Jason Ekstrand <ja...@jlekstrand.net> writes: > Sorry, said that just a little early. Do we really want 4 bits for a > floating-point format? How many bits does nvidia report?
NVidia reports the RGB components as 8/8/8 for the two normalized formats and 32/32/32 for the two half-float formats. I think the 8 makes some sense because the interpolation phase of the decompression is done on an 8-bit value so it you wanted to accurately store all of the potential values of the decompressed image you would need 8 bits. The 32 bits is a bit more weird because the decompressor can only generate half-float values so at most 16 bits should be needed. I don't know what it says for the alpha components because I just now realised that I made a mistake in the quick test code that I wrote late last night when I had access to an NVidia card. It also says 8/8/8 for the DXT3 and DXT5 formats of S3TC so I guess if we wanted to copy what NVidia does we could also change the bits we report for those formats. Regards, - Neil _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
