Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!njin!princeton!taylor!ssr
From: ssr@taylor.Princeton.EDU (Steve S. Roy)
Newsgroups: comp.arch
Subject: Re: Compiler Costs
Message-ID: <1225@idunno.Princeton.EDU>
Date: 15 Jul 90 02:24:51 GMT
References: <40052@mips.mips.COM> <628@dg.dg.com> <64045@sgi.sgi.com>
Sender: news@idunno.Princeton.EDU
Reply-To: ssr@taylor.Princeton.EDU (Steve S. Roy)
Organization: Princeton University
Lines: 29

Burce Karsh writes:

>>In article <628@dg.dg.com> publius@dg-pag.webo.dg.com (Publius) writes:
>>
>>Many "large" programs spend most of the CPU time in one or a few relatively
>>"tiny" routines.
>
>This is an often stated.  Is there any evidence for this?

Oh yes.

Naturally it depends on what type of code you're talking about.  There
is a strong tendency for a person who works on a particular type of
program to take the things s/he does as completely typical and
assumes their problems are the same as everyone else's.  This is very
much not the case -- the TYPES of bottlenecks people find vary a lot.
Scientific number-crunchers are not like editors are not like
compilers etc.

Much of my experience is with scientific number crunchers.  It is VERY
VERY VERY common that >90% of the execution time of such programs is
spent in <<10% of the code, even just a few dozen lines.  It is also
VERY VERY VERY true that speed is often of paramount importance.

For the types of programs I use, those conditions are true more often
than not.

Steve Roy
ssr@acm.princeton.edu