Add regionally-refined grid over TWP

[brhillman]

Add configuration for new atmosphere and land grid with regionally refinement from ne30 to ne120 over the Tropical Western Pacific region. All needed input files to run the new configuration have been uploaded to the input data repository. Supported compsets include only FC5AV1C-04P2 (72 levels).

Fixes #1357

[BFB]

[AaronDonahue]

@singhbalwinder , @rljacob , I've finished nearly everything to push this to Next, but I have a few fails in the acme developer suite:

FAIL SMS_Ln9.ne4_ne4.FC5AV1C-L.anvil_intel.cam-outfrq9s time=59
FAIL SMS_D_Ln5.ne4_ne4.FC5.anvil_intel RUN time=47
FAIL SMS_D_Ln5.ne4_ne4.FC5AV1C-L.anvil_intel RUN time=58
FAIL SMS.f09_g16_a.IGCLM45_MLI.anvil_intel RUN time=35
FAIL ERS_Ln9.ne4_ne4.FC5AV1C-L.anvil_intel.cam-rtm_null RUN time=45

Are these fails that we would expect and I can just go ahead and push to next?

[rljacob]

This PR should not cause any existing tests to fail. Did those tests pass with master for the platform you're using?

[singhbalwinder]

@AaronDonahue : Like @rljacob mentioned, this PR should not break existing tests. I would first look at one of the failing tests to see why it failed. If I can't see an obvious reason, I would run the same test using current master to see if it fails. I would choose SMS_D_Ln5.ne4_ne4.FC5 for this as this is the simplest test case.

It goes without saying but if this test fails using current master then this PR obviously didn't break that test.

[AaronDonahue]

@singhbalwinder , Thank you, I will try master. I did check the errors and I think I understand them. All of the ne4 mesh tests failed because the land model can't handle 720 cores, that is more cores than there are gridpoints in the land mesh. So that has nothing to do with the PR, just with the submission using too many cores.

The final FAIL, SMS.f09_g16_a.IGCLM45_MLI.anvil_intel, had something to do with elevation classes which again I don't think would be associated with this PR.

[singhbalwinder]

@singhbalwinder , Thank you, I will try master. I did check the errors and I think I understand them. All of the ne4 mesh tests failed because the land model can't handle 720 cores, that is more cores than there are gridpoints in the land mesh. So that has nothing to do with the PR, just with the submission using too many cores.

Thanks. Yes, that would break the simulation and it is not related to this PR, so it should be fine. Just to be sure, you can run SMS_D_P32x1_Ln5.ne4_ne4.FC5 (or a variant of this). This test ensures that each component gets only 32 tasks (_P32x1).

The final FAIL, SMS.f09_g16_a.IGCLM45_MLI.anvil_intel, had something to do with elevation classes which again I don't think would be associated with this PR.

I do not know about this failure. @bishtgautam : Do you expect this test to fail due to elevation classes code?

@AaronDonahue : It would be beneficial if you can copy the error message here. Thanks!

[AaronDonahue]

I could be way off on the elevation classes, I'm basing it on seeing the following warning message:

` get_glc_elevation_classes: WARNING, for glc_pt, topo = 33
-570.381878807675
Topographic height below the lower bound of the lowest elevation class

get_glc_elevation_classes: WARNING, for glc_pt, topo = 34
-758.880539182521
Topographic height below the lower bound of the lowest elevation class

get_glc_elevation_classes: WARNING, for glc_pt, topo = 35
-708.216219656158
Topographic height below the lower bound of the lowest elevation class

get_glc_elevation_classes: WARNING, for glc_pt, topo = 36
-653.435608094268
Topographic height below the lower bound of the lowest elevation class

get_glc_elevation_classes: WARNING, for glc_pt, topo = 37
-566.451650397649
Topographic height below the lower bound of the lowest elevation class`

repeated many times in the acme.log file finished by the following:

=================================================================================== = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES = PID 133393 RUNNING AT b565 = EXIT CODE: 233 = CLEANING UP REMAINING PROCESSES = YOU CAN IGNORE THE BELOW CLEANUP MESSAGES ===================================================================================

[rljacob]

@jayeshkrishna does acme_developer pass on anvil with intel?

[jayeshkrishna]

I haven't tried it (anvil+intel+acme_developer) recently.

[rljacob]

@brhillman did you run acme_developer somewhere for this PR?

[AaronDonahue]

I reran "SMS.f09_g16_a.IGCLM45_MLI" on blues and it passed. So it could be something to do with running on Anvil with Intel, or running that particular test with 720 processors.

[rljacob]

So do you testing on blues or you can skip it if Bill says he ran them already.

[AaronDonahue]

Sounds good, I reran all the failed tests on Blues and just have one more pending, but the others passed, so I think we are good. Sorry to cause a stir with my original fails.

[jayeshkrishna]

Looks like we need some machine-specific PE layouts for anvil + ne4_ne4 .

[brhillman]

@rljacob yes I ran acme_developer. All tests that passed on master on skybridge passed on this branch before and after making the changes, and I did baseline comparisons before and after as well and all were bit for bit consistent. There are some tests failing on skybridge on master right now, but those are unrelated to this PR and we're failing before I made these changes, and are not the same fails reported here.

[rljacob]

Good. In that case, Aaron, you can proceed with integrating this. Anvil pe layouts can be done in a separate PR.