[llvm] [AMDGPU] Schedule independent instructions between s_barrier_signal and s_barrier_wait (PR #172057)

[Nicolai Hähnle via llvm-commits]

https://github.com/nhaehnle approved this pull request.

Nice. 16 is a good number, it sounds in the right ballpark for actual barrier latency cycles.

For future reference, it's welcome practice to add new tests first in a separate commit without code changes, so that the effect of your code change is more obvious in code review.

https://github.com/llvm/llvm-project/pull/172057