*BSD News Article 38392


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!yarrina.connect.com.au!werple.apana.org.au!otis.apana.org.au!serval.net.wsu.edu!netnews.nwnet.net!oracle.pnl.gov!osi-east2.es.net!lll-winken.llnl.gov!uwm.edu!math.ohio-state.edu!jussieu.fr!univ-lyon1.fr!ensta!bsdtest.ensta.fr!bouyer
From: bouyer@bsdtest.ensta.fr (Manuel Bouyer)
Newsgroups: comp.os.386bsd.misc
Subject: Intel Pentium FDIV bug (fwd)
Date: 22 Nov 1994 17:02:58 GMT
Organization: Ecole Nationale Superieure de Techniques Avancees, Paris
Lines: 74
Distribution: world
Message-ID: <3at882$74r@ici-paris.ensta.fr>
NNTP-Posting-Host: bsdtest.ensta.fr

This article comes from to comp.os.linux.announce, but i think this may interrest
Net/FreeBSD users too.

>Path: ensta!jussieu.fr!math.ohio-state.edu!usc!hookup!swrinde!pipex!sunic!news.funet.fi!news.csc.fi!news.helsinki.fi!not-for-mail
>From: ig25@fg70.rz.uni-karlsruhe.de (Thomas Koenig)
>Newsgroups: comp.os.linux.announce
>Subject: Intel Pentium FDIV bug
>Followup-To: comp.os.linux.misc
>Date: 21 Nov 1994 09:41:19 +0200
>Organization: University of Karlsruhe, Germany
>Lines: 46
>Sender: wirzeniu@cc.Helsinki.FI
>Approved: linux-announce@tc.cornell.edu (Lars Wirzenius)
>Message-ID: <3apiuv$3is@plootu.Helsinki.FI>
>References: <3a58g0$75m@vkhdsu01.hda.hydro.com>
>Reply-To: Thomas.Koenig@ciw.uni-karlsruhe.de
>NNTP-Posting-Host: plootu.helsinki.fi
>Mime-Version: 1.0
>Content-Type: text/plain; charset=ISO-8859-1
>Content-Transfer-Encoding: 8bit
>Keywords: Pentium, floating point bug, FDIV, division
>
>[ Moderator's note (gosh, it's been a while since I did one of these,
>  hasn't it :-): While this is not strictly Linux-related, I think it's
>  interesting enough for Linuxers.  Besides, Linus has a Pentium on
>  his desk at work. :-)  --liw ]
>
>Recently, a bug has been discovered in the floating point arithmetic
>of Intel's Pentium processors.
>
>This bug affects the FDIV instruction in double precision.  According to
>information on comp.sys.intel, the bug will manifest itself for ceratin
>bit patterns; the odds of hitting it are about 1/2^18, or 1/6E7.  The
>bug reduces the precision of the result to about 8 significant digits
>(from 16).
>
>Apparently, Intel has fixed the problem with its newest chips; it
>is still undecided wether to offer people an upgrade, however.  For
>more information, see comp.sys.intel, the "FDIV bug" thread.
>
>If you have used a Pentium system for numerical work which depends
>heavily on double precision floating point arithmetic, you are
>advised to check your results against those on a 486 system.
>
>Here is a simple test program that will reproduce the bug, if present;
>if this prints "256.000000" as result, your system is affected; on a 486,
>this prints "0.000000".
>
>#include <stdio.h>
>
>int main()
>{
>    double x,y,z;
>
>    x = 4195835.0;
>    y = 3145727.0;
>    z = x - (x / y) * y;
>    printf("%f\n",z);
>    return 0;
>}
>--
>Thomas Koenig, Thomas.Koenig@ciw.uni-karlsruhe.de, ig25@dkauni2.bitnet.
>The joy of engineering is to find a straight line on a double
>logarithmic diagram.
>
>--
>Send submissions for comp.os.linux.announce to: linux-announce@tc.cornell.edu
>PLEASE remember Keywords: and a short description of the software.


--
Manuel Bouyer, Ecole Nationale Superieure de Techniques Avancees, Paris
email: bouyer@ensta.fr
--