Failed to dlopen libcuda.so in WSL environment about cuda-python

[jakirkham]

Maybe similar to issue ( rapidsai/cudf#10308 )?

[kkraus14]

Also, in the ldconfig cache it looks like others load libcuda.so.1 so only that entry is cached which would be another path for dlopen to find the needed library:

...
        libcudart.so (libc6,x86-64) => /usr/local/cuda-11.5/targets/x86_64-linux/lib/libcudart.so
        libcuda.so.1 (libc6,x86-64) => /usr/lib/wsl/lib/libcuda.so.1
        libcublasLt.so.11 (libc6,x86-64) => /usr/local/cuda-11.4/targets/x86_64-linux/lib/libcublasLt.so.11
...

Additionally, looking at the ELF table of libcuda.so it looks like the SONAME is in fact libcuda.so.1 so we should probably be trying to open that instead regardless.

 0x000000000000000e (SONAME)             Library soname: [libcuda.so.1]

[vzhurba01]

I've merged commit 15935e5 that should resolve this.

@kkraus14 do you still get this error?
As of right now this fix is only on the repo and it still needs to be propagated to PyPI/Conda.

[kkraus14]

I've merged commit 15935e5 that should resolve this.

@kkraus14 do you still get this error? As of right now this fix is only on the repo and it still needs to be propagated to PyPI/Conda.

Just built the latest main branch and confirmed that it fixed the problem and passed the pytests. Closing, thanks @vzhurba01!