[FFmpeg-devel] [PATCH][7/8] Add VA API accelerated H.264 decoding (take 4)

[Gwenole Beauchesne]

On Mon, 9 Feb 2009, Michael Niedermayer wrote:

> On Mon, Feb 09, 2009 at 11:39:43AM +0100, Gwenole Beauchesne wrote:
>> Hi,
>>
>> On Fri, 6 Feb 2009, Michael Niedermayer wrote:
>>
>>> C. as stephen suggested, outputting decoded surfaces
>>
>> I had a look at it when Jean-Michel Pour? asked for it. However, it turns
>> out this could be rather suboptimal. In particular, there are cases where
>> NV12 readback might be supported but not YV12. Sure, it doesn't cost that
>> much to convert both, but it still has some cost.
>
> i dont understand the problem you describe, If the hardware can only make
> NV12 available and the user app cannot use that convert is needed anyway
> and if a common format is supported that could be used ...
> what am i missing?

You said "outputting decoded surfaces", this implied a default behaviour 
unlike what I initially suggested (an optional flag). Performing the 
conversion in any case when we usually only want to see the result 
on-screen is not a good use of the underlying resources.

So, an approach based on (C) only is not the best one, IMHO. i.e. (B) + an 
option to retrieve the decoded surfaces is a better way. Though, that 
option is not necessary for an initial commit.