Bill Sommerfeld wrote:

> > Note that this is not designed for fine-grained feature selection, like
> > the alpha amask or x86 CPUID.  
> 
> This is not architecturally correct for x86.
> 
> intel's processor documentation is *very* explicit that, in
> architectures which support the CPUID feature mask, it is never
> correct to assume that a feature is present because the processor type
> is greater than some threshhold.
> 
>  - Certain features have been removed from the architecture entirely
> (e.g., processor serial number)
> 
>  - certain features can be disabled by the BIOS before boot.
> 
>  - given the multiple parallel tracks of development even within
> Intel, features don't appear in a linear order.  If you add in AMD and
> Cyrix and the like, life gets even more nonlinear..

Hmm...  I was thinking more along the lines of gcc cpu targets.
Currently the list appears to be "i386", "i486", "i586", "pentium",
"i686" and "pentiumpro".  I assume that code generated for these
cpu targets follows some common denominator, regardless of the
CPUID feature mask?  And whether or not AMD and Cyrix cpus map
to this is another question again - which I have no idea of the
answer...

Simon.
--
Simon Burge                            <simonb@wasabisystems.com>
NetBSD Sales, Support and Service:  http://www.wasabisystems.com/