||||
@ -31,444 +31,144 @@ else()
|
||||
list(REMOVE_AT CMAKE_MODULE_PATH 0)
|
||||
endif()
|
||||
|
||||
if(CUDA_FOUND)
|
||||
unset(CUDA_nvcuvenc_LIBRARY CACHE)
|
||||
set(HAVE_CUDA 1)
|
||||
if(NOT CUDA_VERSION VERSION_LESS 11.0)
|
||||
# CUDA 11.0 removes nppicom
|
||||
ocv_list_filterout(CUDA_nppi_LIBRARY "nppicom")
|
||||
ocv_list_filterout(CUDA_npp_LIBRARY "nppicom")
|
||||
endif()
|
||||
|
||||
if(WITH_CUFFT)
|
||||
set(HAVE_CUFFT 1)
|
||||
endif()
|
||||
|
||||
if(WITH_CUBLAS)
|
||||
set(HAVE_CUBLAS 1)
|
||||
endif()
|
||||
|
||||
if(WITH_CUDNN)
|
||||
set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
|
||||
find_host_package(CUDNN "${MIN_VER_CUDNN}")
|
||||
list(REMOVE_AT CMAKE_MODULE_PATH 0)
|
||||
|
||||
if(CUDNN_FOUND)
|
||||
set(HAVE_CUDNN 1)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if(WITH_NVCUVID OR WITH_NVCUVENC)
|
||||
macro(ocv_cuda_SEARCH_NVCUVID_HEADER _filename _result)
|
||||
# place header file under CUDA_TOOLKIT_TARGET_DIR or CUDA_TOOLKIT_ROOT_DIR
|
||||
find_path(_header_result
|
||||
${_filename}
|
||||
PATHS "${CUDA_TOOLKIT_TARGET_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}"
|
||||
ENV CUDA_PATH
|
||||
ENV CUDA_INC_PATH
|
||||
PATH_SUFFIXES include
|
||||
NO_DEFAULT_PATH
|
||||
)
|
||||
if("x${_header_result}" STREQUAL "x_header_result-NOTFOUND")
|
||||
set(${_result} 0)
|
||||
else()
|
||||
set(${_result} 1)
|
||||
endif()
|
||||
unset(_header_result CACHE)
|
||||
endmacro()
|
||||
if(WITH_NVCUVID)
|
||||
ocv_cuda_SEARCH_NVCUVID_HEADER("nvcuvid.h" HAVE_NVCUVID_HEADER)
|
||||
ocv_cuda_SEARCH_NVCUVID_HEADER("dynlink_nvcuvid.h" HAVE_DYNLINK_NVCUVID_HEADER)
|
||||
find_cuda_helper_libs(nvcuvid)
|
||||
if(CUDA_nvcuvid_LIBRARY AND (${HAVE_NVCUVID_HEADER} OR ${HAVE_DYNLINK_NVCUVID_HEADER}))
|
||||
# make sure to have both header and library before enabling
|
||||
set(HAVE_NVCUVID 1)
|
||||
endif()
|
||||
endif()
|
||||
if(WITH_NVCUVENC)
|
||||
ocv_cuda_SEARCH_NVCUVID_HEADER("nvEncodeAPI.h" HAVE_NVCUVENC_HEADER)
|
||||
if(WIN32)
|
||||
find_cuda_helper_libs(nvencodeapi)
|
||||
else()
|
||||
find_cuda_helper_libs(nvidia-encode)
|
||||
endif()
|
||||
if((CUDA_nvencodeapi_LIBRARY OR CUDA_nvidia-encode_LIBRARY) AND ${HAVE_NVCUVENC_HEADER})
|
||||
set(HAVE_NVCUVENC 1)
|
||||
endif()
|
||||
endif()
|
||||
endif()
|
||||
|
||||
message(STATUS "CUDA detected: " ${CUDA_VERSION})
|
||||
|
||||
OCV_OPTION(CUDA_ENABLE_DEPRECATED_GENERATION "Enable deprecated generations in the list" OFF)
|
||||
set(_generations "Maxwell" "Pascal" "Volta" "Turing" "Ampere" "Lovelace" "Hopper")
|
||||
if(CUDA_ENABLE_DEPRECATED_GENERATION)
|
||||
set(_generations "Fermi" "${_generations}")
|
||||
set(_generations "Kepler" "${_generations}")
|
||||
endif()
|
||||
set(_arch_fermi "2.0")
|
||||
set(_arch_kepler "3.0;3.5;3.7")
|
||||
set(_arch_maxwell "5.0;5.2")
|
||||
set(_arch_pascal "6.0;6.1")
|
||||
set(_arch_volta "7.0")
|
||||
set(_arch_turing "7.5")
|
||||
set(_arch_ampere "8.0;8.6")
|
||||
set(_arch_lovelace "8.9")
|
||||
set(_arch_hopper "9.0")
|
||||
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 (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html).")
|
||||
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: ${_generations} Generations are supported.")
|
||||
endif()
|
||||
unset(CUDA_ARCH_BIN CACHE)
|
||||
unset(CUDA_ARCH_PTX CACHE)
|
||||
endif()
|
||||
|
||||
if(OPENCV_CUDA_DETECTION_NVCC_FLAGS MATCHES "-ccbin")
|
||||
# already specified by user
|
||||
elseif(CUDA_HOST_COMPILER AND EXISTS "${CUDA_HOST_COMPILER}")
|
||||
get_filename_component(c_compiler_realpath "${CMAKE_C_COMPILER}" REALPATH)
|
||||
# C compiler doesn't work with --run option, forcing C++ compiler instead
|
||||
if(CUDA_HOST_COMPILER STREQUAL c_compiler_realpath OR CUDA_HOST_COMPILER STREQUAL CMAKE_C_COMPILER)
|
||||
if(DEFINED CMAKE_CXX_COMPILER)
|
||||
get_filename_component(cxx_compiler_realpath "${CMAKE_CXX_COMPILER}" REALPATH)
|
||||
LIST(APPEND OPENCV_CUDA_DETECTION_NVCC_FLAGS -ccbin "${cxx_compiler_realpath}")
|
||||
else()
|
||||
message(STATUS "CUDA: CMAKE_CXX_COMPILER is not available. You may need to specify CUDA_HOST_COMPILER.")
|
||||
endif()
|
||||
else()
|
||||
LIST(APPEND OPENCV_CUDA_DETECTION_NVCC_FLAGS -ccbin "${CUDA_HOST_COMPILER}")
|
||||
endif()
|
||||
elseif(WIN32 AND CMAKE_LINKER) # Workaround for VS cl.exe not being in the env. path
|
||||
get_filename_component(host_compiler_bindir ${CMAKE_LINKER} DIRECTORY)
|
||||
LIST(APPEND OPENCV_CUDA_DETECTION_NVCC_FLAGS -ccbin "${host_compiler_bindir}")
|
||||
else()
|
||||
if(CUDA_HOST_COMPILER)
|
||||
message(STATUS "CUDA: CUDA_HOST_COMPILER='${CUDA_HOST_COMPILER}' is not valid, autodetection may not work. Specify OPENCV_CUDA_DETECTION_NVCC_FLAGS with -ccbin option for fix that")
|
||||
endif()
|
||||
endif()
|
||||
|
||||
macro(ocv_filter_available_architecture result_list)
|
||||
set(__cache_key_check "${ARGN} : ${CUDA_NVCC_EXECUTABLE} ${OPENCV_CUDA_DETECTION_NVCC_FLAGS}")
|
||||
if(DEFINED OPENCV_CACHE_CUDA_SUPPORTED_CC AND OPENCV_CACHE_CUDA_SUPPORTED_CC_check STREQUAL __cache_key_check)
|
||||
set(${result_list} "${OPENCV_CACHE_CUDA_SUPPORTED_CC}")
|
||||
else()
|
||||
set(CC_LIST ${ARGN})
|
||||
foreach(target_arch ${CC_LIST})
|
||||
string(REPLACE "." "" target_arch_short "${target_arch}")
|
||||
set(NVCC_OPTION "-gencode;arch=compute_${target_arch_short},code=sm_${target_arch_short}")
|
||||
set(_cmd "${CUDA_NVCC_EXECUTABLE}" ${OPENCV_CUDA_DETECTION_NVCC_FLAGS} ${NVCC_OPTION} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" --compile)
|
||||
execute_process(
|
||||
COMMAND ${_cmd}
|
||||
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
|
||||
RESULT_VARIABLE _nvcc_res
|
||||
OUTPUT_VARIABLE _nvcc_out
|
||||
ERROR_VARIABLE _nvcc_err
|
||||
#ERROR_QUIET
|
||||
OUTPUT_STRIP_TRAILING_WHITESPACE
|
||||
)
|
||||
if(OPENCV_CMAKE_CUDA_DEBUG)
|
||||
message(WARNING "COMMAND: ${_cmd}")
|
||||
message(STATUS "Result: ${_nvcc_res}")
|
||||
message(STATUS "Out: ${_nvcc_out}")
|
||||
message(STATUS "Err: ${_nvcc_err}")
|
||||
endif()
|
||||
if(_nvcc_res EQUAL 0)
|
||||
LIST(APPEND ${result_list} "${target_arch}")
|
||||
endif()
|
||||
endforeach()
|
||||
string(STRIP "${${result_list}}" ${result_list})
|
||||
if(" ${${result_list}}" STREQUAL " ")
|
||||
message(WARNING "CUDA: Autodetection arch list is empty. Please enable OPENCV_CMAKE_CUDA_DEBUG=1 and check/specify OPENCV_CUDA_DETECTION_NVCC_FLAGS variable")
|
||||
endif()
|
||||
|
||||
# cache detected values
|
||||
set(OPENCV_CACHE_CUDA_SUPPORTED_CC ${${result_list}} CACHE INTERNAL "")
|
||||
set(OPENCV_CACHE_CUDA_SUPPORTED_CC_check "${__cache_key_check}" CACHE INTERNAL "")
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_detect_native_cuda_arch status output)
|
||||
set(OPENCV_CUDA_DETECT_ARCHS_COMMAND "${CUDA_NVCC_EXECUTABLE}" ${OPENCV_CUDA_DETECTION_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run")
|
||||
set(__cache_key_check "${OPENCV_CUDA_DETECT_ARCHS_COMMAND}")
|
||||
if(DEFINED OPENCV_CACHE_CUDA_ACTIVE_CC AND OPENCV_CACHE_CUDA_ACTIVE_CC_check STREQUAL __cache_key_check)
|
||||
set(${output} "${OPENCV_CACHE_CUDA_ACTIVE_CC}")
|
||||
set(${status} 0)
|
||||
else()
|
||||
execute_process(
|
||||
COMMAND ${OPENCV_CUDA_DETECT_ARCHS_COMMAND}
|
||||
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
|
||||
RESULT_VARIABLE ${status}
|
||||
OUTPUT_VARIABLE _nvcc_out
|
||||
ERROR_VARIABLE _nvcc_err
|
||||
ERROR_QUIET
|
||||
OUTPUT_STRIP_TRAILING_WHITESPACE
|
||||
)
|
||||
if(OPENCV_CMAKE_CUDA_DEBUG)
|
||||
message(WARNING "COMMAND: ${OPENCV_CUDA_DETECT_ARCHS_COMMAND}")
|
||||
message(STATUS "Result: ${${status}}")
|
||||
message(STATUS "Out: ${_nvcc_out}")
|
||||
message(STATUS "Err: ${_nvcc_err}")
|
||||
endif()
|
||||
string(REGEX REPLACE ".*\n" "" ${output} "${_nvcc_out}") #Strip leading warning messages, if any
|
||||
|
||||
if(${status} EQUAL 0)
|
||||
# cache detected values
|
||||
set(OPENCV_CACHE_CUDA_ACTIVE_CC ${${output}} CACHE INTERNAL "")
|
||||
set(OPENCV_CACHE_CUDA_ACTIVE_CC_check "${__cache_key_check}" CACHE INTERNAL "")
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
set(__cuda_arch_ptx ${CUDA_ARCH_PTX})
|
||||
if(CUDA_GENERATION STREQUAL "Fermi")
|
||||
set(__cuda_arch_bin ${_arch_fermi})
|
||||
elseif(CUDA_GENERATION STREQUAL "Kepler")
|
||||
set(__cuda_arch_bin ${_arch_kepler})
|
||||
elseif(CUDA_GENERATION STREQUAL "Maxwell")
|
||||
set(__cuda_arch_bin ${_arch_maxwell})
|
||||
elseif(CUDA_GENERATION STREQUAL "Pascal")
|
||||
set(__cuda_arch_bin ${_arch_pascal})
|
||||
elseif(CUDA_GENERATION STREQUAL "Volta")
|
||||
set(__cuda_arch_bin ${_arch_volta})
|
||||
elseif(CUDA_GENERATION STREQUAL "Turing")
|
||||
set(__cuda_arch_bin ${_arch_turing})
|
||||
elseif(CUDA_GENERATION STREQUAL "Ampere")
|
||||
set(__cuda_arch_bin ${_arch_ampere})
|
||||
elseif(CUDA_GENERATION STREQUAL "Lovelace")
|
||||
set(__cuda_arch_bin ${_arch_lovelace})
|
||||
elseif(CUDA_GENERATION STREQUAL "Hopper")
|
||||
set(__cuda_arch_bin ${_arch_hopper})
|
||||
elseif(CUDA_GENERATION STREQUAL "Auto")
|
||||
ocv_detect_native_cuda_arch(_nvcc_res _nvcc_out)
|
||||
if(NOT _nvcc_res EQUAL 0)
|
||||
message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
|
||||
else()
|
||||
string(REGEX MATCHALL "[0-9]+\\.[0-9]" __cuda_arch_bin "${_nvcc_out}")
|
||||
endif()
|
||||
elseif(CUDA_ARCH_BIN)
|
||||
message(STATUS "CUDA: Using CUDA_ARCH_BIN=${CUDA_ARCH_BIN}")
|
||||
set(__cuda_arch_bin ${CUDA_ARCH_BIN})
|
||||
endif()
|
||||
|
||||
if(NOT DEFINED __cuda_arch_bin AND NOT DEFINED __cuda_arch_ptx)
|
||||
if(ARM)
|
||||
set(__cuda_arch_bin "3.2")
|
||||
set(__cuda_arch_ptx "")
|
||||
elseif(AARCH64)
|
||||
if(NOT CMAKE_CROSSCOMPILING)
|
||||
ocv_detect_native_cuda_arch(_nvcc_res _nvcc_out)
|
||||
else()
|
||||
set(_nvcc_res -1) # emulate error, see below
|
||||
endif()
|
||||
if(NOT _nvcc_res EQUAL 0)
|
||||
message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
|
||||
# TX1 (5.3) TX2 (6.2) Xavier (7.2) V100 (7.0) Orin (8.7)
|
||||
ocv_filter_available_architecture(__cuda_arch_bin
|
||||
5.3
|
||||
6.2
|
||||
7.2
|
||||
7.0
|
||||
8.7
|
||||
)
|
||||
else()
|
||||
set(__cuda_arch_bin "${_nvcc_out}")
|
||||
endif()
|
||||
set(__cuda_arch_ptx "")
|
||||
else()
|
||||
ocv_filter_available_architecture(__cuda_arch_bin
|
||||
${_arch_fermi}
|
||||
${_arch_kepler}
|
||||
${_arch_maxwell}
|
||||
${_arch_pascal}
|
||||
${_arch_volta}
|
||||
${_arch_turing}
|
||||
${_arch_ampere}
|
||||
${_arch_lovelace}
|
||||
${_arch_hopper}
|
||||
)
|
||||
list(GET __cuda_arch_bin -1 __cuda_arch_ptx)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html)")
|
||||
set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html)")
|
||||
|
||||
string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
|
||||
string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
|
||||
|
||||
# Check if user specified 1.0/2.1 compute capability: we don't support it
|
||||
macro(ocv_wipeout_deprecated_cc target_cc)
|
||||
if(" ${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}" MATCHES " ${target_cc}")
|
||||
message(SEND_ERROR "CUDA: ${target_cc} compute capability is not supported - exclude it from ARCH/PTX list and re-run CMake")
|
||||
endif()
|
||||
endmacro()
|
||||
ocv_wipeout_deprecated_cc("1.0")
|
||||
ocv_wipeout_deprecated_cc("2.1")
|
||||
|
||||
# NVCC flags to be set
|
||||
set(NVCC_FLAGS_EXTRA "")
|
||||
|
||||
# These vars will be passed into the templates
|
||||
set(OPENCV_CUDA_ARCH_BIN "")
|
||||
set(OPENCV_CUDA_ARCH_PTX "")
|
||||
set(OPENCV_CUDA_ARCH_FEATURES "")
|
||||
|
||||
# Tell NVCC to add binaries for the specified GPUs
|
||||
string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${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()
|
||||
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -D_FORCE_INLINES)
|
||||
|
||||
# Tell NVCC to add PTX intermediate code for the specified architectures
|
||||
string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${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")
|
||||
endif()
|
||||
|
||||
# Tell NVCC the maximum number of threads to be used to execute the compilation steps in parallel
|
||||
# (option --threads was introduced in version 11.2)
|
||||
if(NOT CUDA_VERSION VERSION_LESS "11.2")
|
||||
if(CMAKE_GENERATOR MATCHES "Visual Studio" AND NOT $ENV{CMAKE_BUILD_PARALLEL_LEVEL} STREQUAL "")
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "--threads=$ENV{CMAKE_BUILD_PARALLEL_LEVEL}")
|
||||
endif()
|
||||
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 ${CUDA_NVCC_FLAGS} --use_fast_math)
|
||||
endif()
|
||||
|
||||
OCV_OPTION(CUDA_ENABLE_DELAYLOAD "Enable delayed loading of CUDA DLLs" OFF VISIBLE_IF MSVC)
|
||||
|
||||
mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
|
||||
|
||||
macro(ocv_cuda_filter_options)
|
||||
foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
|
||||
set(${var}_backup_in_cuda_compile_ "${${var}}")
|
||||
|
||||
if (CV_CLANG)
|
||||
# we remove -Winconsistent-missing-override and -Qunused-arguments
|
||||
# just in case we are compiling CUDA with gcc but OpenCV with clang
|
||||
string(REPLACE "-Winconsistent-missing-override" "" ${var} "${${var}}")
|
||||
string(REPLACE "-Qunused-arguments" "" ${var} "${${var}}")
|
||||
endif()
|
||||
|
||||
# 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-sign-promo as it generates warnings under linux
|
||||
string(REPLACE "-Wno-sign-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}}")
|
||||
|
||||
string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
|
||||
|
||||
# cc1: warning: command line option '-Wsuggest-override' is valid for C++/ObjC++ but not for C
|
||||
string(REPLACE "-Wsuggest-override" "" ${var} "${${var}}")
|
||||
|
||||
# issue: #11552 (from OpenCVCompilerOptions.cmake)
|
||||
string(REGEX REPLACE "-Wimplicit-fallthrough(=[0-9]+)? " "" ${var} "${${var}}")
|
||||
|
||||
# removal of custom specified options
|
||||
if(OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS)
|
||||
foreach(__flag ${OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS})
|
||||
string(REPLACE "${__flag}" "" ${var} "${${var}}")
|
||||
endforeach()
|
||||
endif()
|
||||
endforeach()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_cuda_compile VAR)
|
||||
ocv_cuda_filter_options()
|
||||
|
||||
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)
|
||||
if(NOT " ${CMAKE_CXX_FLAGS} ${CMAKE_CXX_FLAGS_RELEASE} ${CMAKE_CXX_FLAGS_DEBUG} ${CUDA_NVCC_FLAGS}" MATCHES "-std=")
|
||||
if(CUDA_VERSION VERSION_LESS "11.0")
|
||||
list(APPEND CUDA_NVCC_FLAGS "--std=c++11")
|
||||
else()
|
||||
list(APPEND CUDA_NVCC_FLAGS "--std=c++14")
|
||||
endif()
|
||||
endif()
|
||||
endif()
|
||||
if(APPLE)
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
|
||||
endif()
|
||||
|
||||
if(WIN32 AND NOT (CUDA_VERSION VERSION_LESS "11.2"))
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcudafe --display_error_number --diag-suppress 1394,1388)
|
||||
endif()
|
||||
|
||||
if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xlinker --unresolved-symbols=ignore-in-shared-libs)
|
||||
endif()
|
||||
|
||||
# disabled because of multiple warnings during building nvcc auto generated files
|
||||
if(CV_GCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "4.6.0")
|
||||
ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
|
||||
endif()
|
||||
|
||||
CUDA_COMPILE(${VAR} ${ARGN})
|
||||
|
||||
foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
|
||||
set(${var} "${${var}_backup_in_cuda_compile_}")
|
||||
unset(${var}_backup_in_cuda_compile_)
|
||||
endforeach()
|
||||
endmacro()
|
||||
else()
|
||||
if(NOT CUDA_FOUND)
|
||||
unset(CUDA_ARCH_BIN CACHE)
|
||||
unset(CUDA_ARCH_PTX CACHE)
|
||||
return()
|
||||
endif()
|
||||
|
||||
unset(CUDA_nvcuvenc_LIBRARY CACHE)
|
||||
set(HAVE_CUDA 1)
|
||||
if(NOT CUDA_VERSION VERSION_LESS 11.0)
|
||||
# CUDA 11.0 removes nppicom
|
||||
ocv_list_filterout(CUDA_nppi_LIBRARY "nppicom")
|
||||
ocv_list_filterout(CUDA_npp_LIBRARY "nppicom")
|
||||
endif()
|
||||
|
||||
if(WITH_CUFFT)
|
||||
set(HAVE_CUFFT 1)
|
||||
endif()
|
||||
|
||||
if(WITH_CUBLAS)
|
||||
set(HAVE_CUBLAS 1)
|
||||
endif()
|
||||
|
||||
if(WITH_CUDNN)
|
||||
set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
|
||||
find_host_package(CUDNN "${MIN_VER_CUDNN}")
|
||||
list(REMOVE_AT CMAKE_MODULE_PATH 0)
|
||||
|
||||
if(CUDNN_FOUND)
|
||||
set(HAVE_CUDNN 1)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
include(cmake/OpenCVDetectCUDAUtils.cmake)
|
||||
|
||||
if(WITH_NVCUVID OR WITH_NVCUVENC)
|
||||
set(cuda_toolkit_dirs "${CUDA_TOOLKIT_TARGET_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}")
|
||||
ocv_check_for_nvidia_video_codec_sdk("${cuda_toolkit_dirs}")
|
||||
endif()
|
||||
|
||||
message(STATUS "CUDA detected: " ${CUDA_VERSION})
|
||||
|
||||
ocv_set_cuda_detection_nvcc_flags(CUDA_HOST_COMPILER)
|
||||
ocv_set_cuda_arch_bin_and_ptx(${CUDA_NVCC_EXECUTABLE})
|
||||
|
||||
# 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()
|
||||
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -D_FORCE_INLINES)
|
||||
|
||||
# Tell NVCC to add PTX intermediate code for the specified architectures
|
||||
string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${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")
|
||||
endif()
|
||||
|
||||
ocv_set_nvcc_threads_for_vs()
|
||||
|
||||
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()
|
||||
|
||||
OCV_OPTION(CUDA_ENABLE_DELAYLOAD "Enable delayed loading of CUDA DLLs" OFF VISIBLE_IF MSVC)
|
||||
|
||||
mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
|
||||
|
||||
macro(ocv_check_windows_crt_linkage)
|
||||
# The new MSVC runtime abstraction is only useable if CUDA is a first class language
|
||||
if(WIN32 AND POLICY CMP0091)
|
||||
cmake_policy(GET CMP0091 MSVC_RUNTIME_SET_BY_ABSTRACTION)
|
||||
if(MSVC_RUNTIME_SET_BY_ABSTRACTION STREQUAL "NEW")
|
||||
if(NOT BUILD_SHARED_LIBS AND BUILD_WITH_STATIC_CRT)
|
||||
set(CMAKE_CXX_FLAGS_RELEASE ${CMAKE_CXX_FLAGS_RELEASE} " /MT")
|
||||
set(CMAKE_CXX_FLAGS_DEBUG ${CMAKE_CXX_FLAGS_DEBUG} " /MTd")
|
||||
else()
|
||||
set(CMAKE_CXX_FLAGS_RELEASE ${CMAKE_CXX_FLAGS_RELEASE} " /MD")
|
||||
set(CMAKE_CXX_FLAGS_DEBUG ${CMAKE_CXX_FLAGS_DEBUG} " /MDd")
|
||||
endif()
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_cuda_compile VAR)
|
||||
ocv_cuda_filter_options()
|
||||
ocv_check_windows_crt_linkage()
|
||||
ocv_nvcc_flags()
|
||||
|
||||
if(UNIX OR APPLE)
|
||||
if(NOT " ${CMAKE_CXX_FLAGS} ${CMAKE_CXX_FLAGS_RELEASE} ${CMAKE_CXX_FLAGS_DEBUG} ${CUDA_NVCC_FLAGS}" MATCHES "-std=")
|
||||
if(CUDA_VERSION VERSION_LESS "11.0")
|
||||
list(APPEND CUDA_NVCC_FLAGS "--std=c++11")
|
||||
else()
|
||||
list(APPEND CUDA_NVCC_FLAGS "--std=c++14")
|
||||
endif()
|
||||
endif()
|
||||
endif()
|
||||
|
||||
CUDA_COMPILE(${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()
|
||||
|
||||
if(HAVE_CUDA)
|
||||
set(CUDA_LIBS_PATH "")
|
||||
foreach(p ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
|
||||
@ -525,36 +225,13 @@ if(HAVE_CUDA)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
|
||||
# ----------------------------------------------------------------------------
|
||||
# Add CUDA libraries (needed for apps/tools, samples)
|
||||
# ----------------------------------------------------------------------------
|
||||
if(HAVE_CUDA)
|
||||
# details: https://github.com/NVIDIA/nvidia-docker/issues/775
|
||||
if(" ${CUDA_CUDA_LIBRARY}" MATCHES "/stubs/libcuda.so" AND NOT OPENCV_SKIP_CUDA_STUB_WORKAROUND)
|
||||
set(CUDA_STUB_ENABLED_LINK_WORKAROUND 1)
|
||||
if(EXISTS "${CUDA_CUDA_LIBRARY}" AND NOT OPENCV_SKIP_CUDA_STUB_WORKAROUND_RPATH_LINK)
|
||||
set(CUDA_STUB_TARGET_PATH "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/")
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink "${CUDA_CUDA_LIBRARY}" "${CUDA_STUB_TARGET_PATH}/libcuda.so.1"
|
||||
RESULT_VARIABLE CUDA_STUB_SYMLINK_RESULT)
|
||||
if(NOT CUDA_STUB_SYMLINK_RESULT EQUAL 0)
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different "${CUDA_CUDA_LIBRARY}" "${CUDA_STUB_TARGET_PATH}/libcuda.so.1"
|
||||
RESULT_VARIABLE CUDA_STUB_COPY_RESULT)
|
||||
if(NOT CUDA_STUB_COPY_RESULT EQUAL 0)
|
||||
set(CUDA_STUB_ENABLED_LINK_WORKAROUND 0)
|
||||
endif()
|
||||
endif()
|
||||
if(CUDA_STUB_ENABLED_LINK_WORKAROUND)
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-rpath-link,\"${CUDA_STUB_TARGET_PATH}\"")
|
||||
endif()
|
||||
else()
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--allow-shlib-undefined")
|
||||
endif()
|
||||
if(NOT CUDA_STUB_ENABLED_LINK_WORKAROUND)
|
||||
message(WARNING "CUDA: workaround for stubs/libcuda.so.1 is not applied")
|
||||
endif()
|
||||
endif()
|
||||
ocv_apply_cuda_stub_workaround("${CUDA_CUDA_LIBRARY}")
|
||||
ocv_check_cuda_delayed_load("${CUDA_TOOLKIT_ROOT_DIR}")
|
||||
|
||||
# ----------------------------------------------------------------------------
|
||||
# Add CUDA libraries (needed for apps/tools, samples)
|
||||
# ----------------------------------------------------------------------------
|
||||
set(OPENCV_LINKER_LIBS ${OPENCV_LINKER_LIBS} ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
|
||||
if(HAVE_CUBLAS)
|
||||
set(OPENCV_LINKER_LIBS ${OPENCV_LINKER_LIBS} ${CUDA_cublas_LIBRARY})
|
||||
@ -572,19 +249,4 @@ if(HAVE_CUDA)
|
||||
set(OPENCV_LINKER_LIBS ${OPENCV_LINKER_LIBS} ${CMAKE_LIBRARY_PATH_FLAG}${p})
|
||||
endif()
|
||||
endforeach()
|
||||
|
||||
if(MSVC AND CUDA_ENABLE_DELAYLOAD)
|
||||
set(DELAYFLAGS "delayimp.lib")
|
||||
file(GLOB CUDA_DLLS "${CUDA_TOOLKIT_ROOT_DIR}/bin/*.dll")
|
||||
foreach(d ${CUDA_DLLS})
|
||||
cmake_path(GET "d" FILENAME DLL_NAME)
|
||||
if(NOT ${DLL_NAME} MATCHES "cudart")
|
||||
set(DELAYFLAGS "${DELAYFLAGS} /DELAYLOAD:${DLL_NAME}")
|
||||
endif()
|
||||
endforeach()
|
||||
set(DELAYFLAGS "${DELAYFLAGS} /DELAYLOAD:nvcuda.dll /DELAYLOAD:nvml.dll /IGNORE:4199")
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${DELAYFLAGS}")
|
||||
set(CMAKE_MODULE_LINKER_FLAGS "${CMAKE_MODULE_LINKER_FLAGS} ${DELAYFLAGS}")
|
||||
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${DELAYFLAGS}")
|
||||
endif()
|
||||
endif()
|
||||
|
||||
154
cmake/OpenCVDetectCUDALanguage.cmake
Normal file
154
cmake/OpenCVDetectCUDALanguage.cmake
Normal file
@ -0,0 +1,154 @@
|
||||
#######################
|
||||
# Previously in FindCUDA and still required for FindCUDNN
|
||||
macro(FIND_CUDA_HELPER_LIBS _name)
|
||||
if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
|
||||
set(_cuda_cross_arm_lib_dir "lib/stubs")
|
||||
endif()
|
||||
find_library(CUDA_${_name}_LIBRARY ${_name}
|
||||
NAMES ${_name}
|
||||
PATHS "${CUDAToolkit_LIBRARY_ROOT}"
|
||||
PATH_SUFFIXES "lib/x64" "lib64" ${_cuda_cross_arm_lib_dir} "lib/Win32" "lib"
|
||||
DOC "\"${_name}\" library"
|
||||
)
|
||||
mark_as_advanced(CUDA_${_name}_LIBRARY)
|
||||
endmacro()
|
||||
#######################
|
||||
include(cmake/OpenCVDetectCUDAUtils.cmake)
|
||||
|
||||
if((WIN32 AND NOT MSVC) OR OPENCV_CMAKE_FORCE_CUDA)
|
||||
message(STATUS "CUDA: Compilation is disabled (due to only Visual Studio compiler supported on your platform).")
|
||||
return()
|
||||
endif()
|
||||
|
||||
if((NOT UNIX AND CV_CLANG) OR OPENCV_CMAKE_FORCE_CUDA)
|
||||
message(STATUS "CUDA: Compilation is disabled (due to Clang unsupported on your platform).")
|
||||
return()
|
||||
endif()
|
||||
|
||||
#set(OPENCV_CMAKE_CUDA_DEBUG 1)
|
||||
|
||||
find_package(CUDAToolkit)
|
||||
if(CMAKE_CUDA_COMPILER AND CUDAToolkit_FOUND)
|
||||
set(CUDA_FOUND TRUE)
|
||||
set(CUDA_TOOLKIT_INCLUDE ${CUDAToolkit_INCLUDE_DIRS})
|
||||
set(CUDA_VERSION_STRING ${CUDAToolkit_VERSION})
|
||||
set(CUDA_VERSION ${CUDAToolkit_VERSION})
|
||||
if(NOT CUDA_VERSION VERSION_LESS 11.0)
|
||||
set(CMAKE_CUDA_STANDARD 14)
|
||||
else()
|
||||
set(CMAKE_CUDA_STANDARD 11)
|
||||
endif()
|
||||
if(UNIX AND NOT BUILD_SHARED_LIBS)
|
||||
set(CUDA_LIB_EXT "_static")
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if(NOT CUDA_FOUND)
|
||||
unset(CUDA_ARCH_BIN CACHE)
|
||||
unset(CUDA_ARCH_PTX CACHE)
|
||||
return()
|
||||
endif()
|
||||
|
||||
set(HAVE_CUDA 1)
|
||||
|
||||
if(WITH_CUFFT)
|
||||
set(HAVE_CUFFT 1)
|
||||
endif()
|
||||
|
||||
if(WITH_CUBLAS)
|
||||
set(HAVE_CUBLAS 1)
|
||||
endif()
|
||||
|
||||
if(WITH_CUDNN)
|
||||
set(CMAKE_MODULE_PATH "${OpenCV_SOURCE_DIR}/cmake" ${CMAKE_MODULE_PATH})
|
||||
find_host_package(CUDNN "${MIN_VER_CUDNN}")
|
||||
list(REMOVE_AT CMAKE_MODULE_PATH 0)
|
||||
|
||||
if(CUDNN_FOUND)
|
||||
set(HAVE_CUDNN 1)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if(WITH_NVCUVID OR WITH_NVCUVENC)
|
||||
ocv_check_for_nvidia_video_codec_sdk("${CUDAToolkit_LIBRARY_ROOT}")
|
||||
endif()
|
||||
|
||||
ocv_check_for_cmake_cuda_architectures()
|
||||
ocv_set_cuda_detection_nvcc_flags(CMAKE_CUDA_HOST_COMPILER)
|
||||
ocv_set_cuda_arch_bin_and_ptx(${CUDAToolkit_NVCC_EXECUTABLE})
|
||||
|
||||
# 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(CMAKE_CUDA_ARCHITECTURES ${CMAKE_CUDA_ARCHITECTURES} ${CMAKE_MATCH_2}-virtual;${CMAKE_MATCH_1}-real;)
|
||||
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(CMAKE_CUDA_ARCHITECTURES ${CMAKE_CUDA_ARCHITECTURES} ${ARCH}-real;)
|
||||
set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
|
||||
set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
|
||||
endif()
|
||||
endforeach()
|
||||
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -D_FORCE_INLINES)
|
||||
|
||||
# Tell NVCC to add PTX intermediate code for the specified architectures
|
||||
string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
|
||||
foreach(ARCH IN LISTS ARCH_LIST)
|
||||
set(CMAKE_CUDA_ARCHITECTURES ${CMAKE_CUDA_ARCHITECTURES} ${ARCH}-virtual;)
|
||||
set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
|
||||
set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
|
||||
endforeach()
|
||||
|
||||
ocv_set_nvcc_threads_for_vs()
|
||||
|
||||
# These vars will be processed in other scripts
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
|
||||
set(OpenCV_CUDA_CC "${CMAKE_CUDA_ARCHITECTURES}")
|
||||
|
||||
if(ANDROID)
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
|
||||
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 ${CUDA_NVCC_FLAGS} --use_fast_math)
|
||||
endif()
|
||||
|
||||
OCV_OPTION(CUDA_ENABLE_DELAYLOAD "Enable delayed loading of CUDA DLLs" OFF VISIBLE_IF MSVC)
|
||||
|
||||
mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
|
||||
|
||||
macro(ocv_cuda_unfilter_options)
|
||||
foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
|
||||
set(${var} "${${var}_backup_in_cuda_compile_}")
|
||||
unset(${var}_backup_in_cuda_compile_)
|
||||
endforeach()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_cuda_compile_flags)
|
||||
ocv_cuda_filter_options()
|
||||
ocv_nvcc_flags()
|
||||
set(CMAKE_CXX_FLAGS_CUDA ${CMAKE_CXX_FLAGS})
|
||||
set(CMAKE_CXX_FLAGS_RELEASE_CUDA ${CMAKE_CXX_FLAGS_RELEASE})
|
||||
set(CMAKE_CXX_FLAGS_DEBUG_CUDA ${CMAKE_CXX_FLAGS_DEBUG})
|
||||
ocv_cuda_unfilter_options()
|
||||
endmacro()
|
||||
|
||||
if(HAVE_CUDA)
|
||||
ocv_apply_cuda_stub_workaround("${CUDA_cuda_driver_LIBRARY}")
|
||||
ocv_check_cuda_delayed_load("${cuda_toolkit_root_dir}")
|
||||
endif()
|
||||
442
cmake/OpenCVDetectCUDAUtils.cmake
Normal file
442
cmake/OpenCVDetectCUDAUtils.cmake
Normal file
@ -0,0 +1,442 @@
|
||||
macro(ocv_check_for_nvidia_video_codec_sdk cuda_toolkit_dirs)
|
||||
macro(ocv_cuda_SEARCH_NVCUVID_HEADER _filename _result)
|
||||
# place header file under CUDAToolkit_LIBRARY_ROOT
|
||||
find_path(_header_result
|
||||
${_filename}
|
||||
PATHS ${cuda_toolkit_dirs}
|
||||
PATH_SUFFIXES include
|
||||
NO_DEFAULT_PATH
|
||||
)
|
||||
if("x${_header_result}" STREQUAL "x_header_result-NOTFOUND")
|
||||
set(${_result} 0)
|
||||
else()
|
||||
set(${_result} 1)
|
||||
endif()
|
||||
unset(_header_result CACHE)
|
||||
endmacro()
|
||||
if(WITH_NVCUVID)
|
||||
ocv_cuda_SEARCH_NVCUVID_HEADER("nvcuvid.h" HAVE_NVCUVID_HEADER)
|
||||
# make sure to have both header and library before enabling
|
||||
if(${HAVE_NVCUVID_HEADER})
|
||||
find_cuda_helper_libs(nvcuvid)
|
||||
if(CUDA_nvcuvid_LIBRARY)
|
||||
set(HAVE_NVCUVID 1)
|
||||
message(STATUS "Found NVCUVID: ${CUDA_nvcuvid_LIBRARY}")
|
||||
else()
|
||||
if(WIN32)
|
||||
message(STATUS "NVCUVID: Library not found, WITH_NVCUVID requires Nvidia decoding library nvcuvid.lib to either be inside ${cuda_toolkit_dirs}/lib or its location manually set with CUDA_nvcuvid_LIBRARY, i.e. CUDA_nvcuvid_LIBRARY=${cuda_toolkit_dirs}/lib/nvcuvid.lib")
|
||||
else()
|
||||
message(STATUS "NVCUVID: Library not found, WITH_NVCUVID requires the Nvidia decoding shared library nvcuvid.so from the driver installation or the location of the stub library to be manually set with CUDA_nvcuvid_LIBRARY i.e. CUDA_nvcuvid_LIBRARY=/home/user/Video_Codec_SDK_X.X.X/Lib/linux/stubs/x86_64/nvcuvid.so")
|
||||
endif()
|
||||
endif()
|
||||
else()
|
||||
message(STATUS "NVCUVID: Header not found, WITH_NVCUVID requires Nvidia decoding library header ${cuda_toolkit_dirs}/include/nvcuvid.h")
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if(WITH_NVCUVENC)
|
||||
ocv_cuda_SEARCH_NVCUVID_HEADER("nvEncodeAPI.h" HAVE_NVCUVENC_HEADER)
|
||||
if(${HAVE_NVCUVENC_HEADER})
|
||||
if(WIN32)
|
||||
find_cuda_helper_libs(nvencodeapi)
|
||||
else()
|
||||
find_cuda_helper_libs(nvidia-encode)
|
||||
endif()
|
||||
if(CUDA_nvencodeapi_LIBRARY OR CUDA_nvidia-encode_LIBRARY)
|
||||
set(HAVE_NVCUVENC 1)
|
||||
message(STATUS "Found NVCUVENC: ${CUDA_nvencodeapi_LIBRARY} ${CUDA_nvidia-encode_LIBRARY}")
|
||||
else()
|
||||
if(WIN32)
|
||||
message(STATUS "NVCUVENC: Library not found, WITH_NVCUVENC requires Nvidia encoding library nvencodeapi.lib to either be inside ${cuda_toolkit_dirs}/lib or its location manually set with CUDA_nvencodeapi_LIBRARY, i.e. CUDA_nvencodeapi_LIBRARY=${cuda_toolkit_dirs}/lib/nvencodeapi.lib")
|
||||
else()
|
||||
message(STATUS "NVCUVENC: Library not found, WITH_NVCUVENC requires the Nvidia encoding shared library libnvidia-encode.so from the driver installation or the location of the stub library to be manually set with CUDA_nvidia-encode_LIBRARY i.e. CUDA_nvidia-encode_LIBRARY=/home/user/Video_Codec_SDK_X.X.X/Lib/linux/stubs/x86_64/libnvidia-encode.so")
|
||||
endif()
|
||||
endif()
|
||||
else()
|
||||
message(STATUS "NVCUVENC: Header not found, WITH_NVCUVENC requires Nvidia encoding library header ${cuda_toolkit_dirs}/include/nvEncodeAPI.h")
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
# Use CMAKE_CUDA_ARCHITECTURES if provided: order of preference CMAKE_CUDA_ARCHITECTURES > CUDA_GENERATION > CUDA_ARCH_BIN and/or CUDA_ARCH_PTX
|
||||
function(ocv_check_for_cmake_cuda_architectures)
|
||||
if(NOT CMAKE_CUDA_ARCHITECTURES)
|
||||
return()
|
||||
endif()
|
||||
if(CMAKE_CUDA_ARCHITECTURES STREQUAL "all" OR CMAKE_CUDA_ARCHITECTURES STREQUAL "all-major" OR CMAKE_CUDA_ARCHITECTURES STREQUAL "native")
|
||||
message(WARNING "CUDA: CMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}, special values all, all-major and native are not supported by OpenCV, specify only CUDA real and/or virtual architectures or use combinations of CUDA_ARCH_BIN and CUDA_ARCH_PTX or specify the CUDA_GENERATION where -DCUDA_GENERATION=Auto is equivalent to native!")
|
||||
return()
|
||||
endif()
|
||||
set(internal_ptx "")
|
||||
set(internal_bin "")
|
||||
foreach(ARCH IN LISTS CMAKE_CUDA_ARCHITECTURES)
|
||||
if(ARCH MATCHES "([0-9]+)\-real")
|
||||
set(internal_bin ${internal_bin} ${CMAKE_MATCH_1};)
|
||||
elseif(ARCH MATCHES "([0-9]+)\-virtual")
|
||||
set(internal_ptx ${internal_ptx} ${CMAKE_MATCH_1};)
|
||||
elseif(ARCH MATCHES "([0-9]+)")
|
||||
set(internal_bin ${internal_bin} ${CMAKE_MATCH_1};)
|
||||
set(internal_ptx ${internal_ptx} ${CMAKE_MATCH_1};)
|
||||
endif()
|
||||
endforeach()
|
||||
if(internal_bin OR internal_ptx)
|
||||
unset(CUDA_ARCH_BIN CACHE)
|
||||
unset(CUDA_ARCH_PTX CACHE)
|
||||
endif()
|
||||
if(internal_ptx)
|
||||
set(CUDA_ARCH_PTX ${internal_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html)")
|
||||
endif()
|
||||
if(internal_bin)
|
||||
set(CUDA_ARCH_BIN ${internal_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html)")
|
||||
endif()
|
||||
set(CMAKE_CUDA_ARCHITECTURES "" PARENT)
|
||||
unset(CUDA_GENERATION CACHE)
|
||||
endfunction()
|
||||
|
||||
macro(ocv_initialize_nvidia_device_generations)
|
||||
OCV_OPTION(CUDA_ENABLE_DEPRECATED_GENERATION "Enable deprecated generations in the list" OFF)
|
||||
set(_generations "Maxwell" "Pascal" "Volta" "Turing" "Ampere" "Lovelace" "Hopper")
|
||||
if(CUDA_ENABLE_DEPRECATED_GENERATION)
|
||||
set(_generations "Fermi" "${_generations}")
|
||||
set(_generations "Kepler" "${_generations}")
|
||||
endif()
|
||||
set(_arch_fermi "2.0")
|
||||
set(_arch_kepler "3.0;3.5;3.7")
|
||||
set(_arch_maxwell "5.0;5.2")
|
||||
set(_arch_pascal "6.0;6.1")
|
||||
set(_arch_volta "7.0")
|
||||
set(_arch_turing "7.5")
|
||||
set(_arch_ampere "8.0;8.6")
|
||||
set(_arch_lovelace "8.9")
|
||||
set(_arch_hopper "9.0")
|
||||
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 (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html).")
|
||||
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: ${_generations} Generations are supported.")
|
||||
endif()
|
||||
unset(CUDA_ARCH_BIN CACHE)
|
||||
unset(CUDA_ARCH_PTX CACHE)
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_set_cuda_detection_nvcc_flags cuda_host_compiler_var)
|
||||
if(OPENCV_CUDA_DETECTION_NVCC_FLAGS MATCHES "-ccbin")
|
||||
# already specified by user
|
||||
elseif(${cuda_host_compiler_var} AND EXISTS "${${cuda_host_compiler_var}}")
|
||||
get_filename_component(c_compiler_realpath "${CMAKE_C_COMPILER}" REALPATH)
|
||||
# C compiler doesn't work with --run option, forcing C++ compiler instead
|
||||
if(${cuda_host_compiler_var} STREQUAL c_compiler_realpath OR ${cuda_host_compiler_var} STREQUAL CMAKE_C_COMPILER)
|
||||
if(DEFINED CMAKE_CXX_COMPILER)
|
||||
get_filename_component(cxx_compiler_realpath "${CMAKE_CXX_COMPILER}" REALPATH)
|
||||
LIST(APPEND OPENCV_CUDA_DETECTION_NVCC_FLAGS -ccbin "${cxx_compiler_realpath}")
|
||||
else()
|
||||
message(STATUS "CUDA: CMAKE_CXX_COMPILER is not available. You may need to specify ${cuda_host_compiler_var}.")
|
||||
endif()
|
||||
else()
|
||||
LIST(APPEND OPENCV_CUDA_DETECTION_NVCC_FLAGS -ccbin "${${cuda_host_compiler_var}}")
|
||||
endif()
|
||||
elseif(WIN32 AND CMAKE_LINKER) # Workaround for VS cl.exe not being in the env. path
|
||||
get_filename_component(host_compiler_bindir ${CMAKE_LINKER} DIRECTORY)
|
||||
LIST(APPEND OPENCV_CUDA_DETECTION_NVCC_FLAGS -ccbin "${host_compiler_bindir}")
|
||||
else()
|
||||
if(${cuda_host_compiler_var})
|
||||
message(STATUS "CUDA: ${cuda_host_compiler_var}='${cuda_host_compiler}' is not valid, autodetection may not work. Specify OPENCV_CUDA_DETECTION_NVCC_FLAGS with -ccbin option for fix that")
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_filter_available_architecture nvcc_executable result_list)
|
||||
set(__cache_key_check "${ARGN} : ${nvcc_executable} ${OPENCV_CUDA_DETECTION_NVCC_FLAGS}")
|
||||
if(DEFINED OPENCV_CACHE_CUDA_SUPPORTED_CC AND OPENCV_CACHE_CUDA_SUPPORTED_CC_check STREQUAL __cache_key_check)
|
||||
set(${result_list} "${OPENCV_CACHE_CUDA_SUPPORTED_CC}")
|
||||
else()
|
||||
set(CC_LIST ${ARGN})
|
||||
foreach(target_arch ${CC_LIST})
|
||||
string(REPLACE "." "" target_arch_short "${target_arch}")
|
||||
set(NVCC_OPTION "-gencode;arch=compute_${target_arch_short},code=sm_${target_arch_short}")
|
||||
set(_cmd "${nvcc_executable}" ${OPENCV_CUDA_DETECTION_NVCC_FLAGS} ${NVCC_OPTION} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" --compile)
|
||||
execute_process(
|
||||
COMMAND ${_cmd}
|
||||
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
|
||||
RESULT_VARIABLE _nvcc_res
|
||||
OUTPUT_VARIABLE _nvcc_out
|
||||
ERROR_VARIABLE _nvcc_err
|
||||
#ERROR_QUIET
|
||||
OUTPUT_STRIP_TRAILING_WHITESPACE
|
||||
)
|
||||
if(OPENCV_CMAKE_CUDA_DEBUG)
|
||||
message(WARNING "COMMAND: ${_cmd}")
|
||||
message(STATUS "Result: ${_nvcc_res}")
|
||||
message(STATUS "Out: ${_nvcc_out}")
|
||||
message(STATUS "Err: ${_nvcc_err}")
|
||||
endif()
|
||||
if(_nvcc_res EQUAL 0)
|
||||
LIST(APPEND ${result_list} "${target_arch}")
|
||||
endif()
|
||||
endforeach()
|
||||
string(STRIP "${${result_list}}" ${result_list})
|
||||
if(" ${${result_list}}" STREQUAL " ")
|
||||
message(WARNING "CUDA: Autodetection arch list is empty. Please enable OPENCV_CMAKE_CUDA_DEBUG=1 and check/specify OPENCV_CUDA_DETECTION_NVCC_FLAGS variable")
|
||||
endif()
|
||||
|
||||
# cache detected values
|
||||
set(OPENCV_CACHE_CUDA_SUPPORTED_CC ${${result_list}} CACHE INTERNAL "")
|
||||
set(OPENCV_CACHE_CUDA_SUPPORTED_CC_check "${__cache_key_check}" CACHE INTERNAL "")
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_detect_native_cuda_arch nvcc_executable status output)
|
||||
set(OPENCV_CUDA_DETECT_ARCHS_COMMAND "${nvcc_executable}" ${OPENCV_CUDA_DETECTION_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run")
|
||||
set(__cache_key_check "${OPENCV_CUDA_DETECT_ARCHS_COMMAND}")
|
||||
if(DEFINED OPENCV_CACHE_CUDA_ACTIVE_CC AND OPENCV_CACHE_CUDA_ACTIVE_CC_check STREQUAL __cache_key_check)
|
||||
set(${output} "${OPENCV_CACHE_CUDA_ACTIVE_CC}")
|
||||
set(${status} 0)
|
||||
else()
|
||||
execute_process(
|
||||
COMMAND ${OPENCV_CUDA_DETECT_ARCHS_COMMAND}
|
||||
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
|
||||
RESULT_VARIABLE ${status}
|
||||
OUTPUT_VARIABLE _nvcc_out
|
||||
ERROR_VARIABLE _nvcc_err
|
||||
ERROR_QUIET
|
||||
OUTPUT_STRIP_TRAILING_WHITESPACE
|
||||
)
|
||||
if(OPENCV_CMAKE_CUDA_DEBUG)
|
||||
message(WARNING "COMMAND: ${OPENCV_CUDA_DETECT_ARCHS_COMMAND}")
|
||||
message(STATUS "Result: ${${status}}")
|
||||
message(STATUS "Out: ${_nvcc_out}")
|
||||
message(STATUS "Err: ${_nvcc_err}")
|
||||
endif()
|
||||
string(REGEX REPLACE ".*\n" "" ${output} "${_nvcc_out}") #Strip leading warning messages, if any
|
||||
|
||||
if(${status} EQUAL 0)
|
||||
# cache detected values
|
||||
set(OPENCV_CACHE_CUDA_ACTIVE_CC ${${output}} CACHE INTERNAL "")
|
||||
set(OPENCV_CACHE_CUDA_ACTIVE_CC_check "${__cache_key_check}" CACHE INTERNAL "")
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_set_cuda_arch_bin_and_ptx nvcc_executable)
|
||||
ocv_initialize_nvidia_device_generations()
|
||||
set(__cuda_arch_ptx ${CUDA_ARCH_PTX})
|
||||
if(CUDA_GENERATION STREQUAL "Fermi")
|
||||
set(__cuda_arch_bin ${_arch_fermi})
|
||||
elseif(CUDA_GENERATION STREQUAL "Kepler")
|
||||
set(__cuda_arch_bin ${_arch_kepler})
|
||||
elseif(CUDA_GENERATION STREQUAL "Maxwell")
|
||||
set(__cuda_arch_bin ${_arch_maxwell})
|
||||
elseif(CUDA_GENERATION STREQUAL "Pascal")
|
||||
set(__cuda_arch_bin ${_arch_pascal})
|
||||
elseif(CUDA_GENERATION STREQUAL "Volta")
|
||||
set(__cuda_arch_bin ${_arch_volta})
|
||||
elseif(CUDA_GENERATION STREQUAL "Turing")
|
||||
set(__cuda_arch_bin ${_arch_turing})
|
||||
elseif(CUDA_GENERATION STREQUAL "Ampere")
|
||||
set(__cuda_arch_bin ${_arch_ampere})
|
||||
elseif(CUDA_GENERATION STREQUAL "Lovelace")
|
||||
set(__cuda_arch_bin ${_arch_lovelace})
|
||||
elseif(CUDA_GENERATION STREQUAL "Hopper")
|
||||
set(__cuda_arch_bin ${_arch_hopper})
|
||||
elseif(CUDA_GENERATION STREQUAL "Auto")
|
||||
ocv_detect_native_cuda_arch(${nvcc_executable} _nvcc_res _nvcc_out)
|
||||
if(NOT _nvcc_res EQUAL 0)
|
||||
message(STATUS "CUDA: Automatic detection of CUDA generation failed. Going to build for all known architectures")
|
||||
else()
|
||||
string(REGEX MATCHALL "[0-9]+\\.[0-9]" __cuda_arch_bin "${_nvcc_out}")
|
||||
endif()
|
||||
elseif(CUDA_ARCH_BIN)
|
||||
message(STATUS "CUDA: Using CUDA_ARCH_BIN=${CUDA_ARCH_BIN}")
|
||||
set(__cuda_arch_bin ${CUDA_ARCH_BIN})
|
||||
endif()
|
||||
|
||||
if(NOT DEFINED __cuda_arch_bin AND NOT DEFINED __cuda_arch_ptx)
|
||||
if(ARM)
|
||||
set(__cuda_arch_bin "3.2")
|
||||
set(__cuda_arch_ptx "")
|
||||
elseif(AARCH64)
|
||||
if(NOT CMAKE_CROSSCOMPILING)
|
||||
ocv_detect_native_cuda_arch(${nvcc_executable} _nvcc_res _nvcc_out)
|
||||
else()
|
||||
set(_nvcc_res -1) # emulate error, see below
|
||||
endif()
|
||||
if(NOT _nvcc_res EQUAL 0)
|
||||
message(STATUS "CUDA: Automatic detection of CUDA generation failed. Going to build for all known architectures")
|
||||
# TX1 (5.3) TX2 (6.2) Xavier (7.2) V100 (7.0) Orin (8.7)
|
||||
ocv_filter_available_architecture(${nvcc_executable} __cuda_arch_bin
|
||||
5.3
|
||||
6.2
|
||||
7.2
|
||||
7.0
|
||||
8.7
|
||||
)
|
||||
else()
|
||||
set(__cuda_arch_bin "${_nvcc_out}")
|
||||
endif()
|
||||
set(__cuda_arch_ptx "")
|
||||
else()
|
||||
ocv_filter_available_architecture(${nvcc_executable} __cuda_arch_bin
|
||||
${_arch_fermi}
|
||||
${_arch_kepler}
|
||||
${_arch_maxwell}
|
||||
${_arch_pascal}
|
||||
${_arch_volta}
|
||||
${_arch_turing}
|
||||
${_arch_ampere}
|
||||
${_arch_lovelace}
|
||||
${_arch_hopper}
|
||||
)
|
||||
list(GET __cuda_arch_bin -1 __cuda_arch_ptx)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html)")
|
||||
set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for (see https://docs.opencv.org/4.x/d2/dbc/cuda_intro.html)")
|
||||
string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
|
||||
string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
|
||||
|
||||
# Check if user specified 1.0/2.1 compute capability: we don't support it
|
||||
macro(ocv_wipeout_deprecated_cc target_cc)
|
||||
if(" ${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}" MATCHES " ${target_cc}")
|
||||
message(SEND_ERROR "CUDA: ${target_cc} compute capability is not supported - exclude it from ARCH/PTX list and re-run CMake")
|
||||
endif()
|
||||
endmacro()
|
||||
ocv_wipeout_deprecated_cc("1.0")
|
||||
ocv_wipeout_deprecated_cc("2.1")
|
||||
endmacro()
|
||||
|
||||
macro(ocv_set_nvcc_threads_for_vs)
|
||||
# Tell NVCC the maximum number of threads to be used to execute the compilation steps in parallel
|
||||
# (option --threads was introduced in version 11.2)
|
||||
if(NOT CUDA_VERSION VERSION_LESS "11.2")
|
||||
if(CMAKE_GENERATOR MATCHES "Visual Studio" AND NOT $ENV{CMAKE_BUILD_PARALLEL_LEVEL} STREQUAL "")
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "--threads=$ENV{CMAKE_BUILD_PARALLEL_LEVEL}")
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_cuda_filter_options)
|
||||
foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
|
||||
set(${var}_backup_in_cuda_compile_ "${${var}}")
|
||||
|
||||
if (CV_CLANG)
|
||||
# we remove -Winconsistent-missing-override and -Qunused-arguments
|
||||
# just in case we are compiling CUDA with gcc but OpenCV with clang
|
||||
string(REPLACE "-Winconsistent-missing-override" "" ${var} "${${var}}")
|
||||
string(REPLACE "-Qunused-arguments" "" ${var} "${${var}}")
|
||||
endif()
|
||||
|
||||
# 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-sign-promo as it generates warnings under linux
|
||||
string(REPLACE "-Wno-sign-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}}")
|
||||
|
||||
string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
|
||||
|
||||
# cc1: warning: command line option '-Wsuggest-override' is valid for C++/ObjC++ but not for C
|
||||
string(REPLACE "-Wsuggest-override" "" ${var} "${${var}}")
|
||||
|
||||
# issue: #11552 (from OpenCVCompilerOptions.cmake)
|
||||
string(REGEX REPLACE "-Wimplicit-fallthrough(=[0-9]+)? " "" ${var} "${${var}}")
|
||||
|
||||
# removal of custom specified options
|
||||
if(OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS)
|
||||
foreach(__flag ${OPENCV_CUDA_NVCC_FILTEROUT_OPTIONS})
|
||||
string(REPLACE "${__flag}" "" ${var} "${${var}}")
|
||||
endforeach()
|
||||
endif()
|
||||
endforeach()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_nvcc_flags)
|
||||
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()
|
||||
|
||||
if(WIN32 AND NOT (CUDA_VERSION VERSION_LESS "11.2"))
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcudafe --display_error_number --diag-suppress 1394,1388)
|
||||
endif()
|
||||
|
||||
if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
|
||||
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xlinker --unresolved-symbols=ignore-in-shared-libs)
|
||||
endif()
|
||||
|
||||
# disabled because of multiple warnings during building nvcc auto generated files
|
||||
if(CV_GCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "4.6.0")
|
||||
ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_apply_cuda_stub_workaround cuda_driver_library_path)
|
||||
# details: https://github.com/NVIDIA/nvidia-docker/issues/775
|
||||
if(" ${cuda_driver_library_path}" MATCHES "/stubs/libcuda.so" AND NOT OPENCV_SKIP_CUDA_STUB_WORKAROUND)
|
||||
set(CUDA_STUB_ENABLED_LINK_WORKAROUND 1)
|
||||
if(EXISTS "${cuda_driver_library_path}" AND NOT OPENCV_SKIP_CUDA_STUB_WORKAROUND_RPATH_LINK)
|
||||
set(CUDA_STUB_TARGET_PATH "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/")
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink "${cuda_driver_library_path}" "${CUDA_STUB_TARGET_PATH}/libcuda.so.1"
|
||||
RESULT_VARIABLE CUDA_STUB_SYMLINK_RESULT)
|
||||
if(NOT CUDA_STUB_SYMLINK_RESULT EQUAL 0)
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different "${cuda_driver_library_path}" "${CUDA_STUB_TARGET_PATH}/libcuda.so.1"
|
||||
RESULT_VARIABLE CUDA_STUB_COPY_RESULT)
|
||||
if(NOT CUDA_STUB_COPY_RESULT EQUAL 0)
|
||||
set(CUDA_STUB_ENABLED_LINK_WORKAROUND 0)
|
||||
endif()
|
||||
endif()
|
||||
if(CUDA_STUB_ENABLED_LINK_WORKAROUND)
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-rpath-link,\"${CUDA_STUB_TARGET_PATH}\"")
|
||||
endif()
|
||||
else()
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,--allow-shlib-undefined")
|
||||
endif()
|
||||
if(NOT CUDA_STUB_ENABLED_LINK_WORKAROUND)
|
||||
message(WARNING "CUDA: Workaround for stubs/libcuda.so.1 is not applied")
|
||||
endif()
|
||||
endif()
|
||||
endmacro()
|
||||
|
||||
macro(ocv_check_cuda_delayed_load cuda_toolkit_root_dir)
|
||||
if(MSVC AND CUDA_ENABLE_DELAYLOAD)
|
||||
set(DELAYFLAGS "delayimp.lib")
|
||||
file(GLOB CUDA_DLLS "${cuda_toolkit_root_dir}/bin/*.dll")
|
||||
foreach(d ${CUDA_DLLS})
|
||||
cmake_path(GET "d" FILENAME DLL_NAME)
|
||||
if(NOT ${DLL_NAME} MATCHES "cudart")
|
||||
set(DELAYFLAGS "${DELAYFLAGS} /DELAYLOAD:${DLL_NAME}")
|
||||
endif()
|
||||
endforeach()
|
||||
set(DELAYFLAGS "${DELAYFLAGS} /DELAYLOAD:nvcuda.dll /DELAYLOAD:nvml.dll /IGNORE:4199")
|
||||
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${DELAYFLAGS}")
|
||||
set(CMAKE_MODULE_LINKER_FLAGS "${CMAKE_MODULE_LINKER_FLAGS} ${DELAYFLAGS}")
|
||||
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${DELAYFLAGS}")
|
||||
endif()
|
||||
endmacro()
|
||||
@ -40,7 +40,11 @@ endif()
|
||||
|
||||
# --- CUDA ---
|
||||
if(WITH_CUDA)
|
||||
include("${OpenCV_SOURCE_DIR}/cmake/OpenCVDetectCUDA.cmake")
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
|
||||
include("${OpenCV_SOURCE_DIR}/cmake/OpenCVDetectCUDALanguage.cmake")
|
||||
else()
|
||||
include("${OpenCV_SOURCE_DIR}/cmake/OpenCVDetectCUDA.cmake")
|
||||
endif()
|
||||
if(NOT HAVE_CUDA)
|
||||
message(WARNING "OpenCV is not able to find/configure CUDA SDK (required by WITH_CUDA).
|
||||
CUDA support will be disabled in OpenCV build.
|
||||
|
||||
@ -12,7 +12,11 @@ else()
|
||||
endif()
|
||||
|
||||
if(HAVE_CUDA)
|
||||
ocv_cmake_configure("${CMAKE_CURRENT_LIST_DIR}/templates/OpenCVConfig-CUDA.cmake.in" CUDA_CONFIGCMAKE @ONLY)
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
|
||||
ocv_cmake_configure("${CMAKE_CURRENT_LIST_DIR}/templates/OpenCVConfig-CUDALanguage.cmake.in" CUDA_CONFIGCMAKE @ONLY)
|
||||
else()
|
||||
ocv_cmake_configure("${CMAKE_CURRENT_LIST_DIR}/templates/OpenCVConfig-CUDA.cmake.in" CUDA_CONFIGCMAKE @ONLY)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if(ANDROID)
|
||||
|
||||
@ -1003,7 +1003,7 @@ macro(_ocv_create_module)
|
||||
INTERFACE ${OPENCV_MODULE_${the_module}_DEPS_EXT}
|
||||
)
|
||||
ocv_target_link_libraries(${the_module} PRIVATE ${OPENCV_LINKER_LIBS} ${OPENCV_HAL_LINKER_LIBS} ${IPP_LIBS} ${ARGN})
|
||||
if (HAVE_CUDA)
|
||||
if (NOT ENABLE_CUDA_FIRST_CLASS_LANGUAGE AND HAVE_CUDA)
|
||||
ocv_target_link_libraries(${the_module} PRIVATE ${CUDA_LIBRARIES} ${CUDA_npp_LIBRARY})
|
||||
endif()
|
||||
|
||||
|
||||
@ -1557,13 +1557,23 @@ function(_ocv_append_target_includes target)
|
||||
endif()
|
||||
endfunction()
|
||||
|
||||
macro(ocv_add_cuda_compile_flags)
|
||||
ocv_cuda_compile_flags()
|
||||
target_compile_options(${target} PRIVATE $<$<COMPILE_LANGUAGE:CUDA>: ${CUDA_NVCC_FLAGS}
|
||||
"-Xcompiler=${CMAKE_CXX_FLAGS_CUDA} $<$<CONFIG:Debug>:${CMAKE_CXX_FLAGS_DEBUG_CUDA}> \
|
||||
$<$<CONFIG:Release>:${CMAKE_CXX_FLAGS_RELEASE_CUDA}>" >)
|
||||
endmacro()
|
||||
|
||||
function(ocv_add_executable target)
|
||||
add_executable(${target} ${ARGN})
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE AND HAVE_CUDA)
|
||||
ocv_add_cuda_compile_flags()
|
||||
endif()
|
||||
_ocv_append_target_includes(${target})
|
||||
endfunction()
|
||||
|
||||
function(ocv_add_library target)
|
||||
if(HAVE_CUDA AND ARGN MATCHES "\\.cu")
|
||||
if(NOT ENABLE_CUDA_FIRST_CLASS_LANGUAGE AND HAVE_CUDA AND ARGN MATCHES "\\.cu")
|
||||
ocv_include_directories(${CUDA_INCLUDE_DIRS})
|
||||
ocv_cuda_compile(cuda_objs ${ARGN})
|
||||
set(OPENCV_MODULE_${target}_CUDA_OBJECTS ${cuda_objs} CACHE INTERNAL "Compiled CUDA object files")
|
||||
@ -1571,6 +1581,10 @@ function(ocv_add_library target)
|
||||
|
||||
add_library(${target} ${ARGN} ${cuda_objs})
|
||||
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE AND HAVE_CUDA)
|
||||
ocv_add_cuda_compile_flags()
|
||||
endif()
|
||||
|
||||
if(APPLE_FRAMEWORK AND BUILD_SHARED_LIBS)
|
||||
message(STATUS "Setting Apple target properties for ${target}")
|
||||
|
||||
|
||||
31
cmake/templates/OpenCVConfig-CUDALanguage.cmake.in
Normal file
31
cmake/templates/OpenCVConfig-CUDALanguage.cmake.in
Normal file
@ -0,0 +1,31 @@
|
||||
# Version Compute Capability from which OpenCV has been compiled is remembered
|
||||
set(OpenCV_COMPUTE_CAPABILITIES "@OpenCV_CUDA_CC@")
|
||||
|
||||
set(OpenCV_CUDA_VERSION "@CUDA_VERSION_STRING@")
|
||||
set(OpenCV_USE_CUBLAS "@HAVE_CUBLAS@")
|
||||
set(OpenCV_USE_CUFFT "@HAVE_CUFFT@")
|
||||
set(OpenCV_USE_NVCUVID "@HAVE_NVCUVID@")
|
||||
set(OpenCV_USE_NVCUVENC "@HAVE_NVCUVENC@")
|
||||
set(OpenCV_CUDNN_VERSION "@CUDNN_VERSION@")
|
||||
set(OpenCV_USE_CUDNN "@HAVE_CUDNN@")
|
||||
set(ENABLE_CUDA_FIRST_CLASS_LANGUAGE ON)
|
||||
|
||||
if(NOT CUDAToolkit_FOUND)
|
||||
if(NOT CMAKE_VERSION VERSION_LESS 3.18)
|
||||
if(UNIX AND NOT CMAKE_CUDA_COMPILER AND NOT CUDAToolkit_ROOT)
|
||||
message(STATUS "Checking for CUDAToolkit in default location (/usr/local/cuda)")
|
||||
set(CUDA_PATH "/usr/local/cuda" CACHE INTERNAL "")
|
||||
set(ENV{CUDA_PATH} ${CUDA_PATH})
|
||||
endif()
|
||||
find_package(CUDAToolkit ${OpenCV_CUDA_VERSION} EXACT REQUIRED)
|
||||
else()
|
||||
message(FATAL_ERROR "Using OpenCV compiled with CUDA as first class language requires CMake \>= 3.18.")
|
||||
endif()
|
||||
else()
|
||||
if(CUDAToolkit_FOUND)
|
||||
set(CUDA_VERSION_STRING ${CUDAToolkit_VERSION})
|
||||
endif()
|
||||
if(NOT CUDA_VERSION_STRING VERSION_EQUAL OpenCV_CUDA_VERSION)
|
||||
message(FATAL_ERROR "OpenCV library was compiled with CUDA ${OpenCV_CUDA_VERSION} support. Please, use the same version or rebuild OpenCV with CUDA ${CUDA_VERSION_STRING}")
|
||||
endif()
|
||||
endif()
|
||||
@ -49,13 +49,6 @@ if(DEFINED WINRT AND NOT DEFINED ENABLE_WINRT_MODE_NATIVE)
|
||||
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /ZW")
|
||||
endif()
|
||||
|
||||
if(HAVE_CUDA)
|
||||
if(NOT HAVE_opencv_cudev)
|
||||
message(FATAL_ERROR "CUDA: OpenCV requires enabled 'cudev' module from 'opencv_contrib' repository: https://github.com/opencv/opencv_contrib")
|
||||
endif()
|
||||
ocv_warnings_disable(CMAKE_CXX_FLAGS -Wundef -Wenum-compare -Wunused-function -Wshadow)
|
||||
endif()
|
||||
|
||||
if(CV_TRACE AND HAVE_ITT)
|
||||
add_definitions(-DOPENCV_WITH_ITT=1)
|
||||
endif()
|
||||
@ -153,7 +146,6 @@ elseif(HAVE_CXX11 OR DEFINED OPENCV_ALLOCATOR_STATS_COUNTER_TYPE)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
|
||||
if(PARALLEL_ENABLE_PLUGINS)
|
||||
ocv_append_source_file_compile_definitions(${CMAKE_CURRENT_SOURCE_DIR}/src/parallel/parallel.cpp "PARALLEL_ENABLE_PLUGINS=1")
|
||||
if(OPENCV_DEBUG_POSTFIX)
|
||||
@ -161,6 +153,15 @@ if(PARALLEL_ENABLE_PLUGINS)
|
||||
endif()
|
||||
endif()
|
||||
|
||||
if(HAVE_CUDA)
|
||||
if(NOT HAVE_opencv_cudev)
|
||||
message(FATAL_ERROR "CUDA: OpenCV requires enabled 'cudev' module from 'opencv_contrib' repository: https://github.com/opencv/opencv_contrib")
|
||||
endif()
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
|
||||
ocv_module_include_directories(${CUDAToolkit_INCLUDE_DIRS})
|
||||
endif()
|
||||
ocv_warnings_disable(CMAKE_CXX_FLAGS -Wundef -Wenum-compare -Wunused-function -Wshadow)
|
||||
endif()
|
||||
|
||||
ocv_create_module(${extra_libs})
|
||||
|
||||
|
||||
@ -164,6 +164,12 @@ if(OPENCV_DNN_CUDA AND HAVE_CUDA AND HAVE_CUBLAS AND HAVE_CUDNN)
|
||||
endif()
|
||||
endforeach()
|
||||
unset(CC_LIST)
|
||||
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
|
||||
list(APPEND libs ${CUDNN_LIBRARIES} CUDA::cublas${CUDA_LIB_EXT})
|
||||
if(NOT CUDA_VERSION VERSION_LESS 10.1)
|
||||
list(APPEND libs CUDA::cublasLt${CUDA_LIB_EXT})
|
||||
endif()
|
||||
endif()
|
||||
else()
|
||||
set(sources_options ${sources_options} EXCLUDE_CUDA)
|
||||
endif()
|
||||
|
||||
Loading…
Reference in New Issue
Block a user