Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!bloom-beacon!apple!vsi1!wyse!mips!prls!philabs!linus!munck
From: munck@linus.UUCP (Robert Munck)
Newsgroups: comp.graphics
Subject: Re: PS300 Function Networks
Message-ID: <49041@linus.UUCP>
Date: 11 Apr 89 17:32:05 GMT
References: <7710@batcomputer.tn.cornell.edu> <4751@cloud9.Stratus.COM>
Reply-To: munck@linus.UUCP (Robert Munck)
Organization: The MITRE Corporation, McLean VA
Lines: 20

In article <7710@batcomputer.tn.cornell.edu>, garry@batcomputer.tn.cornell.edu (Garry Wiegand as guest) writes:
> (For those who don't know: a function network is a directed graph
> of arithmetic functions. Data flows in one end, for example from a
> mouse, follows the connections, is transformed in arbitrary ways, 
> and flows out the other end, for example to affect a screen display. 
> The data flow happens *continuously*.  (Carl, did I say it right?))


I recently saw a new Navy processor called the AN/UYS-2 that was designed
to work that way.  It has eight or so arithmetic engines each capable of
about 5 Megaflops, a billion or so bytes of memory, and various other
wonders.  It's intended for signal processing, and you program it by
drawing dataflow diagrams on a workstation.  You can also write Ada programs
that control the topology of the graph while it's running.  A real data-
chewer; I'd love to see it used for graphics.  Size is that of a large 
suitcase, cost is ... if you have to ask, you can't afford it.

BTW, I believe that the first mention of functionnetworks or data-flow
architectures was by Jorge Rodriquez in his MIT thesis in the mid 1960's.
                     -- Bob Munck, MITRE