[llvm] [AMDGPU] Emit a waitcnt instruction after each memory instruction (PR #68932)

[Matt Arsenault via llvm-commits]

arsenm wrote:

I think it would be better to refine the existing flag to control the behavior. It could expand to an enum for wait before/wait after.

Also on some hardware, gdb can force this behavior anyway

https://github.com/llvm/llvm-project/pull/68932