[PATCH] D85530: [AMDGPU] Don't cluster stores

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D85530#2204974 <https://reviews.llvm.org/D85530#2204974>, @critson wrote:

> I tried this change with game traces on GFX10.
>
> I could not convince myself that there was any statically significant changes in performance.
> Some small gains, some small losses, nothing outside the range of variance.
>
> I do however wonder if this would be better as a tuning option?
> The same could also be said for load clustering as essentially you are trading VGPR pressure for VMEM access efficiency (e.g. stalls).

Did you try this with xnack enabled? This will reduce the number of soft clauses formed for stores


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D85530/new/

https://reviews.llvm.org/D85530