[vlc-devel] [PATCH] DxVA2: allow NV12 pixel format all the way to the D3D texture

[Steve Lhomme]

On Wed, Apr 1, 2015 at 2:28 PM, Rémi Denis-Courmont <remi at remlab.net> wrote:
> Le 2015-04-01 14:19, Steve Lhomme a écrit :
>>
>> Do not force YV12 out of DxVA using the costly SplitPlanes as NV12 is the
>> preferred format and will always comes with DxVA and the renderers
>>
>>
>> https://msdn.microsoft.com/en-us/library/windows/desktop/dd206750%28v=vs.85%29.aspx#nv12
>
>
> With the current architecture, the decoder cannot know what is downstream.
> You cannot assume that it is the Direct3D 11 video output. It could be a
> filter (as JB already noted), it could be another video output, it could be
> video splitter or it could even be an encoder (in principles).
>
>> "NV12 is the preferred 4:2:0 pixel format for DirectX VA. It is
>> expected to be an intermediate-term requirement for DirectX VA
>> accelerators supporting 4:2:0 video."
>
>
> I think the preferred texture format for any reasonable video acceleration
> is "opaque". If you want to optimize the data path from the DX VA decoder to
> the DX 3D video output, you should not copy the data back to CPU memory at
> all. The issue of NV12 vs YV12 is thus moot.

That could be done when I add support for ID3D11VideoDevice::CreateVideoDecoder
https://msdn.microsoft.com/en-us/library/windows/desktop/hh447786%28v=vs.85%29.aspx

j-b mentioned that a custom VLC_CODEC_DXVA pixel format could be used
for hardware acceleration, like it's done in Android. But if we want
to use filters that means the DXVA format cannot be opaque.