X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=119e82851ec5de9433d119e2409b1b951067104f;hb=996b0c06e23bcb6b300d7b8799df94993692e07d;hp=d0b1de748c0e2960dc4bc3ed9aeade1c8db2ab00;hpb=9bdd8cc51942a13e360dde4efc04b3ca417c8b94;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index d0b1de748..119e82851 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 @@ -48,6 +50,7 @@ extern "C" { #include "ffmpeg_decoder.h" #include "filter_graph.h" #include "subtitle.h" +#include "audio_buffers.h" #include "i18n.h" @@ -56,6 +59,7 @@ 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; @@ -63,10 +67,10 @@ using libdcp::Size; boost::mutex FFmpegDecoder::_mutex; -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, bool subtitles) : Decoder (f) - , VideoDecoder (f) - , AudioDecoder (f) + , VideoDecoder (f, c) + , AudioDecoder (f, c) , _ffmpeg_content (c) , _format_context (0) , _video_stream (-1) @@ -80,16 +84,11 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptrcodec->sample_rate, s->codec->channel_layout) + shared_ptr ( + 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 (FFmpegSubtitleStream (stream_name (s), i)); + _subtitle_streams.push_back (shared_ptr (new FFmpegSubtitleStream (stream_name (s), i))); } } @@ -221,63 +222,46 @@ 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 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; + + 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 && _decode_subtitles) { int got_subtitle; AVSubtitle sub; @@ -288,19 +272,18 @@ 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); } } - + av_free_packet (&_packet); - return false; } /** @param data pointer to array of pointers to buffers. @@ -422,57 +405,29 @@ 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 { 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_(" "); + + 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; } - n << title->value; } if (n.str().empty()) { @@ -489,114 +444,67 @@ FFmpegDecoder::bytes_per_audio_sample () const } void -FFmpegDecoder::filter_and_emit_video (AVFrame* frame) +FFmpegDecoder::seek (Time t) { - 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 (N_("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 ()); - } + do_seek (t, false, false); } -bool -FFmpegDecoder::seek (double p) +void +FFmpegDecoder::seek_back () { - /* This use of AVSEEK_FLAG_BACKWARD 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. - */ - bool const backwards = (p == last_content_time()); - - 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); + if (next() < (2.5 * TIME_HZ / video_frame_rate())) { + return; } - return r < 0; + do_seek (next() - 2.5 * TIME_HZ / video_frame_rate(), true, true); } void -FFmpegDecoder::out_with_sync () +FFmpegDecoder::seek_forward () { - /* Where we are in the output, in seconds */ - double const out_pts_seconds = video_frame() / video_frame_rate(); - - /* 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 / video_frame_rate(); - - /* 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 (frame_time ()); - _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, video_frame_rate() - ) - ); - } + if (next() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) { + return; } - 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)); - } + do_seek (next() - 0.5 * TIME_HZ / video_frame_rate(), true, true); } void -FFmpegDecoder::film_changed (Film::Property p) +FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate) { - switch (p) { - case Film::CROP: - case Film::FILTERS: - { - boost::mutex::scoped_lock lm (_filter_graphs_mutex); - _filter_graphs.clear (); + 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) { + avcodec_flush_buffers (_subtitle_codec_context); } - break; - default: - break; + 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) { + break; + } + } + } + + av_free_packet (&_packet); + } } + + return; } /** @return Length (in video frames) according to our content's header */ @@ -606,12 +514,6 @@ FFmpegDecoder::video_length () const return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate(); } -double -FFmpegDecoder::frame_time () const -{ - return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base); -} - void FFmpegDecoder::decode_audio_packet () { @@ -625,55 +527,94 @@ FFmpegDecoder::decode_audio_packet () 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 (decode_result >= 0) { + if (frame_finished) { - 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); - } - } + /* 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); 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)); + Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds); } - } - - if (decode_result >= 0) { + copy_packet.data += decode_result; copy_packet.size -= decode_result; } } } + +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; + + 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); + } + + 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 film = _film.lock (); + assert (film); + film->log()->log ("Dropping frame without PTS"); + } + } + + return true; +} + +Time +FFmpegDecoder::next () const +{ + if (_decode_video && _decode_audio) { + return min (_next_video, _next_audio); + } + + if (_decode_audio) { + return _next_audio; + } + + return _next_video; +}