teams: input split stride is relative to parent

[davidozog]

This is a potential fix for #1151

@zhongchen530 - If you are able, can you please check if this change results in the expected behavior for shmem_team_split_strided? I'm also curious if you can please share any details about your use-case that identified this issue. Or perhaps you are simply exploring and/or learning the API?

[zhongchen530]

This is a potential fix for #1151

@zhongchen530 - If you are able, can you please check if this change results in the expected behavior for shmem_team_split_strided? I'm also curious if you can please share any details about your use-case that identified this issue. Or perhaps you are simply exploring and/or learning the API?

I apologize for the delayed reply. I discovered this issue when I tried to write a test where I recursively split teams into even and odd teams, like even of even and odd of even etc. I was going to rebuild shmem using the pull request and was waiting for my virtual cluster to be cloned before doing so. Unfortunately, there was some issue with cloning of the VM on my side and I wouldn't be able to change the shmem version I am using until that is resolved.

[davidozog]

@zhongchen530 - no problem. This change is safe to merge now, so please feel free to test on the updated main branch.

Our testing does not yet cover this case, so feel free to contribute your test to https://github.com/openshmem-org/tests-sos.git. Until then I plan to add a very simple test asap.

[zhongchen530]

Hi @davidozog, I finally got permission to rebuild shmem on my cluster with the updated main branch. The original stride issue related to shmem_team_split is now fixed. However, when I ran make test to run the test suite, shmem_team_split_2d is now failing to pass. I ran the test with 4 PEs. I am afraid the change have broken shmem_team_split_2d call or that the test itself was wrong in the first place.