[PATCH] D133441: [SLP] Look ahead for mutual horizontal reductions.

[Vasileios Porpodas via Phabricator via llvm-commits]

vporpo added a comment.

> Ah, I see what do you mean. But this is not the best place for gathering the reduced values, they must be gathered during an attempt to build the tree, since they might be not vectorized but gathered. If they are gathered, you cannot treat them as a future vector.

Unlike the similar situation with store seeds, the reduction does not have the issue with the "good" lane ordering, and there are no dependency issues that could prohibit them to be scheduled in a specific way, so this looks a lot safer. If the reduction user is connected to a vectorizable TreeEntry and the current tree gets vectorized, isn't it guaranteed that the tree rooted at the reduction will also get vectorized?

> Fair enough. Can you recommend a better solution that could fix both D132773 <https://reviews.llvm.org/D132773> and this one?

One solution could be to collect the seeds that resulted in failed trees due to external users (and perhaps filter them more by checking if the users are actually seeds), and revisit the seeds once again. I think this is more or less what Alexey is describing.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D133441/new/

https://reviews.llvm.org/D133441