-
Notifications
You must be signed in to change notification settings - Fork 28
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Bugfix for regional runs when dycore is compiled in double precision #25
Conversation
…r bit-for-bit identical results on Cheyenne with Intel 19.1 and SGI MPT 2.19
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
buf1-4 are now allocatable variables, but the count argument for irecv is still using the ibufexch parameter. I recommend changing the count in the irecv be the size of buf1 and buf2 respectively.
Good catch. Thanks very much. Will cleanup! |
I fixed this - removed ibufexch entirely and used size(bufX) in the messages (since the count isn't known at the time the first MPI commands occur). I reran the tests on Cheyenne, it compiles and runs in DEBUG mode for both 32bit and 64bit without crashing, and the PROD runs in 64-bit give b4b identical results from run to run (which was the issue that triggered this PR in the first place). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Changes look consistent.
I would recommend some inline comments or even protection logic to ensure buffering/unbuffering goes as planned. Any future modification to the logic needs to understand index changes must be thoroughly tested - especially when packing multiple variables into the same buffer.
@bensonr does this look better? It still works on Cheyenne as expected. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good
Associated PRs: For regression testing, see ufs-community/ufs-weather-model#155. |
updated RELEASE.md for the 2020.02 release (issue #25)
This PR fixes issue #24.
It also adds one additional line to the initial printout, which we found useful to make sure that all constants and tracers are set up correctly.