Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 13 May 2002 17:30:28 +1000 (EST)
From:      Bruce Evans <bde@zeta.org.au>
To:        Bernd Walter <ticso@cicely5.cicely.de>
Cc:        Gerald Pfeifer <pfeifer@dbai.tuwien.ac.at>, <cvs-all@FreeBSD.org>, <cvs-committers@FreeBSD.org>, "David O'Brien" <obrien@FreeBSD.org>, John Baldwin <jhb@FreeBSD.org>
Subject:   Re: GCC bugs (was: cvs commit: src/sys/alpha/include atomic.h)
Message-ID:  <20020513171450.W12629-100000@gamplex.bde.org>
In-Reply-To: <20020512162424.GM30437@cicely5.cicely.de>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 12 May 2002, Bernd Walter wrote:

> On Sun, May 12, 2002 at 11:10:42AM -0500, David W. Chapman Jr. wrote:
> > On Mon, May 13, 2002 at 02:08:31AM +1000, Bruce Evans wrote:
> > > On Sun, 12 May 2002, Gerald Pfeifer wrote:
> > >
> > > > And if the bug you are seeing, *any* bug you are seeing, with GCC 3.1
> > > > or later is a regression from *any* previous version of GCC, please
> > > > drop me a note and I will upgrade the priority of your PR so that it
> > > > has highest chance of being fixed for GCC 3.1.1 or GCC 3.2.
> > >
> > > Do efficiency bugs count?  makeworld is about 50% slower with gcc-3.1
> > > than with gcc-2.9x.
> > >
> > Is that because you can't make with -j or comparing old make
> > buildworld to new make buildworld without -j ?

This is because gcc-3 is about 60% larger and slower and gcc-2.9x.

> I also see increased time on alpha.
> About 5 hours instead of just about 3 hours as usual.

The slowdown is remarkably consistent.  60% again.  The space bloat may be
to be little larger (65% for the i386 cc1's).  The time bloat for compiling
the new gcc on i386's is about 62% user time.  The times for compiling the
new gcc with the old and new gcc using "cd /usr/src/gnu/usr.bin/cc;
make cleandir; make obj; make depend; time make" on an Athlon1400XP were:
with old gcc: 103.32 real  94.22 user 7.00 sys
with new gcc: 162.30 real 152.48 user 8.32 sys.

> Build is done always without -j and on the same machine and OS.
> Not that it's important to me - I just noticed.

Running makeworld fast is important to me.  Anything longer than 5-10
minutes is too long, since it is not reasonable to check every commit
using makeworld if it takes longer.  I was hoping to get a 20 minute
makeworld on the Athlon1400XP, but my best time so far was 23 minutes
42 seconds using my version of makeworld back on Feb 24.  The current
version took about 2 minutes longer.  Now with gcc-3, the buildworld
time for -current has bloated to 37 minutes 48 seconds.

Bruce


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020513171450.W12629-100000>