[PATCH] D64333: AMDGPU: Move waitcnt intrinsic to instruction definition pattern

Thu Jul 11 15:42:28 PDT 2019

arsenm added a comment.

In D64333#1581767 <https://reviews.llvm.org/D64333#1581767>, @arsenm wrote:

> In D64333#1581763 <https://reviews.llvm.org/D64333#1581763>, @hakzsam wrote:
>
> >
>
>
>
>
> >   call void @llvm.amdgcn.s.waitcnt(i32 49279) #2
>
> This isn't a 16-bit value, so this isn't a correct use. Previously it would have been truncated. I'm also surprised there is a user of this that doesn't just use 0? In context it looks like you should be using a fence instruction instead?

Well, it is, just a uint16. Should probably change this to checking if it' unsigned 16

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64333/new/

https://reviews.llvm.org/D64333