cvs commit: src/lib/libc/i386/net htonl.S ntohl.S

[Erik Trulsson]

On Wed, Oct 20, 2004 at 05:24:57PM -0400, David Schultz wrote:
> On Wed, Oct 20, 2004, Peter Jeremy wrote:
> > On Tue, 2004-Oct-19 18:52:31 -0400, David Schultz wrote:

> > >> > Nice.  \me can't wait for the day when developers are no longer
> > >> > required to spend time and effort to support anything older than a PPro.
> > >> 
> > >> That day will hopefully be far in the future. Personally I don't have
> > >> anything as modern as a PPro.
> > >
> > >Don't worry, it will be.  I can dream, can't I?
> > 
> > The 486 and Pentium provide useful new instructions and system controls.
> > What benefits does a PPro provide as far as the kernel and core userland
> > is concerned?
> 
> Now we're really getting sidetracked.  The main new feature of
> interest for me (and admittedly not for most other people) is MMX,
> for a couple of bad reasons.  First, gcc generates slightly more
> correct floating-point code with MMX.  Second, as long as we
> support !MMX, we would have to multilib the math library in order
> to make it work correctly with programs compiled for newer
> processors, in particular with respect to the floating-point
> exception flags.  Granted, a few Pentium I chips have MMX support,
> too.  But when you consider that the PPro also has CMOV and FCMOV,
> that the scheduling is fundamentally different for all processors
> after the Pentium, and that we currently have several hacks for
> Pentium I errata, drawing the line at PPro makes sense.

The PentiumPro does not have MMX so the above argument makes no sense. 
MMX was first supported on the P55C which came out in 1997 (after the
original PentiumPro which came out in 1996.) (And there are probably
more PentiumMMX computers out there than there are PentiumPro ones.)
The Pentium II was the first CPU based on the P6-microarchitecture to
support MMX. (P6 is the basic architecture used for the PentiumPro,
Pentium II, and Pentium III.)

So if you want to have MMX as well as CMOV you will have to draw the
line at the Pentium II.

As for scheduling that is also quite different between the Netburst
architecture (Pentium 4) and the P6-based chips, so that sounds like a
red herring.


-- 
<Insert your favourite quote here.>
Erik Trulsson
ertr1013 at student.uu.se