Some thoughts and quetsions about the data flow infrastracture

[Richard Kenner]

> Let us look at major RTL optimizations: combiner, scheduler, RA.  Do
> we need a global analysis for building def-use use-def chains?   We
> don't need it for combiner (only in bb scope), 

Combine would work *far better* if it had some uniform data structure.  For
one thing, that means it could work across blocks, which would be very
valuable. For another, there's a tremendous amount of highly-complex code
to do the sorts of tracking that is best done with a uniform infrastructure.

> we don't need it for scheduler (only for DAG region)

Yes, but again, it would work better if it didn't.

> we don't need at all for reload.  

Again, a tremendous amount of kludgery in reload would be removed by
having accurate and better-tracked information.