Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watnot!watmath!clyde!rutgers!brl-adm!adm!V125KJG8%UBVMS.BITNET@wiscvm.wisc.edu
From: V125KJG8%UBVMS.BITNET@wiscvm.wisc.edu
Newsgroups: comp.lang.c
Subject: Re: Floating Point Errors
Message-ID: <5056@brl-adm.ARPA>
Date: Fri, 20-Mar-87 13:27:16 EST
Article-I.D.: brl-adm.5056
Posted: Fri Mar 20 13:27:16 1987
Date-Received: Sun, 22-Mar-87 14:32:31 EST
Sender: news@brl-adm.ARPA
Lines: 49

In article <652@sdchema.sdchem.UUCP> tps@sdchem.UUCP (Tom Stockfisch) writes:

>In article <365@bms-at.UUCP> stuart@bms-at.UUCP (Stuart D. Gathman) writes:

>>In article <4804@brl-adm.ARPA>,Schauble@mit-multics.arpa (Paul Schauble)
 writes:

>>>...[claim that with float x,y, the test (x*y==x*y) can fail on many machines

>Every single usage of "==" for floating point can fail?
>What about
>
>    double    dflmax();
>    double    x =    dflmax();
>
>    if ( x == dflmax() )
>or
>    double    x = 0;
>
>    if ( x == 0 )
>        ...
>
> ?

I can't accept the fact that (x * y == x * y) will fail, because the floating
point will work the same way.

Perhaps we mean something like this:

  if (7.0 * 5.0 == 35.0)  /* which can fail */

Or, possibly this will fail:

float x = 5.0, y = 7.0, z;

z = x * y * y;
z /= y;
if (x * y == z)
..

depending on how the CPU handles the floating point.

Have fun and see yuh...
  --Curtis R. Anderson
    State University of New York at Buffalo
    (V125KJG8@UBVMS.bitnet)

[opinions are my own, and do not necessairily represent views of SUNYAB.  If
they happen to do so, then that is strictly coincidence.]