Use default binding of tasks to cores in non-xstrid PE-layouts on Frontier

[amametjanov]

Use default binding of tasks to cores in non-xstrid PE-layouts on Frontier.

Fixes #8557

[BFB]

---

Testing: this reverts to prior default (cyclic/round-robin) distribution at 8 mpi tasks per node; and sets the needed --distribution=*:block at fully occupied node with 56 mpi tasks per node.

[Copilot]

Pull request overview

This PR updates the Frontier machine configuration to avoid forcing Slurm’s
--distribution=*:block in non-fully-occupied PE layouts, restoring the prior
(default/cyclic) placement behavior at lower MPI-task counts while still
enabling *:block when the node is fully occupied (motivated by
#8557).

Changes:

• Make Frontier GPU_BIND_ARGS conditionally include --distribution=*:block
  based on the configured MAX_MPITASKS_PER_NODE.
• Preserve --gpu-bind=closest behavior for HIP builds.

[grnydawn]

@amametjanov , have you seen this error. When I create the test case of SMS_P256.ne256pg2_ne256pg2.F2010-SCREAMv1 on your branch, I got this:

Output directory : /lustre/orion/cli115/scratch/grnydawn/e3sm_tests/AzFix
ERROR: Command: '/usr/bin/xmllint --xinclude --noout --schema /autofs/nccs-svm1_home1/grnydawn/repos/github/E3SM/cime/CIME/data/config/xml_schemas/config_machines.xsd /autofs/nccs-svm1_home1/grnydawn/repos/github/E3SM/cime_config/machines/config_machines.xml' failed with error '/autofs/nccs-svm1_home1/grnydawn/repos/github/E3SM/cime_config/machines/config_machines.xml:1855: element CMAKE_BACKEND: Schemas validity error : Element 'CMAKE_BACKEND': This element is not expected. Expected is one of ( TESTS, NTEST_PARALLEL_JOBS, BATCH_SYSTEM ).
/autofs/nccs-svm1_home1/grnydawn/repos/github/E3SM/cime_config/machines/config_machines.xml fails to validate' from dir '/autofs/nccs-svm1_home1/grnydawn/repos/github/workflow/e3sm'

The above error in config_machines.xml is not related to Frontier. However, it seems that your branch is quite far from the official master branch.

[amametjanov]

Mm, there is this at that line 1855:

1838   <machine MACH="mappy">
...
1854     <BATCHED_BUILD>FALSE</BATCHED_BUILD>
1855     <CMAKE_BACKEND>ninja</CMAKE_BACKEND>

which is coming from PR #8532 . This branch is off latest master from today Jul-20. Is local cime submod pointing to ed580a836, and is there a local binary in externals/ninja/bin/ninja?

[grnydawn]

Tested successfully with the reported case, SMS_P256.ne256pg2_ne256pg2.F2010-SCREAMv1. The e3sm_eamxx_v1_medres test suite also passed. Not all tests in the suite were completed due to a network disconnection, but all of the tests that ran passed. Approved.