Mail Archives: pgcc/2000/09/19/09:47:11
Regarding your question, I think I have some ideas about most of them, so
I'll try to answer... in fact I think most of the questions are in the FAQ
which you should have checked...
(1) You probably should address that question to Redhat/Suse as they
decide which gcc version to include in the distribution. As far as I know,
Mandrake is the only one that ships pgcc and/or have pentium optimization
turned on (and breaking backward 486 compatibility). But a lot of these
decisions is based on stability not on speed; at the beginning (maybe
even now) pgcc miscompiled when doing a self-build (i.e. going through
the bootstrap phrase and uses an older/current version of itself to
compile and build itself for upgrading itself, for example) and that is
certainly not very good for a system compiler - so it is left to the
individuals who want extra speeds for individual applications to try to
see if they can get a speed gain without miscompiling. Some people report
to be able to compile and build a functional linux kernel with it, but it
is still generally not recommended. I myself actually use to keep three
compilers and take them off/on when I need to - egcs 1.1.x for general
use, pgcc for extra speed, and gcc 2.7.2.x for compiling the linux
kernels; these days I still use two, gcc 2.95.x for general use and
dropping back to egcs 1.1.x for kernel compiling, and pgcc only for very
occasional use. It is actually quite easy to take them on and off the
system as needed with any of the rpm/deb/slackware tgz management systems.
2. That depends on your skill-level... most people are quite happy keeping
their installation and just recompile certain performance critical or
frequently used software, e.g. the x server, which does a lot of
rendering, etc. But again, you have to be the person to judge whether a
10% speed gain versus the X server occasionally freezes up on you for no
apparent reason (just for an example) is worth it.
3. Check FAQ, but I remember a 5-20% speed gain being typical - but you
can also have slight speed lost in special cases, I think...
In fact for a desktop home user system, if you are running Gnome/KDE, you
probably would "percieve" a performance gain just by adding more memory
rather than recompiling anything. I (still) use fvwm 1.24r, mostly
slackware 7.0, but my own bootstraped gcc 2.95.2 and kernel 2.2.17 patched
with USB and a few other things.
On Tue, 19 Sep 2000, Malte Gell wrote:
> Hello !
> I just found this interesting pgcc site and have a few questions, but
> just one thing at first: since i have not subscribed to the pgcc mailing
> list, please also answer directly to me, thanx !
>
> Well, now my questions:
> 1. Since so many people use a Pentium class CPU nowadays, why is it not
> a standard, that all distributions or the distributions "for personal
> use" (=standard distros for the home user like RedHat or Caldera or
> SuSE) use pgcc by default ? It makes me a bit angry somehow that i have
> a fast AMD based machine and the code is just optimized for an 486, in
> my eyes this is a waste of performance, or am i wrong ?
>
> 2. I am using SuSE Linux 7.0, how easy would it be to change completely
> to pgcc ? Is pgcc also optimized for AMD CPU´s ? What are your
> experiences with changing to pgcc ? Would i have to re-compile all my
> apps or can i still use them ?
>
> 3. How big would be the gain of performance if i change to pgcc ? (AMD
> K6-III 400 Mhz 128 MB RAM system, Kernel 2.2.16)
>
> Thanx a lot
>
> Malte
>
>
> --
> __,,
> /o // "Why do we have to hide from the police, Daddy ?"
> /_ ) "Because we use vi. They use emacs"
> /__| Dave Fischer, 1995/06/19
- Raw text -