Path: utzoo!attcan!uunet!know!cs.utexas.edu!sun-barr!newstop!sun!concertina.Eng.Sun.COM!fiddler
From: fiddler@concertina.Eng.Sun.COM (Steve Hix)
Newsgroups: comp.sys.mac.hardware
Subject: Re: MAC Co-Processors
Message-ID: <142802@sun.Eng.Sun.COM>
Date: 20 Sep 90 17:49:47 GMT
References: <142737@sun.Eng.Sun.COM> <1104@beguine.UUCP> <1990Sep19.230715.15188@Neon.Stanford.EDU>
Sender: news@sun.Eng.Sun.COM
Lines: 30

In article <1990Sep19.230715.15188@Neon.Stanford.EDU>, philip@pescadero.Stanford.EDU (Philip Machanick) writes:
> In article <142737@sun.Eng.Sun.COM>, fiddler@concertina.Eng.Sun.COM
> (Steve Hix) writes:
> [...]
> > On the other hand, you *could* look at a 68881/68882 as a SANE
> > coprocessor...they do pretty much the same job.
> 
> Sort of - but SANE even with a 68881/6882 is about 20 times slower than
> code run directly on the coprocessor. Apple maintains some of their routines are more
> accurate than the IEEE standard, and this justifies the slowness.

I thought that that went without saying.  Apparently not.

If you can't run a 68881/2 on your machine, SANE is about the only game in
town.  (Even then, you'll end up dealing with one application's or another's
own math usage, if they don't use SANE or a math coprocessor.  Like Microsoft.)

Btw, some of the team that developed SANE at Apple were on the IEEE committee
that developed the floating-point standard.

Don't forget, even under SANE or an '881/2, there will still be some degree
of error in a greater or lesser portion of your math results.  At least with
SANE or some other IEEE-compliant floating point option, you'll know how
much error there is.

--
------------
  The only drawback with morning is that it comes 
    at such an inconvenient time of day.
------------