[PATCH] D113448: [AMDGPU] Check for unneeded shift mask in shift PatFrags.

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp:3887
+
+  const MachineInstr *Opnd1Def =
+      getOpcodeDef(TargetOpcode::G_CONSTANT, MI.getOperand(2).getReg(), *MRI);
----------------
Could use getIConstantVRegVal here to get the APInt value directly?


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp:3887
+
+  const MachineInstr *Opnd1Def =
+      getOpcodeDef(TargetOpcode::G_CONSTANT, MI.getOperand(2).getReg(), *MRI);
----------------
foad wrote:
> Could use getIConstantVRegVal here to get the APInt value directly?
`Opnd1` is a bit confusing because it's MI.getOperand(2). Maybe call it something vague like MaskVal?


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstructions.td:261
+
 foreach width = [16, 32, 64] in {
 defvar mask = !sub(width, 1);
----------------
Maybe change this to `foreach logwidth = [4, 5, 6]` so you can put the definition of csh_mask_#logwidth inside the loop? Or maybe that's impossible because you need to refer to logwidth inside a C++ code fragment?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113448/new/

https://reviews.llvm.org/D113448