[llvm] AMDGPU/GFX12: Do not wait unnecessarily before barriers (PR #154970)

[Nicolai Hähnle via llvm-commits]

nhaehnle wrote:

Ah, I remember now -- is this the issue of VMEM instructions getting re-ordered across barriers because a workgroup release fence doesn't wait for anything? Do you have relevant changes already in upstream review somewhere?

https://github.com/llvm/llvm-project/pull/154970