-
-
Notifications
You must be signed in to change notification settings - Fork 9.6k
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
BUG: SciPy is seeing incorrect warnings on MacOS M1 (M2?) array * scalar
#23996
Comments
Let me know if you want me to test anything specific. |
Thanks, no need... This is a trivial repro that works on my machine:
|
Ping @Developer-Ecosystem-Engineering since maybe you can have a look at this and there is an easy fix. |
* deals with the second part of scipygh-18732 * I was able to reproduce locally, but this one was even flakier--analysis of patch success below, and possibly related to subtle low-level MacOS ARM issues upstream in numpy/numpy#23996 (NumPy `1.25.0`) * before this patch, 4 trials of this command with `pytest-repeat`: `CC=clang CXX=clang++ FC=gfortran-12 python dev.py test -t scipy/signal/tests/test_signaltools.py::TestDecimate::test_complex_iir_dlti -- --count=20000` `81 failed, 19919 passed in 43.68s` `81 failed, 19919 passed in 43.79s` `20000 passed in 33.24s` `43 failed, 19957 passed in 35.17s` * after this patch, 6 trials of the same command: `20000 passed in 30.44s` `20000 passed in 33.08s` `20000 passed in 33.46s` `20000 passed in 33.45s` `20000 passed in 33.56s` `20000 passed in 33.30s` [skip circle]
See also: Flaky behavior, sometimes < 80 failures in 20,000 trials of the same calculation, so reproducing is a bit tricky.
|
* deals with the second part of scipygh-18732 * I was able to reproduce locally, but this one was even flakier--analysis of patch success below, and possibly related to subtle low-level MacOS ARM issues upstream in numpy/numpy#23996 (NumPy `1.25.0`) * before this patch, 4 trials of this command with `pytest-repeat`: `CC=clang CXX=clang++ FC=gfortran-12 python dev.py test -t scipy/signal/tests/test_signaltools.py::TestDecimate::test_complex_iir_dlti -- --count=20000` `81 failed, 19919 passed in 43.68s` `81 failed, 19919 passed in 43.79s` `20000 passed in 33.24s` `43 failed, 19957 passed in 35.17s` * after this patch, 6 trials of the same command: `20000 passed in 30.44s` `20000 passed in 33.08s` `20000 passed in 33.46s` `20000 passed in 33.45s` `20000 passed in 33.56s` `20000 passed in 33.30s` [skip circle]
We understand the issue, please stand by |
Please see #24054 for a proposed fix. |
CC @tylerjereddy. See scipy/scipy#18697, it seems MacOSX on arm optimizes the multiply in a way that leads to loading uninitialized values/zeros probably, so that
0 * inf
triggers a warning.I am not sure we should worry too much about it, since clang on MacOS X for arm doesn't properly support
-ftrapping-math
IIRC, which tends to make things difficult, but it is incorrect. The SciPy test should not show a warning in the specific multiplication.The text was updated successfully, but these errors were encountered: