Skip to content

Commit 156406b

Browse files
tomoaki0705tomoaki0705
committed
select the architecture based on nvcc result
 * cache the result * DRY * brush up based on review
1 parent ea0d142 commit 156406b

File tree

1 file changed

+56
-24
lines changed

1 file changed

+56
-24
lines changed

‎cmake/OpenCVDetectCUDA.cmake

Lines changed: 56 additions & 24 deletions
Original file line numberDiff line numberDiff line change
@@ -53,6 +53,12 @@ if(CUDA_FOUND)
5353
message(STATUS "CUDA detected: " ${CUDA_VERSION})
5454

5555
set(_generations "Fermi" "Kepler" "Maxwell" "Pascal" "Volta" "Turing")
56+
set(_arch_fermi "2.0")
57+
set(_arch_kepler "3.0;3.5;3.7")
58+
set(_arch_maxwell "5.0;5.2")
59+
set(_arch_pascal "6.0;6.1")
60+
set(_arch_volta "7.0")
61+
set(_arch_turing "7.5")
5662
if(NOT CMAKE_CROSSCOMPILING)
5763
list(APPEND _generations "Auto")
5864
endif()
@@ -70,29 +76,57 @@ if(CUDA_FOUND)
7076
unset(CUDA_ARCH_PTX CACHE)
7177
endif()
7278

79+
macro(ocv_filter_available_architecture result_list)
80+
if(DEFINED CUDA_SUPPORTED_CC)
81+
set(${result_list} "${CUDA_SUPPORTED_CC}")
82+
else()
83+
set(CC_LIST ${ARGN})
84+
foreach(target_arch ${CC_LIST})
85+
string(REPLACE "." "" target_arch_short ${target_arch})
86+
set(NVCC_OPTION "-gencode;arch=compute_${target_arch_short},code=sm_${target_arch_short}")
87+
execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${NVCC_OPTION} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu"
88+
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
89+
RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
90+
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
91+
if(_nvcc_res EQUAL 0)
92+
set(${result_list} "${${result_list}} ${target_arch}")
93+
endif()
94+
endforeach()
95+
string(STRIP ${${result_list}} ${result_list})
96+
set(CUDA_SUPPORTED_CC ${${result_list}} CACHE INTERNAL "List of supported compute capability")
97+
endif()
98+
endmacro()
99+
100+
macro(ocv_detect_native_cuda_arch status output)
101+
execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
102+
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
103+
RESULT_VARIABLE ${status} OUTPUT_VARIABLE ${output}
104+
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
105+
endmacro()
106+
107+
macro(ocv_wipeout_deprecated _arch_bin_list)
108+
string(REPLACE "2.1" "2.1(2.0)" ${_arch_bin_list} ${${_arch_bin_list}})
109+
endmacro()
110+
73111
set(__cuda_arch_ptx "")
74112
if(CUDA_GENERATION STREQUAL "Fermi")
75-
set(__cuda_arch_bin "2.0")
113+
set(__cuda_arch_bin ${_arch_fermi})
76114
elseif(CUDA_GENERATION STREQUAL "Kepler")
77-
set(__cuda_arch_bin "3.0 3.5 3.7")
115+
set(__cuda_arch_bin ${_arch_kepler})
78116
elseif(CUDA_GENERATION STREQUAL "Maxwell")
79-
set(__cuda_arch_bin "5.0 5.2")
117+
set(__cuda_arch_bin ${_arch_maxwell})
80118
elseif(CUDA_GENERATION STREQUAL "Pascal")
81-
set(__cuda_arch_bin "6.0 6.1")
119+
set(__cuda_arch_bin ${_arch_pascal})
82120
elseif(CUDA_GENERATION STREQUAL "Volta")
83-
set(__cuda_arch_bin "7.0")
121+
set(__cuda_arch_bin ${_arch_volta})
84122
elseif(CUDA_GENERATION STREQUAL "Turing")
85-
set(__cuda_arch_bin "7.5")
123+
set(__cuda_arch_bin ${_arch_turing})
86124
elseif(CUDA_GENERATION STREQUAL "Auto")
87-
execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
88-
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
89-
RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
90-
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
125+
ocv_detect_native_cuda_arch(_nvcc_res _nvcc_out)
91126
if(NOT _nvcc_res EQUAL 0)
92127
message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
93128
else()
94-
set(__cuda_arch_bin "${_nvcc_out}")
95-
string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
129+
string(REGEX MATCHALL "[0-9]+\\.[0-9]" __cuda_arch_bin "${_nvcc_out}")
96130
endif()
97131
endif()
98132

@@ -101,28 +135,26 @@ if(CUDA_FOUND)
101135
set(__cuda_arch_bin "3.2")
102136
set(__cuda_arch_ptx "")
103137
elseif(AARCH64)
104-
execute_process( COMMAND "${CUDA_NVCC_EXECUTABLE}" ${CUDA_NVCC_FLAGS} "${OpenCV_SOURCE_DIR}/cmake/checks/OpenCVDetectCudaArch.cu" "--run"
105-
WORKING_DIRECTORY "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeTmp/"
106-
RESULT_VARIABLE _nvcc_res OUTPUT_VARIABLE _nvcc_out
107-
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
138+
ocv_detect_native_cuda_arch(_nvcc_res _nvcc_out)
108139
if(NOT _nvcc_res EQUAL 0)
109140
message(STATUS "Automatic detection of CUDA generation failed. Going to build for all known architectures.")
110141
set(__cuda_arch_bin "5.3 6.2 7.2")
111142
else()
112143
set(__cuda_arch_bin "${_nvcc_out}")
113-
string(REPLACE "2.1" "2.1(2.0)" __cuda_arch_bin "${__cuda_arch_bin}")
114144
endif()
115145
set(__cuda_arch_ptx "")
116146
else()
117-
if(CUDA_VERSION VERSION_LESS "9.0")
118-
set(__cuda_arch_bin "2.0 3.0 3.5 3.7 5.0 5.2 6.0 6.1")
119-
elseif(CUDA_VERSION VERSION_LESS "10.0")
120-
set(__cuda_arch_bin "3.0 3.5 3.7 5.0 5.2 6.0 6.1 7.0")
121-
else()
122-
set(__cuda_arch_bin "3.0 3.5 3.7 5.0 5.2 6.0 6.1 7.0 7.5")
123-
endif()
147+
ocv_filter_available_architecture(__cuda_arch_bin
148+
${_arch_fermi}
149+
${_arch_kepler}
150+
${_arch_maxwell}
151+
${_arch_pascal}
152+
${_arch_volta}
153+
${_arch_turing}
154+
)
124155
endif()
125156
endif()
157+
ocv_wipeout_deprecated(__cuda_arch_bin)
126158

127159
set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
128160
set(CUDA_ARCH_PTX ${__cuda_arch_ptx} CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")

0 commit comments

Comments
 (0)