Message ID | tencent_42C2E7D12AB5D3109C6ECB97A62CD08C2606@qq.com |
---|---|
State | New |
Headers | show |
Series | [FFmpeg-devel] avcodec/nvenc: Rework on DTS generation | expand |
Context | Check | Description |
---|---|---|
andriy/make_x86 | success | Make finished |
andriy/make_fate_x86 | success | Make fate finished |
On 17/07/2024 11:47, Zhao Zhili wrote: > From: Zhao Zhili <zhilizhao@tencent.com> > > Before the patch, the method to generate DTS only works with > timebase equal to 1/fps. With timebase like 1/1000 > > ./ffmpeg -i foo.mp4 -an -c:v h264_nvenc -enc_time_base 1/1000 bar.mp4 > > pts 0 dts -3 > pts 160 dts 37 > pts 80 dts 77 > pts 40 dts 117 <-- invalid > pts 120 dts 157 > pts 320 dts 197 > pts 240 dts 237 > pts 200 dts 277 <-- invalid > pts 280 dts 317 <-- invalid > > The generated DTS can be larger than PTS, since it only reorder the > input PTS and minus the number of frame delay, which doesn't take > timebase into account. It should minus the "time" of frame delay. > > 9a245bd trying to fix the issue, but the implementation is incomplete, > which only use time_base.num. Then it got reverted by ac7c265b33b. > > After this patch: > > pts 0 dts -120 > pts 160 dts -80 > pts 80 dts -40 > pts 40 dts 0 > pts 120 dts 40 > pts 320 dts 80 > pts 240 dts 120 > pts 200 dts 160 > pts 280 dts 200 > --- > libavcodec/nvenc.c | 62 ++++++++++++++++++++++++++++++++++++++-------- > libavcodec/nvenc.h | 3 +++ > 2 files changed, 55 insertions(+), 10 deletions(-) > > diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c > index ab92395ed6..34448462f0 100644 > --- a/libavcodec/nvenc.c > +++ b/libavcodec/nvenc.c > @@ -1893,7 +1893,8 @@ static av_cold int nvenc_setup_surfaces(AVCodecContext *avctx) > if (!ctx->frame_data_array) > return AVERROR(ENOMEM); > > - ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces, sizeof(int64_t), 0); > + ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces + ctx->encode_config.frameIntervalP, > + sizeof(int64_t), 0); > if (!ctx->timestamp_list) > return AVERROR(ENOMEM); > > @@ -2347,26 +2348,65 @@ static inline int64_t timestamp_queue_dequeue(AVFifo *queue) > return timestamp; > } > > +static inline int64_t timestamp_queue_peek(AVFifo *queue, size_t index) > +{ > + int64_t timestamp = AV_NOPTS_VALUE; > + av_fifo_peek(queue, ×tamp, 1, index); > + > + return timestamp; > +} > + > static int nvenc_set_timestamp(AVCodecContext *avctx, > NV_ENC_LOCK_BITSTREAM *params, > AVPacket *pkt) > { > NvencContext *ctx = avctx->priv_data; > + int delay; > + int64_t delay_time; > > pkt->pts = params->outputTimeStamp; > > - if (avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER) { > -FF_DISABLE_DEPRECATION_WARNINGS > - pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list) - > -#if FF_API_TICKS_PER_FRAME > - FFMAX(avctx->ticks_per_frame, 1) * > -#endif > - FFMAX(ctx->encode_config.frameIntervalP - 1, 0); > -FF_ENABLE_DEPRECATION_WARNINGS > - } else { > + if (!(avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER)) { > pkt->dts = pkt->pts; > + return 0; > + } > + > + // This can be more than necessary, but we don't know the real reorder delay. > + delay = FFMAX(ctx->encode_config.frameIntervalP - 1, 0); > + if (ctx->output_frame_num >= delay) { > + pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list); > + ctx->output_frame_num++; > + return 0; > } > > + delay_time = ctx->initial_delay_time; > + if (!delay_time) { > + int64_t t1, t2, t3; > + t1 = timestamp_queue_peek(ctx->timestamp_list, delay); > + t2 = timestamp_queue_peek(ctx->timestamp_list, 0); > + t3 = (delay > 1) ? timestamp_queue_peek(ctx->timestamp_list, 1) : t1; > + > + if (t1 != AV_NOPTS_VALUE) { > + delay_time = t1 - t2; > + } else if (avctx->framerate.num > 0 && avctx->framerate.den > 0) { > + delay_time = av_rescale_q(delay, (AVRational) {avctx->framerate.den, avctx->framerate.num}, > + avctx->time_base); > + } else if (t3 != AV_NOPTS_VALUE) { > + delay_time = delay * (t3 - t2); > + } else { > + delay_time = delay; > + } > + ctx->initial_delay_time = delay_time; > + } > + > + /* The following method is simple, but doesn't guarantee monotonic with VFR > + * when delay_time isn't accurate (that is, t1 == AV_NOPTS_VALUE) > + * > + * dts = timestamp_queue_peek(ctx->timestamp_list, ctx->output_frame_num) - delay_time > + */ > + pkt->dts = timestamp_queue_peek(ctx->timestamp_list, 0) - delay_time * (delay - ctx->output_frame_num) / delay; > + ctx->output_frame_num++; > + > return 0; > } > > @@ -2902,4 +2942,6 @@ av_cold void ff_nvenc_encode_flush(AVCodecContext *avctx) > > nvenc_send_frame(avctx, NULL); > av_fifo_reset2(ctx->timestamp_list); > + ctx->output_frame_num = 0; > + ctx->initial_delay_time = 0; > } > diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h > index 09de00badc..dc7fe41951 100644 > --- a/libavcodec/nvenc.h > +++ b/libavcodec/nvenc.h > @@ -206,6 +206,9 @@ typedef struct NvencContext > AVFifo *output_surface_queue; > AVFifo *output_surface_ready_queue; > AVFifo *timestamp_list; > + // This is for DTS calculating, reset after flush > + int64_t output_frame_num; > + int64_t initial_delay_time; > > NV_ENC_SEI_PAYLOAD *sei_data; > int sei_data_size; This looks good on first glance, but I'm away from any nvenc capable PC for the next two weeks, so it'll unfortunately be a while until I can properly test and apply. Timo
applied, thanks
On Fri, 2 Aug 2024, 19:28 Timo Rothenpieler, <timo@rothenpieler.org> wrote: > applied, thanks > This is a bugfix, can it be backported to release 6.x and 7.x? >
On 02.08.2024 18:54, Dennis Mungai wrote: > On Fri, 2 Aug 2024, 19:28 Timo Rothenpieler, <timo@rothenpieler.org> wrote: > >> applied, thanks >> > > > This is a bugfix, can it be backported to release 6.x and 7.x? It's also a quite substantial change in behaviour, so I'd like to wait with that for a while.
Zhao Zhili: > From: Zhao Zhili <zhilizhao@tencent.com> > > Before the patch, the method to generate DTS only works with > timebase equal to 1/fps. With timebase like 1/1000 > > ./ffmpeg -i foo.mp4 -an -c:v h264_nvenc -enc_time_base 1/1000 bar.mp4 > > pts 0 dts -3 > pts 160 dts 37 > pts 80 dts 77 > pts 40 dts 117 <-- invalid > pts 120 dts 157 > pts 320 dts 197 > pts 240 dts 237 > pts 200 dts 277 <-- invalid > pts 280 dts 317 <-- invalid > > The generated DTS can be larger than PTS, since it only reorder the > input PTS and minus the number of frame delay, which doesn't take > timebase into account. It should minus the "time" of frame delay. > > 9a245bd trying to fix the issue, but the implementation is incomplete, > which only use time_base.num. Then it got reverted by ac7c265b33b. > > After this patch: > > pts 0 dts -120 > pts 160 dts -80 > pts 80 dts -40 > pts 40 dts 0 > pts 120 dts 40 > pts 320 dts 80 > pts 240 dts 120 > pts 200 dts 160 > pts 280 dts 200 It looks like you only have two reorder frames here, so shouldn't the first dts be -80? > --- > libavcodec/nvenc.c | 62 ++++++++++++++++++++++++++++++++++++++-------- > libavcodec/nvenc.h | 3 +++ > 2 files changed, 55 insertions(+), 10 deletions(-) > > diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c > index ab92395ed6..34448462f0 100644 > --- a/libavcodec/nvenc.c > +++ b/libavcodec/nvenc.c > @@ -1893,7 +1893,8 @@ static av_cold int nvenc_setup_surfaces(AVCodecContext *avctx) > if (!ctx->frame_data_array) > return AVERROR(ENOMEM); > > - ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces, sizeof(int64_t), 0); > + ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces + ctx->encode_config.frameIntervalP, > + sizeof(int64_t), 0); > if (!ctx->timestamp_list) > return AVERROR(ENOMEM); > > @@ -2347,26 +2348,65 @@ static inline int64_t timestamp_queue_dequeue(AVFifo *queue) > return timestamp; > } > > +static inline int64_t timestamp_queue_peek(AVFifo *queue, size_t index) > +{ > + int64_t timestamp = AV_NOPTS_VALUE; > + av_fifo_peek(queue, ×tamp, 1, index); > + > + return timestamp; > +} > + > static int nvenc_set_timestamp(AVCodecContext *avctx, > NV_ENC_LOCK_BITSTREAM *params, > AVPacket *pkt) > { > NvencContext *ctx = avctx->priv_data; > + int delay; > + int64_t delay_time; > > pkt->pts = params->outputTimeStamp; > > - if (avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER) { > -FF_DISABLE_DEPRECATION_WARNINGS > - pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list) - > -#if FF_API_TICKS_PER_FRAME > - FFMAX(avctx->ticks_per_frame, 1) * > -#endif > - FFMAX(ctx->encode_config.frameIntervalP - 1, 0); > -FF_ENABLE_DEPRECATION_WARNINGS > - } else { > + if (!(avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER)) { > pkt->dts = pkt->pts; > + return 0; > + } > + > + // This can be more than necessary, but we don't know the real reorder delay. > + delay = FFMAX(ctx->encode_config.frameIntervalP - 1, 0); > + if (ctx->output_frame_num >= delay) { > + pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list); > + ctx->output_frame_num++; > + return 0; > } > > + delay_time = ctx->initial_delay_time; > + if (!delay_time) { > + int64_t t1, t2, t3; > + t1 = timestamp_queue_peek(ctx->timestamp_list, delay); > + t2 = timestamp_queue_peek(ctx->timestamp_list, 0); > + t3 = (delay > 1) ? timestamp_queue_peek(ctx->timestamp_list, 1) : t1; > + > + if (t1 != AV_NOPTS_VALUE) { > + delay_time = t1 - t2; > + } else if (avctx->framerate.num > 0 && avctx->framerate.den > 0) { > + delay_time = av_rescale_q(delay, (AVRational) {avctx->framerate.den, avctx->framerate.num}, > + avctx->time_base); > + } else if (t3 != AV_NOPTS_VALUE) { > + delay_time = delay * (t3 - t2); > + } else { > + delay_time = delay; > + } > + ctx->initial_delay_time = delay_time; > + } > + > + /* The following method is simple, but doesn't guarantee monotonic with VFR > + * when delay_time isn't accurate (that is, t1 == AV_NOPTS_VALUE) > + * > + * dts = timestamp_queue_peek(ctx->timestamp_list, ctx->output_frame_num) - delay_time > + */ > + pkt->dts = timestamp_queue_peek(ctx->timestamp_list, 0) - delay_time * (delay - ctx->output_frame_num) / delay; > + ctx->output_frame_num++; > + > return 0; > } > > @@ -2902,4 +2942,6 @@ av_cold void ff_nvenc_encode_flush(AVCodecContext *avctx) > > nvenc_send_frame(avctx, NULL); > av_fifo_reset2(ctx->timestamp_list); > + ctx->output_frame_num = 0; > + ctx->initial_delay_time = 0; > } > diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h > index 09de00badc..dc7fe41951 100644 > --- a/libavcodec/nvenc.h > +++ b/libavcodec/nvenc.h > @@ -206,6 +206,9 @@ typedef struct NvencContext > AVFifo *output_surface_queue; > AVFifo *output_surface_ready_queue; > AVFifo *timestamp_list; > + // This is for DTS calculating, reset after flush > + int64_t output_frame_num; > + int64_t initial_delay_time; > > NV_ENC_SEI_PAYLOAD *sei_data; > int sei_data_size;
> On Aug 7, 2024, at 18:18, Andreas Rheinhardt <andreas.rheinhardt@outlook.com> wrote: > > Zhao Zhili: >> From: Zhao Zhili <zhilizhao@tencent.com> >> >> Before the patch, the method to generate DTS only works with >> timebase equal to 1/fps. With timebase like 1/1000 >> >> ./ffmpeg -i foo.mp4 -an -c:v h264_nvenc -enc_time_base 1/1000 bar.mp4 >> >> pts 0 dts -3 >> pts 160 dts 37 >> pts 80 dts 77 >> pts 40 dts 117 <-- invalid >> pts 120 dts 157 >> pts 320 dts 197 >> pts 240 dts 237 >> pts 200 dts 277 <-- invalid >> pts 280 dts 317 <-- invalid >> >> The generated DTS can be larger than PTS, since it only reorder the >> input PTS and minus the number of frame delay, which doesn't take >> timebase into account. It should minus the "time" of frame delay. >> >> 9a245bd trying to fix the issue, but the implementation is incomplete, >> which only use time_base.num. Then it got reverted by ac7c265b33b. >> >> After this patch: >> >> pts 0 dts -120 >> pts 160 dts -80 >> pts 80 dts -40 >> pts 40 dts 0 >> pts 120 dts 40 >> pts 320 dts 80 >> pts 240 dts 120 >> pts 200 dts 160 >> pts 280 dts 200 > > It looks like you only have two reorder frames here, so shouldn't the > first dts be -80? But NVENC doesn’t provide the reorder delay value. I have added a comment without changing the method to guess a maximum reorder delay. Maybe we can do better by parse the maximum reorder delay from VPS/SPS/PPS, it’s not the patch supposed to do. I do hope Nvidia to fix their SDK issues than workaround in FFmpeg. >> + >> + // This can be more than necessary, but we don't know the real reorder delay. >> + delay = FFMAX(ctx->encode_config.frameIntervalP - 1, 0); > >> --- >> libavcodec/nvenc.c | 62 ++++++++++++++++++++++++++++++++++++++-------- >> libavcodec/nvenc.h | 3 +++ >> 2 files changed, 55 insertions(+), 10 deletions(-) >> >> diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c >> index ab92395ed6..34448462f0 100644 >> --- a/libavcodec/nvenc.c >> +++ b/libavcodec/nvenc.c >> @@ -1893,7 +1893,8 @@ static av_cold int nvenc_setup_surfaces(AVCodecContext *avctx) >> if (!ctx->frame_data_array) >> return AVERROR(ENOMEM); >> >> - ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces, sizeof(int64_t), 0); >> + ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces + ctx->encode_config.frameIntervalP, >> + sizeof(int64_t), 0); >> if (!ctx->timestamp_list) >> return AVERROR(ENOMEM); >> >> @@ -2347,26 +2348,65 @@ static inline int64_t timestamp_queue_dequeue(AVFifo *queue) >> return timestamp; >> } >> >> +static inline int64_t timestamp_queue_peek(AVFifo *queue, size_t index) >> +{ >> + int64_t timestamp = AV_NOPTS_VALUE; >> + av_fifo_peek(queue, ×tamp, 1, index); >> + >> + return timestamp; >> +} >> + >> static int nvenc_set_timestamp(AVCodecContext *avctx, >> NV_ENC_LOCK_BITSTREAM *params, >> AVPacket *pkt) >> { >> NvencContext *ctx = avctx->priv_data; >> + int delay; >> + int64_t delay_time; >> >> pkt->pts = params->outputTimeStamp; >> >> - if (avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER) { >> -FF_DISABLE_DEPRECATION_WARNINGS >> - pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list) - >> -#if FF_API_TICKS_PER_FRAME >> - FFMAX(avctx->ticks_per_frame, 1) * >> -#endif >> - FFMAX(ctx->encode_config.frameIntervalP - 1, 0); >> -FF_ENABLE_DEPRECATION_WARNINGS >> - } else { >> + if (!(avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER)) { >> pkt->dts = pkt->pts; >> + return 0; >> + } >> + >> + // This can be more than necessary, but we don't know the real reorder delay. >> + delay = FFMAX(ctx->encode_config.frameIntervalP - 1, 0); >> + if (ctx->output_frame_num >= delay) { >> + pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list); >> + ctx->output_frame_num++; >> + return 0; >> } >> >> + delay_time = ctx->initial_delay_time; >> + if (!delay_time) { >> + int64_t t1, t2, t3; >> + t1 = timestamp_queue_peek(ctx->timestamp_list, delay); >> + t2 = timestamp_queue_peek(ctx->timestamp_list, 0); >> + t3 = (delay > 1) ? timestamp_queue_peek(ctx->timestamp_list, 1) : t1; >> + >> + if (t1 != AV_NOPTS_VALUE) { >> + delay_time = t1 - t2; >> + } else if (avctx->framerate.num > 0 && avctx->framerate.den > 0) { >> + delay_time = av_rescale_q(delay, (AVRational) {avctx->framerate.den, avctx->framerate.num}, >> + avctx->time_base); >> + } else if (t3 != AV_NOPTS_VALUE) { >> + delay_time = delay * (t3 - t2); >> + } else { >> + delay_time = delay; >> + } >> + ctx->initial_delay_time = delay_time; >> + } >> + >> + /* The following method is simple, but doesn't guarantee monotonic with VFR >> + * when delay_time isn't accurate (that is, t1 == AV_NOPTS_VALUE) >> + * >> + * dts = timestamp_queue_peek(ctx->timestamp_list, ctx->output_frame_num) - delay_time >> + */ >> + pkt->dts = timestamp_queue_peek(ctx->timestamp_list, 0) - delay_time * (delay - ctx->output_frame_num) / delay; >> + ctx->output_frame_num++; >> + >> return 0; >> } >> >> @@ -2902,4 +2942,6 @@ av_cold void ff_nvenc_encode_flush(AVCodecContext *avctx) >> >> nvenc_send_frame(avctx, NULL); >> av_fifo_reset2(ctx->timestamp_list); >> + ctx->output_frame_num = 0; >> + ctx->initial_delay_time = 0; >> } >> diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h >> index 09de00badc..dc7fe41951 100644 >> --- a/libavcodec/nvenc.h >> +++ b/libavcodec/nvenc.h >> @@ -206,6 +206,9 @@ typedef struct NvencContext >> AVFifo *output_surface_queue; >> AVFifo *output_surface_ready_queue; >> AVFifo *timestamp_list; >> + // This is for DTS calculating, reset after flush >> + int64_t output_frame_num; >> + int64_t initial_delay_time; >> >> NV_ENC_SEI_PAYLOAD *sei_data; >> int sei_data_size; > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c index ab92395ed6..34448462f0 100644 --- a/libavcodec/nvenc.c +++ b/libavcodec/nvenc.c @@ -1893,7 +1893,8 @@ static av_cold int nvenc_setup_surfaces(AVCodecContext *avctx) if (!ctx->frame_data_array) return AVERROR(ENOMEM); - ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces, sizeof(int64_t), 0); + ctx->timestamp_list = av_fifo_alloc2(ctx->nb_surfaces + ctx->encode_config.frameIntervalP, + sizeof(int64_t), 0); if (!ctx->timestamp_list) return AVERROR(ENOMEM); @@ -2347,26 +2348,65 @@ static inline int64_t timestamp_queue_dequeue(AVFifo *queue) return timestamp; } +static inline int64_t timestamp_queue_peek(AVFifo *queue, size_t index) +{ + int64_t timestamp = AV_NOPTS_VALUE; + av_fifo_peek(queue, ×tamp, 1, index); + + return timestamp; +} + static int nvenc_set_timestamp(AVCodecContext *avctx, NV_ENC_LOCK_BITSTREAM *params, AVPacket *pkt) { NvencContext *ctx = avctx->priv_data; + int delay; + int64_t delay_time; pkt->pts = params->outputTimeStamp; - if (avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER) { -FF_DISABLE_DEPRECATION_WARNINGS - pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list) - -#if FF_API_TICKS_PER_FRAME - FFMAX(avctx->ticks_per_frame, 1) * -#endif - FFMAX(ctx->encode_config.frameIntervalP - 1, 0); -FF_ENABLE_DEPRECATION_WARNINGS - } else { + if (!(avctx->codec_descriptor->props & AV_CODEC_PROP_REORDER)) { pkt->dts = pkt->pts; + return 0; + } + + // This can be more than necessary, but we don't know the real reorder delay. + delay = FFMAX(ctx->encode_config.frameIntervalP - 1, 0); + if (ctx->output_frame_num >= delay) { + pkt->dts = timestamp_queue_dequeue(ctx->timestamp_list); + ctx->output_frame_num++; + return 0; } + delay_time = ctx->initial_delay_time; + if (!delay_time) { + int64_t t1, t2, t3; + t1 = timestamp_queue_peek(ctx->timestamp_list, delay); + t2 = timestamp_queue_peek(ctx->timestamp_list, 0); + t3 = (delay > 1) ? timestamp_queue_peek(ctx->timestamp_list, 1) : t1; + + if (t1 != AV_NOPTS_VALUE) { + delay_time = t1 - t2; + } else if (avctx->framerate.num > 0 && avctx->framerate.den > 0) { + delay_time = av_rescale_q(delay, (AVRational) {avctx->framerate.den, avctx->framerate.num}, + avctx->time_base); + } else if (t3 != AV_NOPTS_VALUE) { + delay_time = delay * (t3 - t2); + } else { + delay_time = delay; + } + ctx->initial_delay_time = delay_time; + } + + /* The following method is simple, but doesn't guarantee monotonic with VFR + * when delay_time isn't accurate (that is, t1 == AV_NOPTS_VALUE) + * + * dts = timestamp_queue_peek(ctx->timestamp_list, ctx->output_frame_num) - delay_time + */ + pkt->dts = timestamp_queue_peek(ctx->timestamp_list, 0) - delay_time * (delay - ctx->output_frame_num) / delay; + ctx->output_frame_num++; + return 0; } @@ -2902,4 +2942,6 @@ av_cold void ff_nvenc_encode_flush(AVCodecContext *avctx) nvenc_send_frame(avctx, NULL); av_fifo_reset2(ctx->timestamp_list); + ctx->output_frame_num = 0; + ctx->initial_delay_time = 0; } diff --git a/libavcodec/nvenc.h b/libavcodec/nvenc.h index 09de00badc..dc7fe41951 100644 --- a/libavcodec/nvenc.h +++ b/libavcodec/nvenc.h @@ -206,6 +206,9 @@ typedef struct NvencContext AVFifo *output_surface_queue; AVFifo *output_surface_ready_queue; AVFifo *timestamp_list; + // This is for DTS calculating, reset after flush + int64_t output_frame_num; + int64_t initial_delay_time; NV_ENC_SEI_PAYLOAD *sei_data; int sei_data_size;
From: Zhao Zhili <zhilizhao@tencent.com> Before the patch, the method to generate DTS only works with timebase equal to 1/fps. With timebase like 1/1000 ./ffmpeg -i foo.mp4 -an -c:v h264_nvenc -enc_time_base 1/1000 bar.mp4 pts 0 dts -3 pts 160 dts 37 pts 80 dts 77 pts 40 dts 117 <-- invalid pts 120 dts 157 pts 320 dts 197 pts 240 dts 237 pts 200 dts 277 <-- invalid pts 280 dts 317 <-- invalid The generated DTS can be larger than PTS, since it only reorder the input PTS and minus the number of frame delay, which doesn't take timebase into account. It should minus the "time" of frame delay. 9a245bd trying to fix the issue, but the implementation is incomplete, which only use time_base.num. Then it got reverted by ac7c265b33b. After this patch: pts 0 dts -120 pts 160 dts -80 pts 80 dts -40 pts 40 dts 0 pts 120 dts 40 pts 320 dts 80 pts 240 dts 120 pts 200 dts 160 pts 280 dts 200 --- libavcodec/nvenc.c | 62 ++++++++++++++++++++++++++++++++++++++-------- libavcodec/nvenc.h | 3 +++ 2 files changed, 55 insertions(+), 10 deletions(-)