This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: libraries for Pentium 4
- From: Joel Sherrill <joel dot sherrill at OARcorp dot com>
- To: GATMKORN at aol dot com
- Cc: gcc at gcc dot gnu dot org
- Date: Mon, 27 Jan 2003 11:50:59 -0600
- Subject: Re: libraries for Pentium 4
- Organization: OAR Corporation
- References: <23.2b09a9e8.2b66c43b@aol.com>
GATMKORN@aol.com wrote:
>
> - the new gcc 3.2 with -march=pentium4 -O3 speeds up my dynamic-system
> simulations by 5 to 15 percent, congratulations!
>
> But should we not also optimize the math library and the Xlib library? Is
> there any Pentium-4 optimized library?
>
> Incidentally, the newer Athlons still beat the Pentium 4 in our scientific
> applications, probabl because Pentium 4's long pipelines stall on bad branch
> predictions.
>
> Thanks for any information on library improvements - GAK
You don't say what OS you are using but assuming you are on a GNU/Linux
or
*BSD variant, you could always locally compile the packages you feel
would
have the most impact and try them out.
FWIW on RTEMS, we have been multilib'ing our i386 target a while now.
Embedded systems people tend to want things compiled in the most optimal
way for their particular CPU. We have these variants for libc.a:
default (i386 w/FPU)
athlon
k6
m486/soft-float
m486
mpentium
mpentiumpro
soft-float
It might be worth it for other OSes to consider multilibing. UNIX
variants
wouldn't have to include the soft-float variants since they usually
support
an FPU trap and emulate.
> (Prof. G.A. Korn, ex-Uof Arizona)
--
Joel Sherrill, Ph.D. Director of Research & Development
joel@OARcorp.com On-Line Applications Research
Ask me about RTEMS: a free RTOS Huntsville AL 35805
Support Available (256) 722-9985