On Fri, Sep 6, 2024 at 1:57 AM Anton Khirnov <an...@khirnov.net> wrote:
>
> Quoting Lynne via ffmpeg-devel (2024-09-05 23:27:40)
> > On 05/09/2024 07:16, Cameron Gutman wrote:
> > > Signed-off-by: Cameron Gutman <aicomman...@gmail.com>
> > > ---
> > > libavcodec/amfenc.c | 34 +++++++++++++++++++++++++++++++++-
> > > 1 file changed, 33 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/libavcodec/amfenc.c b/libavcodec/amfenc.c
> > > index 41eaef9758..6f2b211d14 100644
> > > --- a/libavcodec/amfenc.c
> > > +++ b/libavcodec/amfenc.c
> > > @@ -766,11 +766,43 @@ int ff_amf_receive_packet(AVCodecContext *avctx,
> > > AVPacket *avpkt)
> > > switch (avctx->codec->id) {
> > > case AV_CODEC_ID_H264:
> > > AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_INSERT_AUD, !!ctx->aud);
> > > + if (frame->flags & AV_FRAME_FLAG_KEY) {
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_INSERT_SPS, 1);
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_INSERT_PPS, 1);
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_FORCE_PICTURE_TYPE, AMF_VIDEO_ENCODER_PICTURE_TYPE_IDR);
> > > + } else {
> > > + switch (frame->pict_type) {
> > > + case AV_PICTURE_TYPE_I:
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_FORCE_PICTURE_TYPE, AMF_VIDEO_ENCODER_PICTURE_TYPE_I);
> > > break;
> > > + case AV_PICTURE_TYPE_P:
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_FORCE_PICTURE_TYPE, AMF_VIDEO_ENCODER_PICTURE_TYPE_P);
> > > break;
> > > + case AV_PICTURE_TYPE_B:
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_FORCE_PICTURE_TYPE, AMF_VIDEO_ENCODER_PICTURE_TYPE_B);
> > > break;
> > > + }
> > > + }
> > > break;
> > > case AV_CODEC_ID_HEVC:
> > > AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_HEVC_INSERT_AUD, !!ctx->aud);
> > > + if (frame->flags & AV_FRAME_FLAG_KEY) {
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_HEVC_INSERT_HEADER, 1);
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_HEVC_FORCE_PICTURE_TYPE,
> > > AMF_VIDEO_ENCODER_HEVC_PICTURE_TYPE_IDR);
> > > + } else {
> > > + switch (frame->pict_type) {
> > > + case AV_PICTURE_TYPE_I:
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_HEVC_FORCE_PICTURE_TYPE,
> > > AMF_VIDEO_ENCODER_HEVC_PICTURE_TYPE_I); break;
> > > + case AV_PICTURE_TYPE_P:
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_HEVC_FORCE_PICTURE_TYPE,
> > > AMF_VIDEO_ENCODER_HEVC_PICTURE_TYPE_P); break;
> > > + }
> > > + }
> > > + break;
> > > + case AV_CODEC_ID_AV1:
> > > + if (frame->flags & AV_FRAME_FLAG_KEY) {
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_AV1_FORCE_INSERT_SEQUENCE_HEADER, 1);
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_AV1_FORCE_FRAME_TYPE,
> > > AMF_VIDEO_ENCODER_AV1_FORCE_FRAME_TYPE_KEY);
> > > + } else if (frame->pict_type == AV_PICTURE_TYPE_I) {
> > > + AMF_ASSIGN_PROPERTY_INT64(res, surface,
> > > AMF_VIDEO_ENCODER_AV1_FORCE_FRAME_TYPE,
> > > AMF_VIDEO_ENCODER_AV1_FORCE_FRAME_TYPE_INTRA_ONLY);
> > > + }
> > > break;
> > > - //case AV_CODEC_ID_AV1 not supported
> > > default:
> > > break;
> > > }
> >
> > You should hook up FFCodec.flush instead. That's what nvenc uses to
> > reset the GOP.
>
> No, the patch does almost the right thing, though most other encoders
> look at pict_type instead.
>
I took a look at what other common encoders do. Most of the ones I looked at
use pict_type as you mentioned, and in some cases there's an additional codec
option called 'forced-idr' or 'forced_idr' that causes AV_PICTURE_TYPE_I to
generate an IDR frame instead of a normal I frame. Some of these encoders that
don't have a 'forced-idr' option may require a closed GOP to get IDR frames via
AV_PICTURE_TYPE_I but I couldn't easily tell that from code review so I'm
ignoring it here.
Here is a brief summary of what I found is required to generate an IDR frame:
- qsvenc: AV_PICTURE_TYPE_I and either forced_idr=1 or AV_FRAME_FLAG_KEY
- nvenc: AV_PICTURE_TYPE_I and forced-idr=1
- libx265: AV_PICTURE_TYPE_I and forced-idr=1
- libx264: AV_PICTURE_TYPE_I and forced-idr=1
- d3d12va: AV_PICTURE_TYPE_I alone
- vaapi: AV_PICTURE_TYPE_I alone
- videotoolbox: AV_PICTURE_TYPE_I alone
- mfenc: AV_PICTURE_TYPE_I alone
Given that it's basically a tie between encoders that give an IDR frame for
any I frame request and those that require a secondary option to do so, I
think I will follow the lead of NVENC and the software encoders and add a
'forced-idr' option to the AMF encoders that will determine the behavior of
AV_PICTURE_TYPE_I for IDR frame generation.
Cameron
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
