[llvm] [MachineScheduler][ScheduleDAG] Add ability to bias scheduling by longest paths (PR #93223)

[via llvm-commits]

csstormq wrote:

> > Is there any way to add a unit test to demonstrate the improved assembly code on RISC-V?
> 
> I have attempted to add a test that does this. However, it seems like such a test would not be stable: on some runs, the critical path version ends up being the same as the longer paths version. Other times, the critical path version is different from the longer paths version.
> 
> This leads the test to pass some times and to fail other times. The benefit of using the test I add in this patch, is that it is deterministic.

Can you attach that test case if possible? I'm curious why it's unstable. It would be great if we could visually demonstrate what the improved code looks like.

https://github.com/llvm/llvm-project/pull/93223