[llvm] [AMDGPU] Fix code sequence for barrier start in GFX10+ CU Mode (PR #160501)

[Pierre van Houtryve via llvm-commits]

Pierre-vh wrote:

> Does this need a change in the memory model docs?

We don't have the documentation for that upstream yet, I'll fix the downstream one

https://github.com/llvm/llvm-project/pull/160501