Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uwm.edu!lll-winken!ames!haven!adm!smoke!gwyn
From: gwyn@smoke.BRL.MIL (Doug Gwyn)
Newsgroups: comp.unix.wizards
Subject: Re: Floating Point Expectations
Keywords: Double Precision Floating Point
Message-ID: <13000@smoke.BRL.MIL>
Date: 25 May 90 16:31:49 GMT
References: <12977@smoke.BRL.MIL> <1990May24.132423.3080@eddie.mit.edu> <995@s8.Morgan.COM>
Organization: U.S. Army Ballistic Research Laboratory, APG, MD.
Lines: 13

In article <995@s8.Morgan.COM> amull@Morgan.COM (Andrew P. Mullhaupt) writes:
>It becomes quite clear that this authoritative and very recent
>reference is written with the point of view that you should _not_
>assume the radix of the floating point representation is 2, ...

It is even more obvious when you consider the wide variation even among
binary-radix floating-point processors with respect to guard bits,
rounding modes, and so on that it is insane to think than any simple
model of floating-point is going to apply accurately universally.
Rather than attempting to model exact behavior of floating-point systems,
I recommend devising algorithms that are robust in the face of
considerable cruft from the floating-point unit.  This is virtually
always possible, if you undertake the task with sufficient suspicion.