[llvm] [SystemZ] Add a SystemZ specific pre-RA scheduling strategy. (PR #135076)

[Andrew Trick via llvm-commits]

atrick wrote:

> So with global liveness and better handling of pressure Excess it should do better, but then again it will not matter in many interesting cases where there will be Excess anyway where it will always be a matter of reducing it as much as possible. @atrick - have you tried this?

The generic scheduler was meant to exercise most of the basic MachineScheduler features without being heavily biased in one direction. It was sort of tuned to not perturb the schedule in most cases. At the time, people were hand-tuning C kernels and didn't want LLVM messing it up. I think you can always find a simpler, better strategy for a particular microarchitecture and the workloads the matter most to your platform. You can also do much more aggressive register pressure reduction with cross-block code motion (before MachineScheduling). And you can do multi-pass scheduling with different strategies each time, etc. As long as you can justify maintaining that extra logic and spending compilation time on it.

https://github.com/llvm/llvm-project/pull/135076