> -----Original Message-----
> From: Xiang, Haihao <haihao.xi...@intel.com>
> Sent: Tuesday, April 6, 2021 2:24 PM
> To: Chen, Wenbin <wenbin.c...@intel.com>; ffmpeg-devel@ffmpeg.org
> Subject: Re: [FFmpeg-devel] [PATCH] libavcodec/qsvdec: use the param from
> decodeHeader to configure surface
>
> On Tue, 2021-03-23 at 06:45 +0000, Chen, Wenbin wrote:
> > > -----Original Message-----
> > > From: Xiang, Haihao <haihao.xi...@intel.com>
> > > Sent: Tuesday, March 23, 2021 12:10 PM
> > > To: ffmpeg-devel@ffmpeg.org
> > > Cc: Chen, Wenbin <wenbin.c...@intel.com>
> > > Subject: Re: [FFmpeg-devel] [PATCH] libavcodec/qsvdec: use the param
> from
> > > decodeHeader to configure surface
> > >
> > > On Mon, 2021-03-22 at 14:31 +0800, wenbin.c...@intel.com wrote:
> > > > From: "Chen,Wenbin" <wenbin.c...@intel.com>
> > > >
> > > > MSDK recognizes both yuv420p10 and yuv420p9 as MFX_FOURCC_P010,
> > >
> > > but param
> > > > are different. When decode yuv420p9 video, ffmpeg-qsv will use
> > > > yuv420p10le to configure surface which is different with param from
> > > > DecoderHeader and this will lead to error. Now change it use
> > > > param from decoderHeader to configure surface.
> > >
> > >
> > > Both yuv420p10 and yuv420p9 have 3 planes (
> > >
> https://github.com/FFmpeg/FFmpeg/blob/master/libavutil/pixdesc.c#L1359-
> > > L1406) ,
> > > but MFX_FOURCC_P010 has 2 planes only. MSDK doesn't recognizes
> > > yuv420p10 and
> > > yuv420p9 as MFX_FOURCC_P010. FFmpeg-qsv uses p010le instead of
> > > yuv420p10le to
> > > configure surface for 10bit video.
> > >
> >
> > Sorry, I didn't describe the problem well. I mean when MSDK
> decodeHeader a
> > 9bit video
> > It will return MFX_FOURCC_P010, but its frameInfo is 9bit. However
> FFmpeg-qsv
> > will use P010's parameter
> > to configure surface which will be 10bit, and if ffmpeg-qsv send 10bit
> surface
> > to decode 9bit video MSDK will report an error.
> >
>
> So it will use yuv420p10le for a 9bit video, right? A potential issue here is
> the depth info will be lost in the pipeline, a filter after qsv decoder will
> take it as 10bit frame instead of 9bit frame.
>
> Thanks
> Haihao
>
>
Yes, there is a risk of losing depth information. Function ff_qsv_map_fourcc()
should also takes
"FrameInfo.BitDepth" into consideration. This issue is not brought by this
patch, we may fix it in a new patch.
Thanks
wenbin
> > >
> > > >
> > > > Signed-off-by Wenbin Chen <wenbin.c...@intel.com>
> > > > ---
> > > > libavcodec/qsvdec.c | 4 ++--
> > > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/libavcodec/qsvdec.c b/libavcodec/qsvdec.c
> > > > index 569ccd4fba..3ab48ea7a2 100644
> > > > --- a/libavcodec/qsvdec.c
> > > > +++ b/libavcodec/qsvdec.c
> > > > @@ -309,13 +309,13 @@ static int alloc_frame(AVCodecContext *avctx,
> > >
> > > QSVContext
> > > > *q, QSVFrame *frame)
> > > > if (frame->frame->format == AV_PIX_FMT_QSV) {
> > > > frame->surface = *(mfxFrameSurface1*)frame->frame->data[3];
> > > > } else {
> > > > - frame->surface.Info = q->frame_info;
> > > > -
> > > > frame->surface.Data.PitchLow = frame->frame->linesize[0];
> > > > frame->surface.Data.Y = frame->frame->data[0];
> > > > frame->surface.Data.UV = frame->frame->data[1];
> > > > }
> > > >
> > > > + frame->surface.Info = q->frame_info;
> > > > +
> > > > if (q->frames_ctx.mids) {
> > > > ret = ff_qsv_find_surface_idx(&q->frames_ctx, frame);
> > > > if (ret < 0)
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
