On Sun, Jul 6, 2014, at 01:05 PM, Krzysztof Kosiński wrote:
The problem is that the dependent area of the filter effect in this case is extremely large, i.e. the Gaussian blur must be computed on this enormous surface in order for the result to be correct.
We could do three things a) put in some logic to arbitrarily stop rendering an effect when the dependent area crosses some threshold b) use stacked box blur instead of true Gaussian blur (though I doubt it would be faster) c) speed up the existing Gaussian blur implementation - it shouldn't be that slow
What about tiling the filtering? At some point to address 'c' we should be looking more into breaking up to sections. However I'm not following that code at the momentenough to know if we're getting close to that point yet or not.
In general focus on tuning that one operation is probably the first step. Once we get up to leveraging multi-core processing better then we can re-address the tiling point.