X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_examiner.cc;h=5683555e4aaffc42b3bce4493b17e9a5f1f3a72b;hb=5f1fdbafc6eef37523250e0b8542a8939a038823;hp=544cc11b91fc545d2d74426b5ab372db4933fdd1;hpb=422be0eece2bf6ee80db1d3c21553cd82efff789;p=dcpomatic.git diff --git a/src/lib/ffmpeg_examiner.cc b/src/lib/ffmpeg_examiner.cc index 544cc11b9..5683555e4 100644 --- a/src/lib/ffmpeg_examiner.cc +++ b/src/lib/ffmpeg_examiner.cc @@ -18,19 +18,24 @@ */ -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 +#include "warnings.h" +DCPOMATIC_DISABLE_WARNINGS +extern "C" { +#include +#include +#include +#include +#include +#include +} +DCPOMATIC_ENABLE_WARNINGS #include #include @@ -39,19 +44,30 @@ extern "C" { using std::string; using std::cout; using std::max; +using std::vector; using boost::shared_ptr; 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]; +DCPOMATIC_DISABLE_WARNINGS 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, @@ -73,7 +89,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr c, shared_ptrcodec->codec->name, s->id, s->codec->sample_rate, - (double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate, + llrint ((double (_format_context->duration) / AV_TIME_BASE) * s->codec->sample_rate), s->codec->channels ) ) @@ -88,31 +104,26 @@ 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); if (r < 0) { @@ -128,9 +139,10 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr c, shared_ptrstreams[_packet.stream_index]->codec; +DCPOMATIC_ENABLE_WARNINGS if (_video_stream && _packet.stream_index == _video_stream.get()) { - video_packet (context); + video_packet (context, temporal_reference); } bool got_all_audio = true; @@ -144,65 +156,71 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr c, shared_ptruses_index (_format_context, _packet.stream_index)) { - subtitle_packet (context, _subtitle_streams[i]); - } - } - av_packet_unref (&_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) { + /* This code taken from get_rotation() in ffmpeg:cmdutils.c */ + AVStream* stream = _format_context->streams[*_video_stream]; + AVDictionaryEntry* 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); + + DCPOMATIC_ASSERT (fabs (*_rotation - 90 * round (*_rotation / 90)) < 2); + } + + 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)"); + } + +#ifdef DCPOMATIC_VARIANT_SWAROOP + AVDictionaryEntry* e = av_dict_get (_format_context->metadata, SWAROOP_ID_TAG, 0, 0); + if (e) { + _id = e->value; } +#endif } + +/** @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, + * ths seconds is 3 or 2 depending on whether "repeat_pict" is true or not. + */ void -FFmpegExaminer::video_packet (AVCodecContext* context) +FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference) { DCPOMATIC_ASSERT (_video_stream); - if (_first_video && !_need_video_length) { + if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) { return; } int frame_finished; +DCPOMATIC_DISABLE_WARNINGS if (avcodec_decode_video2 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { +DCPOMATIC_ENABLE_WARNINGS if (!_first_video) { _first_video = frame_time (_format_context->streams[_video_stream.get()]); } @@ -211,9 +229,14 @@ FFmpegExaminer::video_packet (AVCodecContext* context) _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 += (_frame->top_field_first ? "T" : "B"); + temporal_reference += (_frame->repeat_pict ? "3" : "2"); + } } } + void FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr stream) { @@ -222,89 +245,22 @@ FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr= 0 && frame_finished) { +DCPOMATIC_ENABLE_WARNINGS stream->first_audio = frame_time (stream->stream (_format_context)); } } -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 - } - } - - avsubtitle_free (&sub); - } -} optional FFmpegExaminer::frame_time (AVStream* s) const { optional t; +DCPOMATIC_DISABLE_WARNINGS int64_t const bet = av_frame_get_best_effort_timestamp (_frame); +DCPOMATIC_ENABLE_WARNINGS if (bet != AV_NOPTS_VALUE) { t = ContentTime::from_seconds (bet * av_q2d (s->time_base)); } @@ -312,15 +268,12 @@ FFmpegExaminer::frame_time (AVStream* s) const 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 @@ -351,45 +304,43 @@ FFmpegExaminer::sample_aspect_ratio () const string FFmpegExaminer::subtitle_stream_name (AVStream* s) const { - locked_stringstream n; - - n << stream_name (s); + string 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 { - locked_stringstream n; + string n; if (s->metadata) { AVDictionaryEntry 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); 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 optional(); } AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt); @@ -483,3 +434,16 @@ FFmpegExaminer::has_video () const { return static_cast (_video_stream); } + +VideoRange +FFmpegExaminer::range () const +{ + switch (color_range()) { + case AVCOL_RANGE_MPEG: + case AVCOL_RANGE_UNSPECIFIED: + return VIDEO_RANGE_VIDEO; + case AVCOL_RANGE_JPEG: + default: + return VIDEO_RANGE_FULL; + } +}