[PATCH] D97922: [AMDGPU] Keep skip branch for ds instructions

[Carl Ritson via Phabricator via llvm-commits]

critson added a comment.

In D97922#2603323 <https://reviews.llvm.org/D97922#2603323>, @arsenm wrote:

> If I remember my cycle counts correctly, with an untaken branch this will be about breakeven for a single DS op?

I believe this is about right.
However since we allow 12 instructions and rarely encounter a DS op in isolation, I think the branch is generally cheaper (assuming cache hit).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D97922/new/

https://reviews.llvm.org/D97922