[FFmpeg-devel] [PATCH v2 1/4] lavc/vaapi_encode: add EQUAL_MULTI_ROWS support for slice structure

[Fu, Linjie]

> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> Mark Thompson
> Sent: Tuesday, July 28, 2020 05:52
> To: FFmpeg development discussions and patches <ffmpeg-
> devel at ffmpeg.org>
> Subject: Re: [FFmpeg-devel] [PATCH v2 1/4] lavc/vaapi_encode: add
> EQUAL_MULTI_ROWS support for slice structure
> 
> On 19/07/2020 08:00, Linjie Fu wrote:
> > On Thu, Jun 18, 2020 at 1:36 PM Linjie Fu <linjie.justin.fu at gmail.com>
> wrote:
> >>
> >> On Tue, May 12, 2020 at 9:49 PM Linjie Fu <linjie.fu at intel.com> wrote:
> >>>
> >>> VA_ENC_SLICE_STRUCTURE_EQUAL_MULTI_ROWS is added to in the
> latest
> >>> libva (1.8.0) which matches the hardware behaviour:
> >>>
> >>> /** \brief Driver supports any number of rows per slice but they must
> be the same
> >>> *       for all slices except for the last one, which must be equal or smaller
> >>> *       to the previous slices. */
> >>>
> >>> And VA_ENC_SLICE_STRUCTURE_EQUAL_ROWS is kind of deprecated
> for iHD
> >>> since it's somehow introduced in [1] however misleading from what we
> >>> actually handle, and one row per slice would not get a good quality.
> >>>
> >>> Caps query support in driver is WIP, and this would fix the multi slice
> >>> encoding for VAEntrypointEncSliceLP.
> >>>
> >>>
> [1]<https://github.com/intel/libva/commit/0e6d5441f19bdc674b4da3169d61
> 4d10fd644778>
> >>>
> >>> Signed-off-by: Linjie Fu <linjie.fu at intel.com>
> >>> ---
> >>>   libavcodec/vaapi_encode.c | 3 +++
> >>>   1 file changed, 3 insertions(+)
> >>>
> >>> diff --git a/libavcodec/vaapi_encode.c b/libavcodec/vaapi_encode.c
> >>> index cb05ebd..234618a 100644
> >>> --- a/libavcodec/vaapi_encode.c
> >>> +++ b/libavcodec/vaapi_encode.c
> >>> @@ -1888,6 +1888,9 @@ static av_cold int
> vaapi_encode_init_slice_structure(AVCodecContext *avctx)
> >>>           req_slices = avctx->slices;
> >>>       }
> >>>       if (slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS
> ||
> >>> +#if VA_CHECK_VERSION(1, 8, 0)
> >>> +        slice_structure &
> VA_ENC_SLICE_STRUCTURE_EQUAL_MULTI_ROWS ||
> >>> +#endif
> >>>           slice_structure &
> VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS) {
> >>>           ctx->nb_slices  = req_slices;
> >>>           ctx->slice_size = ctx->slice_block_rows / ctx->nb_slices;
> >>> --
> >>> 2.7.4
> >>
> >> Full support is provided in libva[1] and media-driver[2], and we've
> >> observed it works for multi slice encoding for AVC. (300+ cases)
> >>
> >> Prefer to apply this soon with commit message refined.
> >>
> > Applied, thx.
> Treating EQUAL_MULTI_ROWS in the same way as the arbitrary-size cases is
> just wrong.
> 
> Consider 9 rows, 4 slices - we pick 4 slices with sizes { 3, 2, 2, 2 }, which
> EQUAL_MULTI_ROWS does not allow.
> 
> It isn't possible to split the frame into 4 slices at all with the
> EQUAL_MULTI_ROWS structure - the closest options are 3 slices with sizes { 3,
> 3, 3 } or 5 slices with sizes { 2, 2, 2, 2, 1 }.
> 
> You can see the same problem at 1080p with five slices.

Umm.. saw the conflicts with the definition/notation in LIBVA.

Actually, I doubt whether Hardware still have such limitation which needs the last
slice to be no larger than the previous. In the case of 9 rows, 4 slices, the current
implementation of rounding works well. {3, 2, 2, 2} .

Had sent a mail to the media-driver/libva scrum team to double confirm this, thx.

- Linjie