Return to BSD News archive
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!elroy.jpl.nasa.gov!swrinde!ihnp4.ucsd.edu!library.ucla.edu!news.mic.ucla.edu!kaleida.com!conklin From: conklin@kaleida.com (J.T. Conklin) Newsgroups: comp.os.386bsd.bugs Subject: Re: math_emulate problem NetBSD .9 Date: 07 Mar 1994 22:13:27 GMT Organization: Winning Strategies, Inc. Lines: 18 Message-ID: <CONKLIN.94Mar7141327@ngai.kaleida.com> References: <2l2tun$1qk@news.iastate.edu> <vp.762814710@news.forth.gr> <1994Mar5.011654.26711@dde.dk> Reply-To: conklin@kaleida.com NNTP-Posting-Host: ngai.kaleida.com In-reply-to: kim@dde.dk's message of Sat, 5 Mar 1994 01:16:54 GMT In article <1994Mar5.011654.26711@dde.dk> kim@dde.dk (Kim Andersen) writes: Kim> Another solution might be to upgrade to NetBSD-current which has much Kim> improved FPU emulation. Although there have been some changes to the FPU emulator since 0.9, I believe that the most significant changes wrt. floating point have been to the math library itself. The NetBSD-current math library is derrived from Sun's fdlibm 5.1. It can be configured (at run time) to be ANSI/POSIX, X/OPEN XPG3, SVID, and IEEE 754 compiliant. The new library returns correct values for acos(), asin(), ceil(), floor(), and sqrt() on machines without FP coprocessors. The old library is known to generate wildly inaccurate values for these functions. --jtc