Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!ukma!tut.cis.ohio-state.edu!ucbvax!agate!eris.berkeley.edu!mwm
From: mwm@eris.berkeley.edu (Mike (I'll think of something yet) Meyer)
Newsgroups: comp.sys.amiga.tech
Subject: Re: Lattice 5.02 floating point
Message-ID: <24891@agate.BERKELEY.EDU>
Date: 26 May 89 08:38:58 GMT
References: <6@iisat.UUCP>
Sender: usenet@agate.BERKELEY.EDU
Reply-To: mwm@eris.berkeley.edu (Mike (I'll think of something yet) Meyer)
Distribution: na
Organization: Missionaria Phonibalonica
Lines: 27

In article <6@iisat.UUCP> promac@iisat.UUCP (Promac Systems) writes:
<I seem to be having a problem with accessing the 68881 directly in my 2500.
<I have been in contact with Lattice and just got an ear full of static. The
<problem seems to be in a few of the functions(ie log(xx)) gives me an answer
<that is way off of what it should be(where xx stands for a floating point #).
<ie. I run the savage benchmark using lc -fi -Lm, and I get an answer of 
<    about 2.5E+04. I also get the same answer when using the normal math librari    -es. But when I access the 68881 by using the m68881.h header and -f8
<    option on the compiler I get an answer of 3.483E+308???????

Look in include:m68881.h (be sure to get the text, not compacted,
version). All the math calls are defined as __builtin_fpc(<hexconstant>, d)
or some such. These all generate things to tweak the 68881, using the
constants to determine what operation you're going to do. I've been
told that if you're 68881 savy, you can figure out what they are
yourself, and what they should do.

Some of the constants are wrong. Most notablly, log. Lattice knows
this. Not having a 68881, I didn't get the details. Call the support
BBS and check the Amiga discussion for details.

	<mike

--
Estant assis, de nuit secrette estude,			Mike Meyer
Seul, repose sur la selle d'airain,   			mwm@berkeley.edu
Flambe exigue, sortant de solitude,   			ucbvax!mwm
Fait proferer qui n'est a croire vain.			mwm@ucbjade.BITNET