[llvm] [AMDGPU] AMDGPUIGroupLP: Avoid repeating reachability checks in greedy algorithm (PR #182463)

[Frederik Harwath via llvm-commits]

frederik-h wrote:

This gives a more than 40% performance improvement on a long running example with 800 sched group barriers and did not affect smaller examples too badly. AddedEdges could perhaps be changed to a list if the copying turns out to be problematic.

https://github.com/llvm/llvm-project/pull/182463