Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!news.sprintlink.net!newsfeed.internetmci.com!news.mathworks.com!uhog.mit.edu!nntp.club.cc.cmu.edu!cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!andrew.cmu.edu!mwhite+ From: Matthew Jason White <mwhite+@CMU.EDU> Newsgroups: comp.unix.bsd.freebsd.misc Subject: Re: Packages on a 386? Date: Wed, 16 Aug 1995 05:09:21 -0400 Organization: Senior, Math/Computer Science, Carnegie Mellon, Pittsburgh, PA Lines: 27 Message-ID: <4kAPL1W00YUpA15HMG@andrew.cmu.edu> References: <3vv184$qou@cronkite.cisco.com> <ASAMI.95Aug10151737@forgery.cs.berkeley.edu> <40pb71$jut@cronkite.cisco.com> <DDE4yM.9G1@theatre.pandora.sax.de> NNTP-Posting-Host: po8.andrew.cmu.edu In-Reply-To: <DDE4yM.9G1@theatre.pandora.sax.de> Excerpts from netnews.comp.unix.bsd.freebsd.misc: 16-Aug-95 Re: Packages on a 386? by Martin Welk@theatre.pand > I may be wrong, but as I remember, the way a CPU without co-processor > works is this: if there appears a floating point instruction and there > is no FPU, the CPU generates that ``floating exception'' to leave the > possibility to the software to get around that. That raises an interesting point. Some neural net code I've been developing is getting floating point exceptions due to a divide by zero (the program has a bug). What bothers me is that divide by zero should not cause a floating point exception in an IEEE compliant system (the correct answer should be +/-Inf). Similarly, I seem to be getting an exception due to floating point overflows (I'm not positive about this, it may just be an unseen divide by zero...in any case, that should return NaNQ). So, the question is: does the FreeBSD emulator handle these situations correctly? If I enable the emulator at compile time, will floating point math still go to the FPU, and then to the software if the FPU flames out? I suppose I should just go look at the kernel sources, but I simply have too many other, pressing, concerns. -Matt