[FFmpeg-devel] [PATCH] Allow using primary CUDA device context

[Oleg Dobkin]

On Wed, Nov 20, 2019 at 12:48 PM Timo Rothenpieler
<timo at rothenpieler.org> wrote:
>
> On 20/11/2019 11:29, Oleg Dobkin wrote:
> > On Tue, 2019-11-19 at 23:58 +0100, Timo Rothenpieler wrote:
> >> Only thing I'm still wondering is if it would make sense to use the
> >> primary context by default.
> >> I can't think of any obvious downsides, other than weakened
> >> isolation,
> >> which really shouldn't be a huge concern.
> >> And there apparently are some potential performance gains.
> >
> > I think it's safer to leave the default behavior, in case of some
> > unknown corner cases - there could be some legacy code that assumes the
> > creation of different contexts per each decoder.
> >
>
> Good point. I haven't found the time to give this a test run so far.
> Once I did, and nothing exploded, I'll do a round of releases of
> ffnvcodec and apply the patch.

Is there any progress on this?