[llvm] [AMDGPU][True16][CodeGen] workaround sgpr folding in true16 (PR #128929)

[Joe Nash via llvm-commits]

================
@@ -576,6 +579,11 @@ bool SIFoldOperandsImpl::updateOperand(FoldCandidate &Fold) const {
   }
 
   MachineOperand *New = Fold.OpToFold;
+  // TODO: Temporarily allow folding from SGPRs to 16-bit VGPRs.
----------------
Sisyph wrote:

I would say remove the first line of the comment. We don't have plans to implement 16-bit SGPRs
```suggestion
```

https://github.com/llvm/llvm-project/pull/128929