Evgeni Burovski

Hm... Actually, it did fail like this before: https://github.com/OpenMathLib/OpenBLAS/actions/workflows/codspeed-bench.yml And the single-precision variant has been introduced in https://github.com/OpenMathLib/OpenBLAS/pull/4763, so it's relatively recent. Hm. Am taking a look.

A few quick observations: - SLASCL is an internal LAPACK routine, called from somewhere inside sgesdd (a driver for SVD factorization) so this failure is smoking some possible error in...

This is indeed codspeed-specific: https://github.com/OpenMathLib/OpenBLAS/actions/runs/9765188405/job/26955309472?pr=4777 I asked on their discord (https://discord.com/channels/1065233827569598464/1065686090452828251/threads/1257753281342738502 -- might need to join to see, no idea; will repost any insights here anyway). Meanwhile, do we want...

okay, I repurposed https://github.com/OpenMathLib/OpenBLAS/pull/4777 to ignore the sgesdd failure.

One check that the problem's fixed is to re-enable the `sdgesdd` codspeed benchmark: ```diff $ git diff diff --git a/benchmark/pybench/benchmarks/bench_blas.py b/benchmark/pybench/benchmarks/bench_blas.py index 628c0cb2a..8127dd0c7 100644 --- a/benchmark/pybench/benchmarks/bench_blas.py +++ b/benchmark/pybench/benchmarks/bench_blas.py @@ -234,14...

Yes, changing to `x[sl][i]` seems to work just fine, so low prio. For context, this came up in https://github.com/scipy/scipy/pull/22298, where https://github.com/scipy/scipy/pull/22298#issuecomment-2585628632 looks more like a blocker.

+1 to reporting it upstream. Ideally, there's an upstream issue, plus a reference to in a workaround in `-compat`. EDIT: This isn't a hard requirement, of course.

Hm. Apparently, `cupyx.scipy.signal.sosfilt_zi` is not a drop-in replacement for `scipy.signal.sosfilt_zi`: 1. `sosfilt_zi` returns `zi` of different shapes in scipy and cupyx: in SciPy it's `(n_sections, 2)` and in `cupyx.scipy` it's...

Thanks Edgar! This should probably be great to document, for both scipy and cupy. That said, this does not explain the need to transpose noted in https://github.com/scipy/scipy/pull/21713#discussion_r1802139152 though?

Added a commit with a slew of skips for CuPy, and a separate commit to avoid delegating for functions which deliberately deviate in scipy and cupy. I believe skipping tests...