linux下通过jni编译CUDA代码生成动态链接库遇到的问题 nvcc -fPIC

来源：互联网发布：临沂淘宝拍照详情制作编辑：程序博客网时间：2024/05/22 03:21

错误类型：

relocation R_X86_64_32 against `.rodata' can not be used when making a shared object

一般c++程序编译生成动态链接库时，我们会使用-fPIC 和 -shared编译参数，至于具体参数说明这里不详细介绍。nvcc本身不支持-fPIC编译参数，但是却提供了-Xcompiler

用途上，使用nvcc -h可以发现，这是提供了向低级编译工具(gcc)传递编译参数的功能，因此在编译.cu文件时，在nvcc后加上 -shared -Xcomplier -fPIC 即可链接成功

最后贴上编译写的shell脚本

NVCC=nvccCXX=g++LD=ldINCLUDES="-I/usr/local/cuda/include -I/usr/local/opencv/include -I/usr/local/jdk/include -I/usr/local/jdk/include/linux"LDFLAGS="-shared -fPIC"CXXFLAGS="-g -fPIC"NVCCFLAGS="-g -shared "CUDA_LD_LIBRARYS="-lcudart"CUDA_LD_LIBRARY_PATH="-L/usr/local/cuda/lib64  -L/usr/lib/\"nvidia-352\""OPENCV_LD_LIBRARYS="-lopencv_core -lopencv_highgui -lopencv_imgproc"OPENCV_LD_LIBRARY_PATH="-L/usr/local/opencv/lib"LIBRARYS="$CUDA_LD_LIBRARYS $OPENCV_LD_LIBRARYS -lstdc++"LIBRARY_PATH="$CUDA_LD_LIBRARY_PATH $OPENCV_LD_LIBRARY_PATH"TARGET_LIB="bayes_fge.so"echo "$CXX $INCLUDES $CXXFLAGS -o edu_fudan_lwang_FGEHelper.o -c edu_fudan_lwang_FGEHelper.cpp $LIBRARY_PATH $LIBRARYS"$CXX $INCLUDES $CXXFLAGS -o edu_fudan_lwang_FGEHelper.o -c edu_fudan_lwang_FGEHelper.cpp $LIBRARY_PATH $LIBRARYSecho "$CXX $INCLUDES $CXXFLAGS -o fge_hepler.o -c src/fge_helper.cpp $LIBRARY_PATH $LIBRARYS"$CXX $INCLUDES $CXXFLAGS -o fge_hepler.o -c src/fge_helper.cpp $LIBRARY_PATH $LIBRARYSecho "$NVCC $INCLUDES $NVCCFLAGS -o cuda_utility.o -c src/cuda_utility.cu $LIBRARY_PATH $LIBRARYS"$NVCC $INCLUDES $NVCCFLAGS -o cuda_utility.o -c src/cuda_utility.cu $LIBRARY_PATH $LIBRARYSecho "$LD $LDFLAGS -o $TARGET_LIB cuda_utility.o fge_hepler.o edu_fudan_lwang_FGEHelper.o"$LD $LDFLAGS -o $TARGET_LIB  cuda_utility.o fge_hepler.o edu_fudan_lwang_FGEHelper.o

0 0