home *** CD-ROM | disk | FTP | other *** search
- Path: phoenix.rhein.de!yaps!arno
- From: arno@yaps.rhein.de (Arno Eigenwillig)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: PPC compilers
- Message-ID: <eTzVx*N3f@yaps.rhein.de>
- Date: Tue, 09 Jan 1996 14:33:50 +0100
- References: <4com6v$415@maureen.teleport.com> <jasonb.821098303@cs.uwa.edu.au>
- <volker.0e58@vb.franken.de>
- Organization: Yet Another Private Site in Meckenheim, Germany
- X-Copyright: This article may not be distributed on a CD-ROM
- or in printed form without prior written consent of the author.
- X-Newsreader: Arn V 1.04
-
- In article <volker.0e58@vb.franken.de>, Volker Barthelmann writes:
-
- > You sometimes need to know it in HLLs, too. Numerical calculations almost
- > always need very detailed information on the data types. Just have a look
- > at a C compiler's float.h.
- >
- > In many cases You can forget such things after the declaration, but there
- > are lots of other cases, too.
-
- Once again "The Volker Pathological Cases Show"? :-)
- No offense; and this time, I (partly ;-) agree.
-
- Would you concede that what matters about the properties of floating
- point types for fp calculations is not their actual representation on
- hardware level, but their behaviour, i.e. a strict definition of what
- value a certain operation on two given values produces?
-
- I admit that the current ANSI C Standard gives insufficient guarantees
- for real number-crunching projects. Usually you can only achieve more
- limited portability, say, ANSI C with IEEE-nnn fp.
-
- -- __
- __/// Arno Eigenwillig /\ <arno@yaps.rhein.de> \/ PGP key available.
- \XX/ V+49-2225-5870 /\ <Arnooo @ #amigager> \/ MIME 8bit welcome.
-