[FFmpeg-devel] [PATCH] avcodec: only warn about hwaccel with frame threads

wm4 nfxjfg at googlemail.com
Sat Jan 23 12:29:30 CET 2016


On Sat, 23 Jan 2016 12:10:26 +0100
Andreas Cadhalpun <andreas.cadhalpun at googlemail.com> wrote:

> On 23.01.2016 11:42, Hendrik Leppkes wrote:
> > On Sat, Jan 23, 2016 at 10:12 AM, Andreas Cadhalpun
> > <andreas.cadhalpun at googlemail.com> wrote:  
> >> VLC uses hwaccel with frame threads and it works fine, but returning
> >> an error here made it fail.
> >>
> >> This regression was introduced in commit 31741ae.
> >>  
> > 
> > I'm still opposed to this, and so is everyone else that commented on the issue.  
> 
> That's not true.
> 
> > Also, the vp9 hwaccel will most definitely misbehave with MT, so this
> > one needs to be re-factored before MT is allowed at the very least.  
> 
> Feel free to propose an alternative patch fixing the regression your
> commit introduced. But unless you do so soon, I intend to apply this
> patch to make VLC with hwaccel usable again until you find a better
> solution.

So far you've expressed a strong opinion that it works with vdpau or
vaapi. I don't remember which. (Also keep in mind that API wrapper
drivers for each other can change behavior. And that you didn't
necessarily confirm this for sure - I mean, if we e.g. removed all
malloc checks from the codebase, it would apparently still work.)

This doesn't change that VP9 will be broken, and that VLC will have to
fix this fundamental issue anyway. Also, if I understand this right,
DXVA is normally broken, unless you apply heavy workarounds like VLC
did.

If you really care (and care much more about this than the VLC dev,
which is strange), then maybe you could investigate the wrapper decoder
idea or other ideas?

Also, this is not a regression. It's a change in behavior. Maybe we
forgot to bump the version number, but that's trivial to fix.


More information about the ffmpeg-devel mailing list