search

conda-forge / mumps-feedstock

A conda-smithy repository for mumps.
BSD 3-Clause "New" or "Revised" License
5 stars 24 forks source link

enable GEMMT with openblas, mkl #124

Open minrk opened 1 week ago

minrk commented 1 week ago

adds build variant with openblas, with GEMMT enabled. As a result, installing mumps with openblas should get GEMMT by default. Installing mumps with any other blas should still get the most-compatible build targeting netlib.

closes #120

Draft because I'm still doing some performance testing to see if this is worth doing.

conda-forge-webservices[bot] commented 1 week ago

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

minrk commented 1 week ago

On my laptop (10 core M1 macbook), GEMMT does not appear to be worth it, at least for a simple poisson solve:

visualization

visualization-2

OMP clearly gives a benefit in single-process solves (much less so in multiprocess, as expected), but at least openblas GEMMT doesn't seem to affect performance much, and when it does, the effect is not always positive.

minrk commented 1 week ago

Apparently my test didn't actually exercise the GEMMT path, but when I try cholesky route that actually uses GEMMT, it's way slower, in fact. Not sure what's up with that. Trying with MKL as well.