OK, so you're basically caching recent lookups of various functions so 
you don't have to compute them later.

I still doubt that this would really benefit POV-Ray, because I doubt 
that the various functions such as sqrt or cos are called very often 
with the same exact input.  And if it isn't called with the same input, 
you'd still have to evaluate the function.

Something like a modeller which does object tesselation would probably 
benefit a great deal from this, especially if it tends to evaluate on 
fixed intervals (for algorithms such as marching cubes); however, 
POV-Ray samples many, many more points, all over the spacial area, so 
you'd likely need a much larger cache, resulting in more overhead and 
less performance.

Of course, the only real way to know for sure would be to implement it 
and test; why don't you branch the 3.6 codebase, try it out, and let us 
know if it works?

...Chambers

Post a reply to this message