X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=2d1792390dbdbb2e743124d5b6d0c8d87c832afd;hb=d39bfb9a1360a83f3451b79d01fdc4f465ef21d3;hp=ef7c4487d010ee9543d40c030713b35f836abd1b;hpb=145d1cf483f9ec554b3114db73d5be4def36027b;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index ef7c4487d..2d1792390 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -28,18 +28,12 @@ #include #include #include +#include extern "C" { -#include #include #include -#include -#include } -#include #include "film.h" -#include "format.h" -#include "transcoder.h" -#include "job.h" #include "filter.h" #include "exceptions.h" #include "image.h" @@ -48,6 +42,7 @@ extern "C" { #include "ffmpeg_decoder.h" #include "filter_graph.h" #include "subtitle.h" +#include "audio_buffers.h" #include "i18n.h" @@ -56,177 +51,67 @@ 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; -boost::mutex FFmpegDecoder::_mutex; - -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio, bool subtitles) +FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio) : Decoder (f) , VideoDecoder (f) , AudioDecoder (f) - , _ffmpeg_content (c) - , _format_context (0) - , _video_stream (-1) - , _frame (0) - , _video_codec_context (0) - , _video_codec (0) - , _audio_codec_context (0) - , _audio_codec (0) + , FFmpeg (c) , _subtitle_codec_context (0) , _subtitle_codec (0) , _decode_video (video) , _decode_audio (audio) - , _decode_subtitles (subtitles) + , _pts_offset (0) { - setup_general (); - setup_video (); - setup_audio (); setup_subtitle (); -} - -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); - } - - if (_subtitle_codec_context) { - avcodec_close (_subtitle_codec_context); - } - - av_free (_frame); - - avformat_close_input (&_format_context); -} - -void -FFmpegDecoder::setup_general () -{ - av_register_all (); - - 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 */ - - for (uint32_t i = 0; i < _format_context->nb_streams; ++i) { - AVStream* s = _format_context->streams[i]; - 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 ( - 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 (FFmpegSubtitleStream (stream_name (s), i)); - } - } - - if (_video_stream < 0) { - throw DecodeError (N_("could not find video stream")); - } - - _frame = avcodec_alloc_frame (); - if (_frame == 0) { - throw DecodeError (N_("could not allocate frame")); + if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) { + _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate()); } } -void -FFmpegDecoder::setup_video () +double +FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate) { - boost::mutex::scoped_lock lm (_mutex); + assert (first_video >= 0); + assert (first_audio >= 0); - _video_codec_context = _format_context->streams[_video_stream]->codec; - _video_codec = avcodec_find_decoder (_video_codec_context->codec_id); - - if (_video_codec == 0) { - throw DecodeError (_("could not find video decoder")); - } - - if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) { - throw DecodeError (N_("could not open video decoder")); - } -} - -void -FFmpegDecoder::setup_audio () -{ - boost::mutex::scoped_lock lm (_mutex); + double const old_first_video = first_video; - if (!_ffmpeg_content->audio_stream ()) { - return; - } - - _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) { - throw DecodeError (_("could not find audio decoder")); + /* Round the first video to a frame boundary */ + if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) { + first_video = ceil (first_video * video_frame_rate) / video_frame_rate; } - if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) { - throw DecodeError (N_("could not open audio decoder")); - } + /* Compute the required offset (also removing any common start delay) */ + return first_video - old_first_video - min (first_video, first_audio); } -void -FFmpegDecoder::setup_subtitle () +FFmpegDecoder::~FFmpegDecoder () { - 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[_ffmpeg_content->subtitle_stream()->id]->codec; - _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); - - if (_subtitle_codec == 0) { - throw DecodeError (_("could not find subtitle decoder")); - } - - if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { - throw DecodeError (N_("could not open subtitle decoder")); + if (_subtitle_codec_context) { + avcodec_close (_subtitle_codec_context); } -} - +} -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 film = _film.lock (); + assert (film); + film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r)); } /* Get any remaining frames */ @@ -237,14 +122,17 @@ FFmpegDecoder::pass () /* XXX: should we reset _packet.data and size after each *_decode_* call? */ if (_decode_video) { - decode_video_packet (); + while (decode_video_packet ()); } if (_ffmpeg_content->audio_stream() && _decode_audio) { decode_audio_packet (); } - - return true; + + /* Stop us being asked for any more data */ + _video_position = _ffmpeg_content->video_length (); + _audio_position = _ffmpeg_content->audio_length (); + return; } avcodec_get_frame_defaults (_frame); @@ -253,7 +141,8 @@ FFmpegDecoder::pass () decode_video_packet (); } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) { decode_audio_packet (); - } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) { + } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) { +#if 0 int got_subtitle; AVSubtitle sub; @@ -264,19 +153,19 @@ FFmpegDecoder::pass () if (sub.num_rects > 0) { shared_ptr ts; try { - emit_subtitle (shared_ptr (new TimedSubtitle (sub))); + subtitle (shared_ptr (new TimedSubtitle (sub))); } catch (...) { /* some problem with the subtitle; we probably didn't understand it */ } } else { - emit_subtitle (shared_ptr ()); + subtitle (shared_ptr ()); } avsubtitle_free (&sub); } +#endif } - + av_free_packet (&_packet); - return false; } /** @param data pointer to array of pointers to buffers. @@ -375,89 +264,14 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) return audio; } -float -FFmpegDecoder::video_frame_rate () const -{ - 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 av_q2d (s->r_frame_rate); -} - AVSampleFormat FFmpegDecoder::audio_sample_format () const { - if (_audio_codec_context == 0) { + if (!_ffmpeg_content->audio_stream()) { return (AVSampleFormat) 0; } - return _audio_codec_context->sample_fmt; -} - -libdcp::Size -FFmpegDecoder::native_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 -{ - stringstream n; - - if (s->metadata) { - AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0); - if (lang) { - n << lang->value; - } - - AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0); - if (title) { - if (!n.str().empty()) { - n << N_(" "); - } - n << title->value; - } - } - - if (n.str().empty()) { - n << N_("unknown"); - } - - return n.str (); + return audio_codec_context()->sample_fmt; } int @@ -466,40 +280,29 @@ FFmpegDecoder::bytes_per_audio_sample () const return av_get_bytes_per_sample (audio_sample_format ()); } -bool -FFmpegDecoder::seek (double p) +void +FFmpegDecoder::seek (VideoContent::Frame frame) { - return do_seek (p, false, false); + do_seek (frame, false, false); } -bool +void FFmpegDecoder::seek_back () { - if (last_content_time() < 2.5) { - return true; - } - - return do_seek (last_content_time() - 2.5 / video_frame_rate(), true, true); -} - -bool -FFmpegDecoder::seek_forward () -{ - if (last_content_time() >= (video_length() - video_frame_rate())) { - return true; + if (_video_position == 0) { + return; } - return do_seek (last_content_time() - 0.5 / video_frame_rate(), true, true); + do_seek (_video_position - 1, true, true); } -bool -FFmpegDecoder::do_seek (double p, bool backwards, bool accurate) +void +FFmpegDecoder::do_seek (VideoContent::Frame frame, 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 = frame * _ffmpeg_content->video_frame_rate() / av_q2d (_format_context->streams[_video_stream]->time_base); + av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0); - avcodec_flush_buffers (_video_codec_context); + avcodec_flush_buffers (video_codec_context()); if (_subtitle_codec_context) { avcodec_flush_buffers (_subtitle_codec_context); } @@ -508,14 +311,14 @@ 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); if (_packet.stream_index == _video_stream) { int finished = 0; - int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet); + int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet); if (r >= 0 && finished) { int64_t const bet = av_frame_get_best_effort_timestamp (_frame); if (bet > vt) { @@ -527,32 +330,8 @@ FFmpegDecoder::do_seek (double p, bool backwards, bool accurate) av_free_packet (&_packet); } } - - return r < 0; -} - -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 (); - } - break; - default: - break; - } -} - -/** @return Length (in video frames) according to our content's header */ -ContentVideoFrame -FFmpegDecoder::video_length () const -{ - return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate(); + return; } void @@ -567,20 +346,36 @@ FFmpegDecoder::decode_audio_packet () while (copy_packet.size > 0) { int frame_finished; - int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet); + int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); if (decode_result >= 0) { if (frame_finished) { - - /* Where we are in the source, in seconds */ - double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base) - * av_frame_get_best_effort_timestamp(_frame); + + if (_audio_position == 0) { + /* Where we are in the source, in seconds */ + double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base) + * av_frame_get_best_effort_timestamp(_frame) - _pts_offset; + + if (pts > 0) { + /* Emit some silence */ + shared_ptr silence ( + new AudioBuffers ( + _ffmpeg_content->audio_channels(), + pts * _ffmpeg_content->content_audio_frame_rate() + ) + ); + + silence->make_silent (); + audio (silence, _audio_position); + } + } + int const data_size = av_samples_get_buffer_size ( - 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1 + 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1 ); - assert (_audio_codec_context->channels == _ffmpeg_content->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), _audio_position); } copy_packet.data += decode_result; @@ -589,40 +384,112 @@ FFmpegDecoder::decode_audio_packet () } } -void +bool FFmpegDecoder::decode_video_packet () { int frame_finished; - while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - 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 (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) { + return false; + } - 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 (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format)); - } else { - graph = *i; + 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 ()) { + shared_ptr 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 > images = graph->process (_frame); + + string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second; + + for (list >::iterator i = images.begin(); i != images.end(); ++i) { + + shared_ptr image = *i; + if (!post_process.empty ()) { + image = image->post_process (post_process, true); } - list > images = graph->process (_frame); - - for (list >::iterator i = images.begin(); i != images.end(); ++i) { - int64_t const bet = av_frame_get_best_effort_timestamp (_frame); - if (bet != AV_NOPTS_VALUE) { - emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base)); - } else { - _film->log()->log ("Dropping frame without PTS"); + int64_t const bet = av_frame_get_best_effort_timestamp (_frame); + if (bet != AV_NOPTS_VALUE) { + + double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset; + double const next = _video_position / _ffmpeg_content->video_frame_rate(); + double const one_frame = 1 / _ffmpeg_content->video_frame_rate (); + double delta = pts - next; + + while (delta > one_frame) { + /* This PTS is more than one frame forward in time of where we think we should be; emit + a black frame. + */ + boost::shared_ptr black ( + new SimpleImage ( + static_cast (_frame->format), + libdcp::Size (video_codec_context()->width, video_codec_context()->height), + true + ) + ); + + black->make_black (); + video (image, false, _video_position); + delta -= one_frame; } + + if (delta > -one_frame) { + /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */ + video (image, false, _video_position); + } + } else { + shared_ptr film = _film.lock (); + assert (film); + film->log()->log ("Dropping frame without PTS"); } } + + return true; } + +void +FFmpegDecoder::setup_subtitle () +{ + 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[_ffmpeg_content->subtitle_stream()->id]->codec; + _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); + if (_subtitle_codec == 0) { + throw DecodeError (_("could not find subtitle decoder")); + } + + if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { + throw DecodeError (N_("could not open subtitle decoder")); + } +} + +bool +FFmpegDecoder::done () const +{ + bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length()); + bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length()); + return vd && ad; +}