Kenneth Hoste <[email protected]> writes: > On 03/12/2019 15:50, Loris Bennett wrote: >> Hi, >> >> Installing >> >> TensorFlow-1.13.1-fosscuda-2019a-Python-3.7.2.eb >> >> fails with >> >> >> bazel-out/k8-opt/bin/tensorflow/python/gen_ragged_conversion_ops_py_wrappers_cc: >> error while loading shared libraries: libcuda.so.1: cannot open shared object >> file: No such file or directory >> >> There is a step before which includes: >> >> cp -f >> "/trinity/shared/easybuild/software/CUDA/10.1.105-GCC-8.2.0-2.31.1/targets/x86_64-linux/lib/stubs/libcuda.so" >> "bazel-out/k8-opt/genfiles/external/local_config_cuda/cuda/cuda/lib/libcuda.so" >> >> So it seems to me as if bazel is perhaps failing to add libcuda.so.1 as a >> link to >> libcuda.so. >> >> Does anyone have an idea about how to proceed here? > > The libcuda.so and libcuda.so.1 are only stubs to the actual libcuda.so & > libcuda.so.1 which should be provided through the OS (because they're tied to > the GPU drivers). > > On our GPU system they're provided via an RPM: > > $ rpm -qf /usr/lib64/libcuda.so.1 > nvidia-driver-latest-dkms-cuda-libs-418.87.01-1.el7.x86_64 > > which seems to come from > https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/ .
Thanks, I forgot that I should be compiling on one of our GPU nodes. That raises another question, but I start a separate thread. Cheers, Loris -- Dr. Loris Bennett (Mr.) ZEDAT, Freie Universität Berlin Email [email protected]
