add trabbc to exclusions in NEMO

[addy419]

trabbc.f90 produces fort2: terminated by signal 11 on Nvidia Grace-Hopper (Isambard-AI Phase 2)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.95%. Comparing base (2823f9d) to head (e28bcea).
⚠️ Report is 3 commits behind head on master.

❌ Your project status has failed because you have indirect coverage changes. Learn more about Unexpected Coverage Changes and reasons for indirect coverage changes.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3282      +/-   ##
==========================================
- Coverage   99.95%   99.95%   -0.01%     
==========================================
  Files         376      376              
  Lines       53545    53520      -25     
==========================================
- Hits        53523    53497      -26     
- Misses         22       23       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sergisiso]

@addy419 I am happy merging this but it is concerning that we can't replicate the issue as from time to time I do check if the exclusions are still necessary in our end. Could you also let me know if you use any NEMO build keys or del_keys in your tests?

[addy419]

@sergisiso The only difference I see is that I am using cpp -Dkey_nosignedzero instead of add_key "key_nosignedzero". I am not sure if it makes any difference being used on the pre-processor instead of makenemo.

[sergisiso]

And do these also match?

$ cat tests/BENCH/cpp_BENCH.fcm
bld::tool::fppkeys key_si3 key_top key_qco key_vco_1d

$ cat cfgs/ORCA2_ICE_PISCES/cpp_ORCA2_ICE_PISCES.fcm
bld::tool::fppkeys   key_qco key_vco_1d3d key_RK3 key_mpi_off

[arporter]

If it really is an issue with just one file and only on Grace-Hopper it's probably worth raising it with NVIDIA.

[addy419]

@sergisiso I've made no changes to the upstream repo, so both .fcm files are identical. The only keys added/removed are xios, mpi, sea-ice and pisces related https://github.com/ukri-bench/spack-packages/blob/96b31aa54a45abea18412ff63bbd9c31be7520ba/spack_pkgs/spack_repo/ukri_bench/packages/nemo/package.py#L257

[addy419]

@arporter Technically speaking, any terminated by signal 11 is a critical compiler bug where the compiler is trying to do something that the OS does not like and hence is killed by the system. Since now we have CI and a spack package that is able to reproduce these, it is worthwhile to report them all to Nvidia

PSyclone/examples/nemo/scripts/omp_gpu_trans.py

Line 101 in 2823f9d

# Signal 11 issues

[sergisiso]

Wouldn't depends_on("py-psyclone", type="build", when="+omp") use the latest release instead of master?

[addy419]

@sergisiso Ah that, I am overriding the dependency with py-pysclone@master.