Path: utzoo!utgpu!news-server.csri.toronto.edu!csri.toronto.edu!norvell
Newsgroups: comp.software-eng
From: norvell@csri.toronto.edu (Theo Norvell)
Subject: Re: Cost/Benefit of compiler optimization techniques?
Message-ID: <1990Nov28.224718.27555@jarvis.csri.toronto.edu>
Keywords: optimize, design
Organization: CSRI, University of Toronto
References: <9011280511.aa16546@ICS.UCI.EDU>
Date: 29 Nov 90 03:47:18 GMT
Lines: 24

In article <9011280511.aa16546@ICS.UCI.EDU> Ira Baxter <baxter@zola.ICS.UCI.EDU> writes:
>
>I am interested in finding out the "most useful" optimizations (at both
>levels) for conventional procedural languages (FORTRAN, C, etc.), by
>comparing average quantitative payoffs which rank them (say, reduction in
>execution time measured over some large suite of user programs) to some
>measure of the average effort to implement that optimization (measured in
>man-somethings).  Essentially what I am asking is, "What are the
>cost/benefit ratios of various techniques"?  Additional useful information
>would be something like conditional utility, i.e., if technique A is used,
>then technique B is X% less useful.
                        ^^^^ (or more!)

An ought-to-be classic work on just this question is Fredrick Chow's thesis,
done under John Henessey:

Author:       F.C. Chow
Title:        A Portable Machine-independent Global Optimizer - Design and
              Measurements
Report:       CSL T.R. 83-254
Publisher:    Computer Systems Laboratory, Stanford University
Availability: cost: $8.25

Theo Norvell