Scientific Core Libraries for Numerical Computing

Hi, I’m a code developer at the Lawrence Berkeley National Laboratory. We are developing a finite element code for superconducting magnets, and had to deal with dependencies of commonly used numerical libraries such as PETSc, STRUMPACK, MUMPS, NLOPT and SUNDIALS.

I didn’t want to use docker or spack to distribute our code to our coworkers, as I figured that an RPM repository might be the most natural way for an Enterprise Linux. For that reason, I have created a repository that contains these libraries and I am happy to share the repository under the well known “AS IS” terms.

All libraries are optimized for Intel x86_64 architectures, and are available with either MKL bindings or both MKL and CUDA.

I compiled these libraries under RockyLinux, but they should work on Alma as well. I hope my repository is useful for others too. Comments are welcome.

1 Like

This looks fantastic, and is so great to see. I’ve shared a link to this post in the public chat as well. Do you have the code for the source for the website hosted anywhere that I can submit a pull request?

Thanks! Glad it has a positive resonance! Yes, all the packages including the SRPMs are in a browsable directory Index of /scls

Although probably not as well documented as they should be, my python and bash scripts that I use to create the flavors are in GitHub - cmesse/rpmbuild