Conversation
This PR implements multiple diagnostic bucket reset periods for total totprcpb and convective cnvprcpb preciptitation variables by combining different accumulation amounts into an additional dimension with number_of_diagnostic_buckets. More bucket diagnostics (e.g., snow, ice, freezing ice, graupel) could be added to this list. Note this PR does not affect the accumulation of physics tendencies or radiative fluxes. This PR also removes unused bucket-related variables. No need to pass fhzero and cnvprcpb as arguments in GWD and SCNV schemes, respectively.
…ation/radiation_aerosols.f to fix b4b differences with Intel oneAPI 2025.3.0 (NCAR#36) Disable OpenMP for certain computations in init phase in physics/Radiation/radiation_aerosols.f to fix bit-for-bit differences in NEPTUNE with Intel oneAPI 2025.2.1+. Also: initialize a few more variables to make debugging easier.
grantfirl
requested review from
dustinswales,
lisa-bengtsson,
mdtoyNOAA and
rhaesung
as code owners
|
SCM testing is in here (PASSED with no RT changes): NCAR/ccpp-scm#640 |
|
I'm getting several unexpected failures on UFS RTs when combining this with an update from NCAR/main (see https://github.com/grantfirl/ccpp-physics/tree/NCAR_sync_20251112). This will take some investigating before we can merge this. So far, about halfway through the list, all failures are with the intel compiler (oneapi 2024.2.1) used on Ursa. @climbfuji Is this expected, in your opinion? |
|
List of failed UFS RTs: control_flake intel RT output directory: /scratch3/BMC/gmtb/Grant.Firl/stmp2/Grant.Firl/FV3_RT/rt_3405751 So far, for the tests I've checked, there aren't any compilation/runtime failures; everything looks to be different results. I have not looked at magnitude of differences. |
|
@grantfirl While not expected, I am not surprised either. Given that we are addressing b4b differences when using multiple threads arising from these sections of the code with newer versions of the compiler, it is not surprising to me that older versions (that produced the same results with OpenMP threading) now produce different (albeit at least consistent w.r.t. number of threads) results. |
|
I updated the list of failures. It includes tests from intel, gnu, and intelllvm. All of them use MPI, so the change in radiation_aerosols should affect all test compilers, I guess. |
|
I confirmed that the failing RTs are only due to the removed openmp statements. I'll go ahead and merge and make a note of this in the UWM PR. |
2 participants
Description of Changes:
See #1175 and #1176
Tests Conducted:
SCM RTs, UFS RTs
Dependencies:
None
Documentation:
None
Issue (optional):
Closes #1170
Also related to ESCOMP/ESMStandardNames#130
Contributors (optional):
@matusmartini @climbfuji