Compute dz in the halo for OBC segments #520
Conversation
kshedstrom
commented
Nov 15, 2023
kshedstrom
commented
- It was blowing up with "forrtl: error (65): floating invalid" when accessing dz in the halo at the boundary, but just sometimes. My default layout is trouble while my testing layout of 48 cores is not.
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## dev/gfdl #520 +/- ##
=========================================
Coverage 37.16% 37.17%
=========================================
Files 271 271
Lines 80655 80657 +2
Branches 15046 15047 +1
=========================================
+ Hits 29979 29981 +2
Misses 45102 45102
Partials 5574 5574 ☔ View full report in Codecov by Sentry. |
|
This now takes care of #518 |
|
My Huynh diffs were not at the boundaries, but were definitely not everywhere. It would be worthwhile for someone to figure out why they occur. They show up in hour three of tracer fields with a DT_THERM of an hour, at the bottom. Good point about the pass_var. I just report what I was seeing in the debugger. |
|
I did some additional testing of this PR by comparing runs of the Northwest Atlantic regional model with this version of MOM6 vs. the dev-gfdl version of MOM6. I found that for both this branch and dev/gfdl MOM6 the answers did not change when the default number of halo points (4) was increased to a "very wide halo" (12 points). Further work may be needed to determine the halo/stencil size of different OBC options, but for now I think this PR is tested and can be approved. |
- It was blowing up with "forrtl: error (65): floating invalid" when accessing dz in the halo at the boundary, but just sometimes. My default layout is trouble while my testing layout of 48 cores is not.
-even for PPM advection and OBCs.
Hallberg-NOAA
force-pushed
the
halo_for_dz
branch
from
e530cee to
fc23f99
Compare
Removed allocated tests for the potentially statically allocated arrays CS%uhtr and CS%vhtr in a newly added line testing whether there are OBCs in use that would require a halo update on these arrays. Even in dynamic memory mode, these arrays are always being allocated, so these tests served no purpose, but in static memory mode they led to compile time errors. All answers are bitwise identical.
|
You may delete the comment. |
|
The regression suite is finally back online! Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/22549 ✔️ |
