[llvm] [AMDGPU] Enable serializing of allocated preload kernarg SGPRs info (PR #168374)

[Quentin Colombet via llvm-commits]

================
@@ -718,6 +717,20 @@ convertArgumentInfo(const AMDGPUFunctionArgInfo &ArgInfo,
   Any |= convertArg(AI.WorkItemIDY, ArgInfo.WorkItemIDY);
   Any |= convertArg(AI.WorkItemIDZ, ArgInfo.WorkItemIDZ);
 
+  // Write FirstKernArgPreloadReg separately, since it's a Register,
+  // not ArgDescriptor.
+  if (ArgInfo.FirstKernArgPreloadReg) {
+    Register Reg = ArgInfo.FirstKernArgPreloadReg;
+    if (Reg.isPhysical()) {
+      yaml::SIArgument SA = yaml::SIArgument::createArgument(true);
+      raw_string_ostream OS(SA.RegisterName.Value);
+      OS << printReg(Reg, &TRI);
----------------
qcolombet wrote:

Reuse `convertArg` here instead of inlining the logic.

https://github.com/llvm/llvm-project/pull/168374