Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!bionet!ig!arizona!debray
From: debray@arizona.edu (Saumya K. Debray)
Newsgroups: comp.lang.prolog
Subject: Re: Inline expansion versus threaded code
Summary: paging, code reorganization
Message-ID: <10595@megaron.arizona.edu>
Date: 3 May 89 15:52:07 GMT
References: <1654@kulcs.kulcs.uucp>
Organization: U of Arizona CS Dept, Tucson
Lines: 27



There seems to be a consensus that native code implementations of
Prolog may suffer because of non-locality of code references,
and the excessive paging that results.  However, it may be
possible to reduce the amount of paging, using code reorganization
techniques developed for cache management.  The following is the
abstract of a TR titled "Code Reorganization for Instruction Caches"
by A. Dain Samples and Paul Hilfinger (UC Berkeley, Oct 1988):

   "Program performance can be improved on machines with virtual
    memory by reorganizing the program's address space.  We
    address the question whether reorganization could benefit
    programs on machines with instruction caches.  We have
    performed experiments to determine the efficacy of
    restructuring using simple reordering algorithms and profile
    data, concluding that performance improvement can be
    obtained relatively cheaply.  Experiments show improvements
    in miss rates on the order of 30% to 50%, and sometimes as
    high as 50% to 80%, by performing a simple algorithm that
    relocates only 3% to 8% of the basic blocks of a program."

-- 
Saumya Debray		CS Department, University of Arizona, Tucson

     internet:   debray@arizona.edu
     uucp:       arizona!debray