convolution vs. transform
Jaymz Julian
jaymz@artificial-stupidity.net
Mon, 26 Jan 2004 02:20:27 +1100
a couple of questions to which i currently lack the answers relating to
convolutions.
a) Which would be considered the correct order to do things in,
convole -> transform -> composite, or transform -> convole -> composite?
(I suspect that the latter is the more attractive option, although under the
current implentation of composite+transform, the former is easier to
implement. I did have the idea that maybe we should support both as picture
properties (neither is implemented right now, only matrix convolutions as a
standalone op are), but that just seems like overkill. One point to note,
i guess, is that many, many video cards can accelerate a composite+transform
in one op if the convolution comes first, but it's require 2 graphics card ops
if the convolution has to happen in the middle).
b) which is considered a better interface - one call per differnet type of
convolution (XRenderMatrixConvolution, XRenderGaussianBlur, etc), or just a
single call with an op parameter? (I'm leaning towards the second of these
right now, but that's mostly because that's path of least resistance -
juggling Render, XRender, and xserver is annoying :-p).
-- jj
--
Jaymz Julian aka A Life in Hell / Warriors of the Wasteland / Unreal
Coder, Visionary, Fat Ass.
overdrive, overscan, over awed, over it.
We're heading for a mad-man situation.