X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_examiner.cc;h=46f4f236e889515eca0254da7f7195be74efb5bb;hb=f3617efde1751ac9e55f692369dc4ba3c6350ca8;hp=18a87f40e591ab0e27d66b88d2677ea75043bae4;hpb=97d25da42455d0ed93c2eebe023883767bb12d53;p=dcpomatic.git diff --git a/src/lib/ffmpeg_examiner.cc b/src/lib/ffmpeg_examiner.cc index 18a87f40e..46f4f236e 100644 --- a/src/lib/ffmpeg_examiner.cc +++ b/src/lib/ffmpeg_examiner.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2013-2015 Carl Hetherington + Copyright (C) 2013-2021 Carl Hetherington This file is part of DCP-o-matic. @@ -18,69 +18,86 @@ */ -extern "C" { -#include -#include -#include -#include -} + +#include "dcpomatic_log.h" #include "ffmpeg_examiner.h" #include "ffmpeg_content.h" #include "job.h" #include "ffmpeg_audio_stream.h" #include "ffmpeg_subtitle_stream.h" #include "util.h" -#include "safe_stringstream.h" -#include +#include +LIBDCP_DISABLE_WARNINGS +extern "C" { +#include +#include +#include +#include +#include +#include +#include +} +LIBDCP_ENABLE_WARNINGS #include #include "i18n.h" -using std::string; + using std::cout; +using std::make_shared; using std::max; -using boost::shared_ptr; +using std::shared_ptr; +using std::string; +using std::vector; using boost::optional; +using namespace dcpomatic; + + +/* This is how many frames from the start of any video that we will examine to see if we + * can spot soft 2:3 pull-down ("telecine"). + */ +static const int PULLDOWN_CHECK_FRAMES = 16; + /** @param job job that the examiner is operating in, or 0 */ FFmpegExaminer::FFmpegExaminer (shared_ptr c, shared_ptr job) : FFmpeg (c) , _video_length (0) , _need_video_length (false) + , _pulldown (false) { /* Find 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_AUDIO) { + auto s = _format_context->streams[i]; + auto codec = _codec_context[i] ? _codec_context[i]->codec : nullptr; + if (s->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && codec) { /* 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); + if (s->codecpar->channel_layout == 0) { + s->codecpar->channel_layout = av_get_default_channel_layout (s->codecpar->channels); } DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE); - DCPOMATIC_ASSERT (s->codec->codec); - DCPOMATIC_ASSERT (s->codec->codec->name); + DCPOMATIC_ASSERT (codec->name); _audio_streams.push_back ( - shared_ptr ( - new FFmpegAudioStream ( - stream_name (s), - s->codec->codec->name, - s->id, - s->codec->sample_rate, - (double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate, - s->codec->channels - ) + make_shared( + stream_name (s), + codec->name, + s->id, + s->codecpar->sample_rate, + llrint ((double(_format_context->duration) / AV_TIME_BASE) * s->codecpar->sample_rate), + s->codecpar->channels, + s->codecpar->bits_per_raw_sample ? s->codecpar->bits_per_raw_sample : s->codecpar->bits_per_coded_sample ) ); - } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) { - _subtitle_streams.push_back (shared_ptr (new FFmpegSubtitleStream (subtitle_stream_name (s), s->id))); + } else if (s->codecpar->codec_type == AVMEDIA_TYPE_SUBTITLE) { + _subtitle_streams.push_back (make_shared(subtitle_stream_name (s), s->id)); } } @@ -88,34 +105,32 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr c, shared_ptrduration == AV_NOPTS_VALUE; if (!_need_video_length) { - _video_length = (double (_format_context->duration) / AV_TIME_BASE) * video_frame_rate().get (); + _video_length = llrint ((double (_format_context->duration) / AV_TIME_BASE) * video_frame_rate().get()); } } - if (job) { - if (_need_video_length) { - job->sub (_("Finding length and subtitles")); - } else if (!_subtitle_streams.empty()) { - job->sub (_("Finding subtitles")); - } else { - job->sub (_("Finding length")); - } + if (job && _need_video_length) { + job->sub (_("Finding length")); } /* Run through until we find: * - the first video. * - the first audio for each stream. - * - the subtitle periods for each stream. - * - * We have to note subtitle periods as otherwise we have no way of knowing - * where we should look for subtitles (video and audio are always present, - * so they are ok). + * - the top-field-first and repeat-first-frame values ("temporal_reference") for the first PULLDOWN_CHECK_FRAMES video frames. */ int64_t const len = _file_group.length (); + /* A string which we build up to describe the top-field-first and repeat-first-frame values for the first few frames. + * It would be nicer to use something like vector here but we want to search the array for a pattern later, + * and a string seems a reasonably neat way to do that. + */ + string temporal_reference; while (true) { - int r = av_read_frame (_format_context, &_packet); + auto packet = av_packet_alloc (); + DCPOMATIC_ASSERT (packet); + int r = av_read_frame (_format_context, packet); if (r < 0) { + av_packet_free (&packet); break; } @@ -127,208 +142,176 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr c, shared_ptrstreams[_packet.stream_index]->codec; + auto context = _codec_context[packet->stream_index]; - if (_video_stream && _packet.stream_index == _video_stream.get()) { - video_packet (context); + if (_video_stream && packet->stream_index == _video_stream.get()) { + video_packet (context, temporal_reference, packet); } bool got_all_audio = true; for (size_t i = 0; i < _audio_streams.size(); ++i) { - if (_audio_streams[i]->uses_index (_format_context, _packet.stream_index)) { - audio_packet (context, _audio_streams[i]); + if (_audio_streams[i]->uses_index(_format_context, packet->stream_index)) { + audio_packet (context, _audio_streams[i], packet); } if (!_audio_streams[i]->first_audio) { got_all_audio = false; } } - for (size_t i = 0; i < _subtitle_streams.size(); ++i) { - if (_subtitle_streams[i]->uses_index (_format_context, _packet.stream_index)) { - subtitle_packet (context, _subtitle_streams[i]); - } - } - - av_packet_unref (&_packet); + av_packet_free (&packet); - if (_first_video && got_all_audio && _subtitle_streams.empty ()) { + if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) { /* All done */ break; } } - /* Finish off any hanging subtitles at the end */ - for (LastSubtitleMap::const_iterator i = _last_subtitle_start.begin(); i != _last_subtitle_start.end(); ++i) { - if (i->second) { - if (i->second->image) { - i->first->add_image_subtitle ( - i->second->id, - ContentTimePeriod ( - i->second->time, - ContentTime::from_frames (video_length(), video_frame_rate().get_value_or (24)) - ) - ); - } else { - i->first->add_text_subtitle ( - i->second->id, - ContentTimePeriod ( - i->second->time, - ContentTime::from_frames (video_length(), video_frame_rate().get_value_or (24)) - ) - ); + if (_video_stream) { + auto context = _codec_context[_video_stream.get()]; + while (video_packet(context, temporal_reference, nullptr)) {} + } + + for (auto i: _audio_streams) { + auto context = _codec_context[i->index(_format_context)]; + audio_packet(context, i, nullptr); + } + + if (_video_stream) { + /* This code taken from get_rotation() in ffmpeg:cmdutils.c */ + auto stream = _format_context->streams[*_video_stream]; + auto rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0); + uint8_t* displaymatrix = av_stream_get_side_data (stream, AV_PKT_DATA_DISPLAYMATRIX, 0); + _rotation = 0; + + if (rotate_tag && *rotate_tag->value && strcmp(rotate_tag->value, "0")) { + char *tail; + _rotation = av_strtod (rotate_tag->value, &tail); + if (*tail) { + _rotation = 0; } } - } - /* We just added subtitles to our streams without taking the PTS offset into account; - this is because we might not know the PTS offset when the first subtitle is seen. - Now we know the PTS offset so we can apply it to those subtitles. - */ - if (has_video() && video_frame_rate()) { - BOOST_FOREACH (shared_ptr i, _subtitle_streams) { - i->add_offset (pts_offset (_audio_streams, _first_video, video_frame_rate().get())); + if (displaymatrix && !_rotation) { + _rotation = - av_display_rotation_get ((int32_t*) displaymatrix); } + + _rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360); + } + + LOG_GENERAL("Temporal reference was %1", temporal_reference); + if (temporal_reference.find("T2T3B2B3T2T3B2B3") != string::npos || temporal_reference.find("B2B3T2T3B2B3T2T3") != string::npos) { + /* The magical sequence (taken from mediainfo) suggests that 2:3 pull-down is in use */ + _pulldown = true; + LOG_GENERAL_NC("Suggest that this may be 2:3 pull-down (soft telecine)"); } } -void -FFmpegExaminer::video_packet (AVCodecContext* context) + +/** @param temporal_reference A string to which we should add two characters per frame; + * the first is T or B depending on whether it's top- or bottom-field first, + * the second is 3 or 2 depending on whether "repeat_pict" is true or not. + * @return true if some video was decoded, otherwise false. + */ +bool +FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference, AVPacket* packet) { DCPOMATIC_ASSERT (_video_stream); - if (_first_video && !_need_video_length) { - return; + if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) { + return false; } - int frame_finished; - if (avcodec_decode_video2 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - if (!_first_video) { - _first_video = frame_time (_format_context->streams[_video_stream.get()]); + bool pending = false; + do { + int r = avcodec_send_packet (context, packet); + if (r < 0) { + LOG_WARNING("avcodec_send_packet returned %1 for a video packet", r); } - if (_need_video_length) { - _video_length = frame_time ( - _format_context->streams[_video_stream.get()] - ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ()); + + /* EAGAIN means we should call avcodec_receive_frame and then re-send the same packet */ + pending = r == AVERROR(EAGAIN); + + r = avcodec_receive_frame (context, _video_frame); + if (r == AVERROR(EAGAIN)) { + /* More input is required */ + return true; + } else if (r == AVERROR_EOF) { + /* No more output is coming */ + return false; } + } while (pending); + + if (!_first_video) { + _first_video = frame_time (_video_frame, _format_context->streams[_video_stream.get()]); + } + if (_need_video_length) { + _video_length = frame_time ( + _video_frame, + _format_context->streams[_video_stream.get()] + ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ()); } + if (temporal_reference.size() < (PULLDOWN_CHECK_FRAMES * 2)) { + temporal_reference += (_video_frame->top_field_first ? "T" : "B"); + temporal_reference += (_video_frame->repeat_pict ? "3" : "2"); + } + + return true; } + void -FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr stream) +FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr stream, AVPacket* packet) { if (stream->first_audio) { return; } - int frame_finished; - if (avcodec_decode_audio4 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { - stream->first_audio = frame_time (stream->stream (_format_context)); + int r = avcodec_send_packet (context, packet); + if (r < 0) { + LOG_WARNING("avcodec_send_packet returned %1 for an audio packet", r); + return; } -} -void -FFmpegExaminer::subtitle_packet (AVCodecContext* context, shared_ptr stream) -{ - int frame_finished; - AVSubtitle sub; - if (avcodec_decode_subtitle2 (context, &sub, &frame_finished, &_packet) >= 0 && frame_finished) { - string id = subtitle_id (sub); - FFmpegSubtitlePeriod const period = subtitle_period (sub); - bool const starts_image = subtitle_starts_image (sub); - - LastSubtitleMap::iterator last = _last_subtitle_start.find (stream); - if (last != _last_subtitle_start.end() && last->second) { - /* We have seen the start of a subtitle but not yet the end. Whatever this is - finishes the previous subtitle, so add it */ - if (last->second->image) { - stream->add_image_subtitle (last->second->id, ContentTimePeriod (last->second->time, period.from)); - } else { - stream->add_text_subtitle (last->second->id, ContentTimePeriod (last->second->time, period.from)); - } - if (sub.num_rects == 0) { - /* This is a `proper' end-of-subtitle */ - _last_subtitle_start[stream] = optional (); - } else { - /* This is just another subtitle, so we start again */ - _last_subtitle_start[stream] = SubtitleStart (id, starts_image, period.from); - } - } else if (sub.num_rects == 1) { - if (period.to) { - if (starts_image) { - stream->add_image_subtitle (id, ContentTimePeriod (period.from, period.to.get ())); - } else { - stream->add_text_subtitle (id, ContentTimePeriod (period.from, period.to.get ())); - } - } else { - _last_subtitle_start[stream] = SubtitleStart (id, starts_image, period.from); - } - } - - for (unsigned int i = 0; i < sub.num_rects; ++i) { - if (sub.rects[i]->type == SUBTITLE_BITMAP) { -#ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT - uint32_t* palette = (uint32_t *) sub.rects[i]->pict.data[1]; - for (int j = 0; j < sub.rects[i]->nb_colors; ++j) { - RGBA rgba ( - (palette[j] & 0x00ff0000) >> 16, - (palette[j] & 0x0000ff00) >> 8, - (palette[j] & 0x000000ff) >> 0, - (palette[j] & 0xff000000) >> 24 - ); - - stream->set_colour (rgba, rgba); - } -#else - uint32_t* palette = (uint32_t *) sub.rects[i]->data[1]; - for (int j = 0; j < sub.rects[i]->nb_colors; ++j) { - RGBA rgba ( - (palette[j] & 0x00ff0000) >> 16, - (palette[j] & 0x0000ff00) >> 8, - (palette[j] & 0x000000ff) >> 0, - (palette[j] & 0xff000000) >> 24 - ); - - stream->set_colour (rgba, rgba); - } -#endif - } - } + auto frame = audio_frame (stream); - avsubtitle_free (&sub); + if (avcodec_receive_frame (context, frame) < 0) { + return; } + + stream->first_audio = frame_time (frame, stream->stream(_format_context)); } + optional -FFmpegExaminer::frame_time (AVStream* s) const +FFmpegExaminer::frame_time (AVFrame* frame, AVStream* stream) const { optional t; - int64_t const bet = av_frame_get_best_effort_timestamp (_frame); + int64_t const bet = frame->best_effort_timestamp; if (bet != AV_NOPTS_VALUE) { - t = ContentTime::from_seconds (bet * av_q2d (s->time_base)); + t = ContentTime::from_seconds (bet * av_q2d(stream->time_base)); } return t; } + optional FFmpegExaminer::video_frame_rate () const { DCPOMATIC_ASSERT (_video_stream); - /* This use of r_frame_rate is debateable; there's a few different - * frame rates in the format context, but this one seems to be the most - * reliable. - */ - return av_q2d (av_stream_get_r_frame_rate (_format_context->streams[_video_stream.get()])); + return av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[_video_stream.get()], 0)); } + dcp::Size FFmpegExaminer::video_size () const { return dcp::Size (video_codec_context()->width, video_codec_context()->height); } + /** @return Length according to our content's header */ Frame FFmpegExaminer::video_length () const @@ -336,67 +319,70 @@ FFmpegExaminer::video_length () const return max (Frame (1), _video_length); } + optional FFmpegExaminer::sample_aspect_ratio () const { DCPOMATIC_ASSERT (_video_stream); - AVRational sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0); + auto sar = av_guess_sample_aspect_ratio (_format_context, _format_context->streams[_video_stream.get()], 0); if (sar.num == 0) { /* I assume this means that we don't know */ - return optional (); + return {}; } return double (sar.num) / sar.den; } + string FFmpegExaminer::subtitle_stream_name (AVStream* s) const { - SafeStringStream n; - - n << stream_name (s); + auto n = stream_name (s); - if (n.str().empty()) { - n << _("unknown"); + if (n.empty()) { + n = _("unknown"); } - return n.str (); + return n; } + string FFmpegExaminer::stream_name (AVStream* s) const { - SafeStringStream n; + string n; if (s->metadata) { - AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0); + auto const lang = av_dict_get (s->metadata, "language", 0, 0); if (lang) { - n << lang->value; + n = lang->value; } - AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0); + auto const title = av_dict_get (s->metadata, "title", 0, 0); if (title) { - if (!n.str().empty()) { - n << " "; + if (!n.empty()) { + n += " "; } - n << title->value; + n += title->value; } } - return n.str (); + return n; } -int + +optional FFmpegExaminer::bits_per_pixel () const { if (video_codec_context()->pix_fmt == -1) { - throw DecodeError (_("Could not find pixel format for video.")); + return {}; } - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt); + auto const d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt); DCPOMATIC_ASSERT (d); return av_get_bits_per_pixel (d); } + bool FFmpegExaminer::yuv () const { @@ -478,8 +464,33 @@ FFmpegExaminer::yuv () const } } + bool FFmpegExaminer::has_video () const { - return static_cast (_video_stream); + return static_cast(_video_stream); } + + +VideoRange +FFmpegExaminer::range () const +{ + switch (color_range()) { + case AVCOL_RANGE_MPEG: + case AVCOL_RANGE_UNSPECIFIED: + return VideoRange::VIDEO; + case AVCOL_RANGE_JPEG: + default: + return VideoRange::FULL; + } +} + + +PixelQuanta +FFmpegExaminer::pixel_quanta () const +{ + auto const desc = av_pix_fmt_desc_get(video_codec_context()->pix_fmt); + DCPOMATIC_ASSERT (desc); + return { 1 << desc->log2_chroma_w, 1 << desc->log2_chroma_h }; +} +