------- Additional Comments From law at redhat dot com 2004-11-01 14:09 -------
Subject: Re: [4.0 Regression] jump threading
on trees is slow with switch statements with large # of cases
On Mon, 2004-11-01 at 05:16 +0000, pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org 2004-11-01 05:16
> -------
> We are still way behind 3.3, it takes 15 seconds on my 1.5GHz PPC 7400 with 3.3 but
> with 4.0, well for
> 4.0 time just look at Jeff's data and see that we are way behind still.
Well, I haven't looked at 3.3, but I can make a reasonable guess that
we're still way way way behind due to the way we update case labels
when forwarding edges and splitting critical edges. Some early
experiments I've done with that indicate there's another 30-35%
improvement that can be made by fixing that problem. Then there's
*another* 30% or so we're burning in the RTL branch prediction code
(I haven't looked to see if there's anything we can do with that code
yet).
I doubt it makes much sense to look closely at 3.3 vs 4.0 for this
testcase and similar code until we fix those glaring problems.
It's also not clear to me how much of an improvement those changes
will make in real-world code. ie, we could easily run into a case
where we drastically improve that testcase without improving any
real code, much like what happened recently with my changes to
improve how we find/record equivalences for edges.
Jeff
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15524
