For SCM cases, correct how we store the computed area to write in eam history file #6585
Conversation
…survive until writing to netcdf file
correct what looks like an incorrect if conditional for scm and convert it to positive. only nullify is scm
ndkeen
added
GNU
|
Note the first commit here is cleaning up white-space. I also think there is incorrect logic for when SCM is enabled. I changed Nope that was wrong as I misunderstood what scm_multcols was for. Switched back to way it was. |
|
I think I incorrectly did the logic for scm -- will fix. |
|
…(PR #6585) For SCM cases, we compute area and register it, but as it's a local copy, the value is not available at writing time to netcdf. This results in the code writing a value for area that is uninitialized and different compilers can have different behavior, which could result in a DIFF between runs (and the incorrect area written to the file). BFB for all cases, except SCM. And the results are still BFB for SCM, just affects the area written to file. Fixes #6556
|
merged to next However, it will still show a DIFF. We will have to bless and then see if it stays passing after that... |
|
merged to master, submitted bless request (for pm-cpu with GNU only) |
|
@ndkeen did you bless on mappy? Its still diffing. |
|
No I did not. I was actually not aware that this was an issue on any other machine. I can't even login to mappy. Glad it fixed more than one. Should I make bless request? |
|
Yes please do. |
For SCM cases, we compute area and register it, but as it's a local copy, the value is not available at writing time to netcdf.
This results in the code writing a value for area that is uninitialized and different compilers can have different behavior, which could result in a DIFF between runs (and the incorrect area written to the file).
BFB for all cases, except SCM. And the results are still BFB for SCM, just affects the area written to file.
Fixes #6556