[llvm] [AMDGPU] Support preloading hidden kernel arguments (PR #98861)

[Matt Arsenault via llvm-commits]

================
@@ -2533,6 +2534,23 @@ void SITargetLowering::allocatePreloadKernArgSGPRs(
       unsigned NumAllocSGPRs =
           alignTo(ArgLoc.getLocVT().getFixedSizeInBits(), 32) / 32;
 
+      // Add padding SPGR to fix alignment for hidden arguments.
----------------
arsenm wrote:

Missing the globalisel equvalent of this handling 

https://github.com/llvm/llvm-project/pull/98861