GEOSgcm at C180 does not layout regress

[mathomp4]

As found by @aoloso, if you run GEOSgcm at C180 at two different layouts, the model does not regress. Start-stop is good, but not layout. I can confirm this with runs at 4x24 and 6x36.

Interestingly, C90 at 4x24 and 6x36 does layout regress. So there must be something the model does at C180 it doesn't do at C90? I tried looking at logfiles but n_split was about the only thing that popped out (other than what you'd expect).

CC'ing...everyone @atrayano @wmputman @tclune @narnold1 @bena-nasa @sdrabenh

[Dooruk]

We may have encountered this issue with our recent coupled C180+MOM6+CICE6 runs with @zhaobin74. I would like to provide some feedback with the hopes that it might be helpful.

With GEOSgcm v11.0.3 (pre CICE6), I was able to run the coupled model without any issues, longest runs being a month in simulation time.

With v11.3.1, I used the exact same layout as v11.0.3 (24x30) and although the model run successfully for a short while I had an extreme value error after 1.5 days. Long story short, after many debugging steps and creating identical setups with @bzhao we realized that the issue was with the layout. The model is past that extreme value point with a 5x90 layout so we solved the issue for now. I forced it to be cas nodes instead of requesting cas|sky. I can provide more details if needed.