Thanks for looking into the issue.
The purist in me would argue for
supporting no-SSE2 machines, however a) I
don't have any and b)
your point re:
compromise is a good one.
I would say that unless someone who a) has hardware
and b) is willing to help debug
objects, we
should consider dropping non-SSE2 capable
hardware. Let's
get through negotiations with FESCo
first, then revisit this. Say a
month from now? (assuming the
lights haven't been turned out on
us)
Thoughts? Objects?
jeff
--
Jeff Backus
jeff.backus(a)gmail.com
http://github.com/jsbackus
http://gitlab.com/jsbackus
_______________________________________________
X86
mailing list -- x86(a)lists.fedoraproject.org
To unsubscribe send an email to
x86-leave(a)lists.fedoraproject.org
The irony is that I'm the type who'd
advocate supporting i586 machines (pentium 1 and up, no cmov instructions), since those
are
useful for some VIA systems as well as old AMD K6, if anyone's still using those. I
deliberately
compiled an i586 distro a couple years back from source, and I constantly ran into
packages that
had assembly that required the cmov instructions. I fear we're in for a similar
problem with
SSE2.
The problem I think is that some packages make it very hard to turn off
assembly routines that break compatibility, and then often when you do, you take an
unacceptable
performance hit.
I'm actually questioning the utility of i686 at all if it requires
SSE2, since I feel like there's a fairly slim generation of machines that supported
only i686
and no SSE2. While SSE2 was introduced in 2001, not every chip had it as far as I know. I
have
several Pentium 3 machines that work quite well actually, and I can still get use out of
them in
a pinch.
I know, I'm a crusty luddite hehe.