[llvm] [AMDGPU][DAG] Enable ganging up of memcpy loads/stores for AMDGPU (PR #96185)

[Matt Arsenault via llvm-commits]

================

----------------
arsenm wrote:

Which really means p4, without an invariant annotation we don't handle now. You would need to use a kernel for the p1->p4 promotion, so just test the constant address space uniform source case 

https://github.com/llvm/llvm-project/pull/96185