[FFmpeg-devel] [PATCH] libavcodec/amfenc: Add more pixel formats support

[Evgeny Pavlov]

On Sun, Oct 22, 2023 at 5:01 PM Mark Thompson <sw at jkqxz.net> wrote:

> On 20/10/2023 09:13, Evgeny Pavlov wrote:
> > On Tue, Jul 18, 2023 at 10:32 AM Evgeny Pavlov <lucenticus at gmail.com>
> wrote:
> >
> >> This commit adds BGRA, RGBA and ARGB pixel formats for AMF encoders
> >>
> >> Signed-off-by: Evgeny Pavlov <lucenticus at gmail.com>
> >> ---
> >>   libavcodec/amfenc.c | 3 +++
> >>   1 file changed, 3 insertions(+)
> >>
> >> diff --git a/libavcodec/amfenc.c b/libavcodec/amfenc.c
> >> index cb48f8c273..234cd012ef 100644
> >> --- a/libavcodec/amfenc.c
> >> +++ b/libavcodec/amfenc.c
> >> @@ -74,6 +74,9 @@ static const FormatMap format_map[] =
> >>       { AV_PIX_FMT_NV12,       AMF_SURFACE_NV12 },
> >>       { AV_PIX_FMT_BGR0,       AMF_SURFACE_BGRA },
> >>       { AV_PIX_FMT_RGB0,       AMF_SURFACE_RGBA },
> >> +    { AV_PIX_FMT_BGRA,       AMF_SURFACE_BGRA },
> >> +    { AV_PIX_FMT_RGBA,       AMF_SURFACE_RGBA },
> >> +    { AV_PIX_FMT_ARGB,       AMF_SURFACE_ARGB },
> >>       { AV_PIX_FMT_GRAY8,      AMF_SURFACE_GRAY8 },
> >>       { AV_PIX_FMT_YUV420P,    AMF_SURFACE_YUV420P },
> >>       { AV_PIX_FMT_YUYV422,    AMF_SURFACE_YUY2 },
> >> --
> >> 2.37.3.windows.1
> >>
> >> The purpose of this patch is to fix an issue with feeding ddagrab output
> > to AMF directly. The output of ddagrab might be BGRA, AMF supports this
> > input, but failed to encode due to missing mapping from AV_PIX_BGRA to
> > AMF_SURFACE_BGRA in amfenc.c
> > DXGI isn't firm to distinguish between RGBA & RGBX, so it is safer to
> > support both to avoid failures like with ddagrab.
>
> See patch just sent to fix the bug in ddagrab that it incorrectly
> advertises an alpha channel.
>
> Do you have ddagrab->amfenc working with just something to fix the alpha
> channel?  To make it work I also need to mess with the bind flags because
> amfenc wants to use the textures as shader resources.
>
> This has been noted as a problem before, where a proper fix would require
> large changes in the format negotiation setup and so it hasn't been done.
> I'm wondering whether adding "-bind_flags shader_resource" option to
> ddagrab would be a plausible hack for these cases, or whether that's a bit
> too obscure for an actual user?
>
> Thanks,
>
> - Mark
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".>
>
>
Yes, I have ddagrab->amfenc working with just fixing alpha channel, but AMF
encoders sometimes use shader, sometimes VCN for color conversion - it
depends on HW capabilities. I would suggest to always set bind in ddgrab to
D3D11_BIND_SHADER_RESOURCE, because this is no reason not to do it.

Thanks,
Evgeny