generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added CUDA_ARCH_FEATUERS define (for correct handling of BIN(PTX) cases)

Browse Source
This commit is contained in:
Alexey Spizhevoy 2011-02-01 12:28:39 +00:00
parent 5a166ca963
commit e303b0dd62
2 changed files with 15 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
CMakeLists.txt
View File

@ -711,44 +711,49 @@ if(WITH_CUDA)
set(CUDA_ARCH_BIN "1.3 2.0" CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported") set(CUDA_ARCH_BIN "1.3 2.0" CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
set(CUDA_ARCH_PTX "1.1 1.3" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for") set(CUDA_ARCH_PTX "1.1 1.3" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
# These variables are used in config templates
string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}") string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}") string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
# Ckeck if user specified 1.0 compute capability # Ckeck if user specified 1.0 compute capability: we don't support it
string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}") string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
set(CUDA_ARCH_BIN_OR_PTX_10 0) set(CUDA_ARCH_BIN_OR_PTX_10 0)
if(NOT ${HAS_ARCH_10} STREQUAL "") if(NOT ${HAS_ARCH_10} STREQUAL "")
set(CUDA_ARCH_BIN_OR_PTX_10 1) set(CUDA_ARCH_BIN_OR_PTX_10 1)
endif() endif()
# Flags to be set # NVCC flags to be set
set(NVCC_FLAGS_EXTRA "") set(NVCC_FLAGS_EXTRA "")
# These variables are passed into the template # These vars will be passed into the templates
set(OPENCV_CUDA_ARCH_BIN "") set(OPENCV_CUDA_ARCH_BIN "")
set(OPENCV_CUDA_ARCH_PTX "") set(OPENCV_CUDA_ARCH_PTX "")
set(OPENCV_CUDA_ARCH_FEATURES "")
# Tell nvcc to add binaries for the specified GPUs # Tell NVCC to add binaries for the specified GPUs
string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}") string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
foreach(ARCH IN LISTS ARCH_LIST) foreach(ARCH IN LISTS ARCH_LIST)
if (ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)") 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(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} ${CMAKE_MATCH_1}") set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${CMAKE_MATCH_2}")
else() 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(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}") set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
endif() endif()
endforeach() endforeach()
# Tell nvcc to add PTX intermediate code for the specified architectures # Tell NVCC to add PTX intermediate code for the specified architectures
string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}") string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
foreach(ARCH IN LISTS ARCH_LIST) foreach(ARCH IN LISTS ARCH_LIST)
set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH}) set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}") set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
set(OPENCV_CUDA_ARCH_FEATURES "${OPENCV_CUDA_ARCH_FEATURES} ${ARCH}")
endforeach() endforeach()
# Wil; be processed in other scripts # These vars will be processed in other scripts
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA}) set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}") set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")

4
modules/gpu/src/initialization.cpp
View File

@ -72,9 +72,9 @@ namespace
CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::GpuFeature feature) CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::GpuFeature feature)
{ {
if (feature == NATIVE_DOUBLE) if (feature == NATIVE_DOUBLE)
return hasEqualOrGreater(1, 3); return ::compareToSet(CUDA_ARCH_FEATURES, 13, std::greater_equal<int>());
if (feature == ATOMICS) if (feature == ATOMICS)
return hasEqualOrGreater(1, 1); return ::compareToSet(CUDA_ARCH_FEATURES, 11, std::greater_equal<int>());
return true; return true;
} }