Andrew

Right, erm, from what I can remember (As I'm on the PC now), there are 5 blur/fade types i've stuck in...

1. A simple fade, where 1 is taken away from each pixels component (R,G,B) until it reaches black

2. A fractional fade, where each component is multipied by the fade amount, shared by 256, then cut down to whatever constraints are needed

3. A 9-pixel mean blur (What you're interested in).
The components of the 3x3 source grid are totalled up, then shared by 9. For 16bit I use some quick division code (I think), while for 24 bit I use a lookup table (since fast and accurate code is a bit tricky for 8 bit). The components are then squashed back together, and stored in the output bank. The exact way the code works for each colour depth is complicated, so it's best to look at the source. Each pixel has identical 'weighting' by the way, although making things like the central pixel more prominent could be easily done (Note to self: Add for next release). Except that I think my current code won't work like that

4. A 'plasma' effect. This is essentially the 9 pixel mean, but uses some 'experimental' /9 code, which actually shares by about 7.5. This means that each component (providing it isn't black) increases in colour, until it overflows into the component next to it. Also because it's overflowed it resets to black. This gives you some interesting colours from a white dot, instead of a load of grey.

5. A plain block to block copy. What fun.