-
Notifications
You must be signed in to change notification settings - Fork 6
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Different results depending on availability of AVX-512 instruction set and numpy>1.25 's vectorized sort/argsort #115
Comments
On bb5, I can repro:
Architecture is:
|
This was worked around, for the time being, with: |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
numpy==1.25
introduced simd vectorized version of quicksort with: numpy/numpy#23707That change affects the behavior of the sort and argsort functions when using quicksort (default) based on the existence of the avx-512 instruction set.
In NeuroR, one of the functions that are affected is
scipy.optimize.minimize
:NeuroR/neuror/cut_plane/detection.py
Line 279 in ae52411
which uses internally numpy's argsort in:
https://github.com/scipy/scipy/blob/cfe80116aaa145061246b8aec0e98248fbefb835/scipy/optimize/_optimize.py#L907
See also:
numpy/numpy#24064
Github hosted runners for ubuntu-latest seem to have enabled the aforementioned instruction set as they output different values than the expected ones in the past: https://github.com/BlueBrain/NeuroR/actions/runs/5484236591/jobs/9991722132
The text was updated successfully, but these errors were encountered: