[Mesa-dev] [PATCH] vl/mpeg12: implement inverse scan/quantization steps

[Ilia Mirkin]

On Sun, Jun 23, 2013 at 5:47 PM, Andy Furniss <adf.lists at gmail.com> wrote:
> Ilia Mirkin wrote:
>>
>> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
>> ---
>>
>> These changes make MPEG2 I-frames generate the correct macroblock data (as
>> compared to mplayer via xvmc). Other MPEG2 frames are still misparsed, and
>> MPEG1 I-frames have some errors (but largely match up).
>
>
> This messes up mpeg2 vdpau decode for my radeon - VDPAU_DRIVER=softpipe is
> also affected (and then crashes after a few frames - but that's normal).
>
> xvmc is unaffected.
>
> Normally mpeg2 xvmc and vdpau decode look the same - almost right but not
> quite. The same issues are visible with xvmc softpipe, so I don't think it's
> hw.

Thanks for testing! Perhaps softpipe and radeon implement these stages
inside of decode_macroblock? I'll take a look. My experience (while
developing hardware-accel support for some nvidia cards) is that xvmc
works flawlessly, and I'm just trying to adjust the vl/mpeg12 logic
until the data it passes to decode_macroblock matches what mplayer
passes through xvmc. However while this patch makes i-frames work, the
rest don't. So if the decoding looks at all not-totally-broken for you
with vdpau (and no hardware bitstream support), that might mean that
there's some optional work passed to the decode_macroblock logic that
doesn't happen with xvmc or something. I'll try to work out what's
going on.

  -ilia