*Fix negative zero maximum CFL #231
Conversation
Hallberg-NOAA
force-pushed
the
fix_negative_zero_CFL
branch
from
45f67cf to
d677c1c
Compare
marshallward
left a comment
marshallward
left a comment
There was a problem hiding this comment.
As far as I know, the MPI library can strip the sign of zero during any reduction, making it impossible to reproduce signed zeros, so explicit assignment to (positive) zero is the only way to reliably reproduce the sign.
Normally I would say that this is better handled at the MOM_coms level. But the delayed adoption of MOM6 code in SIS2 justifies taking this right away.
The only minor improvement here would be to replace max_CFL = abs(max_CFL) with max_CFL = 0.. But I will approve as it is, and this line can be removed after a future update to max_across_PEs.
SIS2 reports the global maximum CFL in the seaice.stats files, but in the first timestep when all velocities are zero, it is reporting a maximum CFL of -0.000 with the gnu and Nvidia compilers. This single-line commit replaces any negative zero maximum CFL returned by `max_across_PEs()` with its absolute value so that only non-negative maximum CFL values are returned. All solutions are bitwise identical, but this does change the entries in the seaice.stats files so it may appear to fail automated regression testing in some cases.
Hallberg-NOAA
force-pushed
the
fix_negative_zero_CFL
branch
from
5de7363 to
e1cdc80
Compare
|
This PR completed the pipeline testing at https://gitlab.gfdl.noaa.gov/ogrp/mom6ci/SIS2/-/pipelines/29802 with the expected failure due to the changes in some seaice.stats files that were previously reporting negative zero maximum CFL numbers. |
2 participants
SIS2 reports the global maximum CFL in the seaice.stats files, but in the first timestep when all velocities are zero, it is reporting a maximum CFL of -0.000 with the gnu and Nvidia compilers. This single-line commit replaces any negative zero maximum CFL returned by
max_across_PEs()with its absolute value so that only non-negative maximum CFL values are returned. All solutions are bitwise identical, but this does change the entries in the seaice.stats files so it may appear to fail automated regression testing in some cases.This commit addresses the problem in SIS2 that is noted in #230, but it does not address the underlying question of why this bizarre behavior is happening in the first place.