| Message ID | 0101018dd846432e-958ca9e5-663b-43aa-96b1-8bc9fc5d938c-000000@us-west-2.amazonses.com |
|---|---|
| State | Accepted |
| Commit | 69dd1ce610fcffec453a0663c613c9b13165fd9e |
| Headers | show |
| Series | [FFmpeg-devel] avcodec/libsvtav1: send the EOS signal without a one frame delay to allow for the library to operate in a low-delay mode | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
On 2/23/2024 8:21 PM, Cosmin Stejerean via ffmpeg-devel wrote: > From: Cosmin Stejerean <cosmin@cosmin.at> > > Co-authored-by: Amir Naghdinezhad <amir.naghdinezhad@intel.com> > Signed-off-by: Cosmin Stejerean <cosmin@cosmin.at> > --- > libavcodec/libsvtav1.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/libavcodec/libsvtav1.c b/libavcodec/libsvtav1.c > index 3b41f5a39e..1eda63200c 100644 > --- a/libavcodec/libsvtav1.c > +++ b/libavcodec/libsvtav1.c > @@ -539,6 +539,14 @@ static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) > if (svt_ret == EB_NoErrorEmptyQueue) > return AVERROR(EAGAIN); > > +#if SVT_AV1_CHECK_VERSION(2, 0, 0) SVT-AV1 1.8.0 has this value set to 1.8.0, same as in the current git head commit. Is this in preparation for an upcoming release? > + if (headerPtr->flags & EB_BUFFERFLAG_EOS) { > + svt_enc->eos_flag = EOS_RECEIVED; > + svt_av1_enc_release_out_buffer(&headerPtr); > + return AVERROR_EOF; > + } > +#endif > + > ref = get_output_ref(avctx, svt_enc, headerPtr->n_filled_len); > if (!ref) { > av_log(avctx, AV_LOG_ERROR, "Failed to allocate output packet.\n"); > @@ -573,8 +581,10 @@ static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) > if (headerPtr->pic_type == EB_AV1_NON_REF_PICTURE) > pkt->flags |= AV_PKT_FLAG_DISPOSABLE; > > +#if !(SVT_AV1_CHECK_VERSION(2, 0, 0)) > if (headerPtr->flags & EB_BUFFERFLAG_EOS) > svt_enc->eos_flag = EOS_RECEIVED; > +#endif > > ff_side_data_set_encoder_stats(pkt, headerPtr->qp * FF_QP2LAMBDA, NULL, 0, pict_type); >
> On Feb 27, 2024, at 1:19 PM, James Almer <jamrial@gmail.com> wrote: > > SVT-AV1 1.8.0 has this value set to 1.8.0, same as in the current git head commit. Is this in preparation for an upcoming release? Yes, this is in preparation for release 2.0 which is targeted for next week. https://gitlab.com/AOMediaCodec/SVT-AV1/-/issues/2155 is tracking the status, and this API change is one of the outstanding items. - Cosmin
On 2/27/2024 6:36 PM, Cosmin Stejerean via ffmpeg-devel wrote: > > >> On Feb 27, 2024, at 1:19 PM, James Almer <jamrial@gmail.com> wrote: >> >> SVT-AV1 1.8.0 has this value set to 1.8.0, same as in the current git head commit. Is this in preparation for an upcoming release? > > Yes, this is in preparation for release 2.0 which is targeted for next week. https://gitlab.com/AOMediaCodec/SVT-AV1/-/issues/2155 is tracking the status, and this API change is one of the outstanding items. Without this patch, the command "ffmpeg -lavfi testsrc,format=yuv420p -vframes 101 -c:v libsvtav1 -loglevel debug -f null -" gives this: > Output stream #0:0 (video): 101 frames encoded; 101 packets muxed (17890 bytes); > Total: 101 packets (17890 bytes) muxed > frame= 101 fps=0.0 q=31.0 Lsize=N/A time=00:00:04.00 bitrate=N/A Whereas with it applied, i get: > Output stream #0:0 (video): 101 frames encoded; 100 packets muxed (17885 bytes); > Total: 100 packets (17885 bytes) muxed > frame= 100 fps=0.0 q=35.0 Lsize=N/A time=00:00:03.96 bitrate=N/A If i pass it a single frame, i get no output at all. So the last frame is being lost.
> On Feb 27, 2024, at 1:49 PM, James Almer <jamrial@gmail.com> wrote: > >>> SVT-AV1 1.8.0 has this value set to 1.8.0, same as in the current git head commit. Is this in preparation for an upcoming release? >> Yes, this is in preparation for release 2.0 which is targeted for next week. https://gitlab.com/AOMediaCodec/SVT-AV1/-/issues/2155 is tracking the status, and this API change is one of the outstanding items. > > Without this patch, the command "ffmpeg -lavfi testsrc,format=yuv420p -vframes 101 -c:v libsvtav1 -loglevel debug -f null -" gives this: > >> Output stream #0:0 (video): 101 frames encoded; 101 packets muxed (17890 bytes); >> Total: 101 packets (17890 bytes) muxed >> frame= 101 fps=0.0 q=31.0 Lsize=N/A time=00:00:04.00 bitrate=N/A > > Whereas with it applied, i get: > >> Output stream #0:0 (video): 101 frames encoded; 100 packets muxed (17885 bytes); >> Total: 100 packets (17885 bytes) muxed >> frame= 100 fps=0.0 q=35.0 Lsize=N/A time=00:00:03.96 bitrate=N/A > > If i pass it a single frame, i get no output at all. So the last frame is being lost. This change depends on the API change on the SVT-AV1 side, which is in in https://gitlab.com/AOMediaCodec/SVT-AV1/-/merge_requests/2189 This will get merged in shortly as part of the 2.0 release. If I build this patch against the low-delay-api-change branch (from PR 2189) then I get > Output stream #0:0 (video): 101 frames encoded; 101 packets muxed (17970 bytes); > Total: 101 packets (17970 bytes) muxed - Cosmin
On 2/27/2024 7:22 PM, Cosmin Stejerean via ffmpeg-devel wrote: > > >> On Feb 27, 2024, at 1:49 PM, James Almer <jamrial@gmail.com> wrote: >> >>>> SVT-AV1 1.8.0 has this value set to 1.8.0, same as in the current git head commit. Is this in preparation for an upcoming release? >>> Yes, this is in preparation for release 2.0 which is targeted for next week. https://gitlab.com/AOMediaCodec/SVT-AV1/-/issues/2155 is tracking the status, and this API change is one of the outstanding items. >> >> Without this patch, the command "ffmpeg -lavfi testsrc,format=yuv420p -vframes 101 -c:v libsvtav1 -loglevel debug -f null -" gives this: >> >>> Output stream #0:0 (video): 101 frames encoded; 101 packets muxed (17890 bytes); >>> Total: 101 packets (17890 bytes) muxed >>> frame= 101 fps=0.0 q=31.0 Lsize=N/A time=00:00:04.00 bitrate=N/A >> >> Whereas with it applied, i get: >> >>> Output stream #0:0 (video): 101 frames encoded; 100 packets muxed (17885 bytes); >>> Total: 100 packets (17885 bytes) muxed >>> frame= 100 fps=0.0 q=35.0 Lsize=N/A time=00:00:03.96 bitrate=N/A >> >> If i pass it a single frame, i get no output at all. So the last frame is being lost. > > This change depends on the API change on the SVT-AV1 side, which is in in https://gitlab.com/AOMediaCodec/SVT-AV1/-/merge_requests/2189 > > This will get merged in shortly as part of the 2.0 release. If I build this patch against the low-delay-api-change branch (from PR 2189) then I get > >> Output stream #0:0 (video): 101 frames encoded; 101 packets muxed (17970 bytes); >> Total: 101 packets (17970 bytes) muxed Yes, i can confirm it works after applying !2189. Will apply then. Thanks.
diff --git a/libavcodec/libsvtav1.c b/libavcodec/libsvtav1.c index 3b41f5a39e..1eda63200c 100644 --- a/libavcodec/libsvtav1.c +++ b/libavcodec/libsvtav1.c @@ -539,6 +539,14 @@ static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) if (svt_ret == EB_NoErrorEmptyQueue) return AVERROR(EAGAIN); +#if SVT_AV1_CHECK_VERSION(2, 0, 0) + if (headerPtr->flags & EB_BUFFERFLAG_EOS) { + svt_enc->eos_flag = EOS_RECEIVED; + svt_av1_enc_release_out_buffer(&headerPtr); + return AVERROR_EOF; + } +#endif + ref = get_output_ref(avctx, svt_enc, headerPtr->n_filled_len); if (!ref) { av_log(avctx, AV_LOG_ERROR, "Failed to allocate output packet.\n"); @@ -573,8 +581,10 @@ static int eb_receive_packet(AVCodecContext *avctx, AVPacket *pkt) if (headerPtr->pic_type == EB_AV1_NON_REF_PICTURE) pkt->flags |= AV_PKT_FLAG_DISPOSABLE; +#if !(SVT_AV1_CHECK_VERSION(2, 0, 0)) if (headerPtr->flags & EB_BUFFERFLAG_EOS) svt_enc->eos_flag = EOS_RECEIVED; +#endif ff_side_data_set_encoder_stats(pkt, headerPtr->qp * FF_QP2LAMBDA, NULL, 0, pict_type);