[PATCH] D115261: [LV] Disable runtime unrolling for vectorized loops.

Wed Jan 4 14:48:45 PST 2023

fhahn added a comment.

>> In D115261#3807958 <https://reviews.llvm.org/D115261#3807958>, @xbolva00 wrote:
>>
>>> Are you gonna land this patch? Or any blockers?
>>
>> Not any longer with 9758242046b3 <https://reviews.llvm.org/rG9758242046b3cdce6fb713acb6d3f5bfaa933a47> landed.
>
> (not any longer //what//? :)

As reply to @xbolva00  no pending blocker patches I was aware of :)

In D115261#4023078 <https://reviews.llvm.org/D115261#4023078>, @lebedev.ri wrote:

> Hmm.
> So i *was* thinking of something else.
> It's possible that LV's unroll heuristic
> may need further tuning, but in general
> please proceed with this.

Thanks for taking a look! Interleaving in LV clones vector operations. If interleaving causes spilling, this is a bug in the cost model and should be fixed. There will be some cases where we won't perform interleaving but could do runtime unrolling, but it is unlikely to give performance benefits (at least on recentish OOO CPUs). If there are cases where this causes regressions, I'll take a look!

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115261/new/

https://reviews.llvm.org/D115261