This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RFA: tree DSE bug
On Fri, 2005-11-11 at 00:07 -0500, Daniel Berlin wrote:
> On Thu, 2005-11-10 at 21:41 -0700, Jeffrey A Law wrote:
> > On Thu, 2005-11-10 at 23:19 -0500, Daniel Jacobowitz wrote:
> >
> > > I got a lot of feedback about this on IRC, so:
> > >
> > > What advantage does the simpler test have, if any, now that I've gone
> > > and figured out the correct fix to the existing algorithm?
> > One could easily argue that not walking through PHI arguments associated
> > with backedges *is* fixing the existing algorithm.
>
> Except that your solution requires us keeping the back edge flag up to
> date (which, AFAIK, it is not always, and mark_dfs_back_edges is not
> free),
Yes, I was already well aware of this issue.
> *plus* it's potentially worse for the reasons Daniel pointed out.
And as I pointed out, I doubt anyone has any real world code where
the difference would show up.
> In addition, the patch you are suggesting isn't written or tested yet.
I wrote it in about 15 minutes this morning. 10 lines. But as I
mentioned in a message a few minutes ago, the compile time cost
of my solution is too high to offset the improvement in cleanliness
and simplicity.
jeff