Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!ucla-cs!ucla-ma!euphemia!pmontgom
From: pmontgom@euphemia.math.ucla.edu (Peter Montgomery)
Newsgroups: comp.arch
Subject: Re: IEEE arithmetic (Goldberg paper)
Message-ID: <1991Jun8.224715.18882@math.ucla.edu>
Date: 8 Jun 91 22:47:15 GMT
References: <9106052113.AA14439@ucbvax.Berkeley.EDU> <3652@charon.cwi.nl>
Sender: news@math.ucla.edu
Organization: UCLA Mathematics Dept.
Lines: 27

In article <3652@charon.cwi.nl> dik@cwi.nl (Dik T. Winter) writes:
>I still have a question here: do you know realistic cases where inferior
>code will be generated?  I do not think that 'a = 0.0 * b' is realistic.

	The literal zero will probably not appear, but it will not be
unusual to have

	const double coef_friction = 0.0;	/* Ignore friction */

early in a program and coef_friction multiplied by other quantities
later on.  Macro expansion (or inline expansion of procedures)
is another possibility, if one argument is 0.0.  
A third possibility is loop unrolling:

	do i = 0, 2
	    deriv(i) = REAL(i)*poly(i)
	end do

will expand into deriv(0) = 0.0*poly(0), deriv(1) = 1.0*poly(1),
deriv(2) = 2.0*poly(2).  The benefit of this unrolling
really comes when it simplifies this to deriv(0) = 0, 
deriv(1) = poly(1), deriv(2) = poly(2) + poly(2).
--
        Peter L. Montgomery 
        pmontgom@MATH.UCLA.EDU 
        Department of Mathematics, UCLA, Los Angeles, CA 90024-1555
If I spent as much time on my dissertation as I do reading news, I'd graduate.