Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!decwrl!sun!pitstop!sundc!seismo!uunet!fciva!dag
From: dag@fciva.FRANKLIN.COM (Daniel A. Graifer)
Newsgroups: comp.misc
Subject: Re: The definitive (I hope) answer to Re: Data/memory
Message-ID: <459@fciva.FRANKLIN.COM>
Date: 9 Mar 89 20:53:12 GMT
References: <16410030@hpfcdj.HP.COM> <3708@phri.UUCP> <5988@bsu-cs.UUCP> <2392@scolex.sco.COM>
Reply-To: dag@fciva.UUCP (Daniel A. Graifer)
Organization: Franklin Capital Investments, Inc. McLean, Va.
Lines: 24

As I remember, the old Burroughs Large Systems (which have modern versions
still on the market) (the old 67/68/77/7800 machines) had a 48 bit word
(not counting the 3 hidden"tag" bits) which the hardware was quite happy to
treat as a collection of 4,6,or 8 bit bytes.  In fact there are hardware
instructions for copying from one character pointer to another of different
size, via a third pointer to a translate table.  These pointer move
instructions were the only ones that could address main memory as bytes though. 
The processor data registers were all 1 word long, and (it was a strictly stack
architecture machine) LOAD and STORE instructions were all word-at-a-time.

Wasn't the Burroughs 1700/1800 dynamically microcode-able to support arbitrary
word sizes up 24 bits?  You dynamically loaded different microcode out of a 
fast cache to optimize the processor "architecture" for whatever your current
application/language was (at least that's how it was described to me!)  That
company probably deserves some kind of award for producing machines with 
creative architectures.

This was all more than 10 years ago, so I may be confusing much....

Dan

Daniel A. Graifer			Franklin Capital Investments
uunet!fciva!dag				7900 Westpark Drive, Suite A130
(703)821-3244				McLean, VA  22102