[llvm] [AMDGPU][GlobalISel] Combine for breaking s64 and/or into two s32 insts (PR #151731)

[Matt Arsenault via llvm-commits]

================
@@ -516,3 +517,9 @@ bool AMDGPUCombinerHelper::matchCombineFmulWithSelectToFldexp(
 
   return true;
 }
+
+bool AMDGPUCombinerHelper::matchConstantIs32BitMask(Register Reg) const {
+  const KnownBits &Known = VT->getKnownBits(Reg);
+  return Known.One.extractBits(32, 0).isAllOnes() ||
----------------
arsenm wrote:

Could use isShiftedMask and check offsets 

https://github.com/llvm/llvm-project/pull/151731