[PATCH] D136448: [AMDGPU][GISel] Add llvm.amdgcn.icmp selection

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp:1237
+  //  in i1 handling between GISel and DAG.
+  if (Size == 1)
+    Size = 32;
----------------
Pierre-vh wrote:
> Pierre-vh wrote:
> > This is obviously wrong as we only want to compare the first bit in each VGPR, but I don't know what instruction to use here yet.
> > 
> > Which instruction do I need to use here? V_CNDMASK ?
> > 
> CNDMASK doesn't work because it constrains the operands to 64 bits, but the i1 are widened to 32.
> We need something that would do `dst[threadI] = v0.i1`, if I understand correctly?
Where are you seeing i1 VGPRs? Those should just not happen. At one point I might have been trying to handle that, but I think it's basically wrong for it to see the selector 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136448/new/

https://reviews.llvm.org/D136448