[Marius1311]

With libblas unpinned, the solver resolved a project from this template to BLIS (pthreads) with netlib LAPACK on linux-64; osx-arm64 got OpenBLAS. No package requires BLIS: the choice was the solver's. BLIS threads every call, so on a loaded multi-user node tiny linear algebra crawls: 720 rounds of a 2×2 SVD plus a 5×2 matmul took 38.2 s (load ~6 on 8 cores), against 0.013 s with BLIS_NUM_THREADS=1.

The pin uses conda-forge's documented BLAS switch. In a project from the template it changes only the BLAS stack, in both environments:

• libblas, libcblas and liblapack move to OpenBLAS 0.3.34, which adds libopenblas;
• blis is dropped;
• the OpenMP runtime moves to LLVM's (_openmp_mutex kmp_llvm), which OpenBLAS's OpenMP build brings in.

Same test, same node: 0.29 s, about 130× faster, and a 2000×2000 SVD in 1.46 s instead of 2.63 s. scipy, PETSc/SLEPc, numba (parallel), torch and a scanpy pipeline through Leiden all ran.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-21T11:44:30.171137Z	f0d6d7d	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.