This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Improve unrolled size estimates


On Mon, 11 May 2009, Steven Bosscher wrote:

> On Mon, May 11, 2009 at 3:31 PM, Richard Guenther <rguenther@suse.de> wrote:
> >> would only be a win to do CCP if there are expressions that fold to a
> >> constant an are fed back via a back-edge to a PHI. ?I don't know how
> >> likely this is, but I'd dare a hand-wavey guess "not too common".
> >
> > Ah, but I wouldn't include the backedge to the loop entry in the
> > simulation.
> 
> Then it may be a lot easier to teach tree-ssa-dom to only work on part
> of the dominator tree (i.e. SEME regions like loops).  This should be
> as easy as passing an entry block and a bitmap with exit blocks. What
> do you think about this approach?

Apart from that I'd rather like to get rid of these parts of DOM
that would of course work (and likely be even easier to implement).

Richard.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]