[PATCH] D78003: [AMDGPU] Fix llvm.amdgcn.div.scale lowering

[Austin Kerbow via Phabricator via llvm-commits]

kerbowa added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.div.scale.ll:21
 ; GFX7-NEXT:    s_waitcnt vmcnt(0)
-; GFX7-NEXT:    v_div_scale_f32 v0, s[2:3], v0, v0, v2
+; GFX7-NEXT:    v_div_scale_f32 v0, s[2:3], v2, v0, v2
 ; GFX7-NEXT:    s_mov_b64 s[2:3], s[6:7]
----------------
arsenm wrote:
> kerbowa wrote:
> > arsenm wrote:
> > > I checked these were the same order as the final DAG output, so are these inconsistent now?
> > DAG lowering was wrong. I thought we never used the intrinsic directly so it didn't matter. 
> I think this is really "comment is wrong". In case there are any users, we should maintain the current behavior
I thought we were sure it wasn't being used.. Seems like we would have heard about it otherwise.

Anyway I agree it's safer to change the comment and the GlobalISel FDIV lowering so I'll abandon this.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D78003/new/

https://reviews.llvm.org/D78003