X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=148764162c20fe70bfdd9d4be2e7ed96faaf8d57;hb=ab624eaed4423b1813148e1c395092697b541139;hp=98fd4f576531adcbe44bb673a0b1c0cee1d7e747;hpb=093ad6790e1a5d6656dbf422668739f14a52471f;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 98fd4f576..148764162 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -47,20 +47,25 @@ extern "C" { #include "util.h" #include "log.h" #include "ffmpeg_decoder.h" +#include "filter_graph.h" #include "subtitle.h" using std::cout; using std::string; using std::vector; using std::stringstream; +using std::list; using boost::shared_ptr; - -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, Job* j, bool minimal, bool ignore_length) - : Decoder (f, o, j, minimal, ignore_length) +using boost::optional; +using boost::dynamic_pointer_cast; +using libdcp::Size; + +FFmpegDecoder::FFmpegDecoder (shared_ptr f, DecodeOptions o) + : Decoder (f, o) + , VideoDecoder (f, o) + , AudioDecoder (f, o) , _format_context (0) , _video_stream (-1) - , _audio_stream (-1) - , _subtitle_stream (-1) , _frame (0) , _video_codec_context (0) , _video_codec (0) @@ -73,6 +78,10 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, J setup_video (); setup_audio (); setup_subtitle (); + + if (!o.video_sync) { + _first_video = 0; + } } FFmpegDecoder::~FFmpegDecoder () @@ -88,19 +97,18 @@ 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, _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 ()); } @@ -115,28 +123,30 @@ FFmpegDecoder::setup_general () if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) { _video_stream = i; } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) { - if (_audio_stream == -1) { - _audio_stream = i; + + /* 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 (AudioStream (stream_name (s), i, 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) { - if (_subtitle_stream == -1) { - _subtitle_stream = i; - } - _subtitle_streams.push_back (SubtitleStream (stream_name (s), i)); + _subtitle_streams.push_back ( + shared_ptr ( + new SubtitleStream (stream_name (s), i) + ) + ); } } - /* Now override audio and subtitle streams with those from the Film, if it has any */ - - if (_film->audio_stream_index() != -1) { - _audio_stream = _film->audio_stream().id(); - } - - if (_film->subtitle_stream_index() != -1) { - _subtitle_stream = _film->subtitle_stream().id (); - } - if (_video_stream < 0) { throw DecodeError ("could not find video stream"); } @@ -157,14 +167,7 @@ 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"); } } @@ -172,11 +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) { @@ -186,24 +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 || _subtitle_stream->id() >= _format_context->nb_streams) { 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) { @@ -217,13 +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 */ @@ -235,108 +236,97 @@ FFmpegDecoder::do_pass () int frame_finished; - while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - process_video (_frame); + if (_opt.decode_video) { + while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { + 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 == _film->audio_channels()); - process_audio (_frame->data[0], data_size); + Audio (deinterleave_audio (_frame->data, data_size)); } } return true; } - double const pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base) * _packet.pts; - avcodec_get_frame_defaults (_frame); - - if (_packet.stream_index == _video_stream) { - - if (!_first_video) { - _first_video = pts_seconds; - } - int frame_finished; - if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { + shared_ptr ffa = dynamic_pointer_cast (_audio_stream); - /* Where we are in the output, in seconds */ - double const out_pts_seconds = last_video_frame() / frames_per_second(); + if (_packet.stream_index == _video_stream && _opt.decode_video) { - /* Difference between where we are and where we should be */ - double const delta = pts_seconds - out_pts_seconds; - double const one_frame = 1 / frames_per_second(); + int frame_finished; + int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet); + if (r >= 0 && frame_finished) { - /* 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) { - _log->log (String::compose ("Extra frame inserted at %1s", out_pts_seconds)); - process_video (_frame); - } + if (r != _packet.size) { + _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size)); } - if (delta > -one_frame) { - /* Process this frame */ - process_video (_frame); + if (_opt.video_sync) { + out_with_sync (); } else { - /* Otherwise we are omitting a frame to keep things right */ - _log->log (String::compose ("Frame removed at %1s", out_pts_seconds)); + filter_and_emit_video (_frame); } } - } else if (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio && _first_video && _first_video.get() <= pts_seconds) { + } else if (ffa && _packet.stream_index == ffa->id() && _opt.decode_audio) { - /* Note: 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_audio) { - _first_audio = pts_seconds; - - /* This is our first audio packet, and if we've arrived here we must have had our - first video packet. Push some silence to make up the gap between our first - video packet and our first audio. - */ - - /* frames of silence that we must push */ - int const s = rint ((_first_audio.get() - _first_video.get()) * audio_sample_rate ()); - - _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.get(), _first_audio.get(), s, audio_channels(), bytes_per_audio_sample() - ) - ); - - /* hence bytes */ - int const b = s * audio_channels() * bytes_per_audio_sample(); - - /* XXX: this assumes that it won't be too much, and there are shaky assumptions - that all sound representations are silent with memset()ed zero data. - */ - uint8_t silence[b]; - memset (silence, 0, b); - process_audio (silence, b); - } - 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 - ); + + /* 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); + + /* 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) || !_opt.decode_video) { + + if (!_first_audio && _opt.decode_video) { + _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. + */ - assert (_audio_codec_context->channels == _film->audio_channels()); - process_audio (_frame->data[0], data_size); + /* 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, data_size)); + } } - } else if (_subtitle_stream >= 0 && _packet.stream_index == _subtitle_stream && _opt->decode_subtitles && _first_video) { + } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles && _first_video) { int got_subtitle; AVSubtitle sub; @@ -345,9 +335,14 @@ FFmpegDecoder::do_pass () indicate that the previous subtitle should stop. */ if (sub.num_rects > 0) { - process_subtitle (shared_ptr (new TimedSubtitle (sub, _first_video.get()))); + shared_ptr ts; + try { + emit_subtitle (shared_ptr (new TimedSubtitle (sub))); + } catch (...) { + /* some problem with the subtitle; we probably didn't understand it */ + } } else { - process_subtitle (shared_ptr ()); + emit_subtitle (shared_ptr ()); } avsubtitle_free (&sub); } @@ -357,36 +352,115 @@ FFmpegDecoder::do_pass () return false; } -float -FFmpegDecoder::frames_per_second () const +/** @param data pointer to array of pointers to buffers. + * Only the first buffer will be used for non-planar data, otherwise there will be one per channel. + */ +shared_ptr +FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) { - AVStream* s = _format_context->streams[_video_stream]; + assert (_film->audio_channels()); + assert (bytes_per_audio_sample()); - if (s->avg_frame_rate.num && s->avg_frame_rate.den) { - return av_q2d (s->avg_frame_rate); + 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[0]); + 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[i][j]) / (1 << 15); + } + } + } + break; + + case AV_SAMPLE_FMT_S32: + { + int32_t* p = reinterpret_cast (data[0]); + 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[0]); + 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[i], frames * sizeof(float)); + } + } + break; - return av_q2d (s->r_frame_rate); -} - -int -FFmpegDecoder::audio_channels () const -{ - if (_audio_codec_context == 0) { - return 0; + default: + throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast (audio_sample_format()))); } - return _audio_codec_context->channels; + return audio; } -int -FFmpegDecoder::audio_sample_rate () const +float +FFmpegDecoder::frames_per_second () const { - if (_audio_codec_context == 0) { - return 0; + AVStream* s = _format_context->streams[_video_stream]; + + if (s->avg_frame_rate.num && s->avg_frame_rate.den) { + return av_q2d (s->avg_frame_rate); } - - return _audio_codec_context->sample_rate; + + return av_q2d (s->r_frame_rate); } AVSampleFormat @@ -399,20 +473,10 @@ 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 @@ -445,24 +509,6 @@ FFmpegDecoder::sample_aspect_ratio_denominator () const return _video_codec_context->sample_aspect_ratio.den; } -bool -FFmpegDecoder::has_subtitles () const -{ - return (_subtitle_stream != -1); -} - -vector -FFmpegDecoder::audio_streams () const -{ - return _audio_streams; -} - -vector -FFmpegDecoder::subtitle_streams () const -{ - return _subtitle_streams; -} - string FFmpegDecoder::stream_name (AVStream* s) const { @@ -488,3 +534,211 @@ FFmpegDecoder::stream_name (AVStream* s) const 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::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) +{ + 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); +} +