This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCC 3.4.0 040114/GCC 3.5-tree-ssa 040120 compilation speedcomparison.
- From: Karel Gardas <kgardas at objectsecurity dot com>
- To: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>
- Cc: GCC Mailing List <gcc at gcc dot gnu dot org>
- Date: Tue, 20 Jan 2004 18:17:46 +0100 (CET)
- Subject: Re: GCC 3.4.0 040114/GCC 3.5-tree-ssa 040120 compilation speedcomparison.
On Tue, 20 Jan 2004, Steven Bosscher wrote:
> On Tuesday 20 January 2004 17:52, Karel Gardas wrote:
> > Hello,
> >
> > here is yet another report about compilation speed of tree-ssa against
> > 3.4.0 version. The build time itself doesn't regress,
>
> Clearly tree-ssa is ~10% slower over the whole. I call that a regression.
But it's not ~15% as stated somewhere on gcc website...
> > All builds with tree-ssa are slower than those with 3.4.0 version.
>
> Is the tarball with preprocessed sources already available somewhere? I
> want to see time reports for this and analyze a few things.
Not yet, since I need to find the right files. Of course, if you have some
flaming candidate, then you can help me a lot! Send me its name.
> One thing you should be aware of, is that the parameter "large-function-insns"
> is set to 10000 for tree-ssa, and for 3000 for 3.4.0. With the next merge
> from mainline to the branch, tree-ssa will also be at 3000. Jan Hubicka has
> shown that this parameter change significantly speeds up GCC without a
> significant performance loss in the generated code.
Yes, I was aware that I should test GCC 3.4.0 on which tree-ssa is based
on, but I haven't done it finally since I have already all the timings for
3.4.0 040114...
> So if it is not asked too much, can you try tree-ssa again with an extra
> option "--param large-function-insns=3000"?
> If you have the tarball for me, I can do it myself.
If you can do it, its better since I still need to write all the
bugreport. Just tell me file names and I will send you tarball (and then
create bugreport with it)
Thanks,
Karel
--
Karel Gardas kgardas@objectsecurity.com
ObjectSecurity Ltd. http://www.objectsecurity.com