[FFmpeg-devel] [RFC] huffyuv usage

[Christophe Gisquet]

Hi,

I have been working on improving huffyuv (a lossless video codec)
decoding speed. I haven't checked 32bits system but on my system
(Win64 core i5) the speedup was 15-20%.

Now, I'm thinking of 2 further changes that require potentially format
or behaviour change, so I'm wondering first how much huffyuv is used.

1) 4:2:0 content

It's stored as NV12 yet output is 4:2:0 planar. It's faster to output
NV12 and there may be benefits to users of the output (x264?). I guess
it could be forced through a specific marker or requested output.

Is it worth it at all? I don't expect more than 5% speedup from it.

2) 16bits content

This one is frankly dubious. Some speedup in parsing can be achieved
by changing the bitstream layout (for the interested, modified
interleaving of LSBs fixed-length-code and VLC MSBs). I don't think
it's worth it, but if there are people using 16bit content and
interested...

-- 
Christophe