[FFmpeg-devel] [PATCH v2] lavc/qsvenc: add tile encoding support for VP9

[mypopy at gmail.com]

On Mon, Jan 17, 2022 at 6:57 PM Soft Works <softworkz at hotmail.com> wrote:
>
>
>
> > -----Original Message-----
> > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> > mypopy at gmail.com
> > Sent: Monday, January 17, 2022 11:36 AM
> > To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> > Subject: Re: [FFmpeg-devel] [PATCH v2] lavc/qsvenc: add tile encoding support
> > for VP9
> >
> > On Mon, Jan 17, 2022 at 4:30 PM Xiang, Haihao
> > <haihao.xiang-at-intel.com at ffmpeg.org> wrote:
> > >
> > > On Thu, 2022-01-13 at 13:45 +0800, Haihao Xiang wrote:
> > > > Add -tile_rows and -tile_cols options to specify the number of tile
> > > > rows and columns
> > > >
> > > > Signed-off-by: Haihao Xiang <haihao.xiang at intel.com>
> > > > ---
> > > > v2: add option descriptions in the doc
> > > >
> > > >  doc/encoders.texi       |  6 ++++++
> > > >  libavcodec/qsvenc.c     |  4 ++++
> > > >  libavcodec/qsvenc.h     |  1 +
> > > >  libavcodec/qsvenc_vp9.c | 10 ++++++++++
> > > >  4 files changed, 21 insertions(+)
> > > >
> > > > diff --git a/doc/encoders.texi b/doc/encoders.texi
> > > > index 7cc8be1209..a4176089d5 100644
> > > > --- a/doc/encoders.texi
> > > > +++ b/doc/encoders.texi
> > > > @@ -3457,6 +3457,12 @@ These options are used by vp9_qsv
> > > >  @item profile2
> > > >  @item profile3
> > > >  @end table
> > > > +
> > > > + at item @var{tile_cols}
> > > > +Number of columns for tiled encoding (requires libmfx >= 1.29).
> > > > +
> > > > + at item @var{tile_rows}
> > > > +Number of rows for tiled encoding (requires libmfx  >= 1.29).
> > > >  @end table
> > > >
> > > >  @section snow
> > > > diff --git a/libavcodec/qsvenc.c b/libavcodec/qsvenc.c
> > > > index 4e7a15f060..4cbc9ff4dc 100644
> > > > --- a/libavcodec/qsvenc.c
> > > > +++ b/libavcodec/qsvenc.c
> > > > @@ -939,6 +939,10 @@ static int init_video_param(AVCodecContext *avctx,
> > > > QSVEncContext *q)
> > > >          q->extvp9param.Header.BufferId = MFX_EXTBUFF_VP9_PARAM;
> > > >          q->extvp9param.Header.BufferSz = sizeof(q->extvp9param);
> > > >          q->extvp9param.WriteIVFHeaders = MFX_CODINGOPTION_OFF;
> > > > +#if QSV_HAVE_EXT_VP9_TILES
> > > > +        q->extvp9param.NumTileColumns  = q->tile_cols;
> > > > +        q->extvp9param.NumTileRows     = q->tile_rows;
> > > > +#endif
> > > >          q->extparam_internal[q->nb_extparam_internal++] = (mfxExtBuffer
> > *)&q-
> > > > >extvp9param;
> > > >      }
> > > >  #endif
> > > > diff --git a/libavcodec/qsvenc.h b/libavcodec/qsvenc.h
> > > > index 31516b8e55..00ee52a5d1 100644
> > > > --- a/libavcodec/qsvenc.h
> > > > +++ b/libavcodec/qsvenc.h
> > > > @@ -41,6 +41,7 @@
> > > >
> > > >  #define QSV_HAVE_EXT_HEVC_TILES QSV_VERSION_ATLEAST(1, 13)
> > > >  #define QSV_HAVE_EXT_VP9_PARAM QSV_VERSION_ATLEAST(1, 26)
> > > > +#define QSV_HAVE_EXT_VP9_TILES QSV_VERSION_ATLEAST(1, 29)
> > > >
> > > >  #define QSV_HAVE_TRELLIS QSV_VERSION_ATLEAST(1, 8)
> > > >  #define QSV_HAVE_MAX_SLICE_SIZE QSV_VERSION_ATLEAST(1, 9)
> > > > diff --git a/libavcodec/qsvenc_vp9.c b/libavcodec/qsvenc_vp9.c
> > > > index 9329990d11..1168ddda0e 100644
> > > > --- a/libavcodec/qsvenc_vp9.c
> > > > +++ b/libavcodec/qsvenc_vp9.c
> > > > @@ -73,6 +73,16 @@ static const AVOption options[] = {
> > > >      { "profile2",  NULL, 0,                   AV_OPT_TYPE_CONST, { .i64
> > =
> > > > MFX_PROFILE_VP9_2   },  INT_MIN,  INT_MAX,  VE,  "profile" },
> > > >      { "profile3",  NULL, 0,                   AV_OPT_TYPE_CONST, { .i64
> > =
> > > > MFX_PROFILE_VP9_3   },  INT_MIN,  INT_MAX,  VE,  "profile" },
> > > >
> > > > +#if QSV_HAVE_EXT_VP9_TILES
> > > > +    /* The minimum tile width in luma pixels is 256, set maximum
> > tile_cols to
> > > > 32 for 8K video */
> > > > +    { "tile_cols",  "Number of columns for tiled
> > > > encoding",   OFFSET(qsv.tile_cols),    AV_OPT_TYPE_INT, { .i64 = 0 }, 0,
> > 32,
> > > > VE },
> > > > +    /* Set maximum tile_rows to 4 per VP9 spec */
> > > > +    { "tile_rows",  "Number of rows for tiled
> > > > encoding",      OFFSET(qsv.tile_rows),    AV_OPT_TYPE_INT, { .i64 = 0 },
> > 0, 4,
> > > > VE },
> > > > +#else
> > > > +    { "tile_cols",  "(not
> > > > supported)",                        OFFSET(qsv.tile_cols),
> > AV_OPT_TYPE_INT,
> > > > { .i64 = 0 }, 0, 0, VE },
> > > > +    { "tile_rows",  "(not
> > > > supported)",                        OFFSET(qsv.tile_rows),
> > AV_OPT_TYPE_INT,
> > > > { .i64 = 0 }, 0, 0, VE },
> > > > +#endif
> > > > +
> > perfer one option like "-tile rows x cols" than two options like
>
> The example is invalid. (spaces)
>
> > "-tile_rows  row  -tile_cols  col"
>
> This way, the options are typed, have min, max and default values,
> without needing any code to write and maintain.
>
> if you mean "-tile rowsxcols", there's no option type for this,
> unless you would want to mis-use the video size type for it.
> Otherwise you'd need to make it string and write all the parsing
> and checking code for it.
> Do you think that it would be worth the effort?
You convinced me
>
> Also, in most use cases, you will probably have rows or cols,
> rather than both..
>