[FFmpeg-devel] [PATCH v7 8/8] libavfilter/vf_deinterlace_qsv: enabling d3d11va support, added mfxhdlpair

[Soft Works]

> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> Artem Galin
> Sent: Thursday, February 25, 2021 12:34 PM
> To: FFmpeg development discussions and patches <ffmpeg-
> devel at ffmpeg.org>
> Subject: Re: [FFmpeg-devel] [PATCH v7 8/8] libavfilter/vf_deinterlace_qsv:
> enabling d3d11va support, added mfxhdlpair
> 
> On Wed, 24 Feb 2021 at 17:50, Soft Works <softworkz at hotmail.com> wrote:
> 
> >
> >
> > > -----Original Message-----
> > > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> > > Artem Galin
> > > Sent: Tuesday, November 3, 2020 7:46 PM
> > > To: ffmpeg-devel at ffmpeg.org
> > > Cc: Artem Galin <artem.galin at intel.com>
> > > Subject: [FFmpeg-devel] [PATCH v7 8/8] libavfilter/vf_deinterlace_qsv:
> > > enabling d3d11va support, added mfxhdlpair
> > >
> > > Adding DX11 relevant device type checks and adjusting callback with
> > proper
> > > MediaSDK pair type support.
> > >
> >
> > Hi Artem,
> >
> > I have a few more notes regarding the patch:
> >
> > The switch to using mfxhdlpair will cause a regression in case of
> > VAAPI/OpenCL Interop. In hwcontext_opencl.cs, function
> > opencl_map_from_qsv, you need to cast the MemId to mfxHDLPair for
> > getting the surface id.
> >
> > OpenCL interop on Windows is a whole different story, but it's not a
> > small thing like the above, so you better handle that at a later time.
> > It's not a regression anyway because this functionality doesn't exist
> > yet (for D3D11).
> >
> > At the same time, that's another reason for NOT changing the default
> > to
> > D3D11
> > because somebody using QSV<=>OpenCL interop would see his commands
> > failing when you change the default to D3D11.
> >
> > I don't know what others here are thinking (because nobody says
> > anything), but IMO, a patch that doesn't switch the default impl would
> > have less impact and probably better chances to get merged.
> >
> > Kind regards,
> > softworkz
> >
> >
> Hi Softworkz,
> 
> D3D11VA works with more variety of HW configurations and there is no DX9
> drop in the patch. Feel free to send your version of the patch with interop
> fix.

Hi Artem,

" there is no DX9 drop in the patch" > sure, D3D9 is not dropped, but you 
are changing the default to D3D11, which brings a lot of differences.

Command lines that are working today won't work anymore - and not just for 
a single reason. There's a bunch of cases as I had already laid out earlier.
The missing OpenCL+D3D11 interop is just an additional bullet to that list.


In addition to that, there are at least two bugs in your v7 patch as I've outlined
above.

Kind regards,
Softworkz