Ahh yes, I had forgotten about the 0 data length = RGBA encoding thing, although I'm still trying to get DXT3 and especially DXT5 which is a creation of the purest evil to function properly. Hopefully I will be able to get it working tonight.
As for your experiment, the former 8 bytes of a 16-byte block control the alpha, which should explain your results. Everything else (except the 1-bit alpha mode) is the same as DXT1.
Last edited by eiz; 01-08-2004 at 11:42 PM.