From:	Francois Charton <deef AT pobox DOT oleane DOT com>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: sqrt() problem
Date:	Wed, 11 Dec 1996 18:54:04 +0100
Organization:	CCMSA
Lines:	27
Message-ID:	<32AEF53C.74D2@pobox.oleane.com>
References:	<32aa47fd DOT 8683086 AT nntp DOT southeast DOT net> <32AABA23 DOT 15FC AT cs DOT com>
NNTP-Posting-Host:	deef.pobox.oleane.com
Mime-Version:	1.0
To:	fighteer AT cs DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

John M. Aldrich wrote:
> 
> #include <math.h>
> 
>     double s;
>     s = sqrt( pow( r, 2 ) - pow( m, 2 ) );
> 

This is *very* likely to be extremely slow... Power functions are 
usually much slower than multiplies...
I'd suggest either :
       sqrt(r*r-m*m);
or, better,
	sqrt((r-m)*(r+m));
(only one multiply)

The latter formula has another advantage, it should be a less 
floating-point-error-prone : if r and m are close and very big, 
r*r - m*m will result in incorrect results

Still better, formulae like
r * sqrt((1-m/r)*(1+m/r));
(if r and m are both positive).
Which are less rounding error prone, and will overflow less easily.

Francois

- Raw text -