[FFmpeg-devel] [PATCH 2/5] lavc/qsvdec: Replace current parser with MFXVideoDECODE_DecodeHeader()

Rogozhkin, Dmitry V dmitry.v.rogozhkin at intel.com
Thu Jan 24 23:43:13 EET 2019


On Mon, 2019-01-21 at 20:41 +0800, Zhong Li wrote:
> Using MSDK parser can improve qsv decoder pass rate in some cases
> (E.g:
> sps declares a wrong level_idc, smaller than it should be).
> And it is necessary for adding new qsv decoders such as MJPEG and VP9
> since current parser can't provide enough information.
> Actually using MFXVideoDECODE_DecodeHeader() was disscussed at
> https://ffmpeg.org/pipermail/ffmpeg-devel/2015-July/175734.html and
> merged as commit 1acb19d,
> but was overwritten when merged libav patches (commit: 1f26a23)
> without any explain.
> 
> Signed-off-by: Zhong Li <zhong.li at intel.com>
> ---
>  libavcodec/qsvdec.c | 103 ++++++++++++++++------------------------
> ------------
>  libavcodec/qsvdec.h |   2 +
>  2 files changed, 33 insertions(+), 72 deletions(-)
> 
> diff --git a/libavcodec/qsvdec.c b/libavcodec/qsvdec.c
> index 4a0be81..013400b 100644
> --- a/libavcodec/qsvdec.c
> +++ b/libavcodec/qsvdec.c
> @@ -120,7 +120,7 @@ static inline unsigned int qsv_fifo_size(const
> AVFifoBuffer* fifo)
>      return av_fifo_size(fifo) / qsv_fifo_item_size();
>  }
>  
> -static int qsv_decode_init(AVCodecContext *avctx, QSVContext *q)
> +static int qsv_decode_init(AVCodecContext *avctx, QSVContext *q,
> AVPacket *avpkt)
>  {
>      const AVPixFmtDescriptor *desc;
>      mfxSession session = NULL;
> @@ -129,6 +129,17 @@ static int qsv_decode_init(AVCodecContext
> *avctx, QSVContext *q)
>      int frame_width  = avctx->coded_width;
>      int frame_height = avctx->coded_height;
>      int ret;
> +    mfxBitstream bs = { { { 0 } } };
> +
> +    if (avpkt->size) {
> +        bs.Data       = avpkt->data;
> +        bs.DataLength = avpkt->size;
> +        bs.MaxLength  = bs.DataLength;
> +        bs.TimeStamp  = avpkt->pts;
> +        if (avctx->field_order == AV_FIELD_PROGRESSIVE)
> +            bs.DataFlag   |= MFX_BITSTREAM_COMPLETE_FRAME;
+    } else
> +        return AVERROR_INVALIDDATA;
>  
>      desc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);
>      if (!desc)
> @@ -174,32 +185,19 @@ static int qsv_decode_init(AVCodecContext
> *avctx, QSVContext *q)
>      if (ret < 0)
>          return ret;
>  
> -    param.mfx.CodecId      = ret;
> -    param.mfx.CodecProfile = ff_qsv_profile_to_mfx(avctx->codec_id,
> avctx->profile);
> -    param.mfx.CodecLevel   = avctx->level == FF_LEVEL_UNKNOWN ?
> MFX_LEVEL_UNKNOWN : avctx->level;
> -
> -    param.mfx.FrameInfo.BitDepthLuma   = desc->comp[0].depth;
> -    param.mfx.FrameInfo.BitDepthChroma = desc->comp[0].depth;
> -    param.mfx.FrameInfo.Shift          = desc->comp[0].depth > 8;
> -    param.mfx.FrameInfo.FourCC         = q->fourcc;
> -    param.mfx.FrameInfo.Width          = frame_width;
> -    param.mfx.FrameInfo.Height         = frame_height;
> -    param.mfx.FrameInfo.ChromaFormat   = MFX_CHROMAFORMAT_YUV420;
> -
> -    switch (avctx->field_order) {
> -    case AV_FIELD_PROGRESSIVE:
> -        param.mfx.FrameInfo.PicStruct = MFX_PICSTRUCT_PROGRESSIVE;
> -        break;
> -    case AV_FIELD_TT:
> -        param.mfx.FrameInfo.PicStruct = MFX_PICSTRUCT_FIELD_TFF;
> -        break;
> -    case AV_FIELD_BB:
> -        param.mfx.FrameInfo.PicStruct = MFX_PICSTRUCT_FIELD_BFF;
> -        break;
> -    default:
> -        param.mfx.FrameInfo.PicStruct = MFX_PICSTRUCT_UNKNOWN;
> -        break;
> -    }
> +    param.mfx.CodecId = ret;
> +    ret = MFXVideoDECODE_DecodeHeader(q->session, &bs, &param);
This function may potentially return MFX_ERR_MORE_DATA if provided
bitstream don't contain full header. I am not sure whether ffmpeg will
guarantee that... And the decoding error reported by Artie suggests
that something is wrong around this. That can be ffmpeg or mediasdk
issue - need to check what was the data which ffmpeg really passed to
DecodeHeader.

> +    if (ret < 0)
> +        return ff_qsv_print_error(avctx, ret,
> +                "Error decoding stream header");
> +
> +    avctx->width        = param.mfx.FrameInfo.CropW;
> +    avctx->height       = param.mfx.FrameInfo.CropH;
> +    avctx->coded_width  = param.mfx.FrameInfo.Width;
> +    avctx->coded_height = param.mfx.FrameInfo.Height;
> +    avctx->level        = param.mfx.CodecProfile;
> +    avctx->profile      = param.mfx.CodecLevel;
Typo here. You assign profile to level and level to profile.

> +    avctx->field_order  =
> ff_qsv_map_picstruct(param.mfx.FrameInfo.PicStruct);
>  
>      param.IOPattern   = q->iopattern;
>      param.AsyncDepth  = q->async_depth;
> @@ -521,62 +519,22 @@ int ff_qsv_process_data(AVCodecContext *avctx,
> QSVContext *q,
>                       pkt->data, pkt->size, pkt->pts, pkt->dts,
>                       pkt->pos);
>  
> -    avctx->field_order  = q->parser->field_order;
>      /* TODO: flush delayed frames on reinit */
> -    if (q->parser->format       != q->orig_pix_fmt    ||
> -        FFALIGN(q->parser->coded_width, 16)  != FFALIGN(avctx-
> >coded_width, 16) ||
> -        FFALIGN(q->parser->coded_height, 16) != FFALIGN(avctx-
> >coded_height, 16)) {
> +
> +    if (!q->initialized){
>          enum AVPixelFormat pix_fmts[3] = { AV_PIX_FMT_QSV,
> -                                           AV_PIX_FMT_NONE,
> +                                           AV_PIX_FMT_NV12,
>                                             AV_PIX_FMT_NONE };
> -        enum AVPixelFormat qsv_format;
> -        AVPacket zero_pkt = {0};
> -
> -        if (q->buffered_count) {
> -            q->reinit_flag = 1;
> -            /* decode zero-size pkt to flush the buffered pkt before
> reinit */
> -            q->buffered_count--;
> -            return qsv_decode(avctx, q, frame, got_frame,
> &zero_pkt);
> -        }
> -
> -        q->reinit_flag = 0;
> -
> -        qsv_format = ff_qsv_map_pixfmt(q->parser->format, &q-
> >fourcc);
> -        if (qsv_format < 0) {
> -            av_log(avctx, AV_LOG_ERROR,
> -                   "Decoding pixel format '%s' is not supported\n",
> -                   av_get_pix_fmt_name(q->parser->format));
> -            ret = AVERROR(ENOSYS);
> -            goto reinit_fail;
> -        }
> -
> -        q->orig_pix_fmt     = q->parser->format;
> -        avctx->pix_fmt      = pix_fmts[1] = qsv_format;
> -        avctx->width        = q->parser->width;
> -        avctx->height       = q->parser->height;
> -        avctx->coded_width  = FFALIGN(q->parser->coded_width, 16);
> -        avctx->coded_height = FFALIGN(q->parser->coded_height, 16);
> -        avctx->level        = q->avctx_internal->level;
> -        avctx->profile      = q->avctx_internal->profile;
> -
>          ret = ff_get_format(avctx, pix_fmts);
>          if (ret < 0)
>              goto reinit_fail;
>  
>          avctx->pix_fmt = ret;
>  
> -        desc = av_pix_fmt_desc_get(avctx->pix_fmt);
> -        if (!desc)
> -            goto reinit_fail;
> -
> -         if (desc->comp[0].depth > 8) {
> -            avctx->coded_width =  FFALIGN(q->parser->coded_width,
> 32);
> -            avctx->coded_height = FFALIGN(q->parser->coded_height,
> 32);
> -        }
> -
> -        ret = qsv_decode_init(avctx, q);
> +        ret = qsv_decode_init(avctx, q, pkt);
>          if (ret < 0)
>              goto reinit_fail;
> +        q->initialized = 1;
>      }
>  
>      return qsv_decode(avctx, q, frame, got_frame, pkt);
> @@ -589,4 +547,5 @@ reinit_fail:
>  void ff_qsv_decode_flush(AVCodecContext *avctx, QSVContext *q)
>  {
>      q->orig_pix_fmt = AV_PIX_FMT_NONE;
> +    q->initialized = 0;
>  }
> diff --git a/libavcodec/qsvdec.h b/libavcodec/qsvdec.h
> index 111536c..4812fb2 100644
> --- a/libavcodec/qsvdec.h
> +++ b/libavcodec/qsvdec.h
> @@ -63,6 +63,8 @@ typedef struct QSVContext {
>      uint32_t fourcc;
>      mfxFrameInfo frame_info;
>  
> +    int initialized;
> +
>      // options set by the caller
>      int async_depth;
>      int iopattern;


More information about the ffmpeg-devel mailing list