X-Git-Url: https://git.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=acaf149f43ade599dff78d95c109d36c86133097;hp=09f699543c790c6755b59cb6d54530833c793039;hb=0b781dd65a90e617866810ecb3dee27267449f1f;hpb=09fcd95ff5125d7238188c59aa40430acb45ecbc diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 09f699543..acaf149f4 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -47,20 +47,24 @@ 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; +using boost::optional; +using boost::dynamic_pointer_cast; -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, Job* j, bool minimal, bool ignore_length) - : Decoder (f, o, j, minimal, ignore_length) +FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr 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) @@ -116,28 +120,30 @@ 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"); } @@ -158,14 +164,7 @@ 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"); } } @@ -173,11 +172,14 @@ 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) { @@ -187,24 +189,16 @@ 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) { @@ -218,7 +212,7 @@ FFmpegDecoder::setup_subtitle () bool -FFmpegDecoder::do_pass () +FFmpegDecoder::pass () { int r = av_read_frame (_format_context, &_packet); @@ -240,17 +234,17 @@ FFmpegDecoder::do_pass () 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 == _film->audio_channels()); - process_audio (_frame->data[0], data_size); + Audio (deinterleave_audio (_frame->data[0], data_size)); } } @@ -258,6 +252,8 @@ FFmpegDecoder::do_pass () } avcodec_get_frame_defaults (_frame); + + shared_ptr ffa = dynamic_pointer_cast (_audio_stream); if (_packet.stream_index == _video_stream) { @@ -270,12 +266,17 @@ FFmpegDecoder::do_pass () } /* Where we are in the output, in seconds */ - double const out_pts_seconds = video_frame_index() / frames_per_second(); + 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; } @@ -291,8 +292,8 @@ FFmpegDecoder::do_pass () repeat_last_video (); _film->log()->log ( String::compose ( - "Extra frame inserted at %1s; DCP frame %2, source PTS %3", - out_pts_seconds, video_frame_index(), source_pts_seconds + "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() ) ); } @@ -300,14 +301,14 @@ FFmpegDecoder::do_pass () if (delta > -one_frame) { /* Process this frame */ - process_video (_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)); } } - } else if (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio) { + } 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) { @@ -331,25 +332,19 @@ FFmpegDecoder::do_pass () */ /* frames of silence that we must push */ - int const s = rint ((_first_audio.get() - _first_video.get()) * audio_sample_rate ()); + 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 frames of silence for %4 channels (%5 bytes per sample)", - _first_video.get(), _first_audio.get(), s, audio_channels(), bytes_per_audio_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() ) ); if (s) { - /* 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); + shared_ptr audio (new AudioBuffers (ffa->channels(), s)); + audio->make_silent (); + Audio (audio); } } @@ -358,11 +353,11 @@ FFmpegDecoder::do_pass () ); assert (_audio_codec_context->channels == _film->audio_channels()); - process_audio (_frame->data[0], data_size); + Audio (deinterleave_audio (_frame->data[0], 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; @@ -371,9 +366,14 @@ 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, _first_video.get()))); + } catch (...) { + /* some problem with the subtitle; we probably didn't understand it */ + } } else { - process_subtitle (shared_ptr ()); + emit_subtitle (shared_ptr ()); } avsubtitle_free (&sub); } @@ -383,36 +383,84 @@ FFmpegDecoder::do_pass () return false; } -float -FFmpegDecoder::frames_per_second () const +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 = (int16_t *) 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_S32: + { + int32_t* p = (int32_t *) data; + int sample = 0; + int channel = 0; + for (int i = 0; i < total_samples; ++i) { + audio->data(channel)[sample] = float(*p++) / (1 << 31); + + ++channel; + if (channel == _film->audio_channels()) { + channel = 0; + ++sample; + } + } } - return av_q2d (s->r_frame_rate); -} + 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; -int -FFmpegDecoder::audio_channels () const -{ - if (_audio_codec_context == 0) { - return 0; + default: + assert (false); } - 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 @@ -425,16 +473,6 @@ 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 FFmpegDecoder::native_size () const { @@ -471,24 +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 { @@ -514,3 +534,106 @@ 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 (); +} + +void +FFmpegDecoder::filter_and_emit_video (AVFrame* frame) +{ + shared_ptr graph; + + list >::iterator i = _filter_graphs.begin(); + while (i != _filter_graphs.end() && !(*i)->can_process (Size (frame->width, frame->height), (AVPixelFormat) frame->format)) { + ++i; + } + + if (i == _filter_graphs.end ()) { + graph.reset (new FilterGraph (_film, this, _opt->apply_crop, 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); + } +} + +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); +} + +/** @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(); +}