X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;ds=sidebyside;f=src%2Flib%2Fffmpeg_decoder.cc;h=c93012608406cab6aeaf8c78414f5215366898df;hb=d50bc4c21be3e0766916b47a6e688cf896bed059;hp=81f40564460a035856332ae46f25cacf14d1a9dd;hpb=32fc1ddb0ee004d18c36155ddcf4d9b3998a7061;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 81f405644..c93012608 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -27,357 +27,204 @@ #include #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 "options.h" #include "exceptions.h" #include "image.h" #include "util.h" #include "log.h" #include "ffmpeg_decoder.h" #include "filter_graph.h" -#include "subtitle.h" +#include "audio_buffers.h" +#include "ffmpeg_content.h" + +#include "i18n.h" using std::cout; using std::string; using std::vector; using std::stringstream; using std::list; +using std::min; +using std::pair; using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; using libdcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr f, DecodeOptions o, Job* j) - : Decoder (f, o, j) - , VideoDecoder (f, o, j) - , AudioDecoder (f, o, j) - , _format_context (0) - , _video_stream (-1) - , _frame (0) - , _video_codec_context (0) - , _video_codec (0) - , _audio_codec_context (0) - , _audio_codec (0) +FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr c, bool video, bool audio) + : Decoder (f) + , VideoDecoder (f, c) + , AudioDecoder (f, c) + , SubtitleDecoder (f) + , FFmpeg (c) , _subtitle_codec_context (0) , _subtitle_codec (0) + , _decode_video (video) + , _decode_audio (audio) + , _pts_offset (0) + , _just_sought (false) { - setup_general (); - setup_video (); - setup_audio (); setup_subtitle (); - if (!o.video_sync) { - _first_video = 0; - } -} + /* Audio and video frame PTS values may not start with 0. We want + to fiddle them so that: -FFmpegDecoder::~FFmpegDecoder () -{ - if (_audio_codec_context) { - avcodec_close (_audio_codec_context); - } - - if (_video_codec_context) { - avcodec_close (_video_codec_context); - } + 1. One of them starts at time 0. + 2. The first video PTS value ends up on a frame boundary. - if (_subtitle_codec_context) { - avcodec_close (_subtitle_codec_context); - } + Then we remove big initial gaps in PTS and we allow our + insertion of black frames to work. - av_free (_frame); - - avformat_close_input (&_format_context); -} + We will do: + audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset; + video_pts_to_use = video_pts_from_ffmpeg + pts_offset; + */ -void -FFmpegDecoder::setup_general () -{ - av_register_all (); + bool const have_video = video && c->first_video(); + bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio; - if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) { - throw OpenFileError (_film->content_path ()); - } + /* First, make one of them start at 0 */ - if (avformat_find_stream_info (_format_context, 0) < 0) { - throw DecodeError ("could not find stream information"); + if (have_audio && have_video) { + _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get()); + } else if (have_video) { + _pts_offset = - c->first_video().get(); + } else if (have_audio) { + _pts_offset = - c->audio_stream()->first_audio.get(); } - /* Find video, audio and subtitle streams and choose the first of each */ - - 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->channel_layout) - ) - ); - - } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) { - _subtitle_streams.push_back ( - shared_ptr ( - new SubtitleStream (stream_name (s), i) - ) - ); + /* Now adjust both so that the video pts starts on a frame */ + if (have_video && have_audio) { + double first_video = c->first_video().get() + _pts_offset; + double const old_first_video = first_video; + + /* Round the first video up to a frame boundary */ + if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) { + first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate (); } - } - - if (_video_stream < 0) { - throw DecodeError ("could not find video stream"); - } - _frame = avcodec_alloc_frame (); - if (_frame == 0) { - throw DecodeError ("could not allocate frame"); + _pts_offset += first_video - old_first_video; } } -void -FFmpegDecoder::setup_video () +FFmpegDecoder::~FFmpegDecoder () { - _video_codec_context = _format_context->streams[_video_stream]->codec; - _video_codec = avcodec_find_decoder (_video_codec_context->codec_id); + boost::mutex::scoped_lock lm (_mutex); - if (_video_codec == 0) { - throw DecodeError ("could not find video decoder"); - } - - if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) { - throw DecodeError ("could not open video decoder"); + if (_subtitle_codec_context) { + avcodec_close (_subtitle_codec_context); } } void -FFmpegDecoder::setup_audio () +FFmpegDecoder::flush () { - if (!_audio_stream) { - return; + /* Get any remaining frames */ + + _packet.data = 0; + _packet.size = 0; + + /* XXX: should we reset _packet.data and size after each *_decode_* call? */ + + if (_decode_video) { + while (decode_video_packet ()) {} } - - shared_ptr ffa = dynamic_pointer_cast (_audio_stream); - assert (ffa); - _audio_codec_context = _format_context->streams[ffa->id()]->codec; - _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id); - - if (_audio_codec == 0) { - throw DecodeError ("could not find audio decoder"); + if (_ffmpeg_content->audio_stream() && _decode_audio) { + decode_audio_packet (); } - if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) { - throw DecodeError ("could not open audio decoder"); - } + /* Stop us being asked for any more data */ + _video_position = _ffmpeg_content->video_length_after_3d_combine (); + _audio_position = _ffmpeg_content->audio_length (); } void -FFmpegDecoder::setup_subtitle () -{ - if (!_subtitle_stream) { - return; - } - - _subtitle_codec_context = _format_context->streams[_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 ("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 ("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; - - while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - filter_and_emit_video (_frame); + shared_ptr film = _film.lock (); + assert (film); + film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r)); } - 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()); - Audio (deinterleave_audio (_frame->data[0], data_size)); - } - } - - return true; + flush (); + return; } - avcodec_get_frame_defaults (_frame); - - shared_ptr ffa = dynamic_pointer_cast (_audio_stream); - - if (_packet.stream_index == _video_stream) { - - 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 ("Used only %1 bytes of %2 in packet", r, _packet.size)); - } - - if (_opt.video_sync) { - out_with_sync (); - } else { - filter_and_emit_video (_frame); - } - } - - } 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) { - - /* 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); - - /* 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) { - - if (!_first_audio) { - _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()) * ffa->sample_rate ()); - - _film->log()->log ( - String::compose ( - "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) { - shared_ptr audio (new AudioBuffers (ffa->channels(), s)); - audio->make_silent (); - Audio (audio); - } - } + shared_ptr film = _film.lock (); + assert (film); - 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()); - Audio (deinterleave_audio (_frame->data[0], data_size)); - } - } - - } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles && _first_video) { - - int got_subtitle; - AVSubtitle sub; - if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) { - /* Sometimes we get an empty AVSubtitle, which is used by some codecs to - indicate that the previous subtitle should stop. - */ - if (sub.num_rects > 0) { - shared_ptr ts; - try { - emit_subtitle (shared_ptr (new TimedSubtitle (sub))); - } catch (...) { - /* some problem with the subtitle; we probably didn't understand it */ - } - } else { - emit_subtitle (shared_ptr ()); - } - avsubtitle_free (&sub); - } - } + int const si = _packet.stream_index; + if (si == _video_stream && _decode_video) { + decode_video_packet (); + } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) { + decode_audio_packet (); + } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) { + decode_subtitle_packet (); + } + av_free_packet (&_packet); - return false; } +/** @param data pointer to array of pointers to buffers. + * Only the first buffer will be used for non-planar data, otherwise there will be one per channel. + */ shared_ptr -FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) +FFmpegDecoder::deinterleave_audio (uint8_t** data, int size) { - assert (_film->audio_channels()); + assert (_ffmpeg_content->audio_channels()); assert (bytes_per_audio_sample()); - shared_ptr ffa = dynamic_pointer_cast (_audio_stream); - assert (ffa); - /* Deinterleave and convert to float */ - assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0); + assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_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)); + int const frames = total_samples / _ffmpeg_content->audio_channels(); + shared_ptr audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames)); switch (audio_sample_format()) { + case AV_SAMPLE_FMT_U8: + { + uint8_t* p = reinterpret_cast (data[0]); + int sample = 0; + int channel = 0; + for (int i = 0; i < total_samples; ++i) { + audio->data(channel)[sample] = float(*p++) / (1 << 23); + + ++channel; + if (channel == _ffmpeg_content->audio_channels()) { + channel = 0; + ++sample; + } + } + } + break; + case AV_SAMPLE_FMT_S16: { - int16_t* p = reinterpret_cast (data); + int16_t* p = reinterpret_cast (data[0]); 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()) { + if (channel == _ffmpeg_content->audio_channels()) { channel = 0; ++sample; } @@ -387,10 +234,10 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) case AV_SAMPLE_FMT_S16P: { - int16_t* p = reinterpret_cast (data); - for (int i = 0; i < _film->audio_channels(); ++i) { + int16_t** p = reinterpret_cast (data); + for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { for (int j = 0; j < frames; ++j) { - audio->data(i)[j] = static_cast(*p++) / (1 << 15); + audio->data(i)[j] = static_cast(p[i][j]) / (1 << 15); } } } @@ -398,14 +245,14 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) case AV_SAMPLE_FMT_S32: { - int32_t* p = reinterpret_cast (data); + int32_t* p = reinterpret_cast (data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { audio->data(channel)[sample] = static_cast(*p++) / (1 << 31); ++channel; - if (channel == _film->audio_channels()) { + if (channel == _ffmpeg_content->audio_channels()) { channel = 0; ++sample; } @@ -415,14 +262,14 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) case AV_SAMPLE_FMT_FLT: { - float* p = reinterpret_cast (data); + float* p = reinterpret_cast (data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { audio->data(channel)[sample] = *p++; ++channel; - if (channel == _film->audio_channels()) { + if (channel == _ffmpeg_content->audio_channels()) { channel = 0; ++sample; } @@ -432,309 +279,355 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) 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; + float** p = reinterpret_cast (data); + for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) { + memcpy (audio->data(i), p[i], frames * sizeof(float)); } } break; default: - throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast (audio_sample_format()))); + throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast (audio_sample_format()))); } 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; + return audio_codec_context()->sample_fmt; } -libdcp::Size -FFmpegDecoder::native_size () const +int +FFmpegDecoder::bytes_per_audio_sample () const { - return libdcp::Size (_video_codec_context->width, _video_codec_context->height); + return av_get_bytes_per_sample (audio_sample_format ()); } -PixelFormat -FFmpegDecoder::pixel_format () const +void +FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate) { - return _video_codec_context->pix_fmt; -} + double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base); -int -FFmpegDecoder::time_base_numerator () const -{ - return _video_codec_context->time_base.num; -} + /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being + a number plucked from the air) earlier than we want to end up. The loop below + will hopefully then step through to where we want to be. + */ + int initial = frame; -int -FFmpegDecoder::time_base_denominator () const -{ - return _video_codec_context->time_base.den; -} + if (accurate) { + initial -= 5; + } -int -FFmpegDecoder::sample_aspect_ratio_numerator () const -{ - return _video_codec_context->sample_aspect_ratio.num; -} + if (initial < 0) { + initial = 0; + } -int -FFmpegDecoder::sample_aspect_ratio_denominator () const -{ - return _video_codec_context->sample_aspect_ratio.den; -} + /* Initial seek time in the stream's timebase */ + int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base; -string -FFmpegDecoder::stream_name (AVStream* s) const -{ - stringstream n; - - AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0); - if (lang) { - n << lang->value; + av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD); + + avcodec_flush_buffers (video_codec_context()); + if (_subtitle_codec_context) { + avcodec_flush_buffers (_subtitle_codec_context); + } + + /* This !accurate is piling hack upon hack; setting _just_sought to true + even with accurate == true defeats our attempt to align the start + of the video and audio. Here we disable that defeat when accurate == true + i.e. when we are making a DCP rather than just previewing one. + Ewww. This should be gone in 2.0. + */ + if (!accurate) { + _just_sought = true; } - AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0); - if (title) { - if (!n.str().empty()) { - n << " "; - } - n << title->value; + _video_position = frame; + + if (frame == 0 || !accurate) { + /* We're already there, or we're as close as we need to be */ + return; } - if (n.str().empty()) { - n << "unknown"; - } + while (1) { + int r = av_read_frame (_format_context, &_packet); + if (r < 0) { + return; + } - return n.str (); -} + if (_packet.stream_index != _video_stream) { + av_free_packet (&_packet); + continue; + } + + int finished = 0; + r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet); + if (r >= 0 && finished) { + _video_position = rint ( + (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate() + ); -int -FFmpegDecoder::bytes_per_audio_sample () const -{ - return av_get_bytes_per_sample (audio_sample_format ()); + if (_video_position >= (frame - 1)) { + av_free_packet (&_packet); + break; + } + } + + av_free_packet (&_packet); + } } void -FFmpegDecoder::set_audio_stream (shared_ptr s) +FFmpegDecoder::decode_audio_packet () { - AudioDecoder::set_audio_stream (s); - setup_audio (); -} + /* 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) { -void -FFmpegDecoder::set_subtitle_stream (shared_ptr s) -{ - VideoDecoder::set_subtitle_stream (s); - setup_subtitle (); - OutputChanged (); + int frame_finished; + int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet); + if (decode_result < 0) { + shared_ptr film = _film.lock (); + assert (film); + film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result)); + return; + } + + 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); + } + } + + int const data_size = av_samples_get_buffer_size ( + 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1 + ); + + audio (deinterleave_audio (_frame->data, data_size), _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 ("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, int64_t> > images = graph->process (_frame); - 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); -} + for (list, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) { -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); -} + shared_ptr image = i->first; + + if (i->second != AV_NOPTS_VALUE) { -bool -FFmpegDecoder::do_seek (double p, bool backwards) -{ - int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base); + double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset; - 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); - } - - return r < 0; -} + if (_just_sought) { + /* We just did a seek, so disable any attempts to correct for where we + are / should be. + */ + _video_position = rint (pts * _ffmpeg_content->video_frame_rate ()); + _just_sought = false; + } -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)); - } + 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. + */ + + /* XXX: I think this should be a copy of the last frame... */ + boost::shared_ptr black ( + new Image ( + 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; + } - stringstream s (t); - string type; - s >> type; - if (type != "ffmpeg") { - return shared_ptr (); + 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 shared_ptr (new FFmpegAudioStream (t, v)); + return true; } -FFmpegAudioStream::FFmpegAudioStream (string t, optional version) + +void +FFmpegDecoder::setup_subtitle () { - stringstream n (t); + boost::mutex::scoped_lock lm (_mutex); - 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"); + if (!_ffmpeg_content->subtitle_stream()) { + return; } - for (int i = 0; i < name_index; ++i) { - size_t const s = t.find (' '); - if (s != string::npos) { - t = t.substr (s + 1); - } + _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec; + if (_subtitle_codec_context == 0) { + throw DecodeError (N_("could not find subtitle stream")); } - _name = t; -} - -string -FFmpegAudioStream::to_string () const -{ - return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name); -} + _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); -void -FFmpegDecoder::out_with_sync () -{ - /* Where we are in the output, in seconds */ - 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; - } - - /* 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 ( - "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 (_subtitle_codec == 0) { + throw DecodeError (N_("could not find subtitle decoder")); } - 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 ("Frame removed at %1s", out_pts_seconds)); + 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; +} + void -FFmpegDecoder::film_changed (Film::Property p) +FFmpegDecoder::decode_subtitle_packet () { - switch (p) { - case Film::CROP: - case Film::FILTERS: - { - boost::mutex::scoped_lock lm (_filter_graphs_mutex); - _filter_graphs.clear (); + int got_subtitle; + AVSubtitle sub; + if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) { + return; } - OutputChanged (); - break; - default: - break; + /* Sometimes we get an empty AVSubtitle, which is used by some codecs to + indicate that the previous subtitle should stop. + */ + if (sub.num_rects <= 0) { + subtitle (shared_ptr (), dcpomatic::Rect (), 0, 0); + return; + } else if (sub.num_rects > 1) { + throw DecodeError (_("multi-part subtitles not yet supported")); } -} + + /* Subtitle PTS in seconds (within the source, not taking into account any of the + source that we may have chopped off for the DCP) + */ + double const packet_time = (static_cast (sub.pts ) / AV_TIME_BASE) + _pts_offset; -/** @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(); -} + /* hence start time for this sub */ + Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ; + Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ; -double -FFmpegDecoder::frame_time () const -{ - return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base); -} + AVSubtitleRect const * rect = sub.rects[0]; + if (rect->type != SUBTITLE_BITMAP) { + throw DecodeError (_("non-bitmap subtitles not yet supported")); + } + + /* Note RGBA is expressed little-endian, so the first byte in the word is R, second + G, third B, fourth A. + */ + shared_ptr image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true)); + + /* Start of the first line in the subtitle */ + uint8_t* sub_p = rect->pict.data[0]; + /* sub_p looks up into a BGRA palette which is here + (i.e. first byte B, second G, third R, fourth A) + */ + uint32_t const * palette = (uint32_t *) rect->pict.data[1]; + /* Start of the output data */ + uint32_t* out_p = (uint32_t *) image->data()[0]; + + for (int y = 0; y < rect->h; ++y) { + uint8_t* sub_line_p = sub_p; + uint32_t* out_line_p = out_p; + for (int x = 0; x < rect->w; ++x) { + uint32_t const p = palette[*sub_line_p++]; + *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000); + } + sub_p += rect->pict.linesize[0]; + out_p += image->stride()[0] / sizeof (uint32_t); + } + + libdcp::Size const vs = _ffmpeg_content->video_size (); + + subtitle ( + image, + dcpomatic::Rect ( + static_cast (rect->x) / vs.width, + static_cast (rect->y) / vs.height, + static_cast (rect->w) / vs.width, + static_cast (rect->h) / vs.height + ), + from, + to + ); + + + avsubtitle_free (&sub); +}