This applies to both IppiFilter & IppiFilter, & probably other non-in-place ones.
I was getting troubles getting a blur to work (the filter box was working, though), because I kept gettinga yellowish border around my blur. Sometimes it was just black, meaning: it was some non-initialized parameter somewhere. Then I realized that the yellow was coming from my *destination* bitmap! Problem is that the help file says that the function blurs the input, then *sets* the pixel to the destination buffer. It doesn't set, it blends. It will blend the blurred source with the destination one, meaning that if you blur a rectangle from a white source buffer to a destination buffer filled with yellow, you're gonna get yellow around your blurred result.
So you have to first copy your source bitmap to the destination one, I think the help file should mention that.
Btw, these 2 functions seem to work as in-place too. I wish Intel was making clear which non-inplace functions are in-place safe, if any. It hurts to have to allocate huge buffers for nothing.
ok please ignore this, it was my fault (stacking blur passes, I realize it was then taking the uninitialized border around the target frame for the second pass, obviously)
(but I would still like to know if we have to assume that all non-inplace functions cannot be used with the same buffers, even ifsome can in practice)