local refs

[Stephen Weeks]

> On the other hand, we aren't that far off of commonSubexp:
>             commonSubexp starting
>             commonSubexp finished in 4.51 + 4.44 (50% GC)
> which is the other analysis that requires computing the dominator tree for
> each function.  (restore needs to compute the dominator tree; but, we are
> only restore ing 126 out of 1101 functions, so it's still taking too
> long.)

There is still a lot of easy fat (I'd guess >2X) in the dominator
computation.  It's been on my todo list for ages.  It would be nice to
know for both restore and commonSubexp how much of the time is
dominators.  I'd guess a lot.