X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_examiner.cc;h=632508e0ed3670f0e9fa89272751d4d1a9a28dec;hp=cc6a62ff68e180af2118e0e0a23c30e0a6392e52;hb=e29ce33a36c2e20444d57196defc86d5072bce81;hpb=35409488fd54aacefa6858dce4b02a576170e76e diff --git a/src/lib/ffmpeg_examiner.cc b/src/lib/ffmpeg_examiner.cc index cc6a62ff6..632508e0e 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. @@ -18,6 +18,8 @@ */ + +#include "dcpomatic_log.h" #include "ffmpeg_examiner.h" #include "ffmpeg_content.h" #include "job.h" @@ -35,59 +37,67 @@ extern "C" { #include } DCPOMATIC_ENABLE_WARNINGS -#include #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]; -DCPOMATIC_DISABLE_WARNINGS - if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) { + auto s = _format_context->streams[i]; + if (s->codecpar->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); + if (s->codecpar->channel_layout == 0) { + s->codecpar->channel_layout = av_get_default_channel_layout (s->codecpar->channels); } + auto codec = _codec_context[i]->codec; + DCPOMATIC_ASSERT (_format_context->duration != AV_NOPTS_VALUE); - DCPOMATIC_ASSERT (s->codec->codec); - DCPOMATIC_ASSERT (s->codec->codec->name); + DCPOMATIC_ASSERT (codec); + 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, - llrint ((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 ) ); - } 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)); } } @@ -106,12 +116,21 @@ DCPOMATIC_DISABLE_WARNINGS /* Run through until we find: * - the first video. * - the first audio for each stream. + * - 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; } @@ -123,36 +142,45 @@ DCPOMATIC_DISABLE_WARNINGS } } - AVCodecContext* context = _format_context->streams[_packet.stream_index]->codec; -DCPOMATIC_ENABLE_WARNINGS + 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; } } - av_packet_unref (&_packet); + av_packet_free (&packet); - if (_first_video && got_all_audio) { + if (_first_video && got_all_audio && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) { /* All done */ break; } } + 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 */ - AVStream* stream = _format_context->streams[*_video_stream]; - AVDictionaryEntry* rotate_tag = av_dict_get (stream->metadata, "rotate", 0, 0); + 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; @@ -169,57 +197,86 @@ DCPOMATIC_ENABLE_WARNINGS } _rotation = *_rotation - 360 * floor (*_rotation / 360 + 0.9 / 360); - - DCPOMATIC_ASSERT (fabs (*_rotation - 90 * round (*_rotation / 90)) < 2); } -#ifdef DCPOMATIC_VARIANT_SWAROOP - AVDictionaryEntry* e = av_dict_get (_format_context->metadata, SWAROOP_ID_TAG, 0, 0); - if (e) { - _id = e->value; + 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)"); } -#endif } -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; -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()]); - } - if (_need_video_length) { - _video_length = frame_time ( - _format_context->streams[_video_stream.get()] - ).get_value_or (ContentTime ()).frames_round (video_frame_rate().get ()); - } + int r = avcodec_send_packet (context, packet); + if (r < 0 && !(r == AVERROR_EOF && !packet)) { + /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen. + * AVERROR_EOF can happen during flush if we've already sent a flush packet. + */ + throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegExaminer::video_packet"), r); + } + + r = avcodec_receive_frame (context, _frame); + if (r == AVERROR(EAGAIN)) { + /* More input is required */ + return true; + } else if (r == AVERROR_EOF) { + /* No more output is coming */ + return false; + } + + if (!_first_video) { + _first_video = frame_time (_format_context->streams[_video_stream.get()]); + } + if (_need_video_length) { + _video_length = frame_time ( + _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"); } + + 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; -DCPOMATIC_DISABLE_WARNINGS - if (avcodec_decode_audio4 (context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) { -DCPOMATIC_ENABLE_WARNINGS - stream->first_audio = frame_time (stream->stream (_format_context)); + int r = avcodec_send_packet (context, packet); + if (r < 0 && !(r == AVERROR_EOF && !packet) && r != AVERROR(EAGAIN)) { + /* We could cope with AVERROR(EAGAIN) and re-send the packet but I think it should never happen. + * AVERROR_EOF can happen during flush if we've already sent a flush packet. + * EAGAIN means we need to do avcodec_receive_frame, so just carry on and do that. + */ + throw DecodeError (N_("avcodec_send_packet"), N_("FFmpegExaminer::audio_packet"), r); + } + + if (avcodec_receive_frame (context, _frame) < 0) { + return; } + + stream->first_audio = frame_time (stream->stream(_format_context)); } @@ -228,9 +285,7 @@ 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 + int64_t const bet = _frame->best_effort_timestamp; if (bet != AV_NOPTS_VALUE) { t = ContentTime::from_seconds (bet * av_q2d (s->time_base)); } @@ -246,12 +301,14 @@ FFmpegExaminer::video_frame_rate () const 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 @@ -259,22 +316,24 @@ 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 { - string n = stream_name (s); + auto n = stream_name (s); if (n.empty()) { n = _("unknown"); @@ -283,18 +342,19 @@ FFmpegExaminer::subtitle_stream_name (AVStream* s) const return n; } + string FFmpegExaminer::stream_name (AVStream* s) const { 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; } - 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.empty()) { n += " "; @@ -306,18 +366,20 @@ FFmpegExaminer::stream_name (AVStream* s) const return n; } + optional FFmpegExaminer::bits_per_pixel () const { if (video_codec_context()->pix_fmt == -1) { - return optional(); + 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 { @@ -399,21 +461,23 @@ 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 VIDEO_RANGE_VIDEO; + return VideoRange::VIDEO; case AVCOL_RANGE_JPEG: default: - return VIDEO_RANGE_FULL; + return VideoRange::FULL; } }