[XviD-devel] RE: possible vfw YV12 stride bug

[Dirk Knop]

Good morning everyone,

Radek Czyz schrieb:

> Hi,
>
> gruel wrote:
>
>> XVID_BS_VERSION is 27 now. I have no idea how renaming could screw
>> anything up, but then... I know me...
>
>
> Well... they did. This is because the old definitions were equal to 
> the colorspace definitions used in VfW and directshow.
>
> Do I understand correctly that I have to map "YV12" fourCC to 
> XVID_CSP_I420 colourspace and "I420" to XVID_CSP_YV12 now?
>
> Currently both VfW and dshow have them the "normal" way and this 
> creates wrong encodes, with planes swapped. Same on decoding.
>
> What now?
> Radek

I don't know if my mind tricks me, but I think I read some references to 
microsoft documentation in this discussion. M$ documentation is flawed, 
as is directX with some colourspaces (namely M$ yuv dll is doing things 
wrong).  Unfortunately I don't find the thread over at Doom9's forums 
anymore where we discovered the problem and developed the fix for it 
(back in 2002 that was).

Now the xvid bitstream gets stored with swapped UV planes - verified 
even with Cruncher's standalone mpeg4 capable player.

We should fix that ASAP.

Suxendrol suggested to use XVID_CSP_INTERNAL (.._USER) and swap the 
planes like we need them.

Regards
Koepi