Re[6]: [XviD-devel] image pre-processing

Milan Cutka xvid-devel@xvid.org
Sat, 7 Dec 2002 10:25:44 +0100


>> Many image processing routines in ffdshow are written using inline
>> assembly in C++ source files with MSVC syntax. To have them portable
>> it would be required to separate them to asm files and use NASM to
>> compile. BTW I'm not very good in MMX optimizing and anyone with
>> greater experiences could optimize my filters much more.

MF> i'm not better in MMX, but i can always help ^_^

MF> BTW, since we'll need to use a fixed API, ffdshow's filtering structure
MF> is good, but we should add pitch != width support too.

ffdshow image filters of course support this. What should be probably
supported is the case when input pitch != output pitch - as pointed by
sh0dan when he was converting warpsharp to AviSynth filter.

BTW ffdshow image filters are able to change every setting during
playback: not just filters properties, but even filters order. This
requires small overhead which wouldn't be required if using them
during encoding.