Christoph Hormann wrote:
> Let me get this straight - you can't compare the speed because you can 
> only compile your patched version but not the unmodified code?

I can compile the original and the modified version of POV-Ray with 
VS.NET 2003... but the results are inconclusive.  In my limited testing, 
both builds look exactly the same, perform exactly the same (even with 
the Vista Buffer off), and still lose to the Official version.

Despite much effort with compiler tweaking, I've never gotten a VS.NET 
2003 build of POV-Ray 3.6 that defeated an official release.

> Concerning your code - you do:
> 
>   divx = 1/D[X]
> 
> without checking if D[X] is zero for example (the paper you cited 
> clearly states that it relies on the IEEE floating point standard for 
> this).

Yes.  The result of a 1.0/0.0 in IEEE floating point is 
PositiveInfinity.  The math related to infinite values is interesting, 
but doesn't break the algorithm in any way.

-Ryan

Post a reply to this message