[FFmpeg-devel] [PATCH] vp9: add hwaccel hooks

[Hendrik Leppkes]

On Fri, Dec 4, 2015 at 2:52 PM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
> Hi,
>
> On Thu, Dec 3, 2015 at 3:26 PM, Hendrik Leppkes <h.leppkes at gmail.com> wrote:
>
>> On Thu, Dec 3, 2015 at 6:17 PM, Ronald S. Bultje <rsbultje at gmail.com>
>> wrote:
>> > Hi,
>> >
>> > On Thu, Dec 3, 2015 at 11:29 AM, Hendrik Leppkes <h.leppkes at gmail.com>
>> > wrote:
>> >
>> >> On Thu, Dec 3, 2015 at 5:21 PM, Ronald S. Bultje <rsbultje at gmail.com>
>> >> wrote:
>> >> > Hi,
>> >> >
>> >> > On Thu, Dec 3, 2015 at 4:55 AM, Hendrik Leppkes <h.leppkes at gmail.com>
>> >> wrote:
>> >> >
>> >> >> +    if (s->pix_fmt == AV_PIX_FMT_YUV420P) {
>> >> >>
>> >> > +#if CONFIG_VP9_DXVA2_HWACCEL
>> >> >> +        *fmtp++ = AV_PIX_FMT_DXVA2_VLD;
>> >> >> +#endif
>> >> >> +#if CONFIG_VP9_D3D11VA_HWACCEL
>> >> >> +        *fmtp++ = AV_PIX_FMT_D3D11VA_VLD;
>> >> >> +#endif
>> >> >> +    }
>> >> >
>> >> >
>> >> > So ... this suggests that hwaccels will ever only support 420, 8bpp.
>> That
>> >> > may currently be true, but is that likely to hold forever? What
>> happens
>> >> if
>> >> > we just do this regardless of pixfmt and let the hwaccel figure out if
>> >> the
>> >> > hw supports 420 or something else?
>> >> >
>> >>
>> >> At the very least the DXVA2 hwaccel spec is limited to profile 0, so
>> >> maybe it should even check the profile on top of this.
>> >> In general, there is some disagreement if other chroma formats should
>> >> use other hwaccel pixfmts, as the calling code can otherwise get
>> >> confused, as it may need special handling to handle other chroma
>> >> formats. However, since no hwaccel supports anything else yet (except
>> >> hevc, which can do 420 10-bit, but I didn't push that to ffmpeg yet),
>> >> it was not a real problem yet.
>> >
>> >
>> > I'm ok with deferring it until such hardware comes along I guess, so
>> > consider the comment ignorable...
>> >
>>
>> Any comments on the vp9 changes themself? :)
>>
>> Most of the changes are not very invasive, just some extra allocations
>> and error handling, the default hwaccel hooks, and whatnot.
>> I moved the reference frame check down as we discussed on IRC the
>> other day, so that it can happen after the decoder has been switched
>> into or out of HWAccel mode.
>
>
> Yeah, it looked pretty straightforward. Sorry, I now realized I never said
> it was OK. So: LGTM.
>

Set applied with changelog and version bump.