On Mon, 2018-10-29 at 21:34 +0000, Mark Thompson wrote:
> On 29/10/18 21:29, Rogozhkin, Dmitry V wrote:
> > On Mon, 2018-10-29 at 21:06 +0000, Mark Thompson wrote:
> > > On 25/10/18 13:36, Zhong Li wrote:
> > > > This option can be used to repect original input I/IDR frame
> > > > type.
> > > >
> > > > Signed-off-by: Zhong Li <zhong...@intel.com>
> > > > ---
> > > > libavcodec/qsvenc.c | 7 +++++++
> > > > libavcodec/qsvenc.h | 2 ++
> > > > 2 files changed, 9 insertions(+)
> > > >
> > > > diff --git a/libavcodec/qsvenc.c b/libavcodec/qsvenc.c
> > > > index 948751d..e534dcf 100644
> > > > --- a/libavcodec/qsvenc.c
> > > > +++ b/libavcodec/qsvenc.c
> > > > @@ -1192,6 +1192,13 @@ static int encode_frame(AVCodecContext
> > > > *avctx, QSVEncContext *q,
> > > > if (qsv_frame) {
> > > > surf = &qsv_frame->surface;
> > > > enc_ctrl = &qsv_frame->enc_ctrl;
> > > > +
> > > > + if (q->forced_idr >= 0 && frame->pict_type ==
> > > > AV_PICTURE_TYPE_I) {
> > > > + enc_ctrl->FrameType = MFX_FRAMETYPE_I |
> > > > MFX_FRAMETYPE_REF;
> > > > + if (q->forced_idr || frame->key_frame)
> > > > + enc_ctrl->FrameType |= MFX_FRAMETYPE_IDR;
> > > > + } else
> > > > + enc_ctrl->FrameType = MFX_FRAMETYPE_UNKNOWN;
> > > > }
> > > >
> > > > ret = av_new_packet(&new_pkt, q->packet_size);
> > > > diff --git a/libavcodec/qsvenc.h b/libavcodec/qsvenc.h
> > > > index 055b4a6..1f97f77 100644
> > > > --- a/libavcodec/qsvenc.h
> > > > +++ b/libavcodec/qsvenc.h
> > > > @@ -87,6 +87,7 @@
> > > > { "adaptive_i", "Adaptive I-frame
> > > > placement", OFFSET(qsv.adaptive_i), AV_OPT_TYPE
> > > > _INT
> > > > , { .i64 = -1 }, -1, 1, VE
> > > > }, \
> > > > { "adaptive_b", "Adaptive B-frame
> > > > placement", OFFSET(qsv.adaptive_b), AV_OPT_TYPE
> > > > _INT
> > > > , { .i64 = -1 }, -1, 1, VE
> > > > }, \
> > > > { "b_strategy", "Strategy to choose between I/P/B-frames",
> > > > OFFSET(qsv.b_strategy), AV_OPT_TYPE_INT, { .i64 = -1 },
> > > > -1, 1, VE }, \
> > > > +{ "forced_idr", "Forcing I frames as IDR
> > > > frames", OFFSET(qsv.forced_idr), AV_OPT_TYPE_INT, {
> > > > .i64 = -1 }, -1, 1, VE }, \
> > > >
> > > > typedef int SetEncodeCtrlCB (AVCodecContext *avctx,
> > > > const AVFrame *frame,
> > > > mfxEncodeCtrl*
> > > > enc_ctrl);
> > > > @@ -168,6 +169,7 @@ typedef struct QSVEncContext {
> > > > #endif
> > > > char *load_plugins;
> > > > SetEncodeCtrlCB *set_encode_ctrl_cb;
> > > > + int forced_idr;
> > > > } QSVEncContext;
> > > >
> > > > int ff_qsv_enc_init(AVCodecContext *avctx, QSVEncContext *q);
> > > >
> > >
> > > This seems confusing, because it doesn't match what forced_idr
> > > does
> > > in any other encoder.
> > >
> > > Checking of pict_type for AV_PICTURE_TYPE_I in order to get a key
> > > frame (of whatever kind) is always enabled if supported (many
> > > encoders). The "forced_idr" option to H.26[45] encoders
> > > (libx264,
> > > libx265) then forces that to be an IDR frame, not just an I
> > > frame.
> >
> > Is there an option to disable/override this behavior? Will "-
> > force_key_frames" do the trick?
>
> Not in libavcodec; encoders always looks at pict_type (e.g.
> <http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavcodec/libx264.c;
> h=d6367bf557953ec4e91d057d551848b3b2ba9d36;hb=HEAD#l300>).
Where gop structure is being set in SW encoders? I mean, is pict_type
set by infrastructure according to -g and -bf options + some
information on the incoming stream (passing thru I frames probably?) or
pict_type is in a way recommendation to follow, but general gop
structure is calculated/maintained inside the encoder?
>
> "-force_key_frames" is an option to the ffmpeg utility which sets
> pict_type dependent on some expression, making use of the above
> behaviour.
>
> - Mark
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
