[llvm] [AMDGPU] Move kernarg preload logic to AMDGPU Attributor (PR #123547)

[Shilei Tian via llvm-commits]

================
@@ -1400,6 +1654,11 @@ static bool runImpl(Module &M, AnalysisGetter &AG, TargetMachine &TM,
   }
 
   ChangeStatus Change = A.run();
+
+  // Mark kernel arguments with 'inreg' attribute to indicate that they should
+  // be preloaded into SGPRs.
+  markKernelArgsAsInreg(Functions, TM);
----------------
shiltian wrote:

This doesn't reflect the change of the pass or not.

https://github.com/llvm/llvm-project/pull/123547