Path: utzoo!utgpu!water!watmath!clyde!rutgers!cmcl2!nrl-cmf!ames!pasteur!ucbvax!ucsfcgl!cca.ucsf.edu!ccb.ucsf.edu!dick
From: dick@ccb.ucsf.edu (Dick Karpinski)
Newsgroups: comp.arch
Subject: Re: FREQUENCY statements: fill a much-needed gap
Summary: programmers are wrong
Message-ID: <1136@ucsfcca.ucsf.edu>
Date: 19 Jan 88 21:09:10 GMT
References: <839@ima.ISC.COM> <28200085@ccvaxa> <1267@winchester.UUCP> <2861@clash.rutgers.edu>
Sender: root@cca.ucsf.edu
Reply-To: dick@ucsfccb.UUCP (Dick Karpinski)
Organization: UCSF Computer Center
Lines: 26

In article <2861@clash.rutgers.edu> segall@clash.rutgers.edu (Ed Segall) writes:
>Yes, it it true that profiling (with appropriate input data, for
>programs with data-dependent profiles) is the best way to do this.
>However, predicting the most likely branch is not the only reason ...
>
>I think the missle example showed it best (to paraphrase: you don't
>want to optimize for the most common case, when no missles are coming
>in.  You'd be better off optimizing the very uncommon case when they
>are).  

Computer programs exhibit unexpected behavior very often.  Even in this
odd missle case, I would rather try to simulate the environment in which
the missles are coming and instruct the compiler to respect the observed
frequencies rather than expect the programmer to guess correctly.  For
example, it may well be that the usual case is still the usual case even
when the missles are in view.  Only a well designed optimiser can be 
expected to do a good job.  People are seldom much like well designed
optimisers.

Dick

Dick Karpinski  Manager of Minicomputer Services, UCSF Computer Center
UUCP:  ...!ucbvax!ucsfcgl!cca.ucsf!dick        (415) 476-4529 (11-7)
BITNET:  dick@ucsfcca or dick@ucsfvm           Compuserve: 70215,1277  
USPS:  U-76 UCSF, San Francisco, CA 94143-0704   Telemail: RKarpinski   
Domain: dick@cca.ucsf.edu  Home (415) 658-6803  Ans 658-3797