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

[Christudasan Devadasan via llvm-commits]

================
@@ -1317,6 +1318,10 @@ void AMDGPUPassConfig::addIRPasses() {
 }
 
 void AMDGPUPassConfig::addCodeGenPrepare() {
+  if (TM->getTargetTriple().getArch() == Triple::amdgcn &&
----------------
cdevadas wrote:

Realized that it is the common base class for both amdgcn and r600. 

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