Segfault with openblas 0.3.10
Closed this issue · 3 comments
ogrisel commented
I have observed a regression with openblas 0.3.10:
import numpy as np
np.ones(shape=(300000, 5)) @ np.ones(shape=(5, 5))
=> segfaults.
No problem with 0.3.9.
I will report upstream but maybe we might want to mark 0.3.10 as a bad release in conda-forge? Are we might do that once a bugfix release is available from upstream?
Original bug in scikit-learn: scikit-learn/scikit-learn#17980
h-vetinari commented
Hello @ogrisel, what platform / OS are you on?
This issue is very likely a duplicate of #101; we have already backported the upstream fix that's (in all likelihood) the underlying issue for this since yesterday, so could you please try updating to the latest build (openblas-0.3.10-<...>_4
) and checking if the issue persists?
ogrisel commented
Sorry: Linux 64bit. Let me check.
ogrisel commented
I confirm your fix works. That was quick!
Thanks.