[PATCH] D82227: SLP: honor requested max vector size merging PHIs

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added a comment.

In D82227#2414446 <https://reviews.llvm.org/D82227#2414446>, @rampitec wrote:

> WRT the testcase, it is really the testcase in the patch with a single modification right before ret void:
>
>   store float %phi31, float* undef
>   ret void
>
> That makes the value not dead and then if I feed opt's output into llc I either get spilling with wide vectors (first case) or not (second case):
>
>   opt -slp-vectorizer -slp-max-reg-size=1024 -S < slp-max-phi-size.ll | ~/work/llvm/rel/bin/llc -march=amdgcn -mcpu=gfx900
>   opt -slp-vectorizer -slp-max-reg-size=256 -S < slp-max-phi-size.ll | ~/work/llvm/rel/bin/llc -march=amdgcn -mcpu=gfx900

You can check it here: D92047 <https://reviews.llvm.org/D92047>
Not sure how appropriate is it to commit a target test using llc into the Transforms directory.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82227/new/

https://reviews.llvm.org/D82227