-
Notifications
You must be signed in to change notification settings - Fork 900
Find libcuda.so automatically if --with-cuda-lib is not passed #12378
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Find libcuda.so automatically if --with-cuda-lib is not passed #12378
Conversation
Hello! The Git Commit Checker CI bot found a few problems with this PR: 5bcaba0: Find libcuda.so automatically if --with-cuda-lib i...
Please fix these problems and, if necessary, force-push new commits back up to the PR branch. Thanks! |
f718881
to
c8e885d
Compare
this works for me at nersc perlmutter. I load cudatoolkit module which sets CUDA_HOME=/opt/nvidia/hpc_sdk/Linux_x86_64/23.9/cuda/12.2. In my configure line I have
then when doing the libtool link think its linking in libcuda.so at /opt/nvidia/hpc_sdk/Linux_x86_64/23.9/cuda/12.2/targets/x86_64-linux/lib/stubs/libcuda.so,
so this patch is doing what we'd like on perlmutter but the fixes for protecting for cases where just |
Hello! The Git Commit Checker CI bot found a few problems with this PR: cfec4a9: Update to check /usr/local/cuda in case with_cuda ...
Please fix these problems and, if necessary, force-push new commits back up to the PR branch. Thanks! |
af62cc6
to
5204853
Compare
Hi @hppritcha, I have updated the PR to handle the case where $with_cuda is not a directory. |
Finding CUDA libraries without having to specify both --with-cuda and --with-cuda-lib was requested in github issue open-mpi#12264 Signed-off-by: Nick Sarkauskas <[email protected]>
5204853
to
cad3d9a
Compare
Ahh, I did not see the approval before force pushing again. The reason I force pushed was to add redirection to /dev/null in case that |
this tweak looks good to me. |
In newer OpenMPI versions, it's required to pass both
--with-cuda
and--with-cuda-libdir
in order for CUDA to be recognized by the build system. Ideally, just--with-cuda
should be enough and the build system should be able to detect where the CUDA libraries are.This PR allows removes the requirement that
--with-cuda-libdir
be passed and instead makes the build system search for libcuda.so inside of the directory passed with--with-cuda
.See issue: #12264