> That's odd because when I was converting Quake to the ARM I managed to
> change all the floating points maths, with the exception of the QuakeC
> stuff, to use the 64 bit multiply. When it came to dividing the results
> achieved were more accurate in some places than the original floating point.
> I did use a reciprocal table that made use of all 32 bits though and used a
> shift to gain the required accuracy for each stage of the calculation.
> Using the 64 bit mul and accumulate also helped a lot when dealing with 64
> bit numbers.
we also dont had any accuracy problems (why should there be any? :).
Only the shift eats accuracy when the table is not big enough for
the number range.. but i think 16:16 fixed point math should do on AGB
and so the table would be just 128kb.
--
Manfred Linzner
(Project Manager)
Shin'en Multimedia
http://www.shinen.com
Tel.: ++49 (0)89 785 82 565
Fax.: ++49 (0)89 785 82 535
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
Send Email