I received this kind email from Gwenole about hardware decoding using
AVHWAccell infrastructure and access to surface.
This seems a very cool approach for FFmpeg / MLT / Kdenlive.
MLT handles YV12. Does it handle NV12?
-------- Forwarded Message --------
From: Gwenole B
To: Jean-Michel Pouré
Subject: Re: [FFmpeg-devel] [RFC] HWAccel sample implementation: VA API
Date: Tue, 24 Feb 2009 18:14:31 +0100 (CET)
On Tue, 24 Feb 2009, Jean-Michel Pouré wrote:
> On Tue, 2009-02-24 at 14:42 +0100, Gwenole Beauchesne wrote:
>> Here is the combined patch for VA API support through the new
>> infrastructure. A first pass is appreciated prior to going into the
>> trouble of splitting them all...
> Do you any possibility to one day have access to the decoded frames
> getting them back through PCI-e. This is for Kdenlive.
However, if someone manages to push patches ASAP, this saves me time, time
that could probably be spent into doing that.
Besides, note that VA API (Intel implementations) only manages NV12 (and
some other weird formats). VDPAU seems to be able to do YV12. Does NV12
even suit your needs? Otherwise, I fear that the bottleneck would become
the pixfmt conversion. However, if VDPAU handles YV12, so would the