Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uunet!zephyr.ens.tek.com!uw-beaver!ubc-cs!eric!joplin!hui
From: hui@joplin.mpr.ca (Michael Hui)
Newsgroups: comp.arch
Subject: Re: Integer Multiply/Divide on Sparc
Message-ID: <1979@eric.mpr.ca>
Date: 27 Dec 89 18:08:07 GMT
References: <84768@linus.UUCP> <8840004@hpfcso.HP.COM> <1804@l.cc.purdue.edu>
Sender: news@eric.mpr.ca
Reply-To: hui@joplin.mpr.ca
Organization: Microtel Pacific Research Ltd., Burnaby, B.C., Canada
Lines: 15


For a frame of reference, T.I.'s DSP chip TMS320C25 does a 16x16
multiply in a single cycle.

AMD's Am29000 uses multiple instructions to accomplish a general 32x32
multiply. I say general because, quoating from 7.1.6 of the User's
Manual (c)1987:

"It may be beneficial to precede a full multiply procedure with a routine
to discover whether or not the number of multiply steps may be reduced."

In other words, the compiler is given more room for optimization here
when faced with two arguments of different precision.

Michael Hui  604-985-4214  hui@joplin.mpr.ca