[PATCH] D134723: [AMDGPU] Set memory bound occupancy based on addressable VGPRs

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added a comment.

In D134723#3819661 <https://reviews.llvm.org/D134723#3819661>, @critson wrote:

> So this was based on dealing with the temporary 128 VGPR limit on GFX11, which has now been removed.
> On that basis I will probably abandon this revision; however, I do still have concerns about this magic number 4.

The magic number was borrowed from SC and is empirical. The experience is that increasing occupancy beyond 4 for memory bound kernels does not help performance in average. But it is not linked to spilling in any way, just to cache thrashing. That is likely this number may differ for later targets, but again not linked to spilling.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134723/new/

https://reviews.llvm.org/D134723