1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-11-21 10:55:51 +02:00
FFmpeg/fftools/ffmpeg_mux.c

676 lines
18 KiB
C
Raw Normal View History

/*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <stdatomic.h>
#include <stdio.h>
#include <string.h>
#include "ffmpeg.h"
#include "objpool.h"
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
#include "sync_queue.h"
#include "thread_queue.h"
#include "libavutil/fifo.h"
#include "libavutil/intreadwrite.h"
#include "libavutil/log.h"
#include "libavutil/mem.h"
#include "libavutil/timestamp.h"
#include "libavutil/thread.h"
#include "libavcodec/packet.h"
#include "libavformat/avformat.h"
#include "libavformat/avio.h"
typedef struct MuxStream {
/* the packets are buffered here until the muxer is ready to be initialized */
AVFifo *muxing_queue;
/*
* The size of the AVPackets' buffers in queue.
* Updated when a packet is either pushed or pulled from the queue.
*/
size_t muxing_queue_data_size;
/* dts of the last packet sent to the muxer, in the stream timebase
* used for making up missing dts values */
int64_t last_mux_dts;
} MuxStream;
struct Muxer {
AVFormatContext *fc;
pthread_t thread;
ThreadQueue *tq;
MuxStream *streams;
AVDictionary *opts;
int thread_queue_size;
/* filesize limit expressed in bytes */
int64_t limit_filesize;
atomic_int_least64_t last_filesize;
int header_written;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
AVPacket *sq_pkt;
};
static int want_sdp = 1;
static int queue_packet(OutputFile *of, OutputStream *ost, AVPacket *pkt)
{
MuxStream *ms = &of->mux->streams[ost->index];
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
AVPacket *tmp_pkt = NULL;
int ret;
if (!av_fifo_can_write(ms->muxing_queue)) {
size_t cur_size = av_fifo_can_read(ms->muxing_queue);
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
size_t pkt_size = pkt ? pkt->size : 0;
unsigned int are_we_over_size =
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
(ms->muxing_queue_data_size + pkt_size) > ost->muxing_queue_data_threshold;
size_t limit = are_we_over_size ? ost->max_muxing_queue_size : SIZE_MAX;
size_t new_size = FFMIN(2 * cur_size, limit);
if (new_size <= cur_size) {
av_log(NULL, AV_LOG_ERROR,
"Too many packets buffered for output stream %d:%d.\n",
ost->file_index, ost->st->index);
return AVERROR(ENOSPC);
}
ret = av_fifo_grow2(ms->muxing_queue, new_size - cur_size);
if (ret < 0)
return ret;
}
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
if (pkt) {
2022-06-10 12:55:05 +02:00
ret = av_packet_make_refcounted(pkt);
if (ret < 0)
return ret;
2022-06-10 12:55:05 +02:00
tmp_pkt = av_packet_alloc();
if (!tmp_pkt)
return AVERROR(ENOMEM);
2022-06-10 12:55:05 +02:00
av_packet_move_ref(tmp_pkt, pkt);
ms->muxing_queue_data_size += tmp_pkt->size;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
}
av_fifo_write(ms->muxing_queue, &tmp_pkt, 1);
return 0;
}
static int64_t filesize(AVIOContext *pb)
{
int64_t ret = -1;
if (pb) {
ret = avio_size(pb);
if (ret <= 0) // FIXME improve avio_size() so it works with non seekable output too
ret = avio_tell(pb);
}
return ret;
}
static int write_packet(OutputFile *of, OutputStream *ost, AVPacket *pkt)
{
MuxStream *ms = &of->mux->streams[ost->index];
AVFormatContext *s = of->mux->fc;
AVStream *st = ost->st;
int64_t fs;
int ret;
fs = filesize(s->pb);
atomic_store(&of->mux->last_filesize, fs);
if (fs >= of->mux->limit_filesize)
return AVERROR_EOF;
if ((st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO && ost->vsync_method == VSYNC_DROP) ||
(st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && audio_sync_method < 0))
pkt->pts = pkt->dts = AV_NOPTS_VALUE;
if (st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO) {
if (ost->frame_rate.num && ost->is_cfr) {
if (pkt->duration > 0)
av_log(NULL, AV_LOG_WARNING, "Overriding packet duration by frame rate, this should not happen\n");
pkt->duration = av_rescale_q(1, av_inv_q(ost->frame_rate),
ost->mux_timebase);
}
}
av_packet_rescale_ts(pkt, ost->mux_timebase, ost->st->time_base);
if (!(s->oformat->flags & AVFMT_NOTIMESTAMPS)) {
if (pkt->dts != AV_NOPTS_VALUE &&
pkt->pts != AV_NOPTS_VALUE &&
pkt->dts > pkt->pts) {
av_log(s, AV_LOG_WARNING, "Invalid DTS: %"PRId64" PTS: %"PRId64" in output stream %d:%d, replacing by guess\n",
pkt->dts, pkt->pts,
ost->file_index, ost->st->index);
pkt->pts =
pkt->dts = pkt->pts + pkt->dts + ms->last_mux_dts + 1
- FFMIN3(pkt->pts, pkt->dts, ms->last_mux_dts + 1)
- FFMAX3(pkt->pts, pkt->dts, ms->last_mux_dts + 1);
}
if ((st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO || st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO || st->codecpar->codec_type == AVMEDIA_TYPE_SUBTITLE) &&
pkt->dts != AV_NOPTS_VALUE &&
ms->last_mux_dts != AV_NOPTS_VALUE) {
int64_t max = ms->last_mux_dts + !(s->oformat->flags & AVFMT_TS_NONSTRICT);
if (pkt->dts < max) {
int loglevel = max - pkt->dts > 2 || st->codecpar->codec_type == AVMEDIA_TYPE_VIDEO ? AV_LOG_WARNING : AV_LOG_DEBUG;
if (exit_on_error)
loglevel = AV_LOG_ERROR;
av_log(s, loglevel, "Non-monotonous DTS in output stream "
"%d:%d; previous: %"PRId64", current: %"PRId64"; ",
ost->file_index, ost->st->index, ms->last_mux_dts, pkt->dts);
if (exit_on_error)
return AVERROR(EINVAL);
av_log(s, loglevel, "changing to %"PRId64". This may result "
"in incorrect timestamps in the output file.\n",
max);
if (pkt->pts >= pkt->dts)
pkt->pts = FFMAX(pkt->pts, max);
pkt->dts = max;
}
}
}
ms->last_mux_dts = pkt->dts;
ost->data_size += pkt->size;
atomic_fetch_add(&ost->packets_written, 1);
pkt->stream_index = ost->index;
if (debug_ts) {
av_log(NULL, AV_LOG_INFO, "muxer <- type:%s "
"pkt_pts:%s pkt_pts_time:%s pkt_dts:%s pkt_dts_time:%s duration:%s duration_time:%s size:%d\n",
av_get_media_type_string(ost->enc_ctx->codec_type),
av_ts2str(pkt->pts), av_ts2timestr(pkt->pts, &ost->st->time_base),
av_ts2str(pkt->dts), av_ts2timestr(pkt->dts, &ost->st->time_base),
av_ts2str(pkt->duration), av_ts2timestr(pkt->duration, &ost->st->time_base),
pkt->size
);
}
ret = av_interleaved_write_frame(s, pkt);
if (ret < 0) {
print_error("av_interleaved_write_frame()", ret);
return ret;
}
return 0;
}
static int sync_queue_process(OutputFile *of, OutputStream *ost, AVPacket *pkt)
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
{
if (ost->sq_idx_mux >= 0) {
int ret = sq_send(of->sq_mux, ost->sq_idx_mux, SQPKT(pkt));
if (ret < 0)
return ret;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
while (1) {
ret = sq_receive(of->sq_mux, -1, SQPKT(of->mux->sq_pkt));
if (ret < 0)
return (ret == AVERROR_EOF || ret == AVERROR(EAGAIN)) ? 0 : ret;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
ret = write_packet(of, output_streams[of->ost_index + ret],
of->mux->sq_pkt);
if (ret < 0)
return ret;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
}
} else if (pkt)
return write_packet(of, ost, pkt);
return 0;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
}
static void *muxer_thread(void *arg)
{
OutputFile *of = arg;
Muxer *mux = of->mux;
AVPacket *pkt = NULL;
int ret = 0;
pkt = av_packet_alloc();
if (!pkt) {
ret = AVERROR(ENOMEM);
goto finish;
}
while (1) {
OutputStream *ost;
int stream_idx;
ret = tq_receive(mux->tq, &stream_idx, pkt);
if (stream_idx < 0) {
av_log(NULL, AV_LOG_VERBOSE,
"All streams finished for output file #%d\n", of->index);
ret = 0;
break;
}
ost = output_streams[of->ost_index + stream_idx];
ret = sync_queue_process(of, ost, ret < 0 ? NULL : pkt);
av_packet_unref(pkt);
if (ret == AVERROR_EOF)
tq_receive_finish(mux->tq, stream_idx);
else if (ret < 0) {
av_log(NULL, AV_LOG_ERROR,
"Error muxing a packet for output file #%d\n", of->index);
break;
}
}
finish:
av_packet_free(&pkt);
for (unsigned int i = 0; i < mux->fc->nb_streams; i++)
tq_receive_finish(mux->tq, i);
av_log(NULL, AV_LOG_VERBOSE, "Terminating muxer thread %d\n", of->index);
return (void*)(intptr_t)ret;
}
static int print_sdp(void)
{
char sdp[16384];
int i;
int j, ret;
AVIOContext *sdp_pb;
AVFormatContext **avc;
for (i = 0; i < nb_output_files; i++) {
if (!output_files[i]->mux->header_written)
return 0;
}
avc = av_malloc_array(nb_output_files, sizeof(*avc));
if (!avc)
return AVERROR(ENOMEM);
for (i = 0, j = 0; i < nb_output_files; i++) {
if (!strcmp(output_files[i]->format->name, "rtp")) {
avc[j] = output_files[i]->mux->fc;
j++;
}
}
if (!j) {
av_log(NULL, AV_LOG_ERROR, "No output streams in the SDP.\n");
ret = AVERROR(EINVAL);
goto fail;
}
ret = av_sdp_create(avc, j, sdp, sizeof(sdp));
if (ret < 0)
goto fail;
if (!sdp_filename) {
printf("SDP:\n%s\n", sdp);
fflush(stdout);
} else {
ret = avio_open2(&sdp_pb, sdp_filename, AVIO_FLAG_WRITE, &int_cb, NULL);
if (ret < 0) {
av_log(NULL, AV_LOG_ERROR, "Failed to open sdp file '%s'\n", sdp_filename);
goto fail;
}
avio_print(sdp_pb, sdp);
avio_closep(&sdp_pb);
av_freep(&sdp_filename);
}
// SDP successfully written, allow muxer threads to start
ret = 1;
fail:
av_freep(&avc);
return ret;
}
static int submit_packet(OutputFile *of, OutputStream *ost, AVPacket *pkt)
{
Muxer *mux = of->mux;
int ret = 0;
if (!pkt || ost->finished & MUXER_FINISHED)
goto finish;
ret = tq_send(mux->tq, ost->index, pkt);
if (ret < 0)
goto finish;
return 0;
finish:
if (pkt)
av_packet_unref(pkt);
ost->finished |= MUXER_FINISHED;
tq_send_finish(mux->tq, ost->index);
return ret == AVERROR_EOF ? 0 : ret;
}
int of_submit_packet(OutputFile *of, AVPacket *pkt, OutputStream *ost)
{
int ret;
if (of->mux->tq) {
return submit_packet(of, ost, pkt);
} else {
/* the muxer is not initialized yet, buffer the packet */
ret = queue_packet(of, ost, pkt);
if (ret < 0) {
av_packet_unref(pkt);
return ret;
}
}
return 0;
}
static int thread_stop(OutputFile *of)
{
Muxer *mux = of->mux;
void *ret;
if (!mux || !mux->tq)
return 0;
for (unsigned int i = 0; i < mux->fc->nb_streams; i++)
tq_send_finish(mux->tq, i);
pthread_join(mux->thread, &ret);
tq_free(&mux->tq);
return (int)(intptr_t)ret;
}
static void pkt_move(void *dst, void *src)
{
av_packet_move_ref(dst, src);
}
static int thread_start(OutputFile *of)
{
Muxer *mux = of->mux;
AVFormatContext *fc = mux->fc;
ObjPool *op;
int ret;
op = objpool_alloc_packets();
if (!op)
return AVERROR(ENOMEM);
mux->tq = tq_alloc(fc->nb_streams, mux->thread_queue_size, op, pkt_move);
if (!mux->tq) {
objpool_free(&op);
return AVERROR(ENOMEM);
}
ret = pthread_create(&mux->thread, NULL, muxer_thread, (void*)of);
if (ret) {
tq_free(&mux->tq);
return AVERROR(ret);
}
/* flush the muxing queues */
for (int i = 0; i < fc->nb_streams; i++) {
MuxStream *ms = &of->mux->streams[i];
OutputStream *ost = output_streams[of->ost_index + i];
AVPacket *pkt;
/* try to improve muxing time_base (only possible if nothing has been written yet) */
if (!av_fifo_can_read(ms->muxing_queue))
ost->mux_timebase = ost->st->time_base;
while (av_fifo_read(ms->muxing_queue, &pkt, 1) >= 0) {
ret = submit_packet(of, ost, pkt);
if (pkt) {
ms->muxing_queue_data_size -= pkt->size;
av_packet_free(&pkt);
}
if (ret < 0)
return ret;
}
}
return 0;
}
/* open the muxer when all the streams are initialized */
int of_check_init(OutputFile *of)
{
AVFormatContext *fc = of->mux->fc;
int ret, i;
for (i = 0; i < fc->nb_streams; i++) {
OutputStream *ost = output_streams[of->ost_index + i];
if (!ost->initialized)
return 0;
}
ret = avformat_write_header(fc, &of->mux->opts);
if (ret < 0) {
av_log(NULL, AV_LOG_ERROR,
"Could not write header for output file #%d "
"(incorrect codec parameters ?): %s\n",
of->index, av_err2str(ret));
return ret;
}
//assert_avoptions(of->opts);
of->mux->header_written = 1;
av_dump_format(fc, of->index, fc->url, 1);
nb_output_dumped++;
if (sdp_filename || want_sdp) {
ret = print_sdp();
if (ret < 0) {
av_log(NULL, AV_LOG_ERROR, "Error writing the SDP.\n");
return ret;
} else if (ret == 1) {
/* SDP is written only after all the muxers are ready, so now we
* start ALL the threads */
for (i = 0; i < nb_output_files; i++) {
ret = thread_start(output_files[i]);
if (ret < 0)
return ret;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
}
}
} else {
ret = thread_start(of);
if (ret < 0)
return ret;
}
return 0;
}
int of_write_trailer(OutputFile *of)
{
AVFormatContext *fc = of->mux->fc;
int ret;
if (!of->mux->tq) {
av_log(NULL, AV_LOG_ERROR,
"Nothing was written into output file %d (%s), because "
"at least one of its streams received no packets.\n",
of->index, fc->url);
return AVERROR(EINVAL);
}
ret = thread_stop(of);
if (ret < 0)
main_return_code = ret;
ret = av_write_trailer(fc);
if (ret < 0) {
av_log(NULL, AV_LOG_ERROR, "Error writing trailer of %s: %s\n", fc->url, av_err2str(ret));
return ret;
}
of->mux->last_filesize = filesize(fc->pb);
if (!(of->format->flags & AVFMT_NOFILE)) {
ret = avio_closep(&fc->pb);
if (ret < 0) {
av_log(NULL, AV_LOG_ERROR, "Error closing file %s: %s\n",
fc->url, av_err2str(ret));
return ret;
}
}
return 0;
}
static void fc_close(AVFormatContext **pfc)
{
AVFormatContext *fc = *pfc;
if (!fc)
return;
if (!(fc->oformat->flags & AVFMT_NOFILE))
avio_closep(&fc->pb);
avformat_free_context(fc);
*pfc = NULL;
}
static void mux_free(Muxer **pmux)
{
Muxer *mux = *pmux;
if (!mux)
return;
for (int i = 0; i < mux->fc->nb_streams; i++) {
MuxStream *ms = &mux->streams[i];
AVPacket *pkt;
if (!ms->muxing_queue)
continue;
while (av_fifo_read(ms->muxing_queue, &pkt, 1) >= 0)
av_packet_free(&pkt);
av_fifo_freep2(&ms->muxing_queue);
}
av_freep(&mux->streams);
av_dict_free(&mux->opts);
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
av_packet_free(&mux->sq_pkt);
fc_close(&mux->fc);
av_freep(pmux);
}
void of_close(OutputFile **pof)
{
OutputFile *of = *pof;
if (!of)
return;
thread_stop(of);
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
sq_free(&of->sq_encode);
sq_free(&of->sq_mux);
mux_free(&of->mux);
av_freep(pof);
}
int of_muxer_init(OutputFile *of, AVFormatContext *fc,
AVDictionary *opts, int64_t limit_filesize,
int thread_queue_size)
{
Muxer *mux = av_mallocz(sizeof(*mux));
int ret = 0;
if (!mux) {
fc_close(&fc);
return AVERROR(ENOMEM);
}
mux->streams = av_calloc(fc->nb_streams, sizeof(*mux->streams));
if (!mux->streams) {
fc_close(&fc);
av_freep(&mux);
return AVERROR(ENOMEM);
}
of->mux = mux;
mux->fc = fc;
for (int i = 0; i < fc->nb_streams; i++) {
MuxStream *ms = &mux->streams[i];
ms->muxing_queue = av_fifo_alloc2(8, sizeof(AVPacket*), 0);
if (!ms->muxing_queue) {
ret = AVERROR(ENOMEM);
goto fail;
}
ms->last_mux_dts = AV_NOPTS_VALUE;
}
mux->thread_queue_size = thread_queue_size > 0 ? thread_queue_size : 8;
mux->limit_filesize = limit_filesize;
mux->opts = opts;
if (strcmp(of->format->name, "rtp"))
want_sdp = 0;
fftools/ffmpeg: rework -shortest implementation The -shortest option (which finishes the output file at the time the shortest stream ends) is currently implemented by faking the -t option when an output stream ends. This approach is fragile, since it depends on the frames/packets being processed in a specific order. E.g. there are currently some situations in which the output file length will depend unpredictably on unrelated factors like encoder delay. More importantly, the present work aiming at splitting various ffmpeg components into different threads will make this approach completely unworkable, since the frames/packets will arrive in effectively random order. This commit introduces a "sync queue", which is essentially a collection of FIFOs, one per stream. Frames/packets are submitted to these FIFOs and are then released for further processing (encoding or muxing) when it is ensured that the frame in question will not cause its stream to get ahead of the other streams (the logic is similar to libavformat's interleaving queue). These sync queues are then used for encoding and/or muxing when the -shortest option is specified. A new option – -shortest_buf_duration – controls the maximum number of queued packets, to avoid runaway memory usage. This commit changes the results of the following tests: - copy-shortest[12]: the last audio frame is now gone. This is correct, since it actually outlasts the last video frame. - shortest-sub: the video packets following the last subtitle packet are now gone. This is also correct.
2022-06-10 14:21:42 +02:00
if (of->sq_mux) {
mux->sq_pkt = av_packet_alloc();
if (!mux->sq_pkt) {
ret = AVERROR(ENOMEM);
goto fail;
}
}
/* write the header for files with no streams */
if (of->format->flags & AVFMT_NOSTREAMS && fc->nb_streams == 0) {
ret = of_check_init(of);
if (ret < 0)
goto fail;
}
fail:
if (ret < 0)
mux_free(&of->mux);
return ret;
}
int64_t of_filesize(OutputFile *of)
{
return atomic_load(&of->mux->last_filesize);
}
AVChapter * const *
of_get_chapters(OutputFile *of, unsigned int *nb_chapters)
{
*nb_chapters = of->mux->fc->nb_chapters;
return of->mux->fc->chapters;
}