[PATCH] D109760: [GVN] Simple GVN hoist

Wed Sep 15 06:33:02 PDT 2021

xbolva00 added a subscriber: fhahn.
xbolva00 added a comment.

In D109760#3001467 <https://reviews.llvm.org/D109760#3001467>, @chill wrote:

> In D109760#3000216 <https://reviews.llvm.org/D109760#3000216>, @xbolva00 wrote:
>
>> I believe the latest issues with gvnhoist were regressions, what about to take a look at this pass again and possibly enable (some “obviously” profitable) parts of gvnhoist again?
>>
>> Did you run spec benchmarks with trunk+gvnhoist?
>
> I did and indeed I got even better speedup, but with `-O2 -mllvm -enable-gvnhoist`. With `-O3` there was no improvement (and maybe a regression) (due to unrolling, see thread on the mailing list).
>
> In D109760#3001010 <https://reviews.llvm.org/D109760#3001010>, @junparser wrote:
>
>> As for mcf, I guess it may casued by large register pressure due to hoist some expression from some huge successor.  So simple cost model is necessary.
>
>
>
> In D109760#3000184 <https://reviews.llvm.org/D109760#3000184>, @SjoerdMeijer wrote:
>
>> Quick question on the perf numbers. Do you know what the problem is with MCF? Would be good if we can avoid this regression, then the numbers are even better....
>
> No idea why MCF regresses. My plan was to look at MCF regressions and why `-O2 -enable-gvnhoist` gives better numbers than `-O3` + this patch.

There is ongoing work to enhance slp with memory versioning by @fhahn - it should help

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D109760/new/

https://reviews.llvm.org/D109760