X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=bf094913082104a1392048f2b8003daf245c825e;hb=89115db77729a2c99f1a09ff6a461720e16f889e;hp=3a185bd6af3e9d9aca23d8a5a3fa7868ffd64f8f;hpb=9c3e4462d32c726a6c257b0a40e642ab23d9526a;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 3a185bd6a..bf0949130 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" @@ -56,218 +51,97 @@ 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 f, shared_ptr c, bool video, bool audio, bool subtitles, bool video_sync) +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) - , _video_sync (video_sync) + , _pts_offset (0) { - setup_general (); - setup_video (); - setup_audio (); setup_subtitle (); - if (!video_sync) { - _first_video = 0; + 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()); } } -FFmpegDecoder::~FFmpegDecoder () +double +FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate) { - if (_audio_codec_context) { - avcodec_close (_audio_codec_context); - } + double const old_first_video = first_video; - 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 ()); + /* 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 (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->channel_layout) - ); - - } 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")); - } + /* Compute the required offset (also removing any common start delay) */ + return first_video - old_first_video - min (first_video, first_audio); } -void -FFmpegDecoder::setup_video () -{ - _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 () +FFmpegDecoder::~FFmpegDecoder () { - 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); + boost::mutex::scoped_lock lm (_mutex); - if (_audio_codec == 0) { - throw DecodeError (_("could not find audio decoder")); - } - - if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) { - throw DecodeError (N_("could not open audio decoder")); + if (_subtitle_codec_context) { + avcodec_close (_subtitle_codec_context); } -} +} void -FFmpegDecoder::setup_subtitle () -{ - 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::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 */ _packet.data = 0; _packet.size = 0; - + /* XXX: should we reset _packet.data and size after each *_decode_* call? */ - - int frame_finished; - + if (_decode_video) { - while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - filter_and_emit_video (_frame); - } + 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); if (_packet.stream_index == _video_stream && _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)); - } - - if (_video_sync) { - out_with_sync (); - } else { - filter_and_emit_video (_frame); - } - } - + 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 && _first_video) { + } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) { +#if 0 int got_subtitle; AVSubtitle sub; @@ -278,19 +152,19 @@ 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. @@ -389,291 +263,225 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) return audio; } -float -FFmpegDecoder::frames_per_second () 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; + return audio_codec_context()->sample_fmt; } int -FFmpegDecoder::time_base_numerator () const +FFmpegDecoder::bytes_per_audio_sample () const { - return _video_codec_context->time_base.num; + return av_get_bytes_per_sample (audio_sample_format ()); } -int -FFmpegDecoder::time_base_denominator () const +void +FFmpegDecoder::seek (VideoContent::Frame frame) { - return _video_codec_context->time_base.den; + do_seek (frame, false, false); } -int -FFmpegDecoder::sample_aspect_ratio_numerator () const +void +FFmpegDecoder::seek_back () { - return _video_codec_context->sample_aspect_ratio.num; + if (_video_position == 0) { + return; + } + + do_seek (_video_position - 1, true, true); } -int -FFmpegDecoder::sample_aspect_ratio_denominator () const +void +FFmpegDecoder::do_seek (VideoContent::Frame frame, bool backwards, bool accurate) { - return _video_codec_context->sample_aspect_ratio.den; -} + 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); + _video_position = frame; -string -FFmpegDecoder::stream_name (AVStream* s) const -{ - stringstream n; - - 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; + avcodec_flush_buffers (video_codec_context()); + if (_subtitle_codec_context) { + avcodec_flush_buffers (_subtitle_codec_context); } - if (n.str().empty()) { - n << N_("unknown"); + if (accurate) { + while (1) { + int r = av_read_frame (_format_context, &_packet); + if (r < 0) { + 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); + if (r >= 0 && finished) { + int64_t const bet = av_frame_get_best_effort_timestamp (_frame); + if (bet > vt) { + _video_position = (bet * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset) + * _ffmpeg_content->video_frame_rate(); + break; + } + } + } + + av_free_packet (&_packet); + } } - - return n.str (); } -int -FFmpegDecoder::bytes_per_audio_sample () const +void +FFmpegDecoder::decode_audio_packet () { - return av_get_bytes_per_sample (audio_sample_format ()); + /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 + several times. + */ + + AVPacket copy_packet = _packet; + + while (copy_packet.size > 0) { + + int frame_finished; + int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); + if (decode_result >= 0) { + if (frame_finished) { + + 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); + } + } + + copy_packet.data += decode_result; + copy_packet.size -= decode_result; + } + } + } } -void -FFmpegDecoder::filter_and_emit_video (AVFrame* frame) +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 graph; + 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)) { + 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)); + 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)); + + 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); + 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) { - 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); + shared_ptr 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) { + + 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; + } - 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); + 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 r < 0; + + return true; } + void -FFmpegDecoder::out_with_sync () +FFmpegDecoder::setup_subtitle () { - /* Where we are in the output, in seconds */ - double const out_pts_seconds = video_frame() / frames_per_second(); + boost::mutex::scoped_lock lm (_mutex); - /* 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 (N_("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 ( - N_("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 (N_("Frame removed at %1s"), out_pts_seconds)); + if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) { + return; } -} -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; + _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec; + _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); - default: - break; + 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")); } } -/** @return Length (in video frames) according to our content's header */ -ContentVideoFrame -FFmpegDecoder::video_length () const -{ - return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second(); -} - -double -FFmpegDecoder::frame_time () const +bool +FFmpegDecoder::done () const { - return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base); + 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; } - -void -FFmpegDecoder::decode_audio_packet () -{ - /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4 - several times. - */ - AVPacket copy_packet = _packet; - - while (copy_packet.size > 0) { - - int frame_finished; - int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet); - if (decode_result >= 0 && 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); - - /* 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) || !_decode_video) { - - if (!_first_audio && _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. - */ - - /* frames of silence that we must push */ - int const s = rint ((_first_audio.get() - _first_video.get()) * _ffmpeg_content->audio_frame_rate ()); - - _film->log()->log ( - String::compose ( - N_("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, _ffmpeg_content->audio_channels(), bytes_per_audio_sample() - ) - ); - - if (s) { - shared_ptr audio (new AudioBuffers (_ffmpeg_content->audio_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 == _ffmpeg_content->audio_channels()); - Audio (deinterleave_audio (_frame->data, data_size)); - } - } - - if (decode_result >= 0) { - copy_packet.data += decode_result; - copy_packet.size -= decode_result; - } - } -}