[tree-ssa] [PATCH]: Loop invariant code motion for trees

[Daniel Berlin]

> I just bootstrapped it with no problems (turning off LICM, of course).
> Diego, offhand, did you actually change the way default defs are 
> handled, or just added a default_def field?
> LICM assumes that a use whose def was an empty statement (which is 
> what we got with default defs) did not block hoisting.
> Is this still a valid assumption?
>
Apparently not.
Stopping it from hoisting those makes it work again.

The changes are about 4 lines.
Want just a new tree-loop.c or a whole new patch?