[llvm] [AMDGPU][GlobalISel] Fold zext(carry) into add/sub-with-carry (PR #224093)

[Matt Arsenault via llvm-commits]

================
@@ -393,6 +398,35 @@ bool AMDGPUPostLegalizerCombinerImpl::matchCombine_s_mul_u64(
   return false;
 }
 
+bool AMDGPUPostLegalizerCombinerImpl::isCarryOut(Register Reg) const {
+  const MachineInstr *Def = MRI.getVRegDef(Reg);
+  if (!Def)
+    return false;
+
+  switch (Def->getOpcode()) {
+  case TargetOpcode::G_UADDO:
+  case TargetOpcode::G_UADDE:
+  case TargetOpcode::G_USUBO:
+  case TargetOpcode::G_USUBE:
+    return Def->getOperand(1).getReg() == Reg;
+  default:
+    return false;
+  }
+}
+
+bool AMDGPUPostLegalizerCombinerImpl::shouldFoldCarryIntoAdd(
+    Register Src) const {
+  // Let right_identity_zero combine delete this instead.
+  if (mi_match(Src, MRI, m_SpecificICst(0)))
+    return false;
+
+  // When both operands are a carry only the right hand one is folded, so that
+  // the two carries never have to be live at the same time.
+  Register Carry;
+  return !(mi_match(Src, MRI, m_GZExt(m_Reg(Carry))) &&
----------------
arsenm wrote:

extra parens 

https://github.com/llvm/llvm-project/pull/224093