This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: C++ compile time (again)
- From: Karel Gardas <kgardas at objectsecurity dot com>
- To: Mike Stump <mrs at apple dot com>
- Cc: "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Date: Wed, 18 Jun 2003 09:29:42 +0200 (CEST)
- Subject: Re: C++ compile time (again)
On Tue, 17 Jun 2003, Mike Stump wrote:
> On Tue, 2003-06-17 at 09:35, Karel Gardas wrote:
> > Options used to compile all cc files are:
> >
> > -I../include -Wall -fpermissive -DPIC -fPIC -c
> >
> > times (user only) for cd mico/orb; time make are:
> >
> > gcc2.95.4: 6m4s
> > gcc3.4: 9m24s
> > gcc3.5: 11m17s
> >
> >
> > Anything else what should I test for you?
>
> As time goes on, the numbers from PCH will be more and more important
> to not regress, also the 2m slowdown will go from 20% to something like
> 2x slowdown. Some folks might be tempted to put in a 20% slowdown,
> but, they should have a harder time (from us) for a 2x slowdown.
Good idea.
> So, adding a column with good PCH numbers would help prevent C++
> slowdowns.
At the moment I'm only able to obtain numbers from gcc3.4, since gcc3.5 is
missing small important patch in except.c which solves problems with
already defined symbols in generated assembler file.
So gcc3.4 + PCH: 3m41s
if we keep difference between both versions
from non-PCH run, then we can IMHO estimate:
gcc3.5 + PCH: 5m34s
differences between gcc3.4 and gcc3.5 then looks:
non-PCH: 20%
PCH: 51%
Cheers,
Karel
PS: options used to compile cc file were:
-I../include -Winvalid-pch -DFAST_PCH -Wall -fpermissive -DPIC -fPIC -c
--
Karel Gardas kgardas@objectsecurity.com
ObjectSecurity Ltd. http://www.objectsecurity.com