bug fixes for dla networks and ported optimization from different pull request

This commit is contained in:
perseusdg
2021-11-13 02:00:49 +05:30
parent 9e328c0daa
commit ee5000ccca
7 changed files with 122 additions and 192 deletions
-27
View File
@@ -6,36 +6,9 @@ set(CMAKE_CXX_STANDARD 14)
option(ENABLE_OPENCV_CUDA_CONTRIB "Enable OpenCV CUDA Contrib" OFF )
find_package(CUDA 9.0 REQUIRED)
if (CUDA_FOUND)
set(OUTPUTFILE ${CMAKE_CURRENT_SOURCE_DIR}/cmake/cuda_script) # No suffix required
execute_process(COMMAND "rm ${OUTPUTFILE}")
set(CUDAFILE ${CMAKE_CURRENT_SOURCE_DIR}/cmake/getCudaArch.cu)
execute_process(COMMAND ${CUDA_NVCC_EXECUTABLE} -lcuda ${CUDAFILE} -o ${OUTPUTFILE})
execute_process(COMMAND ${OUTPUTFILE}
RESULT_VARIABLE CUDA_RETURN_CODE
OUTPUT_VARIABLE ARCH)
if(${CUDA_RETURN_CODE} EQUAL 0)
set(CUDA_SUCCESS "TRUE")
else()
set(CUDA_SUCCESS "FALSE")
endif()
if (${CUDA_SUCCESS})
message(STATUS "CUDA Architecture: ${ARCH}")
message(STATUS "CUDA Version: ${CUDA_VERSION_STRING}")
message(STATUS "CUDA Path: ${CUDA_TOOLKIT_ROOT_DIR}")
message(STATUS "CUDA Libararies: ${CUDA_LIBRARIES}")
message(STATUS "CUDA Performance Primitives: ${CUDA_npp_LIBRARY}")
set(CUDA_NVCC_FLAGS "${ARCH}")
else()
message(WARNING ${ARCH})
endif()
endif()
SET(CUDA_SEPARABLE_COMPILATION ON)
if(UNIX)
if(CMAKE_BUILD_TYPE MATCHES Release)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC -Wno-deprecated-declarations -Wno-unused-variable -O3")