[PATCH] D48431: AMDGPU: Force skip over s_sendmsg and exp instructions

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In https://reviews.llvm.org/D48431#1141832, @nhaehnle wrote:

> In https://reviews.llvm.org/D48431#1139259, @arsenm wrote:
>
> > We really need to invert how this pass works
>
>
> Yes. Although the code itself is really fine, isn't it? One way to think about it is that SI_MASK_BRANCH has different semantics before this pass and after this pass, and we should split it into two pseudo instructions to express this difference in semantics.
>
> Anyway, I'd say that should be a separate change.


I can see it being useful in the future to factor this condition into a TII->hasEffectsWhenEXECEmpty() or something like that predicate function


Repository:
  rL LLVM

https://reviews.llvm.org/D48431