Re: architecture names (i386-linux, etc.)

["H. Peter Anvin" <hpa@transmeta.com>]

Alan Cox wrote:
> 
> > On the other issue, it might be helpful if there was some sort of
> > standard for x86 architecture names.
> >
> > Does "i686-linux" mean "only runs on a Pentium Pro or above" or does
> > "i686-linux" mean optimized for Pentium Pro, but still runs on i386?
> > I thought it meant the former, not the latter.  Unfortunately, a some
> > number of packages just assign the number according to the system the
> > package was compiled on, not the target system.
> 
> Lets follow the intel recommendations. We then have
> 
>         i386    -       all cpus
>         i486    -       requires i486 or higher (uses bswap)
> 
> People following the rules and writing correct applications will not
> have a problem with just the two types. Intel AMD and Cyrix all tell people
> to use CPUID to check the mmx/3dnow!/kni/etc facilities are present and
> software fallback.
> 
> We should mandate compliant packages do this. We should not mandate what a
> package is optimised for either IMHO, but what it runs on. "Recommended CPU"
> is a marketing issue not a standards one.
> 

Oh yes... and unlike what Intel Application Note AP-485 states, it
should not be a requirement for any software that CPUID returns
"GenuineIntel" unless they really are looking for Intel-specific quirks
(for example, the Pentinum Pro SEP bug.)

	-hpa

-- 
<hpa@transmeta.com> at work, <hpa@zytor.com> in private!