[llvm] [AMDGPU] Add backward compatibility layer for kernarg preloading (PR #119167)

[Matt Arsenault via llvm-commits]

================
@@ -259,6 +259,8 @@ SmallVectorImpl<MCRegister> *SIMachineFunctionInfo::addPreloadedKernArg(
   // If the available register tuples are aligned with the kernarg to be
   // preloaded use that register, otherwise we need to use a set of SGPRs and
   // merge them.
+  if (ArgInfo.FirstKernArgPreloadReg == AMDGPU::NoRegister)
----------------
arsenm wrote:

```suggestion
  if (!ArgInfo.FirstKernArgPreloadReg)
```

https://github.com/llvm/llvm-project/pull/119167