[llvm] [AMDGPU] Model ordered XDL writes in expert scheduling (PR #218326)

[Jay Foad via llvm-commits]

================
@@ -11831,3 +11831,17 @@ bool SIInstrInfo::isXDL(const MachineInstr &MI) const {
 
   return AMDGPU::getMAIIsGFX940XDL(Opcode);
 }
+
+bool SIInstrInfo::isVAVDSTOrderedXDL(const MachineInstr &MI) const {
+  constexpr unsigned MinOrderedXDLVAVDSTVGPRs = 8;
+  if (!AMDGPU::isGFX1250(ST) || !isXDL(MI))
+    return false;
+
+  int DstIdx = AMDGPU::getNamedOperandIdx(MI.getOpcode(), AMDGPU::OpName::vdst);
+  if (DstIdx < 0)
+    return false;
----------------
jayfoad wrote:

Can this happen? If not, you can use `getNamedOperand`

https://github.com/llvm/llvm-project/pull/218326