[PATCH] D129501: Redefine get.active.lane.mask to allow a more scalar lowering
Philip Reames via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 12 17:19:10 PDT 2022
reames abandoned this revision.
reames added a comment.
As pointed out by @efriedma, this change is unsound. The proposed lowering works for the first unrolled copy of a vectorized loop, but in a tail folded loop the second unrolled copy can validly have a starting IV value greater than the trip count. The expected result is a zero vector. This change lost that.
I'm going to explore other options for achieving the lowering for the first unrolled iteration while being correct for all unrolled copies, but that's probably going to require knowledge from the vectorizer. Regardless, doing it here is wrong since we don't have enough information to tell which unrolled iteration we're dealing with.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D129501/new/
https://reviews.llvm.org/D129501
More information about the llvm-commits
mailing list