John Summerfield wrote:
> [EMAIL PROTECTED] said:
> > vendor, Red Hat, still targets the '386 by default! The decision to
> > abandon
>
> That hasn't been entirely true for some time. Some packages are compiled
> differently for different CPUs.
Under Linux certainly, but that is not at all prevalent in the Windows universe,
still the dominant x86 PC environment by several orders of magnitude. And that is
probably the code Plex86 will be running the most.
When it comes to NT on non-x86 platforms, yes, multiple binaries were (and are)
shipped by a few vendors. But now that NT is becoming increasingly x86 specific,
you can expect legacy non-x86 systems to soon lose application support (mainly
upgrades).
For the Mac, yes, some companies still produce "fat binaries", but that is a
rapidly vanishing population as well.
> For most people there is no practical difference anyway; if your CPU is idle
> 90% of the time, does optimisation really matter?
See my other post: There are increasing instances on contemporary PCs where it
DOES matter! Things like WinModems, DirectX, Real Audio/Video, MS MediaPlayer,
NetMeeting and the like.
> If you're down to the last 10% of performance you should have upgraded to
> something faster.
Or use properly optimized software, if available.
> There can be cost advantages to having better code, but I suggest actual
> instances are hard to isolate.
Like postponing upgrades? I was simply advocating moving the "upgrade window" a
bit into the future by targeting a relatively new architecture now, one that
itself will be dated by the time Plex86 1.0 comes out.
A look at the Wine project is informative in this area: Much of the tuning being
done in preparation for Wine 1.0 has involved, in effect, abandoning
("anti-optimizing") slower processor architectures. Some of the speed-ups seen
have been due to major architectural changes, and others have been due to
improving locality of code and data (and thus improving cache performance).
-BobC
