+REMAP_AUX needs at least one more halo update.#496
Conversation
kshedstrom
commented
Oct 6, 2023
kshedstrom
commented
- Only when REMAP_AUXILIARY_VARS is true.
- Addresses OBC problem in tangential_vel #495
- There is still a problem with diffu and diffv when REMAP_AUXILIARY_VARS is true. The first call to horizontal_viscosity after the first REMAP returns with a tile boundary mismatch.
- This one is for CS%u_av, CS%v_av, which need to be updated coming into step_MOM_dyn_split_RK2.
Codecov Report
@@ Coverage Diff @@
## dev/gfdl #496 +/- ##
============================================
- Coverage 37.82% 37.82% -0.01%
============================================
Files 270 270
Lines 78340 78350 +10
Branches 14503 14507 +4
============================================
+ Hits 29636 29638 +2
- Misses 43300 43307 +7
- Partials 5404 5405 +1
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
Hallberg-NOAA
left a comment
Hallberg-NOAA
left a comment
There was a problem hiding this comment.
I agree that these halo updates are necessary and correct.
As written, there will be two separate points of synchronization for the two pass vector calls. This could be reduced to a single point of synchronization (and made more efficient) by adding the argument complete=.false. on the first call and (optionally) complete=.true. on the second, or they could be combined into a group_pass.
|
Thanks for the comments, @Hallberg-NOAA. I will submit a fix. Meanwhile, I will also attempt a fix to the diffu issue, which is due to needing a halo update on r[xy]_normal after the call to remap_OBC_fields. |
- This fixes the Bering ORLANSKI OBCs for differing processor counts. - This is either the wrong way to do group_pass for OBLIQUE OBC's or there is more wrong with them.
- Problem is in tangential_vel at tile boundaries. It matches right at the boundary, but needs some halo points to match too.
- Without this, u_av and v_av don't update a wide enough halo to get answers to reproduce across different proocessor counts with oblique OBCs.
|
@kshedstrom Is this ready to go in? Or did you want to continue working on it? |
|
You can merge this now. The Bering domain now behaves, but Mehmet and I are seeing other lingering issues with different processor counts for both the North Atlantic and the Arctic. But I think the REMAP_AUXILIARY_VARS and the OBCs should both be good now. |
|
Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/20915 ✔️ |
3 participants
