X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=bf094913082104a1392048f2b8003daf245c825e;hb=89115db77729a2c99f1a09ff6a461720e16f889e;hp=7ebb31084799b252e8d6cf3cd6d3c34c68c2fa0e;hpb=6e52f83418c64d5cbd0e613a5fc96ea32881d4d9;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 7ebb31084..bf0949130 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -1,5 +1,3 @@ -/* -*- c-basic-offset: 8; default-tab-width: 8; -*- */ - /* Copyright (C) 2012 Carl Hetherington @@ -30,13 +28,11 @@ #include #include #include +#include extern "C" { #include #include -#include -#include } -#include #include "film.h" #include "filter.h" #include "exceptions.h" @@ -61,161 +57,46 @@ 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) : Decoder (f) - , VideoDecoder (f, c) - , AudioDecoder (f, c) - , _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) + , VideoDecoder (f) + , AudioDecoder (f) + , FFmpeg (c) , _subtitle_codec_context (0) , _subtitle_codec (0) , _decode_video (video) , _decode_audio (audio) + , _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 ( - 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 (shared_ptr (new 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); + double const old_first_video = first_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")); + /* 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 (_video_codec_context, _video_codec, 0) < 0) { - throw DecodeError (N_("could not open video 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_audio () -{ - boost::mutex::scoped_lock lm (_mutex); - - 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")); - } - - if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) { - throw DecodeError (N_("could not open audio decoder")); - } -} - -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); } -} - +} void FFmpegDecoder::pass () @@ -248,7 +129,8 @@ FFmpegDecoder::pass () } /* Stop us being asked for any more data */ - _next_video = _next_audio = _ffmpeg_content->length (); + _video_position = _ffmpeg_content->video_length (); + _audio_position = _ffmpeg_content->audio_length (); return; } @@ -259,6 +141,7 @@ FFmpegDecoder::pass () } 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) { +#if 0 int got_subtitle; AVSubtitle sub; @@ -278,8 +161,7 @@ FFmpegDecoder::pass () } avsubtitle_free (&sub); } - } else { - cout << "[ffmpeg] other packet.\n"; +#endif } av_free_packet (&_packet); @@ -381,59 +263,14 @@ 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::video_size () const -{ - return libdcp::Size (_video_codec_context->width, _video_codec_context->height); -} - -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 @@ -443,41 +280,29 @@ FFmpegDecoder::bytes_per_audio_sample () const } void -FFmpegDecoder::seek (Time t) +FFmpegDecoder::seek (VideoContent::Frame frame) { - do_seek (t, false, false); - VideoDecoder::seek (t); + do_seek (frame, false, false); } void FFmpegDecoder::seek_back () { - if (next() < (2.5 * TIME_HZ / video_frame_rate())) { + if (_video_position == 0) { return; } - do_seek (next() - 2.5 * TIME_HZ / video_frame_rate(), true, true); - VideoDecoder::seek_back (); + do_seek (_video_position - 1, true, true); } void -FFmpegDecoder::seek_forward () +FFmpegDecoder::do_seek (VideoContent::Frame frame, bool backwards, bool accurate) { - if (next() >= (_ffmpeg_content->length() - 0.5 * TIME_HZ / video_frame_rate())) { - return; - } - - do_seek (next() - 0.5 * TIME_HZ / video_frame_rate(), true, true); - VideoDecoder::seek_forward (); -} - -void -FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate) -{ - int64_t const vt = t / (av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ); + 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; - avcodec_flush_buffers (_video_codec_context); + avcodec_flush_buffers (video_codec_context()); if (_subtitle_codec_context) { avcodec_flush_buffers (_subtitle_codec_context); } @@ -493,10 +318,12 @@ FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate) 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) { + _video_position = (bet * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset) + * _ffmpeg_content->video_frame_rate(); break; } } @@ -505,15 +332,6 @@ FFmpegDecoder::do_seek (Time t, bool backwards, bool accurate) av_free_packet (&_packet); } } - - return; -} - -/** @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(); } void @@ -528,24 +346,32 @@ 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); - - 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), source_pts_seconds); + + 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; } - - copy_packet.data += decode_result; - copy_packet.size -= decode_result; } } } @@ -554,7 +380,7 @@ bool FFmpegDecoder::decode_video_packet () { int frame_finished; - if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) { + if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) { return false; } @@ -592,11 +418,33 @@ FFmpegDecoder::decode_video_packet () 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); + + 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); @@ -607,23 +455,33 @@ FFmpegDecoder::decode_video_packet () return true; } -Time -FFmpegDecoder::next () const + +void +FFmpegDecoder::setup_subtitle () { - if (_decode_video && _decode_audio && _audio_codec_context) { - return min (_next_video, _next_audio); + boost::mutex::scoped_lock lm (_mutex); + + if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) { + return; } - if (_decode_audio && _audio_codec_context) { - return _next_audio; - } + _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec; + _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); - return _next_video; + 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 { - return (!_decode_audio || !_audio_codec_context || audio_done()) && (!_decode_video || video_done()); + 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; }