[llvm] [AMDGPU] Use range-based for loops. NFC. (PR #99047)

[Matt Arsenault via llvm-commits]

================
@@ -3642,8 +3641,8 @@ SDValue SITargetLowering::LowerCall(CallLoweringInfo &CLI,
 
   if (Callee.isUndef() || isNullConstant(Callee)) {
     if (!CLI.IsTailCall) {
-      for (unsigned I = 0, E = CLI.Ins.size(); I != E; ++I)
-        InVals.push_back(DAG.getUNDEF(CLI.Ins[I].VT));
+      for (auto &Arg : CLI.Ins)
----------------
arsenm wrote:

I would appreciate not using auto here 

https://github.com/llvm/llvm-project/pull/99047