https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121290

--- Comment #23 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
(In reply to Dhruv Chawla from comment #18)
> Hi Tamar, here is a (somewhat-)minimized repro for the RAJAPerf kernel that
> Ramana mentioned: https://godbolt.org/z/jh8Ke6hPx - I'll also attach the
> code / assembly before and after.

Thanks, the deep loop nesting makes the heuristic apply to the "outer" loop as
well as that one is also a nested loop. I think the penalty heuristic should
only apply to the inner most nested loop.  At the moment the penalty is applied
to both loops it considers for vectorization making both unprofitable.

I'll adjust the condition when back on Monday.

Reply via email to