Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!samsung!emory!stiatl!srchtec!johnb
From: johnb@srchtec.UUCP (John Baldwin)
Newsgroups: comp.lang.c
Subject: Re: Floating point non-exactness
Message-ID: <160@srchtec.UUCP>
Date: 1 Aug 90 21:47:07 GMT
References: <622@.tetrauk.UUCP> <1990Jul31.101803.852@urz.unibas.ch>
Organization: search technology, inc.
Lines: 23

In article <1990Jul31.101803.852@urz.unibas.ch> carlos@urz.unibas.ch writes:
>
>> Can you ensure that "fpcmp (a, b)" and "fpcmp (a-b, 0.0)" yield the same
           ^^^^^^[emphasis mine]
>> result when a and b are very close?
>
>Yes, if your fp system conforms IEEE Std. 754 due to the existence of 
>denormalized numbers.

Actually, it depends on what the previous writer means by "ensure."
If the code has to be portable, one can't rely on IEEE-754 representation
of floating point values, since that's NOT part of the C standard.

Portably, you should be able to "get away with" using the values built in
to <float.h> (a required part of the standard library) for the epsilon
of your floating point representation.  [Just in case this slipped past
anyone, epsilon is the smallest quantity "e" such that x + e > x]

-- 
John T. Baldwin                      |  johnb@srchtec.uucp
Search Technology, Inc.              |  johnb%srchtec.uucp@mathcs.emory.edu
standard disclaimer:                 |  ...uunet!samsung!emory!stiatl!srchtec..
opinions and mistakes purely my own. |  ...mailrus!gatech!stiatl!srchtec...