[FFmpeg-devel] [PATCH 6/6] hwcontext_vaapi: Add support for mapping to DRM objects

wm4 nfxjfg at googlemail.com
Mon Oct 9 14:25:04 EEST 2017


On Sun, 8 Oct 2017 17:13:24 +0100
Derek Buitenhuis <derek.buitenhuis at gmail.com> wrote:

> On 10/8/2017 5:11 PM, Mark Thompson wrote:
> > This is just how hardware surfaces are stored in AVFrames - they have their own API-specific handles in the data[] pointers because that's the only place to put them.
> > 
> > See
> > <http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavutil/pixfmt.h;h=24889c8e5288cd0e6b6dc59bc4455e87747fa7e8;hb=HEAD#l132>
> > and
> > <http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavutil/pixfmt.h;h=24889c8e5288cd0e6b6dc59bc4455e87747fa7e8;hb=HEAD#l337>
> > (and others).
> > 
> > Thanks,  
> 
> Eugh, well OK. That's arguably not exactly OK by the C standard. Oh well.

How come? As long as you're strictly casting through intptr_t it should
be fine.

But yeah, personally I'd prefer having hw surfaces point to some sort
of struct instead. Would make more sense with AVBufferRef semantics as
well. But a bit too late for that.


More information about the ffmpeg-devel mailing list