[PATCH] D134862: [AMDGPU][GISel] Update `isCanonicalized`

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh added a comment.

In D134862#3823116 <https://reviews.llvm.org/D134862#3823116>, @foad wrote:

>> Fixes some regressions introduced in D134857 <https://reviews.llvm.org/D134857> for fdiv.f16 too.
>
> I don't see any real regressions in D134857 <https://reviews.llvm.org/D134857>, just cases where we optimize the "ieee" cases but not the "flush" cases.

Yes, that's what I meant - we had a split in the ieee/flush codegen but this fixes it :p



================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:10202
+    return true;
+  case AMDGPU::G_FNEG:
   case AMDGPU::G_FMINNUM_IEEE:
----------------
foad wrote:
> It seems wrong for FNEG to fall into the "supportsMinMaxDenormModes" check below.
Indeed it was. Fixed it and also added the remaining unary operations while I was at it.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134862/new/

https://reviews.llvm.org/D134862