Chambers wrote:
> The problem is that some "thread blocks" have a noticeably different 
> brightness than the neighboring blocks. 

Actually, this observation while appearing plausible, is not the reason. 
That the problem appears at the thread blocks is due to the way POV-Ray 
decides when to do supersampling (it does that for each block). It so 
happens that there is a bug in computing the average of supersampled pixels 
that carries on as soon as the first supersample is taken. This results in a 
rather interesting chain reaction causing all remaining pixels in a block to 
appear different. And the checker pattern happens to be well at exposing 
that bug. That said, I am still searching for the actual cause in the source 
code, but no doubt will find it eventually ;-)

	Thorsten

Post a reply to this message