[FFmpeg-devel] [PATCH v2 3/3] libavcodec/qsvdec: using suggested num to set init_pool_size

[Xiang, Haihao]

On Sat, 2022-04-02 at 08:52 +0000, Xiang, Haihao wrote:
> On Mon, 2022-03-28 at 02:26 +0000, Xiang, Haihao wrote:
> > On Fri, 2022-03-18 at 07:40 +0000, Soft Works wrote:
> > > > -----Original Message-----
> > > > From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of
> > > > Wenbin Chen
> > > > Sent: Friday, March 18, 2022 7:25 AM
> > > > To: ffmpeg-devel at ffmpeg.org
> > > > Subject: [FFmpeg-devel] [PATCH v2 3/3] libavcodec/qsvdec: using
> > > > suggested num to set init_pool_size
> > > > 
> > > > The init_pool_size is set to be 64 and it is too many.
> > > > Use IOSurfQuery to get NumFrameSuggest which is the suggested
> > > > number of frame that needed to be allocated when initializing the
> > > > decoder.
> > > > Considering that the hevc_qsv encoder uses the  most frame buffer,
> > > > async is 4 (default) and max_b_frames is 8 (default) and decoder
> > > > may followed by VPP, use NumFrameSuggest + 16 to set init_pool_size.
> > > > 
> > > > Signed-off-by: Wenbin Chen <wenbin.chen at intel.com>
> > > > Signed-off-by: Guangxin Xu <guangxin.xu at intel.com>
> > > > ---
> > > >  libavcodec/qsvdec.c | 14 ++++++++++++--
> > > >  1 file changed, 12 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/libavcodec/qsvdec.c b/libavcodec/qsvdec.c
> > > > index 210bd0c1d5..9875d3d632 100644
> > > > --- a/libavcodec/qsvdec.c
> > > > +++ b/libavcodec/qsvdec.c
> > > > @@ -88,7 +88,7 @@ typedef struct QSVContext {
> > > >      uint32_t fourcc;
> > > >      mfxFrameInfo frame_info;
> > > >      AVBufferPool *pool;
> > > > -
> > > > +    int suggest_pool_size;
> > > >      int initialized;
> > > > 
> > > >      // options set by the caller
> > > > @@ -275,7 +275,7 @@ static int qsv_decode_preinit(AVCodecContext
> > > > *avctx, QSVContext *q, enum AVPixel
> > > >          hwframes_ctx->height            = FFALIGN(avctx-
> > > > > coded_height, 32);
> > > > 
> > > >          hwframes_ctx->format            = AV_PIX_FMT_QSV;
> > > >          hwframes_ctx->sw_format         = avctx->sw_pix_fmt;
> > > > -        hwframes_ctx->initial_pool_size = 64 + avctx-
> > > > > extra_hw_frames;
> > > > 
> > > > +        hwframes_ctx->initial_pool_size = q->suggest_pool_size + 16 +
> > > > avctx->extra_hw_frames;
> > > >          frames_hwctx->frame_type        =
> > > > MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET;
> > > > 
> > > >          ret = av_hwframe_ctx_init(avctx->hw_frames_ctx);
> > > > @@ -793,6 +793,9 @@ static int qsv_process_data(AVCodecContext *avctx,
> > > > QSVContext *q,
> > > >      }
> > > > 
> > > >      if (q->reinit_flag || !q->session || !q->initialized) {
> > > > +        mfxFrameAllocRequest request;
> > > > +        memset(&request, 0, sizeof(request));
> > > > +
> > > >          q->reinit_flag = 0;
> > > >          ret = qsv_decode_header(avctx, q, pkt, pix_fmt, &param);
> > > >          if (ret < 0) {
> > > > @@ -802,12 +805,19 @@ static int qsv_process_data(AVCodecContext
> > > > *avctx, QSVContext *q,
> > > >                  av_log(avctx, AV_LOG_ERROR, "Error decoding
> > > > header\n");
> > > >              goto reinit_fail;
> > > >          }
> > > > +        param.IOPattern = q->iopattern;
> > > > 
> > > >          q->orig_pix_fmt = avctx->pix_fmt = pix_fmt =
> > > > ff_qsv_map_fourcc(param.mfx.FrameInfo.FourCC);
> > > > 
> > > >          avctx->coded_width  = param.mfx.FrameInfo.Width;
> > > >          avctx->coded_height = param.mfx.FrameInfo.Height;
> > > > 
> > > > +        ret = MFXVideoDECODE_QueryIOSurf(q->session, &param,
> > > > &request);
> > > > +        if (ret < 0)
> > > > +            return ff_qsv_print_error(avctx, ret, "Error querying IO
> > > > surface");
> > > > +
> > > > +        q->suggest_pool_size = request.NumFrameSuggested;
> > > > +
> > > >          ret = qsv_decode_preinit(avctx, q, pix_fmt, &param);
> > > >          if (ret < 0)
> > > >              goto reinit_fail;
> > > > --
> > > 
> > > Thanks for the patch! I have that on my list for quite a while.
> > > Will look at it shortly.
> > 
> > Hi Softworz,
> > 
> > This patchset LGTM and works well, do you have any comment ? 
> 
> Ping, I'll apply this patchset in a few days if no more comment.
> 

Applied, thx

-Haihao