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

[Matt Arsenault via llvm-commits]

================
@@ -1164,6 +1170,8 @@
 ; GCN-O3-NEXT:      Lazy Block Frequency Analysis
 ; GCN-O3-NEXT:      Optimization Remark Emitter
 ; GCN-O3-NEXT:      Global Value Numbering
+; GCN-O3-NEXT:    AMDGPU Preload Kernel Arguments
+; GCN-O3-NEXT:    FunctionPass Manager
----------------
arsenm wrote:

Can this be moved up with the other module passes, or is there some benefit to running the other function passes first 

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