[Ffmpeg-devel] Re: Using Intel's fDCT
Sun Nov 20 22:41:10 CET 2005
Aurelien Jacobs <aurel <at> gnuage.org> writes:
> Here, the Intel result is exactly the same as the ffmpeg one,
> but quantized by a factor of 8.
> So presumably, Intel's fDCT include the quantization step.
> (Note that this is just rough guess)
Thanks for your suggestion but no, the Intel function is pure fDCT with no
quantization. There are other functions for that. However, maybe the ffmpeg
version does the opposite: scales its results by 8 (and adds a bias?).
There seems to be something strange with permutations of the quantization
stage as well as I see permute being called all over the place.
More information about the ffmpeg-devel