After PR #11592, the Chapel compiler only offers vectorization hints for parallel loops with reductions if the reduction is a + reduction over a scalar.
This could be improved:
Work with the RV project to describe the custom reductions in terms RV understands
Note that the LLVM LoopVectorizer could be taught about custom reductions too, it views these as a kind of recurrence, so it's RecurrenceDescriptor that would need to be updated.
After PR #11592, the Chapel compiler only offers vectorization hints for parallel loops with reductions if the reduction is a
+
reduction over a scalar.This could be improved: