X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=a19f26ad79bdfde7dc66e3d795d87254e5796d43;hb=7e25d487cba3c23d847aedee84a8e5143f4a209c;hp=075420b7ca91f5c5537959ba9d9c2ac18eefeed3;hpb=6bb8df551b23d23f67a9391a37f6170814662b42;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 075420b7c..a19f26ad7 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -59,7 +59,7 @@ using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, Job* j) +FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, Job* j) : Decoder (f, o, j) , VideoDecoder (f, o, j) , AudioDecoder (f, o, j) @@ -77,6 +77,10 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, J setup_video (); setup_audio (); setup_subtitle (); + + if (!o->video_sync) { + _first_video = 0; + } } FFmpegDecoder::~FFmpegDecoder () @@ -101,11 +105,9 @@ FFmpegDecoder::~FFmpegDecoder () void FFmpegDecoder::setup_general () { - int r; - av_register_all (); - if ((r = avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0)) != 0) { + if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) { throw OpenFileError (_film->content_path ()); } @@ -120,11 +122,21 @@ FFmpegDecoder::setup_general () if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) { _video_stream = i; } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) { + + /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up, + so bodge it here. No idea why we should have to do this. + */ + + if (s->codec->channel_layout == 0) { + s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels); + } + _audio_streams.push_back ( shared_ptr ( new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout) ) ); + } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) { _subtitle_streams.push_back ( shared_ptr ( @@ -154,14 +166,7 @@ FFmpegDecoder::setup_video () throw DecodeError ("could not find video decoder"); } - /* I think this prevents problems with green hash on decodes and - "changing frame properties on the fly is not supported by all filters" - messages with some content. Although I'm not sure; needs checking. - */ - AVDictionary* opts = 0; - av_dict_set (&opts, "threads", "1", 0); - - if (avcodec_open2 (_video_codec_context, _video_codec, &opts) < 0) { + if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) { throw DecodeError ("could not open video decoder"); } } @@ -186,14 +191,6 @@ FFmpegDecoder::setup_audio () if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) { throw DecodeError ("could not open audio decoder"); } - - /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up, - so bodge it here. No idea why we should have to do this. - */ - - if (_audio_codec_context->channel_layout == 0) { - _audio_codec_context->channel_layout = av_get_default_channel_layout (ffa->channels()); - } } void @@ -259,7 +256,7 @@ FFmpegDecoder::pass () avcodec_get_frame_defaults (_frame); shared_ptr ffa = dynamic_pointer_cast (_audio_stream); - + if (_packet.stream_index == _video_stream) { int frame_finished; @@ -270,41 +267,10 @@ FFmpegDecoder::pass () _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size)); } - /* Where we are in the output, in seconds */ - double const out_pts_seconds = video_frame() / frames_per_second(); - - /* Where we are in the source, in seconds */ - double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base) - * av_frame_get_best_effort_timestamp(_frame); - - if (!_first_video) { - _first_video = source_pts_seconds; - } - - /* Difference between where we are and where we should be */ - double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds; - double const one_frame = 1 / frames_per_second(); - - /* Insert frames if required to get out_pts_seconds up to pts_seconds */ - if (delta > one_frame) { - int const extra = rint (delta / one_frame); - for (int i = 0; i < extra; ++i) { - repeat_last_video (); - _film->log()->log ( - String::compose ( - "Extra frame inserted at %1s; source frame %2, source PTS %3", - out_pts_seconds, video_frame(), source_pts_seconds - ) - ); - } - } - - if (delta > -one_frame) { - /* Process this frame */ - filter_and_emit_video (_frame); + if (_opt->video_sync) { + out_with_sync (); } else { - /* Otherwise we are omitting a frame to keep things right */ - _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds)); + filter_and_emit_video (_frame); } } @@ -336,7 +302,7 @@ FFmpegDecoder::pass () _film->log()->log ( String::compose ( - "First video at %1, first audio at %2, pushing %3 frames of silence for %4 channels (%5 bytes per sample)", + "First video at %1, first audio at %2, pushing %3 audio frames of silence for %4 channels (%5 bytes per sample)", _first_video.get(), _first_audio.get(), s, ffa->channels(), bytes_per_audio_sample() ) ); @@ -368,7 +334,7 @@ FFmpegDecoder::pass () if (sub.num_rects > 0) { shared_ptr ts; try { - emit_subtitle (shared_ptr (new TimedSubtitle (sub, _first_video.get()))); + emit_subtitle (shared_ptr (new TimedSubtitle (sub))); } catch (...) { /* some problem with the subtitle; we probably didn't understand it */ } @@ -403,7 +369,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) switch (audio_sample_format()) { case AV_SAMPLE_FMT_S16: { - int16_t* p = (int16_t *) data; + int16_t* p = reinterpret_cast (data); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { @@ -418,13 +384,24 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) } break; + case AV_SAMPLE_FMT_S16P: + { + int16_t* p = reinterpret_cast (data); + for (int i = 0; i < _film->audio_channels(); ++i) { + for (int j = 0; j < frames; ++j) { + audio->data(i)[j] = static_cast(*p++) / (1 << 15); + } + } + } + break; + case AV_SAMPLE_FMT_S32: { - int32_t* p = (int32_t *) data; + int32_t* p = reinterpret_cast (data); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { - audio->data(channel)[sample] = float(*p++) / (1 << 31); + audio->data(channel)[sample] = static_cast(*p++) / (1 << 31); ++channel; if (channel == _film->audio_channels()) { @@ -433,7 +410,25 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) } } } + break; + case AV_SAMPLE_FMT_FLT: + { + float* p = reinterpret_cast (data); + int sample = 0; + int channel = 0; + for (int i = 0; i < total_samples; ++i) { + audio->data(channel)[sample] = *p++; + + ++channel; + if (channel == _film->audio_channels()) { + channel = 0; + ++sample; + } + } + } + break; + case AV_SAMPLE_FMT_FLTP: { float* p = reinterpret_cast (data); @@ -445,7 +440,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) break; default: - assert (false); + throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast (audio_sample_format()))); } return audio; @@ -552,11 +547,14 @@ FFmpegDecoder::set_subtitle_stream (shared_ptr s) { VideoDecoder::set_subtitle_stream (s); setup_subtitle (); + OutputChanged (); } void FFmpegDecoder::filter_and_emit_video (AVFrame* frame) { + boost::mutex::scoped_lock lm (_filter_graphs_mutex); + shared_ptr graph; list >::iterator i = _filter_graphs.begin(); @@ -565,7 +563,7 @@ FFmpegDecoder::filter_and_emit_video (AVFrame* frame) } if (i == _filter_graphs.end ()) { - graph.reset (new FilterGraph (_film, this, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format)); + graph.reset (new FilterGraph (_film, this, Size (frame->width, frame->height), (AVPixelFormat) frame->format)); _filter_graphs.push_back (graph); _film->log()->log (String::compose ("New graph for %1x%2, pixel format %3", frame->width, frame->height, frame->format)); } else { @@ -575,10 +573,41 @@ FFmpegDecoder::filter_and_emit_video (AVFrame* frame) list > images = graph->process (frame); for (list >::iterator i = images.begin(); i != images.end(); ++i) { - emit_video (*i); + emit_video (*i, frame_time ()); } } +bool +FFmpegDecoder::seek (double p) +{ + return do_seek (p, false); +} + +bool +FFmpegDecoder::seek_to_last () +{ + /* This AVSEEK_FLAG_BACKWARD in do_seek is a bit of a hack; without it, if we ask for a seek to the same place as last time + (used when we change decoder parameters and want to re-fetch the frame) we end up going forwards rather than + staying in the same place. + */ + return do_seek (last_source_time(), true); +} + +bool +FFmpegDecoder::do_seek (double p, bool backwards) +{ + int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base); + + int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0); + + avcodec_flush_buffers (_video_codec_context); + if (_subtitle_codec_context) { + avcodec_flush_buffers (_subtitle_codec_context); + } + + return r < 0; +} + shared_ptr FFmpegAudioStream::create (string t, optional v) { @@ -631,3 +660,80 @@ FFmpegAudioStream::to_string () const return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name); } +void +FFmpegDecoder::out_with_sync () +{ + /* Where we are in the output, in seconds */ + double const out_pts_seconds = video_frame() / frames_per_second(); + + /* Where we are in the source, in seconds */ + double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base) + * av_frame_get_best_effort_timestamp(_frame); + + _film->log()->log ( + String::compose ("Source video frame ready; source at %1, output at %2", source_pts_seconds, out_pts_seconds), + Log::VERBOSE + ); + + if (!_first_video) { + _first_video = source_pts_seconds; + } + + /* Difference between where we are and where we should be */ + double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds; + double const one_frame = 1 / frames_per_second(); + + /* Insert frames if required to get out_pts_seconds up to pts_seconds */ + if (delta > one_frame) { + int const extra = rint (delta / one_frame); + for (int i = 0; i < extra; ++i) { + repeat_last_video (); + _film->log()->log ( + String::compose ( + "Extra video frame inserted at %1s; source frame %2, source PTS %3 (at %4 fps)", + out_pts_seconds, video_frame(), source_pts_seconds, frames_per_second() + ) + ); + } + } + + if (delta > -one_frame) { + /* Process this frame */ + filter_and_emit_video (_frame); + } else { + /* Otherwise we are omitting a frame to keep things right */ + _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds)); + } +} + +void +FFmpegDecoder::film_changed (Film::Property p) +{ + switch (p) { + case Film::CROP: + case Film::FILTERS: + { + boost::mutex::scoped_lock lm (_filter_graphs_mutex); + _filter_graphs.clear (); + } + OutputChanged (); + break; + + default: + break; + } +} + +/** @return Length (in video frames) according to our content's header */ +SourceFrame +FFmpegDecoder::length () const +{ + return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second(); +} + +double +FFmpegDecoder::frame_time () const +{ + return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base); +} +