diff options
Diffstat (limited to 'src/lib/ffmpeg_decoder.cc')
| -rw-r--r-- | src/lib/ffmpeg_decoder.cc | 412 |
1 files changed, 161 insertions, 251 deletions
diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index bcfbea431..1d000b62b 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,3 +1,5 @@ +/* -*- c-basic-offset: 8; default-tab-width: 8; -*- */ + /* Copyright (C) 2012 Carl Hetherington <cth@carlh.net> @@ -36,11 +38,7 @@ extern "C" { } #include <sndfile.h> #include "film.h" -#include "format.h" -#include "transcoder.h" -#include "job.h" #include "filter.h" -#include "options.h" #include "exceptions.h" #include "image.h" #include "util.h" @@ -48,6 +46,7 @@ extern "C" { #include "ffmpeg_decoder.h" #include "filter_graph.h" #include "subtitle.h" +#include "audio_buffers.h" #include "i18n.h" @@ -56,15 +55,19 @@ using std::string; using std::vector; using std::stringstream; using std::list; +using std::min; using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; using libdcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o) - : Decoder (f, o) - , VideoDecoder (f, o) - , AudioDecoder (f, o) +boost::mutex FFmpegDecoder::_mutex; + +FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles) + : Decoder (f) + , VideoDecoder (f, c) + , AudioDecoder (f, c) + , _ffmpeg_content (c) , _format_context (0) , _video_stream (-1) , _frame (0) @@ -74,6 +77,9 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o) , _audio_codec (0) , _subtitle_codec_context (0) , _subtitle_codec (0) + , _decode_video (video) + , _decode_audio (audio) + , _decode_subtitles (subtitles) { setup_general (); setup_video (); @@ -83,10 +89,12 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, DecodeOptions o) FFmpegDecoder::~FFmpegDecoder () { + boost::mutex::scoped_lock lm (_mutex); + if (_audio_codec_context) { avcodec_close (_audio_codec_context); } - + if (_video_codec_context) { avcodec_close (_video_codec_context); } @@ -105,15 +113,15 @@ FFmpegDecoder::setup_general () { av_register_all (); - if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) { - throw OpenFileError (_film->content_path ()); + if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) { + throw OpenFileError (_ffmpeg_content->file().string ()); } 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 */ + /* Find video, audio and subtitle streams */ for (uint32_t i = 0; i < _format_context->nb_streams; ++i) { AVStream* s = _format_context->streams[i]; @@ -130,17 +138,13 @@ FFmpegDecoder::setup_general () } _audio_streams.push_back ( - shared_ptr<AudioStream> ( - new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout) + shared_ptr<FFmpegAudioStream> ( + new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels) ) ); - + } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) { - _subtitle_streams.push_back ( - shared_ptr<SubtitleStream> ( - new SubtitleStream (stream_name (s), i) - ) - ); + _subtitle_streams.push_back (shared_ptr<FFmpegSubtitleStream> (new FFmpegSubtitleStream (stream_name (s), i))); } } @@ -157,6 +161,8 @@ FFmpegDecoder::setup_general () void FFmpegDecoder::setup_video () { + boost::mutex::scoped_lock lm (_mutex); + _video_codec_context = _format_context->streams[_video_stream]->codec; _video_codec = avcodec_find_decoder (_video_codec_context->codec_id); @@ -172,14 +178,13 @@ FFmpegDecoder::setup_video () void FFmpegDecoder::setup_audio () { - if (!_audio_stream) { + boost::mutex::scoped_lock lm (_mutex); + + if (!_ffmpeg_content->audio_stream ()) { return; } - shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream); - assert (ffa); - - _audio_codec_context = _format_context->streams[ffa->id()]->codec; + _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec; _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id); if (_audio_codec == 0) { @@ -194,11 +199,13 @@ FFmpegDecoder::setup_audio () void FFmpegDecoder::setup_subtitle () { - if (!_subtitle_stream || _subtitle_stream->id() >= int (_format_context->nb_streams)) { + boost::mutex::scoped_lock lm (_mutex); + + if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) { return; } - _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec; + _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec; _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); if (_subtitle_codec == 0) { @@ -211,17 +218,19 @@ FFmpegDecoder::setup_subtitle () } -bool +void FFmpegDecoder::pass () { int r = av_read_frame (_format_context, &_packet); - + if (r < 0) { if (r != AVERROR_EOF) { /* 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 (N_("error on av_read_frame (%1) (%2)"), buf, r)); + shared_ptr<const Film> film = _film.lock (); + assert (film); + film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r)); } /* Get any remaining frames */ @@ -231,41 +240,24 @@ FFmpegDecoder::pass () /* XXX: should we reset _packet.data and size after each *_decode_* call? */ - int frame_finished; - - if (_opt.decode_video) { - while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - filter_and_emit_video (); - } + if (_decode_video) { + while (decode_video_packet ()); } - - if (_audio_stream && _opt.decode_audio) { + + if (_ffmpeg_content->audio_stream() && _decode_audio) { decode_audio_packet (); } - return true; + return; } avcodec_get_frame_defaults (_frame); - shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream); - - if (_packet.stream_index == _video_stream && _opt.decode_video) { - - int frame_finished; - 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 (N_("Used only %1 bytes of %2 in packet"), r, _packet.size)); - } - - filter_and_emit_video (); - } - - } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) { + if (_packet.stream_index == _video_stream && _decode_video) { + decode_video_packet (); + } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) { decode_audio_packet (); - } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles) { + } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) { int got_subtitle; AVSubtitle sub; @@ -276,19 +268,18 @@ FFmpegDecoder::pass () if (sub.num_rects > 0) { shared_ptr<TimedSubtitle> ts; try { - emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub))); + subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub))); } catch (...) { /* some problem with the subtitle; we probably didn't understand it */ } } else { - emit_subtitle (shared_ptr<TimedSubtitle> ()); + subtitle (shared_ptr<TimedSubtitle> ()); } avsubtitle_free (&sub); } } - + av_free_packet (&_packet); - return false; } /** @param data pointer to array of pointers to buffers. @@ -297,19 +288,16 @@ FFmpegDecoder::pass () shared_ptr<AudioBuffers> FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) { - assert (_film->audio_channels()); + assert (_ffmpeg_content->audio_channels()); assert (bytes_per_audio_sample()); - shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream); - assert (ffa); - /* Deinterleave and convert to float */ - assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0); + assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0); int const total_samples = size / bytes_per_audio_sample(); - int const frames = total_samples / _film->audio_channels(); - shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames)); + int const frames = total_samples / _ffmpeg_content->audio_channels(); + shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames)); switch (audio_sample_format()) { case AV_SAMPLE_FMT_S16: @@ -321,7 +309,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = float(*p++) / (1 << 15); ++channel; - if (channel == _film->audio_channels()) { + if (channel == _ffmpeg_content->audio_channels()) { channel = 0; ++sample; } @@ -332,7 +320,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) case AV_SAMPLE_FMT_S16P: { int16_t** p = reinterpret_cast<int16_t **> (data); - for (int i = 0; i < _film->audio_channels(); ++i) { + for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { for (int j = 0; j < frames; ++j) { audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15); } @@ -349,7 +337,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31); ++channel; - if (channel == _film->audio_channels()) { + if (channel == _ffmpeg_content->audio_channels()) { channel = 0; ++sample; } @@ -366,7 +354,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) audio->data(channel)[sample] = *p++; ++channel; - if (channel == _film->audio_channels()) { + if (channel == _ffmpeg_content->audio_channels()) { channel = 0; ++sample; } @@ -377,7 +365,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) case AV_SAMPLE_FMT_FLTP: { float** p = reinterpret_cast<float**> (data); - for (int i = 0; i < _film->audio_channels(); ++i) { + for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { memcpy (audio->data(i), p[i], frames * sizeof(float)); } } @@ -391,7 +379,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) } float -FFmpegDecoder::frames_per_second () const +FFmpegDecoder::video_frame_rate () const { AVStream* s = _format_context->streams[_video_stream]; @@ -413,41 +401,11 @@ FFmpegDecoder::audio_sample_format () const } libdcp::Size -FFmpegDecoder::native_size () const +FFmpegDecoder::video_size () const { return libdcp::Size (_video_codec_context->width, _video_codec_context->height); } -PixelFormat -FFmpegDecoder::pixel_format () const -{ - return _video_codec_context->pix_fmt; -} - -int -FFmpegDecoder::time_base_numerator () const -{ - return _video_codec_context->time_base.num; -} - -int -FFmpegDecoder::time_base_denominator () const -{ - return _video_codec_context->time_base.den; -} - -int -FFmpegDecoder::sample_aspect_ratio_numerator () const -{ - return _video_codec_context->sample_aspect_ratio.num; -} - -int -FFmpegDecoder::sample_aspect_ratio_denominator () const -{ - return _video_codec_context->sample_aspect_ratio.den; -} - string FFmpegDecoder::stream_name (AVStream* s) const { @@ -482,89 +440,36 @@ FFmpegDecoder::bytes_per_audio_sample () const } void -FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s) +FFmpegDecoder::seek (Time t) { - AudioDecoder::set_audio_stream (s); - setup_audio (); -} - -void -FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s) -{ - VideoDecoder::set_subtitle_stream (s); - setup_subtitle (); - OutputChanged (); + do_seek (t, false, false); } void -FFmpegDecoder::filter_and_emit_video () +FFmpegDecoder::seek_back () { - int64_t const bet = av_frame_get_best_effort_timestamp (_frame); - if (bet == AV_NOPTS_VALUE) { - _film->log()->log ("Dropping frame without PTS"); + if (next() < (2.5 * TIME_HZ / video_frame_rate())) { return; } - shared_ptr<FilterGraph> graph; - - { - boost::mutex::scoped_lock lm (_filter_graphs_mutex); - - list<shared_ptr<FilterGraph> >::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 = filter_graph_factory (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format); - _filter_graphs.push_back (graph); - _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format)); - } else { - graph = *i; - } - } - - list<shared_ptr<Image> > images = graph->process (_frame); - - for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) { - emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base)); - } -} - -bool -FFmpegDecoder::seek (double p) -{ - return do_seek (p, false, 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, false); -} - -void -FFmpegDecoder::seek_back () -{ - do_seek (last_source_time() - 2.5 / frames_per_second (), true, true); + do_seek (next() - 2.5 * TIME_HZ / video_frame_rate(), true, true); } void FFmpegDecoder::seek_forward () { - do_seek (last_source_time() - 0.5 / frames_per_second(), true, true); + if (next() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) { + return; + } + + do_seek (next() - 0.5 * TIME_HZ / video_frame_rate(), true, true); } -bool -FFmpegDecoder::do_seek (double p, bool backwards, bool accurate) +void +FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate) { - 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); + int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ); + av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0); avcodec_flush_buffers (_video_codec_context); if (_subtitle_codec_context) { @@ -575,7 +480,7 @@ FFmpegDecoder::do_seek (double p, bool backwards, bool accurate) while (1) { int r = av_read_frame (_format_context, &_packet); if (r < 0) { - return true; + return; } avcodec_get_frame_defaults (_frame); @@ -594,93 +499,20 @@ FFmpegDecoder::do_seek (double p, bool backwards, bool accurate) av_free_packet (&_packet); } } - - return r < 0; -} - -shared_ptr<FFmpegAudioStream> -FFmpegAudioStream::create (string t, optional<int> v) -{ - if (!v) { - /* version < 1; no type in the string, and there's only FFmpeg streams anyway */ - return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v)); - } - - stringstream s (t); - string type; - s >> type; - if (type != N_("ffmpeg")) { - return shared_ptr<FFmpegAudioStream> (); - } - - return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v)); -} - -FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> 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 == N_("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 (N_("ffmpeg %1 %2 %3 %4"), _id, _sample_rate, _channel_layout, _name); -} - -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; } /** @return Length (in video frames) according to our content's header */ -SourceFrame -FFmpegDecoder::length () const +ContentVideoFrame +FFmpegDecoder::video_length () const { - return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second(); + return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate(); } void FFmpegDecoder::decode_audio_packet () { - shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream); - assert (ffa); - /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 several times. */ @@ -702,8 +534,8 @@ FFmpegDecoder::decode_audio_packet () 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, data_size), source_pts_seconds); + assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels()); + audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds); } copy_packet.data += decode_result; @@ -711,3 +543,81 @@ FFmpegDecoder::decode_audio_packet () } } } + +bool +FFmpegDecoder::decode_video_packet () +{ + int frame_finished; + if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) { + return false; + } + + boost::mutex::scoped_lock lm (_filter_graphs_mutex); + + shared_ptr<FilterGraph> graph; + + list<shared_ptr<FilterGraph> >::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 ()) { + shared_ptr<const Film> film = _film.lock (); + assert (film); + + graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)); + _filter_graphs.push_back (graph); + + film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format)); + } else { + graph = *i; + } + + list<shared_ptr<Image> > images = graph->process (_frame); + + string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second; + + for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) { + + shared_ptr<Image> image = *i; + if (!post_process.empty ()) { + image = image->post_process (post_process, true); + } + + int64_t const bet = av_frame_get_best_effort_timestamp (_frame); + if (bet != AV_NOPTS_VALUE) { + /* XXX: may need to insert extra frames / remove frames here ... + (as per old Matcher) + */ + Time const t = bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ; + video (image, false, t); + } else { + shared_ptr<const Film> film = _film.lock (); + assert (film); + film->log()->log ("Dropping frame without PTS"); + } + } + + return true; +} + +Time +FFmpegDecoder::next () const +{ + if (_decode_video && _decode_audio && _audio_codec_context) { + return min (_next_video, _next_audio); + } + + if (_decode_audio && _audio_codec_context) { + return _next_audio; + } + + return _next_video; +} + +bool +FFmpegDecoder::done () const +{ + return (!_decode_audio || !_audio_codec_context || audio_done()) && (!_decode_video || video_done()); +} + |
