[PATCH] D119425: [AMDGPU] Make enable-flat-scratch a subtarget feature

Stanislav Mekhanoshin via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Thu Feb 10 08:55:46 PST 2022


rampitec accepted this revision.
rampitec added a comment.
This revision is now accepted and ready to land.

In D119425#3310662 <https://reviews.llvm.org/D119425#3310662>, @sebastian-ne wrote:

> In D119425#3310638 <https://reviews.llvm.org/D119425#3310638>, @foad wrote:
>
>> Just curious: why is it not enabled by default on subtargets that support it? Then you would only need the attribute if you wanted to turn it off for some reason.
>
> If we do so, I think it only makes sense for gfx10.3+ (there are a couple hardware bugs before that). It would need benchmarking on the compute side.
> For graphics there are almost no differences when using flat scratch, but there is one improvement in a case where we spill a lot.

Right, support is incomplete without _ST addressing mode, so before gfx1030 it can fail under a high pressure.

LGTM


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D119425/new/

https://reviews.llvm.org/D119425



More information about the llvm-commits mailing list