Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!umnd-cs!umn-cs!hall!pmk
From: pmk@hall.UUCP
Newsgroups: comp.lang.c,comp.lang.forth,comp.lang.misc
Subject: Re: The winner!
Message-ID: <410@hall.cray.com>
Date: Fri, 31-Jul-87 16:21:25 EDT
Article-I.D.: hall.410
Posted: Fri Jul 31 16:21:25 1987
Date-Received: Sun, 2-Aug-87 10:36:57 EDT
References: <398@sugar.UUCP> <8326@utzoo.UUCP> <1946@aw.sei.cmu.edu>
Organization: Cray Research, Inc., Mendota Heights, MN
Lines: 32
Xref: utgpu comp.lang.c:3255 comp.lang.forth:141 comp.lang.misc:580
Summary: An even faster interpreter

In article <1946@aw.sei.cmu.edu>, firth@sei.cmu.edu (Robert Firth) writes:
> In article <8326@utzoo.UUCP> henry@utzoo.UUCP (Henry Spencer) writes:
> >> 
> >> NEXT:
> >> 	jmp (IP)+
> >> 
> >> DOCOL:
> >> 	jsr (IP),<first word of definition>
> >
> >Given that that one instruction takes 1.68 us in the first case and 3.7 in
> >the second, this being on a relatively fast 11 (the 44), I would say it is
> >pretty easy to come up with a faster interpreter on something like a MIPS
> >machine...
> 
> On our M/500, the time taken is about 250 ns for each sequence.
> 
> Can anyone beat THAT?

Well, yes.  Cray-2 code for NEXT is:

	a1	0
	a2	1
	a1	a1+a.IP
	a.IP	a.IP+a2
	j	a1

and DOCOL is:

	a1	<first word of defn>
	r,a.IP	a1

It's faster than the M/500, but the hardware is a little more expensive.