For pm-cpu, upgrade Intel compiler to 2023.2.0 as well as other modules for Intel only
#6596
Conversation
|
|
Is this BFB ? |
|
I ran e3sm_prod and compared with baselines. I also run e3m_integration with older version of repo -- both showed all tests passing. But I might not expect all cases to be BFB. Should we: |
|
Yes run the production suite with a comparison against the existing baselines. |
|
I just reminded myself that I had already done that -- running with |
…t (PR #6596) For pm-cpu, move from intel/2023.1.0 to intel/2023.2.0. Updating to this version allows us to also update several other module versions. These are the updates to other modules we are doing at the same just for Intel compiler for now: PrgEnv-intel/8.3.3 PrgEnv-intel/8.5.0 craype/2.7.20 craype/2.7.30 cray-mpich/8.1.25 cray-mpich/8.1.28 cray-hdf5-parallel/1.12.2.3 cray-hdf5-parallel/1.12.2.9 cray-netcdf-hdf5parallel/4.9.0.3 cray-netcdf-hdf5parallel/4.9.0.9 cray-parallel-netcdf/1.12.3.3 cray-parallel-netcdf/1.12.3.9 While this change does not address a known issue, and the versions are higher than machine defaults, this is in preparation for upcoming SW changes. Also do not expect any significant performance changes, but more testing warranted. So far, testing shows the results are BFB, but would rather not assume the PR is BFB as it changes compiler version.
|
merged to next |
|
I think you can merge this to master. |
|
ok. yea I was trying to test more cases with it. In fact, I think a scream case was not BFB after this change. But we can cross that bridge when we merge. |
|
OK, for a basic I can merge this as-is and continue testing? |
|
I think so since probably no one is using pm-cpu for SCREAM v1 simulations. If they are, they're probably using the scream repo. |
|
I tried a recent HR case Will merge and we will have to remember to address non-bfb issues with scream repo. |
For pm-cpu, move from
intel/2023.1.0tointel/2023.2.0.Updating to this version allows us to also update several other module versions.
These are the updates to other modules we are doing at the same just for Intel compiler for now:
While this change does not address a known issue, and the versions are higher than machine defaults, this is in preparation for upcoming SW changes. Also do not expect any significant performance changes, but more testing warranted.
So far, testing shows the results are BFB, but would rather not assume the PR is BFB as it changes compiler version.