[MISched] Add statistics for heuristics

[c-rhodes]

When diagnosing scheduling issues it can be useful to know which heuristics are driving the scheduler. This adds pre-RA and post-RA statistics for all heuristics.

[c-rhodes]

I've been investigating a couple of scheduler issues recently and while it's possible to see what heuristics are driving it in debug output and with grep ("Pick "), I figured it would be good if we just had stats for this, perhaps others will also find this useful.

[mshockwave]

On a related note, I've been thinking about improving the trace output of MachineScheduler because I alway found some messages a bit difficult to comprehend, I'm wondering what's your thoughts?

[c-rhodes]

On a related note, I've been thinking about improving the trace output of MachineScheduler because I alway found some messages a bit difficult to comprehend, I'm wondering what's your thoughts?

I think that's a great idea I also found it difficult, particularly when scheduling is enabled both pre and post register-allocation!

[davemgreen]

This looks like a fair number of statistics but they sound useful. LGTM

[mshockwave]

LGTM. I think these counters will be useful during development.
It's true that the sheer number of new statistic counters looks a little scary, but IMHO users could always opt-out so I think it's fine.

[c-rhodes]

precommit keeps failing but I dont believe the test failures are because of this patch, so going to land.