Path: utzoo!utgpu!watmath!clyde!att!osu-cis!killer!ames!oliveb!amiga!boing!dale
From: dale@boing.UUCP (Dale Luck)
Newsgroups: comp.sys.amiga
Subject: Re: 68881 and the PA was:(none)
Message-ID: <573@boing.UUCP>
Date: 5 Jan 89 12:54:12 GMT
References: <6018@louie.udel.EDU> <533@boing.UUCP> <1434@percival.UUCP> <547@boing.UUCP> <13012@cup.portal.com> <1441@percival.UUCP> <13145@cup.portal.com> <1443@percival.UUCP>
Reply-To: dale@boing.UUCP (Dale Luck)
Distribution: na
Organization: Boing, Milpitas, Ca.
Lines: 38

In article <1443@percival.UUCP> billc@percival.UUCP (William J. Coldwell) writes:
>In article <13145@cup.portal.com> DMasterson@cup.portal.com (David Scott Masterson) writes:
>[Stuff deleted concerning PA and MathChip]
>>Can software be compiled in such a way as to leave it up to the consumer which
>>math library (or any libraries with competing types) is used in a generic
>>sort of fashion?
>No.
More information is helpful in this case. The format of the floating point
numbers is different. Compilers will generate the floating point constants
in the format that it needs. This makes it difficult to make the decision
at runtime unless you have the constants compiled once for each format
as well as stash them away in a structure so you can easily pass pointers
around.
   The other way to make it retargetable in terms of floating point is
to isolate all your floating point in a couple of files that are compiled
multiple times and you use a table of functions to pick the appropriate
precision and speed.

>Since most people do not have mathchips, most developers will only put a
>(n) FFP version on the disk - even if a 68881 will speed things up.  On
>the other hand, people won't buy MCs if there aren't any programs out
>there that use it. Catch 22.

The new single precision IEEE libraries that CA is working on should help
solve this dilemma. They are currently in alpha test, support both style
access to the 68881(like the 1.3 dbl precision library) as well as the
goal of the software emulation is to be as fast as ffp. The single precision
ieee format also has an exponent range twice as big as ffp.

So maybe we can have our cake and eat it too.

>>David Masterson
>-- 
>               William J. Coldwell - Amiga Attitude Adjuster

-- 
Dale Luck     GfxBase/Boing, Inc.
{uunet!cbmvax|pyramid}!amiga!boing!dale