[llvm] AMDGPU: Fix assembler asserting on expressions in vop3 instructions (PR #100103)

[Jay Foad via llvm-commits]

================
@@ -8649,7 +8649,7 @@ void AMDGPUAsmParser::cvtVOP3(MCInst &Inst, const OperandVector &Operands,
       Op.addRegOrImmWithFPInputModsOperands(Inst, 2);
     } else if (Op.isImmModifier()) {
       OptionalIdx[Op.getImmTy()] = I;
-    } else if (Op.isRegOrImm()) {
+    } else if (Op.isRegOrImm() || Op.isExpr()) {
       Op.addRegOrImmOperands(Inst, 1);
     } else {
       llvm_unreachable("unhandled operand type");
----------------
jayfoad wrote:

Maybe don't bother checking the type here, just leave it to assert inside `addRegOrImmOperands`?
```suggestion
    } else {
      Op.addRegOrImmOperands(Inst, 1);
```

https://github.com/llvm/llvm-project/pull/100103