[PATCH] D57681: [InstCombine] Cleanup the TFE/LWE check in AMDGPU SimplifyDemanded

Matt Arsenault via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Feb 4 13:15:32 PST 2019


arsenm added a comment.

In D57681#1383911 <https://reviews.llvm.org/D57681#1383911>, @nhaehnle wrote:

> In D57681#1383030 <https://reviews.llvm.org/D57681#1383030>, @arsenm wrote:
>
> > Does it actually matter? I thought since this needs to be a constant, this just needs to not crash
>
>
> It's certainly a bit on the paranoid side. I was considering the possibility that somebody passes a value that only becomes constant after inlining.


I've been hoping to start working soon on a way to define which intrinsic operands are required to be constants to avoid hacks in global isel, so hopefully we won't need to worry about this much longer.



================
Comment at: test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll:2420
+
+declare {<4 x float>, i32} @llvm.amdgcn.image.load.1d.sl_v4f32i32s.i32(i32, i32, <8 x i32>, i32, i32) #1
+
----------------
nhaehnle wrote:
> arsenm wrote:
> > There's an _ in the name?
> You mean the `sl_v4f32i32s` part? That's what the intrinsic name got canonicalized to for some reason. I'm not really uptodate on the mangling of struct types...
Yes, I wasn't sure how that works


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57681/new/

https://reviews.llvm.org/D57681





More information about the llvm-commits mailing list