Hybrid coordinate enhancements 3 #316
Conversation
…faces for hybrid vertical coordinate to make it less sensitive to bathymetry variations.
…ope vector times Brunt-Vaisala frequency in the vicinity of bathymetry to make Eady growth rate estimates more robust.
| @@ -65,8 +65,8 @@ | |||
| <type>char</type> | |||
| <valid_values>isopyc_bulkml,cntiso_hybrid</valid_values> | |||
| <default_value>isopyc_bulkml</default_value> | |||
| <group>run_component_blom</group> | |||
| <file>env_run.xml</file> | |||
| <group>build_component_blom</group> | |||
There was a problem hiding this comment.
Do we do any tests that verify that OMP works - such as that changing the number of threads gives bfb results?
I have run into some perplexing differences when I ran an MPI case with threads = 1 and with threads = 2.
I need to do some more testing before raising an issue. But I'm wondering what tests are done to verify the functionality of OMP threading?
There was a problem hiding this comment.
Does this mean that hybrid coordinates would not work in release v.1.4.0?
There was a problem hiding this comment.
It should work - my comment is only that the xml variable BLOM_VCOORD should be in env_build.xml and not in env_run.xml - since it sets dimensions.
That said - I really think we need to expand the test list to include hybrid coordinates and more testing of OMP and MPI tasks variations.
There was a problem hiding this comment.
Hybrid coordinates worked in v1.4.0 by modifying BLOM_VCOORD in env_run.xml, but you would have to remember to do it before building.
|
Good news - I have run a test which successfully passed and checks bfb across different thread counts: |
|
All CI tests with intel compiler fails, but don't see anything in this PR that should cause this change. The |
|
Also the PET test PET_Ld3.T62_tn14.NOINYOC.betzy_intel.blom-hamocc1 passes. |
|
Note - that I did these tests with dev1.4.0.1 and not this PR. |
I also saw the failed CI tests. Many of the tests failed with "configure: error: Fortran could not compile .f90 files". It seems unlikely that this PR is causing this and that the issue is on the CI system. |
|
I had the exact same CI failures in my BLOM refactor pushes to my fork. So its definitely not this PR. |
|
Who will be making the v1.5.0 tag? That would be really helpful to have as soon as possible. |
|
I can make a draft release. |
Hybrid coordinate enhancements that includes:
Model results change when using hybrid coordinate, but are unchanged with isopycnic coordinate.