Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!bpa!cbmvax!daveh
From: daveh@cbmvax.commodore.com (Dave Haynie)
Newsgroups: comp.sys.next
Subject: Re: 68040 = 3.5 MFLOPS ?????
Message-ID: <9850@cbmvax.commodore.com>
Date: 27 Feb 90 08:20:54 GMT
References: <16342@oregon.uoregon.edu>
Reply-To: daveh@cbmvax.cbm.commodore.com (Dave Haynie)
Distribution: usa
Organization: Commodore, West Chester, PA
Lines: 23

In article <16342@oregon.uoregon.edu> KUO@oregon.uoregon.edu (Shijong Kuo) writes:
>According to an article on 68040 in lateset DDJ, Motorola omitted
>floating point intructions for transcendental functions with their new CPU. 

Yup, the transcendentals are serviced by instruction trap routines.  Their claim
is that it'll still be better than twice the speed of an '882 for these ops.  The
FPU logic apparently does some cooking of the operands before trapping.

>Perhaps the 3.5 MFLOPS rating touted by Motorola shouldn't be taken seriously!

They are calling for a 3 clock FP add and 5 clock FP multiply.  That's 120ns and
200ns, respectively.  If you just ran additions, that would put you somewhere
around 8 MFLOPS.  Sounds like the instruction mix is pretty important, though
if you think about it, hasn't it always been?  The 3.5 figure was taken from some
benchmark code.  As with most things, a real life 68040 machine running a real
OS and some real benchmarks will do wonders to clear these things up.

>kuo@oregon.uoregon.edu

-- 
Dave Haynie Commodore-Amiga (Systems Engineering) "The Crew That Never Rests"
   {uunet|pyramid|rutgers}!cbmvax!daveh      PLINK: hazy     BIX: hazy
                    Too much of everything is just enough