Newsgroups: comp.lang.scheme
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!ix.netcom.com!netcom.com!NewsWatcher!user
From: hbaker@netcom.com (Henry Baker)
Subject: Re: Locality of reference
Message-ID: <hbaker-0703951344480001@192.0.2.1>
Sender: hbaker@netcom19.netcom.com
Organization: nil
References: <3jfv86$134@agate.berkeley.edu>
Date: Tue, 7 Mar 1995 21:43:05 GMT
Lines: 29

In article <3jfv86$134@agate.berkeley.edu>, bh@anarres.CS.Berkeley.EDU
(Brian Harvey) wrote:

> Today at faculty lunch, Dave Patterson gave a talk about the current
> trends in architecture.  His key point was that with superscalar
> architectures and 12-stage-deep pipelines, it's getting to where the
> cost of executing an instruction is near zero; the remaining
> bottleneck is DRAM speed, so what matters is cache hit rate, more
> than anything else.
> 
> There has been a lot of discussion on c.l.s lately about compiler
> optimization, but what I remember of it has mainly been about reducing
> the number of instructions executed, e.g., by type inference and
> inlining of primitives.
> 
> What is the state of the art in memory allocation for lists?  Can a
> smart compiler achieve locality of reference for the typical sort of
> Scheme code CDRing down a list?

Check out the proceedings of the Lisp & Funct. Prog. 1994 conf.  ML
type inference can be used to create lists in which every pair of
elements is a single node.  Presumably this could be useful in
doubling the speed of list traversal.  This could be done in Lisp as
well, but ML+type inference allows this to be completely hidden from the
programmer.

-- 
www/ftp directory:
ftp://ftp.netcom.com/pub/hb/hbaker/home.html
