[vlc-devel] [PATCH 3/3] snapshot: fix copy of opaque pictures

Thomas Guillem thomas at gllm.fr
Mon Jun 5 18:54:17 CEST 2017



On Mon, Jun 5, 2017, at 18:50, Rémi Denis-Courmont wrote:
> Le maanantaina 5. kesäkuuta 2017, 18.19.13 EEST Thomas Guillem a écrit :
> > On Mon, Jun 5, 2017, at 18:10, Rémi Denis-Courmont wrote:
> > > Le maanantaina 5. kesäkuuta 2017, 19.02.22 EEST Rémi Denis-Courmont a
> > > 
> > > écrit :
> > > > Stating the obvious here, but the patch does not do what the commit
> > > > title
> > > > implies.
> > > > 
> > > > AFAICT, there are essentially only two way to handle copying of pictures
> > > > with an opaque context: the very obvious but very slow way and the
> > > > rather
> > > > obvious and very fast way. And that means either adding a copy callback
> > > > (slow), or replacing the destroy callback with a pair hold/release
> > > > callbacks.
> > > 
> > > (The "correct" way is to perform a shallow copy of the context. A pure
> > > reference counter would interfere with context changes in VDPAU filters.)
> > 
> > This means that the same context will be used simultaneously by the
> > video output and by a video filter (via the video converter needed for
> > the image encoder).
> > 
> > This will create issues with some hardwares.
> 
> No. The picture context has always had reference counting internally. It
> just 
> was never exposed to the core.
> 
> In other words, a back-end without reference counting can´t use the
> picture 
> context trick. And it has always been but a trick to make up for the lack
> of 
> converter, filter and splitter pools.

Ok but why no ? How do you prevent the vout and a filter to use the same
context for display/reading at the same time ?

> 
> -- 
> 雷米‧德尼-库尔蒙
> https://www.remlab.net/
> 
> _______________________________________________
> vlc-devel mailing list
> To unsubscribe or modify your subscription options:
> https://mailman.videolan.org/listinfo/vlc-devel


More information about the vlc-devel mailing list