[llvm-dev] [Proposal][RFC] Epilog loop vectorization

[Michael Kuperstein via llvm-dev]

>
>
> broadcasts that get hoisted out of the loop, etc.
>
>
> Can you elaborate? Do you mean that we're hoisting broadcasts outside of
> the trip-count check (i.e. speculatively executing them) and these are
> actually expensive operations? Is this is a cost-model problem being
> exposed by SimplifyCFG (which is hoisting these things outside of the
> trip-count check)?
>

I'm pretty sure I've seen this happen. We're definitely hoisting them out
of the loop (and relying on CGP and/or Loop Sinking to sink them back when
profitable), but I also remember seeing the broadcasts executed in cases
where the the iteration count was 0.

I'll see if I can come up with an example.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170314/0154a762/attachment.html>