Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!uakari.primate.wisc.edu!pikes!aspen.craycos.com!dmk
From: dmk@craycos.com (David Keaton)
Newsgroups: comp.arch
Subject: Re: Page size and linkers (was: Re: SunMMU history)
Message-ID: <1991Jan29.033024.1516@craycos.com>
Date: 29 Jan 91 03:30:24 GMT
References: <MOSS.91Jan21172107@ibis.cs.umass.edu> <45242@mips.mips.COM> <1991Jan27.214522.24408@watdragon.waterloo.edu>
Organization: Cray Computer Corporation
Lines: 25

In article <1991Jan27.214522.24408@watdragon.waterloo.edu> tbray@watsol.waterloo.edu (Tim Bray) writes:
>mash@mips.COM (John Mashey) writes:
> (Consider the kind of
> program that's >200MB of code, much of it in a giant single loop,
> leading to a high I-cache miss rate.)
>
>PLEASE tell us that's a thought experiment, John.  Does the world really
>contain such appalling creations?  If so, I think a taxonomy would be of
>general interest to the readers here...

     I can imagine 200MB outer loops;  there are certainly at least
near megabyte inner loops in real code out there.  Back at Multiflow I
remember an actual user's code that simulated the aerodynamic
properties of a cyclical device.  (To protect the guilty, I won't say
what the device was.)  There was one enormous inner loop.  When the
device got past a certain point in its cyclical behavior, you could
actually watch the cache-miss lights come on in the back of the
machines as the loop extended way beyond the length of the I-cache.
That VLIW I-cache held the equivalent of about 56k RISC instructions
on Multiflow's smallest machine, with a memory representation of the
cache size taking about 512kB.

					David Keaton
					dmk@craycos.com
					uunet!dmk3b1!dmk