This is the mail archive of the
mailing list for the GCC project.
> > is someone currently working on SSA based value range propagation? If
> > not, I will give it a try.
> Have you seen that Jeff Law just commited a limited dominator based VRP
> for tree-ssa?
> I worked a bit on the starts of SSA VRP as described in Patterson's
> thesis, but then I decided it is probably only worth the effort if we
> can use it to compute branch probabilities; just propagating the value
> ranges that accurately is not going to give much benefit. Especially
> with the new dominator VRP...
I admit I had mostly the branch prediction in mind. Even in the current
state it should be usable -- the predictions may be sent to rtl via
builtins (and of course I still hope we will have a transition between
tree-ssa and rtl that does not destroy cfg some day).