Xref: utzoo comp.arch:12336 comp.lang.prolog:2102 comp.ai:5030
Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!gem.mps.ohio-state.edu!tut.cis.ohio-state.edu!ucbvax!bellatrix.uucp!vanroy
From: vanroy@bellatrix.uucp
Newsgroups: comp.arch,comp.lang.prolog,comp.ai
Subject: Re: Looking for Prolog Chip
Message-ID: <32468@ucbvax.BERKELEY.EDU>
Date: 10 Nov 89 02:05:48 GMT
Sender: usenet@ucbvax.BERKELEY.EDU
Reply-To: vanroy@bellatrix.uucp ()
Organization: University of California at Berkeley
Lines: 30

Jim Hunt (hunt@ernie.berkeley.edu) mentions:
>The original PLM group has all but disbanded here.  
>Al Despain is now at U.S.C., despain@troy.usc.edu
>Yale Patt is now at U. Michigan, patt@crim.eecs.umich.edu
>
>Peter Van Roy is still here, vanroy@ernie.Berkeley.EDU

  This is true; Al Despain and Yale Patt have both moved away from Berkeley.
However, the current Aquarius group at Berkeley is still active in Prolog
architecture research, working with Al Despain at USC, not yet disbanded (as
far as I can tell ;-) ).  Some of the original PLM people have graduated,
including Tep Dobry, Barry Fagin, and Wayne Citrin.
  We are currently building a new Prolog chip, the BAM chip, trying to learn
from the past.  The idea is to find the minimal extension to a general-purpose
architecture that's necessary for high performance of Prolog, without
compromising general-purpose speed.  The new chip was designed together with
the compiler, and we are trying to do as much as possible in software.
There's been lots of interaction between compiler and architecture design.
It seems to be that much is possible in software, but that a small but
crucial 'kernel' of hardware is needed for the highest performance.
  We have been working on this project for about a year now, and we should
have some hard numbers in a few months to corroborate or to disprove this
statement.
  The design itself is independent of the PLM and Xenologic, but we have done
lots of measurements of the VLSI-PLM.  The problem is that some of the
statistics change when we move to a different execution model.  But lots of
useful information is still there for the picking.

	Peter Van Roy
	Aquarius Project