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

[Austin Kerbow via llvm-commits]

================
@@ -523,6 +525,8 @@
 ; GCN-O1-OPTS-NEXT:      Scalarize Masked Memory Intrinsics
 ; GCN-O1-OPTS-NEXT:      Expand reduction intrinsics
 ; GCN-O1-OPTS-NEXT:      Early CSE
+; GCN-O1-OPTS-NEXT:    AMDGPU Preload Kernel Arguments
----------------
kerbowa wrote:

The pass mostly wants to run before function attributes are finalized that may influence the number of available user SGPRs. It also needs to be before LowerKernelArguments. Beyond that there may be some small benefit to lowering the hidden arguments earlier, but only if it doesn't run into the first issue.

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