Repository navigation
FIX Only skip the "sequential_blas_under_openmp" limit for OpenBLAS with OpenMP - #283
Merged
Merged
Conversation
…OpenMP Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
itamarst
approved these changes
Oct 28, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
limits="sequential_blas_under_openmp"doesn't limit anything as soon as OpenBLAS is loaded, whatever its threading layer.select()keeps the libraries matching at least one of its kwargs (as documented), soselect(internal_api="openblas", threading_layer="openmp")also matches pthreads OpenBLAS, i.e. the one of the numpy and scipy wheels. It also matches any library with the OpenMP threading layer (e.g. BLIS).So when I switched to it in scikit-learn/scikit-learn#35180 KMeans' Lloyd iterations got ~2.8x slower on main with 2 cores (29 ms -> 81 ms per iteration, 200k x 15 float64, 150 clusters, Xeon 6787P, PyPI numpy/scipy). The per-chunk GEMMs called from the OpenMP loop are multi-threaded again.
OPENBLAS_NUM_THREADS=1brings it back to 29 ms.The test computed its expected value with the same
select()call, so it couldn't catch this. It now checks the library info directly: on upstream master it fails with pthreads OpenBLAS, and it passes with this fix.cc @itamarst @ogrisel