MPI: fix haloupdate inside factor conditional #2395
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #2395 +/- ##
==========================================
- Coverage 86.69% 78.27% -8.42%
==========================================
Files 234 234
Lines 44383 44410 +27
Branches 8216 8221 +5
==========================================
- Hits 38477 34764 -3713
- Misses 5185 8869 +3684
- Partials 721 777 +56 ☔ View full report in Codecov by Sentry. |
devito/passes/iet/mpi.py
Outdated
| @@ -182,7 +183,12 @@ def rule2(dep, hs, loc_indices): | |||
| # within the same Conditional, otherwise we would break the control | |||
| # flow semantics | |||
| if cond_mapper.get(hs0) != cond_mapper.get(hs1): | |||
| continue | |||
| cond = cond_mapper.get(hs1).symmetric_difference(cond_mapper.get(hs0)) | |||
There was a problem hiding this comment.
would it be neater to de-nest it by re-writing it as (pseudocode):
cond = cond_mapper.get(hs1).symmetric_difference(cond_mapper.get(hs0))
if len(cond) > 1:
<explanation>
continue
elif len(cond) == 1 and isinstance(cond.pop().condition, GuardFactorEq):
<explanation>
continue
There was a problem hiding this comment.
Not sure it'd be neater. Would need either a vague else or do the Guard first but make a copy so that the pop doesn't break the >1 I'll think about it
There was a problem hiding this comment.
cleaner to just filter when making the cond_mapper
On top of #2394