}
int audio_stream_index = find_stream_index(format_ctx.get(), AVMEDIA_TYPE_AUDIO);
+ int subtitle_stream_index = find_stream_index(format_ctx.get(), AVMEDIA_TYPE_SUBTITLE);
+ has_last_subtitle = false;
// Open video decoder.
const AVCodecParameters *video_codecpar = format_ctx->streams[video_stream_index]->codecpar;
if (process_queued_commands(format_ctx.get(), pathname, last_modified, /*rewound=*/nullptr)) {
return true;
}
+ if (should_interrupt.load()) {
+ // Check as a failsafe, so that we don't need to rely on avio if we don't have to.
+ return false;
+ }
UniqueFrame audio_frame = audio_frame_allocator->alloc_frame();
AudioFormat audio_format;
int64_t audio_pts;
bool error;
AVFrameWithDeleter frame = decode_frame(format_ctx.get(), video_codec_ctx.get(), audio_codec_ctx.get(),
- pathname, video_stream_index, audio_stream_index, audio_frame.get(), &audio_format, &audio_pts, &error);
+ pathname, video_stream_index, audio_stream_index, subtitle_stream_index, audio_frame.get(), &audio_format, &audio_pts, &error);
if (error) {
return false;
}
if (frame == nullptr) {
// EOF. Loop back to the start if we can.
+ if (format_ctx->pb != nullptr && format_ctx->pb->seekable == 0) {
+ // Not seekable (but seemingly, sometimes av_seek_frame() would return 0 anyway,
+ // so don't try).
+ return true;
+ }
if (av_seek_frame(format_ctx.get(), /*stream_index=*/-1, /*timestamp=*/0, /*flags=*/0) < 0) {
fprintf(stderr, "%s: Rewind failed, not looping.\n", pathname.c_str());
return true;
if (last_pts == 0 && pts_origin == 0) {
pts_origin = frame->pts;
}
- next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
- if (first_frame && last_frame_was_connected) {
- // If reconnect took more than one second, this is probably a live feed,
- // and we should reset the resampler. (Or the rate is really, really low,
- // in which case a reset on the first frame is fine anyway.)
- if (duration<double>(next_frame_start - last_frame).count() >= 1.0) {
- last_frame_was_connected = false;
+ steady_clock::time_point now = steady_clock::now();
+ if (play_as_fast_as_possible) {
+ video_frame->received_timestamp = now;
+ audio_frame->received_timestamp = now;
+ next_frame_start = now;
+ } else {
+ next_frame_start = compute_frame_start(frame->pts, pts_origin, video_timebase, start, rate);
+ if (first_frame && last_frame_was_connected) {
+ // If reconnect took more than one second, this is probably a live feed,
+ // and we should reset the resampler. (Or the rate is really, really low,
+ // in which case a reset on the first frame is fine anyway.)
+ if (duration<double>(next_frame_start - last_frame).count() >= 1.0) {
+ last_frame_was_connected = false;
+ }
+ }
+ video_frame->received_timestamp = next_frame_start;
+
+ // The easiest way to get all the rate conversions etc. right is to move the
+ // audio PTS into the video PTS timebase and go from there. (We'll get some
+ // rounding issues, but they should not be a big problem.)
+ int64_t audio_pts_as_video_pts = av_rescale_q(audio_pts, audio_timebase, video_timebase);
+ audio_frame->received_timestamp = compute_frame_start(audio_pts_as_video_pts, pts_origin, video_timebase, start, rate);
+
+ if (audio_frame->len != 0) {
+ // The received timestamps in Nageru are measured after we've just received the frame.
+ // However, pts (especially audio pts) is at the _beginning_ of the frame.
+ // If we have locked audio, the distinction doesn't really matter, as pts is
+ // on a relative scale and a fixed offset is fine. But if we don't, we will have
+ // a different number of samples each time, which will cause huge audio jitter
+ // and throw off the resampler.
+ //
+ // In a sense, we should have compensated by adding the frame and audio lengths
+ // to video_frame->received_timestamp and audio_frame->received_timestamp respectively,
+ // but that would mean extra waiting in sleep_until(). All we need is that they
+ // are correct relative to each other, though (and to the other frames we send),
+ // so just align the end of the audio frame, and we're fine.
+ size_t num_samples = (audio_frame->len * 8) / audio_format.bits_per_sample / audio_format.num_channels;
+ double offset = double(num_samples) / OUTPUT_FREQUENCY -
+ double(video_format.frame_rate_den) / video_format.frame_rate_nom;
+ audio_frame->received_timestamp += duration_cast<steady_clock::duration>(duration<double>(offset));
}
- }
- video_frame->received_timestamp = next_frame_start;
-
- // The easiest way to get all the rate conversions etc. right is to move the
- // audio PTS into the video PTS timebase and go from there. (We'll get some
- // rounding issues, but they should not be a big problem.)
- int64_t audio_pts_as_video_pts = av_rescale_q(audio_pts, audio_timebase, video_timebase);
- audio_frame->received_timestamp = compute_frame_start(audio_pts_as_video_pts, pts_origin, video_timebase, start, rate);
-
- if (audio_frame->len != 0) {
- // The received timestamps in Nageru are measured after we've just received the frame.
- // However, pts (especially audio pts) is at the _beginning_ of the frame.
- // If we have locked audio, the distinction doesn't really matter, as pts is
- // on a relative scale and a fixed offset is fine. But if we don't, we will have
- // a different number of samples each time, which will cause huge audio jitter
- // and throw off the resampler.
- //
- // In a sense, we should have compensated by adding the frame and audio lengths
- // to video_frame->received_timestamp and audio_frame->received_timestamp respectively,
- // but that would mean extra waiting in sleep_until(). All we need is that they
- // are correct relative to each other, though (and to the other frames we send),
- // so just align the end of the audio frame, and we're fine.
- size_t num_samples = (audio_frame->len * 8) / audio_format.bits_per_sample / audio_format.num_channels;
- double offset = double(num_samples) / OUTPUT_FREQUENCY -
- double(video_format.frame_rate_den) / video_format.frame_rate_nom;
- audio_frame->received_timestamp += duration_cast<steady_clock::duration>(duration<double>(offset));
- }
- steady_clock::time_point now = steady_clock::now();
- if (duration<double>(now - next_frame_start).count() >= 0.1) {
- // If we don't have enough CPU to keep up, or if we have a live stream
- // where the initial origin was somehow wrong, we could be behind indefinitely.
- // In particular, this will give the audio resampler problems as it tries
- // to speed up to reduce the delay, hitting the low end of the buffer every time.
- fprintf(stderr, "%s: Playback %.0f ms behind, resetting time scale\n",
- pathname.c_str(),
- 1e3 * duration<double>(now - next_frame_start).count());
- pts_origin = frame->pts;
- start = next_frame_start = now;
- timecode += MAX_FPS * 2 + 1;
+ if (duration<double>(now - next_frame_start).count() >= 0.1) {
+ // If we don't have enough CPU to keep up, or if we have a live stream
+ // where the initial origin was somehow wrong, we could be behind indefinitely.
+ // In particular, this will give the audio resampler problems as it tries
+ // to speed up to reduce the delay, hitting the low end of the buffer every time.
+ fprintf(stderr, "%s: Playback %.0f ms behind, resetting time scale\n",
+ pathname.c_str(),
+ 1e3 * duration<double>(now - next_frame_start).count());
+ pts_origin = frame->pts;
+ start = next_frame_start = now;
+ timecode += MAX_FPS * 2 + 1;
+ }
+ }
+ bool finished_wakeup;
+ if (play_as_fast_as_possible) {
+ finished_wakeup = !producer_thread_should_quit.should_quit();
+ } else {
+ finished_wakeup = producer_thread_should_quit.sleep_until(next_frame_start);
}
- bool finished_wakeup = producer_thread_should_quit.sleep_until(next_frame_start);
if (finished_wakeup) {
if (audio_frame->len > 0) {
assert(audio_pts != -1);
start = compute_frame_start(last_pts, pts_origin, video_timebase, start, rate);
pts_origin = last_pts;
rate = cmd.new_rate;
+ play_as_fast_as_possible = (rate >= 10.0);
break;
}
}
} // namespace
AVFrameWithDeleter FFmpegCapture::decode_frame(AVFormatContext *format_ctx, AVCodecContext *video_codec_ctx, AVCodecContext *audio_codec_ctx,
- const std::string &pathname, int video_stream_index, int audio_stream_index,
+ const std::string &pathname, int video_stream_index, int audio_stream_index, int subtitle_stream_index,
FrameAllocator::Frame *audio_frame, AudioFormat *audio_format, int64_t *audio_pts, bool *error)
{
*error = false;
*error = true;
return AVFrameWithDeleter(nullptr);
}
+ } else if (pkt.stream_index == subtitle_stream_index) {
+ last_subtitle = string(reinterpret_cast<const char *>(pkt.data), pkt.size);
+ has_last_subtitle = true;
}
} else {
eof = true; // Or error, but ignore that for the time being.
projects / nageru / blobdiff