[llvm] [AMDGPU] Move kernarg preload logic to separate pass (PR #130434)

[Shilei Tian via llvm-commits]

================
@@ -1294,6 +1295,10 @@ void AMDGPUPassConfig::addIRPasses() {
 }
 
 void AMDGPUPassConfig::addCodeGenPrepare() {
+  if (TM->getTargetTriple().getArch() &&
----------------
shiltian wrote:

what is this check used for?

https://github.com/llvm/llvm-project/pull/130434