X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=8834f28ed1c4f90152589bc2f200a9b4c49b9069;hb=0d3f5aae5b99832b7c5d55f32f5bccb365caa3fd;hp=3c607a7fb260f741392b42d221177e9857a72848;hpb=513ee257dfcac679b346a872e4446ab2ef453f67;p=dcpomatic.git diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc index 3c607a7fb..8834f28ed 100644 --- a/src/lib/ffmpeg_decoder.cc +++ b/src/lib/ffmpeg_decoder.cc @@ -50,6 +50,8 @@ extern "C" { #include "filter_graph.h" #include "subtitle.h" +#include "i18n.h" + using std::cout; using std::string; using std::vector; @@ -58,11 +60,12 @@ using std::list; using boost::shared_ptr; using boost::optional; using boost::dynamic_pointer_cast; +using libdcp::Size; -FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, Job* j) - : Decoder (f, o, j) - , VideoDecoder (f, o, j) - , AudioDecoder (f, o, j) +FFmpegDecoder::FFmpegDecoder (shared_ptr f, DecodeOptions o) + : Decoder (f, o) + , VideoDecoder (f, o) + , AudioDecoder (f, o) , _format_context (0) , _video_stream (-1) , _frame (0) @@ -77,6 +80,10 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr f, shared_ptr o, J setup_video (); setup_audio (); setup_subtitle (); + + if (!o.video_sync) { + _first_video = 0; + } } FFmpegDecoder::~FFmpegDecoder () @@ -108,7 +115,7 @@ FFmpegDecoder::setup_general () } if (avformat_find_stream_info (_format_context, 0) < 0) { - throw DecodeError ("could not find stream information"); + throw DecodeError (_("could not find stream information")); } /* Find video, audio and subtitle streams and choose the first of each */ @@ -122,7 +129,7 @@ FFmpegDecoder::setup_general () /* 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); } @@ -143,12 +150,12 @@ FFmpegDecoder::setup_general () } if (_video_stream < 0) { - throw DecodeError ("could not find video stream"); + throw DecodeError (N_("could not find video stream")); } _frame = avcodec_alloc_frame (); if (_frame == 0) { - throw DecodeError ("could not allocate frame"); + throw DecodeError (N_("could not allocate frame")); } } @@ -159,18 +166,11 @@ FFmpegDecoder::setup_video () _video_codec = avcodec_find_decoder (_video_codec_context->codec_id); if (_video_codec == 0) { - throw DecodeError ("could not find video decoder"); + throw DecodeError (_("could not find video decoder")); } - /* I think this prevents problems with green hash on decodes and - "changing frame properties on the fly is not supported by all filters" - messages with some content. Although I'm not sure; needs checking. - */ - AVDictionary* opts = 0; - av_dict_set (&opts, "threads", "1", 0); - - if (avcodec_open2 (_video_codec_context, _video_codec, &opts) < 0) { - throw DecodeError ("could not open video decoder"); + if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) { + throw DecodeError (N_("could not open video decoder")); } } @@ -188,11 +188,11 @@ FFmpegDecoder::setup_audio () _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id); if (_audio_codec == 0) { - throw DecodeError ("could not find audio decoder"); + throw DecodeError (_("could not find audio decoder")); } if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) { - throw DecodeError ("could not open audio decoder"); + throw DecodeError (N_("could not open audio decoder")); } } @@ -207,11 +207,11 @@ FFmpegDecoder::setup_subtitle () _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id); if (_subtitle_codec == 0) { - throw DecodeError ("could not find subtitle decoder"); + throw DecodeError (_("could not find subtitle decoder")); } if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) { - throw DecodeError ("could not open subtitle decoder"); + throw DecodeError (N_("could not open subtitle decoder")); } } @@ -226,7 +226,7 @@ FFmpegDecoder::pass () /* 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)); + _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r)); } /* Get any remaining frames */ @@ -242,14 +242,14 @@ FFmpegDecoder::pass () filter_and_emit_video (_frame); } - if (_audio_stream && _opt->decode_audio) { + 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)); + Audio (deinterleave_audio (_frame->data, data_size)); } } @@ -259,7 +259,7 @@ FFmpegDecoder::pass () avcodec_get_frame_defaults (_frame); shared_ptr ffa = dynamic_pointer_cast (_audio_stream); - + if (_packet.stream_index == _video_stream) { int frame_finished; @@ -267,53 +267,17 @@ FFmpegDecoder::pass () 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)); + _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size)); } - /* 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 (delta > -one_frame) { - /* Process this frame */ - filter_and_emit_video (_frame); + if (_opt.video_sync) { + out_with_sync (); } else { - /* Otherwise we are omitting a frame to keep things right */ - _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds)); + filter_and_emit_video (_frame); } } - } else if (ffa && _packet.stream_index == ffa->id() && _opt->decode_audio) { + } 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) { @@ -341,7 +305,7 @@ FFmpegDecoder::pass () _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)", + 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, ffa->channels(), bytes_per_audio_sample() ) ); @@ -358,11 +322,11 @@ FFmpegDecoder::pass () ); assert (_audio_codec_context->channels == _film->audio_channels()); - Audio (deinterleave_audio (_frame->data[0], data_size)); + Audio (deinterleave_audio (_frame->data, data_size)); } } - } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt->decode_subtitles && _first_video) { + } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt.decode_subtitles && _first_video) { int got_subtitle; AVSubtitle sub; @@ -373,7 +337,7 @@ FFmpegDecoder::pass () if (sub.num_rects > 0) { shared_ptr ts; try { - emit_subtitle (shared_ptr (new TimedSubtitle (sub, _first_video.get()))); + emit_subtitle (shared_ptr (new TimedSubtitle (sub))); } catch (...) { /* some problem with the subtitle; we probably didn't understand it */ } @@ -388,8 +352,11 @@ FFmpegDecoder::pass () 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 (bytes_per_audio_sample()); @@ -408,7 +375,7 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) switch (audio_sample_format()) { case AV_SAMPLE_FMT_S16: { - int16_t* p = (int16_t *) data; + int16_t* p = reinterpret_cast (data[0]); int sample = 0; int channel = 0; for (int i = 0; i < total_samples; ++i) { @@ -423,13 +390,24 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) } break; + case AV_SAMPLE_FMT_S16P: + { + int16_t** p = reinterpret_cast (data); + for (int i = 0; i < _film->audio_channels(); ++i) { + for (int j = 0; j < frames; ++j) { + audio->data(i)[j] = static_cast(p[i][j]) / (1 << 15); + } + } + } + break; + case AV_SAMPLE_FMT_S32: { - int32_t* p = (int32_t *) 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] = float(*p++) / (1 << 31); + audio->data(channel)[sample] = static_cast(*p++) / (1 << 31); ++channel; if (channel == _film->audio_channels()) { @@ -442,7 +420,7 @@ 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) { @@ -459,16 +437,15 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size) case AV_SAMPLE_FMT_FLTP: { - float* p = reinterpret_cast (data); + float** p = reinterpret_cast (data); for (int i = 0; i < _film->audio_channels(); ++i) { - memcpy (audio->data(i), p, frames * sizeof(float)); - p += frames; + 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; @@ -496,10 +473,10 @@ FFmpegDecoder::audio_sample_format () const return _audio_codec_context->sample_fmt; } -Size +libdcp::Size FFmpegDecoder::native_size () const { - return Size (_video_codec_context->width, _video_codec_context->height); + return libdcp::Size (_video_codec_context->width, _video_codec_context->height); } PixelFormat @@ -537,21 +514,21 @@ FFmpegDecoder::stream_name (AVStream* s) const { stringstream n; - AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0); + 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, "title", 0, 0); + AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0); if (title) { if (!n.str().empty()) { - n << " "; + n << N_(" "); } n << title->value; } if (n.str().empty()) { - n << "unknown"; + n << N_("unknown"); } return n.str (); @@ -575,22 +552,25 @@ FFmpegDecoder::set_subtitle_stream (shared_ptr s) { VideoDecoder::set_subtitle_stream (s); setup_subtitle (); + OutputChanged (); } void FFmpegDecoder::filter_and_emit_video (AVFrame* frame) { + 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 (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, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format)); + 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 ("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; } @@ -598,8 +578,39 @@ FFmpegDecoder::filter_and_emit_video (AVFrame* frame) list > images = graph->process (frame); for (list >::iterator i = images.begin(); i != images.end(); ++i) { - emit_video (*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); + + 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; } shared_ptr @@ -613,7 +624,7 @@ FFmpegAudioStream::create (string t, optional v) stringstream s (t); string type; s >> type; - if (type != "ffmpeg") { + if (type != N_("ffmpeg")) { return shared_ptr (); } @@ -635,7 +646,7 @@ FFmpegAudioStream::FFmpegAudioStream (string t, optional version) string type; /* Current (marked version 1) */ n >> type >> _id >> _sample_rate >> _channel_layout; - assert (type == "ffmpeg"); + assert (type == N_("ffmpeg")); } for (int i = 0; i < name_index; ++i) { @@ -651,6 +662,83 @@ FFmpegAudioStream::FFmpegAudioStream (string t, optional version) string FFmpegAudioStream::to_string () const { - return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name); + return String::compose (N_("ffmpeg %1 %2 %3 %4"), _id, _sample_rate, _channel_layout, _name); +} + +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 (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)); + } +} + +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 (); + } + OutputChanged (); + break; + + default: + break; + } +} + +/** @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(); +} + +double +FFmpegDecoder::frame_time () const +{ + return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base); }