dwi@manta.NOSC.MIL (Steve Stamper) (05/03/89)
I have recompiled the QRT version 1.5 ray trace program using Lattice C
5.0X and am posting the results of running the 8 sphere connected
by 12 reflecting cylinders example that the author provided.
Due to some small technical difficulties with the inline 881
code option in lattice C 5.0X I am unable to post results with
that option at this time, but based on my experience with that
option times will probably be around 25-28 mins. You can see
QRT is MUCH faster when compiled with lattice C 5.0X but
what this also illustrates is that unless you use
a very fast 68881, FFP math is about as fast as 68881 math! The FFP
library is loaded into 32 bit RAM using SETCPU to get this performance.
An FFP INLINE code option may outperform inline 881 math
on a stock amiga 2500 in all except transcendental functions
(sin, cos, sqrt, etc). FFP will also perform well on stock amigas.
For applications that do not need a lot of accuracy, like ray traceing
FFP may still be the way to go!
all tests On an A2620 w/ 3MB RAM (stock Amiga 2500)
using cylinder example
Original QRT v 1.5 program compiled with whatever Steve Koren used:
Qrt 55 mins 53 secs
My compiles using Lattice C 5.0X (QRT v 1.5)
QRTFFP 38 mins 18 secs ; ffp math
QRTLIB 45 mins 23 secs ; amigados 1.3 double precision math libraries using
; the 68881 chip.
-Roger