Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!cs.utexas.edu!usc!elroy.jpl.nasa.gov!ucla-cs!frazier
From: frazier@oahu.cs.ucla.edu (Greg Frazier)
Newsgroups: comp.arch
Subject: Re: Benchmarks for Register Windows Study
Message-ID: <33137@shemp.CS.UCLA.EDU>
Date: 17 Mar 90 00:16:54 GMT
References: <6977@uhccux.uhcc.hawaii.edu>
Sender: news@CS.UCLA.EDU
Reply-To: frazier@oahu.UUCP (Greg Frazier)
Organization: UCLA Computer Science Department
Lines: 23

In article <6977@uhccux.uhcc.hawaii.edu> karlton@wiliki.eng.hawaii.edu (Karlton Lau) writes:
>
>I'm working on a research project which will try to optimize the number
>of frames that are bumped/restored during overflows/underflows in an
>architecture with overlapping multiple register sets.  In order to
>do this I need some procedure-intensive benchmarks on which I can do 
>trace studies.

Make sure you do a good literature search before you try to
publish anything.  A lot of work has been done in this area.
If I recall what I have read, without doing anything tricky
the optimal number of frames to save/restore is 1.  Of course,
if you are doing something tricky, or if it is real expesive
to load the code to save/restore regs on whatever architecture
you are exploring, then your mileage will vary.  Unfortunately,
I do not have the reference handy, but if I find it, I'll post
it.  Oh, and if your interested in procedure-intensive
benchmarks, try Towers of Hanoi.  Not a very realistic benchmark,
but there you go.
+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
Greg Frazier				"Big A, little a / What begins with A?
frazier@CS.UCLA.EDU			Aunt Annie's Alligator / A ... a ... A"
!{ucbvax,rutgers}!ucla-cs!frazier			_Dr._Seuss's_ABCs_