Re: --OMG_OPTIMIZED
On Fri, Apr 4, 2008 at 4:59 PM, Paul Johnson <baloo@ursine.ca> wrote:
> On Friday 04 April 2008 01:50:02 am Ivan Savcic wrote:
> > On Fri, Apr 4, 2008 at 2:11 AM, Chris Walters <cjw2004d@comcast.net> wrote:
> > > -----BEGIN PGP SIGNED MESSAGE-----
> > > Hash: SHA512
> > >
> > > Ivan Savcic wrote:
> > > | Sorry for that personal message, I misclicked. It wasn't aimed at you
> > > | specifically.
> > >
> > > Apology accepted. I am sure most everyone, myself included, has made
> > > similar
> > > mistakes.
> >
> > Thanks.
> >
> > > | When Debian Etch was released, I wanted to give Debian a shot again in
> > > | some server scenarios, because of it's stability, security and ease of
> > > | upgrading. I now deeply respect the concept of "stable", having been
> > > | through security-through-bleeding-edge concept of Gentoo, for example.
> > > | Long End of Life of stable Debian seems priceless. Yet, on the other
> > > | hand, Backports filled the gap caused by some oldish packages and in
> > > | general there are a lot of packages for people to use.
> > >
> > > I remember the days of Sarge. I used backports then, as well as
> > > compiling source. Why? Is it that I have a lot of time on my hands?
> > > No. It is/was to
> > > streamline the package, and optimize it for my processor. The main
> > > problem with precompiled distros, IMHO, is that because the packages,
> > > especially the
> > > kernel, have to run on a multitude of different systems, they tend to be
> > > larger
> > > and slower than if you compile those packages, optimized for your
> > > system.
> >
> > Luckily, there are AMD64 and IA64 flavors of Debian. Shame there
> > aren't (stable?) versions for i686, Athlon and P3/P4.
>
> Do you have evidence that would justify this thinking? Debian already has
> packages optimized for sub-architectures, but only for the packages it
> actually makes a difference on. Optimizing the entire distribution is a
> waste of DD time, and mirror diskspace for truly epsilon gains.
Long time passed since I was messing around with that, but give it a
shot yourself with Acovea[1], or if you manage to fix it, ccbench[2].
Way back, on my Gentoo Linux, I have concluded that -O2 makes a
difference compared to -O1 and i686 --mcpu/--march makes a difference
compared to i386. --fomit-frame-pointer also produced faster binaries.
[1] http://www.coyotegulch.com/products/acovea/
[2] http://www.rocklinux.net/people/clifford/ccbench/ccbench-0.2.tar.bz2
Reply to: