(towards #2671) Parallel array privatisation

stfc / PSyclone

Domain-specific compiler and code transformation system for Finite Difference/Volume/Element Earth-system models in Fortran

BSD 3-Clause "New" or "Revised" License

104 stars 27 forks source link

(towards #2671) Parallel array privatisation #2697

Open sergisiso opened 1 month ago

codecov[bot] commented 1 month ago

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 99.86%. Comparing base (652ad61) to head (0d03431).

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #2697 +/- ## ======================================= Coverage 99.86% 99.86% ======================================= Files 354 354 Lines 49046 49108 +62 ======================================= + Hits 48982 49044 +62 Misses 64 64 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

sergisiso commented 3 weeks ago

@arporter It turns out that the compare_output.py that I added in my previous PR fails to properly compare the run.stat because the scientific notation with "D" is not converted to floats by python, which expects an "E". This is not a problem for passthrough, but the OpenMP workflows return different results (I was surprised that we didn't need to exclude any file as ecmwf needed some for reproducibility!). I will fix it in this PR as the OpenMP results are the main focus for this one.

sergisiso commented 1 day ago

Unfortunately the new excluded files for NEMOv5 GPU cause a performance degradation in NEMOv4, so I will wait a bit for #2725 and #2685 as these remove some of the exclusions.

Additionally I can filter by routine name instead of file, as now all of them complete the passthrough.