X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=81f40564460a035856332ae46f25cacf14d1a9dd;hb=32fc1ddb0ee004d18c36155ddcf4d9b3998a7061;hp=e85439f6ed1adf190eea24435b5a749c39e7715b;hpb=068bb5b95652c022419c64f3d29cc42c4610ae93;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index e85439f6e..81f405644 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -41,24 +41,31 @@ extern "C" { #include "transcoder.h" #include "job.h" #include "filter.h" -#include "film_state.h" #include "options.h" #include "exceptions.h" #include "image.h" #include "util.h" #include "log.h" #include "ffmpeg_decoder.h" +#include "filter_graph.h" #include "subtitle.h" -using namespace std; -using namespace boost; - -FFmpegDecoder::FFmpegDecoder (boost::shared_ptr s, boost::shared_ptr o, Job* j, Log* l, bool minimal, bool ignore_length) - : Decoder (s, o, j, l, minimal, ignore_length) +using std::cout; +using std::string; +using std::vector; +using std::stringstream; +using std::list; +using boost::shared_ptr; +using boost::optional; +using boost::dynamic_pointer_cast; +using libdcp::Size; + +FFmpegDecoder::FFmpegDecoder (shared_ptr f, DecodeOptions o, Job* j) + : Decoder (f, o, j) + , VideoDecoder (f, o, j) + , AudioDecoder (f, o, j) , _format_context (0) , _video_stream (-1) - , _audio_stream (-1) - , _subtitle_stream (-1) , _frame (0) , _video_codec_context (0) , _video_codec (0) @@ -71,6 +78,10 @@ FFmpegDecoder::FFmpegDecoder (boost::shared_ptr s, boost::share setup_video (); setup_audio (); setup_subtitle (); + + if (!o.video_sync) { + _first_video = 0; + } } FFmpegDecoder::~FFmpegDecoder () @@ -86,33 +97,53 @@ FFmpegDecoder::~FFmpegDecoder () if (_subtitle_codec_context) { avcodec_close (_subtitle_codec_context); } - + av_free (_frame); + avformat_close_input (&_format_context); } void FFmpegDecoder::setup_general () { - int r; - av_register_all (); - if ((r = avformat_open_input (&_format_context, _fs->content_path().c_str(), 0, 0)) != 0) { - throw OpenFileError (_fs->content_path ()); + if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) { + throw OpenFileError (_film->content_path ()); } if (avformat_find_stream_info (_format_context, 0) < 0) { throw DecodeError ("could not find stream information"); } + /* Find video, audio and subtitle streams and choose the first of each */ + for (uint32_t i = 0; i < _format_context->nb_streams; ++i) { - if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_VIDEO) { + AVStream* s = _format_context->streams[i]; + if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) { _video_stream = i; - } else if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO) { - _audio_stream = i; - } else if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) { - _subtitle_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 ( + new SubtitleStream (stream_name (s), i) + ) + ); } } @@ -135,7 +166,7 @@ FFmpegDecoder::setup_video () if (_video_codec == 0) { throw DecodeError ("could not find video decoder"); } - + if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) { throw DecodeError ("could not open video decoder"); } @@ -144,11 +175,14 @@ FFmpegDecoder::setup_video () void FFmpegDecoder::setup_audio () { - if (_audio_stream < 0) { + if (!_audio_stream) { return; } + + shared_ptr ffa = dynamic_pointer_cast (_audio_stream); + assert (ffa); - _audio_codec_context = _format_context->streams[_audio_stream]->codec; + _audio_codec_context = _format_context->streams[ffa->id()]->codec; _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id); if (_audio_codec == 0) { @@ -158,24 +192,16 @@ 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 (audio_channels ()); - } } void FFmpegDecoder::setup_subtitle () { - if (_subtitle_stream < 0) { + if (!_subtitle_stream) { return; } - _subtitle_codec_context = _format_context->streams[_subtitle_stream]->codec; + _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec; _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); if (_subtitle_codec == 0) { @@ -189,12 +215,16 @@ FFmpegDecoder::setup_subtitle () bool -FFmpegDecoder::do_pass () +FFmpegDecoder::pass () { int r = av_read_frame (_format_context, &_packet); + if (r < 0) { if (r != AVERROR_EOF) { - throw DecodeError ("error on av_read_frame"); + /* Maybe we should fail here, but for now we'll just finish off instead */ + char buf[256]; + av_strerror (r, buf, sizeof(buf)); + _film->log()->log (String::compose ("error on av_read_frame (%1) (%2)", buf, r)); } /* Get any remaining frames */ @@ -202,53 +232,116 @@ FFmpegDecoder::do_pass () _packet.data = 0; _packet.size = 0; + /* XXX: should we reset _packet.data and size after each *_decode_* call? */ + int frame_finished; while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - process_video (_frame); + filter_and_emit_video (_frame); } - if (_audio_stream >= 0 && _opt->decode_audio) { + if (_audio_stream && _opt.decode_audio) { while (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { int const data_size = av_samples_get_buffer_size ( 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1 ); - - assert (_audio_codec_context->channels == _fs->audio_channels); - process_audio (_frame->data[0], data_size); + + assert (_audio_codec_context->channels == _film->audio_channels()); + Audio (deinterleave_audio (_frame->data[0], data_size)); } } return true; } + avcodec_get_frame_defaults (_frame); + + shared_ptr ffa = dynamic_pointer_cast (_audio_stream); + if (_packet.stream_index == _video_stream) { int frame_finished; - if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - process_video (_frame); + int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet); + if (r >= 0 && frame_finished) { + + if (r != _packet.size) { + _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size)); + } + + if (_opt.video_sync) { + out_with_sync (); + } else { + filter_and_emit_video (_frame); + } } - } else if (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio) { - - avcodec_get_frame_defaults (_frame); - + } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) { + int frame_finished; if (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - int const data_size = av_samples_get_buffer_size ( - 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1 - ); - assert (_audio_codec_context->channels == _fs->audio_channels); - process_audio (_frame->data[0], data_size); - } + /* 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); - } else if (_subtitle_stream >= 0 && _packet.stream_index == _subtitle_stream && _opt->decode_subtitles) { + /* We only decode audio if we've had our first video packet through, and if it + was before this packet. Until then audio is thrown away. + */ + + if (_first_video && _first_video.get() <= source_pts_seconds) { + + if (!_first_audio) { + _first_audio = source_pts_seconds; + + /* This is our first audio frame, and if we've arrived here we must have had our + first video frame. Push some silence to make up any gap between our first + video frame and our first audio. + */ + + /* frames of silence that we must push */ + int const s = rint ((_first_audio.get() - _first_video.get()) * ffa->sample_rate ()); + + _film->log()->log ( + String::compose ( + "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() + ) + ); + + if (s) { + shared_ptr audio (new AudioBuffers (ffa->channels(), s)); + audio->make_silent (); + Audio (audio); + } + } + + int const data_size = av_samples_get_buffer_size ( + 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1 + ); + + assert (_audio_codec_context->channels == _film->audio_channels()); + Audio (deinterleave_audio (_frame->data[0], data_size)); + } + } + + } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles && _first_video) { int got_subtitle; AVSubtitle sub; if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) { - process_subtitle (shared_ptr (new TimedSubtitle (sub))); + /* Sometimes we get an empty AVSubtitle, which is used by some codecs to + indicate that the previous subtitle should stop. + */ + if (sub.num_rects > 0) { + shared_ptr ts; + try { + emit_subtitle (shared_ptr (new TimedSubtitle (sub))); + } catch (...) { + /* some problem with the subtitle; we probably didn't understand it */ + } + } else { + emit_subtitle (shared_ptr ()); + } avsubtitle_free (&sub); } } @@ -257,10 +350,101 @@ FFmpegDecoder::do_pass () return false; } -int -FFmpegDecoder::length_in_frames () const +shared_ptr +FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) { - return (_format_context->duration / AV_TIME_BASE) * frames_per_second (); + assert (_film->audio_channels()); + assert (bytes_per_audio_sample()); + + shared_ptr ffa = dynamic_pointer_cast (_audio_stream); + assert (ffa); + + /* Deinterleave and convert to float */ + + assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0); + + int const total_samples = size / bytes_per_audio_sample(); + int const frames = total_samples / _film->audio_channels(); + shared_ptr audio (new AudioBuffers (ffa->channels(), frames)); + + switch (audio_sample_format()) { + case AV_SAMPLE_FMT_S16: + { + int16_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 << 15); + + ++channel; + if (channel == _film->audio_channels()) { + channel = 0; + ++sample; + } + } + } + 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 = reinterpret_cast (data); + int sample = 0; + int channel = 0; + for (int i = 0; i < total_samples; ++i) { + audio->data(channel)[sample] = static_cast(*p++) / (1 << 31); + + ++channel; + if (channel == _film->audio_channels()) { + channel = 0; + ++sample; + } + } + } + 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); + for (int i = 0; i < _film->audio_channels(); ++i) { + memcpy (audio->data(i), p, frames * sizeof(float)); + p += frames; + } + } + break; + + default: + throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast (audio_sample_format()))); + } + + return audio; } float @@ -275,26 +459,6 @@ FFmpegDecoder::frames_per_second () const return av_q2d (s->r_frame_rate); } -int -FFmpegDecoder::audio_channels () const -{ - if (_audio_codec_context == 0) { - return 0; - } - - return _audio_codec_context->channels; -} - -int -FFmpegDecoder::audio_sample_rate () const -{ - if (_audio_codec_context == 0) { - return 0; - } - - return _audio_codec_context->sample_rate; -} - AVSampleFormat FFmpegDecoder::audio_sample_format () const { @@ -305,20 +469,10 @@ FFmpegDecoder::audio_sample_format () const return _audio_codec_context->sample_fmt; } -int64_t -FFmpegDecoder::audio_channel_layout () const -{ - if (_audio_codec_context == 0) { - return 0; - } - - return _audio_codec_context->channel_layout; -} - -Size +libdcp::Size FFmpegDecoder::native_size () const { - return Size (_video_codec_context->width, _video_codec_context->height); + return libdcp::Size (_video_codec_context->width, _video_codec_context->height); } PixelFormat @@ -351,8 +505,236 @@ FFmpegDecoder::sample_aspect_ratio_denominator () const return _video_codec_context->sample_aspect_ratio.den; } +string +FFmpegDecoder::stream_name (AVStream* s) const +{ + stringstream n; + + AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0); + if (lang) { + n << lang->value; + } + + AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0); + if (title) { + if (!n.str().empty()) { + n << " "; + } + n << title->value; + } + + if (n.str().empty()) { + n << "unknown"; + } + + return n.str (); +} + +int +FFmpegDecoder::bytes_per_audio_sample () const +{ + return av_get_bytes_per_sample (audio_sample_format ()); +} + +void +FFmpegDecoder::set_audio_stream (shared_ptr s) +{ + AudioDecoder::set_audio_stream (s); + setup_audio (); +} + +void +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(); + while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (frame->width, frame->height), (AVPixelFormat) frame->format)) { + ++i; + } + + if (i == _filter_graphs.end ()) { + graph.reset (new FilterGraph (_film, this, libdcp::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 { + graph = *i; + } + + list > images = graph->process (frame); + + for (list >::iterator i = images.begin(); i != images.end(); ++i) { + emit_video (*i, frame_time ()); + } +} + bool -FFmpegDecoder::has_subtitles () const +FFmpegDecoder::seek (double p) { - return (_subtitle_stream != -1); + 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) +{ + if (!v) { + /* version < 1; no type in the string, and there's only FFmpeg streams anyway */ + return shared_ptr (new FFmpegAudioStream (t, v)); + } + + stringstream s (t); + string type; + s >> type; + if (type != "ffmpeg") { + return shared_ptr (); + } + + return shared_ptr (new FFmpegAudioStream (t, v)); +} + +FFmpegAudioStream::FFmpegAudioStream (string t, optional version) +{ + stringstream n (t); + + int name_index = 4; + if (!version) { + name_index = 2; + int channels; + n >> _id >> channels; + _channel_layout = av_get_default_channel_layout (channels); + _sample_rate = 0; + } else { + string type; + /* Current (marked version 1) */ + n >> type >> _id >> _sample_rate >> _channel_layout; + assert (type == "ffmpeg"); + } + + for (int i = 0; i < name_index; ++i) { + size_t const s = t.find (' '); + if (s != string::npos) { + t = t.substr (s + 1); + } + } + + _name = t; +} + +string +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); +} +