ubuntu - Compiling CUDA looks for wrong CUDA version -
on ubuntu, had installation of cuda 6.5, , wanted upgrade cuda 7.0. so, deleted directory @ /usr/local/cuda-6.5, , installed cuda 7.0 /usr/local/cuda-7.0. changed symbolic link @ /usr/local/cuda point /usr/local/cuda-7.0. in bash.rc file, updated environment variables accordingly:
export cuda_home=/usr/local/cuda-7.0 export ld_library_path=${cuda_home}/lib64 export path=${cuda_home}/bin:${path} if type in "nvcc --version", following expected:
cuda compilation tools, release 7.0, v7.0.27 however, compiling code (the caffe deep learning library, precise) uses cuda, , getting following error message:
error while loading shared libraries: libcudart.so.6.5: cannot open shared object file: no such file or directory so reason, still looking cuda 6.5 libraries, rather cuda 7.0 libraries. why this? how tell compiler 7.0 libraries? cannot find reference libcudart.so.6.5 in source code compiling, cuda compiler looking wrong version.
obviously, didn't follow right uninstallation method; please note if installing different toolkit versions there won't conflict between them , can keep both of them. asked during installation process link /usr/local/cuda-x.y /usr/local/cuda. check section 2.6 of cuda 7.0 getting started on linux.
the proper way uninstall mentioned in above link use commands below depending on how did installation (i.e., run method or rpm method):
$ sudo /usr/local/cuda-x.y/bin/uninstall_cuda_x.y.pl use following command uninstall driver runfile installation: $ sudo /usr/bin/nvidia-uninstall use following commands uninstall rpm/deb installation: $ sudo apt-get --purge remove <package_name> # ubuntu $ sudo yum remove <package_name> # fedora/redhat/centos $ sudo zypper remove <package_name> # opensuse/sles i hope works you; don't know caffe deep learning library, assuming didn't configure before providing path of cuda 6.5 compiler , libraries. if case, first try uninstall previous cuda 6.5 properly, configure library scratch , make it.
Comments
Post a Comment