And the follow up question is, do Flash Player images have
32-bits per channel (internally), or 32-bits per pixel. At the
ActionScript API level, you only have access to 8-bits per channel,
so is the internal representation larger? If I run two kernels on a
Flash object in series, will I lose precision over combining the
filter operation into a single kernel?
According to Tinic in Flash, Pixel Bender processing is done
in 32 bpc. Bitmaps in Flash are stored as 8-bit ARGB with Alpha
pre-multiplied. Conversion to/from 8-bit is done on the way in and
out. So with FP 10, there would be a conversion back to 8bpc in
between two shaders and it might be better to combine them
depending on what you are doing.