Path: utzoo!attcan!uunet!husc6!bbn!mit-eddie!uw-beaver!cornell!rochester!pt.cs.cmu.edu!k.gp.cs.cmu.edu!lindsay
From: lindsay@k.gp.cs.cmu.edu (Donald Lindsay)
Newsgroups: comp.arch
Subject: Re: Japanese 32-bit micro can be a 68020 or 80386
Keywords: maybe
Message-ID: <1700@pt.cs.cmu.edu>
Date: 18 May 88 03:54:41 GMT
References: <2006@sugar.UUCP>
Sender: netnews@pt.cs.cmu.edu
Organization: Carnegie-Mellon University, CS/RI
Lines: 39

In article <2006@sugar.UUCP> karl@sugar.UUCP (Karl Lehenbauer) writes:
>There is an article in this week's "PC Week" magazine about a 32-bit micro
>developed in Japan ...
>...that has a writable control store 
>and thus can have different microprograms loaded into it to emulate the
>80386, 68020 and others ...
>It would be a boon to
>researchers and others trying to implement oddball language architectures
>and get them to execute efficiently on a CISC-type machine (Smalltalk, LISP, 
>etc.)

Microprogrammable hardware has been for sale for forever. (Well, almost
forever. I helped evaluate one in 1969. ) The only new aspect here is
having the writable control store on-chip.

It's flexible, all right, but it doesn't necessarily run preexisting
instruction sets as fast as dedicated designs can. It will also be slow when
doing things that it wasn't quite designed for. (I remember a Microdata (?)
box that was programmed to be a PDP-11. Throughput was awful, simply because
the box did not want to believe in three-bit fields.)

There used to be three big wins. One was that the microcode often had access
to extra registers. Another was that there were fewer instruction fetches,
either because your macroinstruction set was application specific, or else
because your most crucial subroutines were written in microcode.

Neither of these is as big a win anymore. The newer machines have more
registers, and the Harvard machines don't do their instruction fetches over
the data bus.

The third  (and least) win came when the low-level muck-with-the-bits coding
allowed some result to come out of the hardware in fewer clocks.  This win
can still happen, but it tends to happen when you have special data types,
or when the hardware supports e.g. tags. Most of these wins are the
territory of the Lisp engines and the graphics engines.


-- 
	Don		lindsay@k.gp.cs.cmu.edu    CMU Computer Science