[llvm] [WIP][AMDGPU] Support the RF inline asm constraint (PR #195123)

[Matt Arsenault via llvm-commits]

================
@@ -10658,8 +10658,14 @@ static bool prepareDAGLevelOperands(ConstraintDecisionInfo &Info,
         assert((OpInfo.isIndirect ||
                 OpInfo.ConstraintType != TargetLowering::C_Memory) &&
                "Operand must be indirect to be a mem!");
+
         assert(InOperandVal.getValueType() ==
-                   TLI.getPointerTy(DAG.getDataLayout()) &&
+                   TLI.getPointerTy(DAG.getDataLayout(),
+                                    OpInfo.CallOperandVal
+                                        ? dyn_cast<PointerType>(
----------------
arsenm wrote:

```suggestion
                                        ? cast<PointerType>(
```

Unchecked dyn_cast 

https://github.com/llvm/llvm-project/pull/195123