[PATCH] D93963: [GlobalISel][AMDGPU] Lower G_UMULO/G_SMULO

[Pushpinder Singh via Phabricator via llvm-commits]

pdhaliwal added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp:606
 
+  getActionDefinitionsBuilder({G_UMULO, G_SMULO}).lower();
+
----------------
pdhaliwal wrote:
> arsenm wrote:
> > Would you get better results if you:
> > 1. scalarized vectors first
> > 2. Promoted small scalar types first?
> In both cases, the MULO is somehow considered legal by compiler. For e.g. even for s32, expansion does not occur, compiler is directly using S/UMULO instructions. I am investigating this.
It was due to missing corresponding definition of widenScalar for this operation. I tried implementing one for UMULO as it was easier. Didn't see any better result.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D93963/new/

https://reviews.llvm.org/D93963