Zero Ray Direction?? back

(L) [2006/09/13] [Michael77] [Zero Ray Direction??] Wayback!

Hi,


I have my raytracer with a BIH up and running right now, but it a ray is 0 on any axis, the results get simply wrong. I guess 1/0 is not really a good value to work with [SMILEY Sad] So I wonder, how are you handling these cases? My current approach is to avoid 0 by slightly modifying the ray direction (to 1.0e-8f) which works in all cases but is not really what I want (especially since the ray generation gets more expensive).


So how can I handle these cases correctly?

(L) [2006/09/13] [funky] [Zero Ray Direction??] Wayback!

It should be no problem. You have to check if all possible cases are handled right:

1 / 0 = INFINITY ( the biggest floating point number) INFINITY >= x is always true and

INFINITY < x is allway false ... the same with -INFINITY.

NaN (Not a Number) <=, >= is allways false ( i.e. INFINITY - INFINITY = NaN )

(L) [2006/09/14] [Michael77] [Zero Ray Direction??] Wayback!

Ok, you´re right, it was a NaN caused by the newton-rhapson iteration for the reciprocal computation. Using a real divide instruction fixes the problems.

(L) [2006/09/14] [tbp] [Zero Ray Direction??] Wayback!

Or bunkerize your approximated reciprocal.