Compiler optimization flags in building LFS-5.1.1

Matthew Burgess matthew at linuxfromscratch.org
Fri Sep 10 10:40:07 PDT 2004


On Fri, 10 Sep 2004 00:59:16 +0200
"Matthias B." <msbREMOVE-THIS at winterdrache.de> wrote:

> On Wed,  8 Sep 2004 14:11:12 -0700 "LEE,CHRISTIAN WELDON"
> <clee1750 at ucla.edu> wrote:
> 
> > Doh... I'd hate to not be able to do any optimization on glibc,
> > considering every program I install or write is pretty likely to link to
> > it. That's the only one I REALLY had my heart set on optimizing.
> 
> There is NO package in base LFS that is worth optimizing, NOT A SINGLE
> ONE. Even glibc is not worth optimizing. No application whose speed you'll
> ever be worried about will get faster that way.
> Extra optimization on base LFS packages is a waste of time and only
> creates instability. And besides, -O3 is known to PESSIMIZE code in a lot
> of situations. It's foolish to use -O3 blindly.

Indeed - this reminds me of a project called ACOVEA (http://www.coyotegulch.com/products/acovea/).  If you're serious about figuring out the most beneficial optimisations to use you could try it out, but only if you've got serious time to waste - at which point, the time you've spent figuring out the optimisation flags will be _much_ greater than any time saved in running your system I'd imagine.  Each and every package would have to be built _several_ times using that package to analyse compile-time & runtime performance differences of each compile!

Matt.



More information about the lfs-support mailing list