wip: hacks
[dcpomatic.git] / src / lib / ffmpeg_examiner.cc
index 632508e0ed3670f0e9fa89272751d4d1a9a28dec..2abeec4937f8ce1d84b956551cc1abf9c6160f01 100644 (file)
 #include "ffmpeg_audio_stream.h"
 #include "ffmpeg_subtitle_stream.h"
 #include "util.h"
-#include "warnings.h"
-DCPOMATIC_DISABLE_WARNINGS
+#include <dcp/warnings.h>
+LIBDCP_DISABLE_WARNINGS
 extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 #include <libavutil/pixfmt.h>
 #include <libavutil/pixdesc.h>
-#include <libavutil/eval.h>
+#include <libavutil/channel_layout.h>
 #include <libavutil/display.h>
+#include <libavutil/eval.h>
 }
-DCPOMATIC_ENABLE_WARNINGS
+LIBDCP_ENABLE_WARNINGS
 #include <iostream>
 
 #include "i18n.h"
@@ -65,11 +66,19 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
        , _need_video_length (false)
        , _pulldown (false)
 {
-       /* Find audio and subtitle streams */
+       /* Find streams */
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
                auto s = _format_context->streams[i];
-               if (s->codecpar->codec_type == AVMEDIA_TYPE_AUDIO) {
+               auto codec = _codec_context[i] ? _codec_context[i]->codec : nullptr;
+               if (s->codecpar->codec_type == AVMEDIA_TYPE_VIDEO) {
+                       _video_streams.push_back(
+                               make_shared<VideoStream>(
+                                       stream_name(s),
+                                       s->id,
+                                       av_q2d(av_guess_frame_rate(_format_context, _format_context->streams[i], 0))
+                                       ));
+               } else 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.
@@ -79,10 +88,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                                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 (codec);
                        DCPOMATIC_ASSERT (codec->name);
 
                        _audio_streams.push_back (
@@ -109,6 +115,30 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                }
        }
 
+       /* Choose the best video stream */
+       using ScoredStream = pair<shared_ptr<FFmpegVideoStream>, int>;
+       vector<ScoredStream> streams_with_scores;
+       for (auto stream: _video_streams) {
+               auto s = stream->stream(_format_context);
+               auto const frame_rate = av_q2d(s->avg_frame_rate);
+               int score = 0;
+               if (1 < frame_rate && frame_rate < 1000) {
+                       ++score;
+               }
+               if (s->avg_frame_rate.num > 0 && s->avg_frame_rate.den > 0) {
+                       ++score;
+               }
+               streams_with_scores.push_back({stream, score});
+       }
+
+       streams_with_scores.sort([](ScoredStream const& a, ScoredStream const& b) {
+               return a.second > b.second;
+       });
+
+       if (!streams_with_scores.empty()) {
+               _best_video_stream = streams_with_scores[0].first;
+       }
+
        if (job && _need_video_length) {
                job->sub (_("Finding length"));
        }
@@ -144,7 +174,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 
                auto context = _codec_context[packet->stream_index];
 
-               if (_video_stream && packet->stream_index == _video_stream.get()) {
+               if (_best_video_stream && packet->stream_index == _best_video_stream->index(_format_context)) {
                        video_packet (context, temporal_reference, packet);
                }
 
@@ -167,8 +197,8 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                }
        }
 
-       if (_video_stream) {
-               auto context = _codec_context[_video_stream.get()];
+       if (_best_video_stream) {
+               auto context = _codec_context[_best_video_stream->index(_format_context)];
                while (video_packet(context, temporal_reference, nullptr)) {}
        }
 
@@ -177,9 +207,9 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                audio_packet(context, i, nullptr);
        }
 
-       if (_video_stream) {
+       if (_best_video_stream) {
                /* This code taken from get_rotation() in ffmpeg:cmdutils.c */
-               auto stream = _format_context->streams[*_video_stream];
+               auto stream = _best_video_stream->stream(_format_context);
                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;
@@ -216,40 +246,44 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 bool
 FFmpegExaminer::video_packet (AVCodecContext* context, string& temporal_reference, AVPacket* packet)
 {
-       DCPOMATIC_ASSERT (_video_stream);
+       DCPOMATIC_ASSERT(_best_video_stream);
 
        if (_first_video && !_need_video_length && temporal_reference.size() >= (PULLDOWN_CHECK_FRAMES * 2)) {
                return false;
        }
 
-       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);
-       }
+       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);
+               }
 
-       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;
-       }
+               /* 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 (_format_context->streams[_video_stream.get()]);
+               _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 += (_frame->top_field_first ? "T" : "B");
-               temporal_reference += (_frame->repeat_pict ? "3" : "2");
+               temporal_reference += (_video_frame->top_field_first ? "T" : "B");
+               temporal_reference += (_video_frame->repeat_pict ? "3" : "2");
        }
 
        return true;
@@ -264,30 +298,29 @@ FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStr
        }
 
        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 (r < 0) {
+               LOG_WARNING("avcodec_send_packet returned %1 for an audio packet", r);
+               return;
        }
 
-       if (avcodec_receive_frame (context, _frame) < 0) {
+       auto frame = audio_frame (stream);
+
+       if (avcodec_receive_frame (context, frame) < 0) {
                return;
        }
 
-       stream->first_audio = frame_time (stream->stream(_format_context));
+       stream->first_audio = frame_time (frame, stream->stream(_format_context));
 }
 
 
 optional<ContentTime>
-FFmpegExaminer::frame_time (AVStream* s) const
+FFmpegExaminer::frame_time (AVFrame* frame, AVStream* stream) const
 {
        optional<ContentTime> t;
 
-       int64_t const bet = _frame->best_effort_timestamp;
+       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;
@@ -465,7 +498,7 @@ FFmpegExaminer::yuv () const
 bool
 FFmpegExaminer::has_video () const
 {
-       return static_cast<bool>(_video_stream);
+       return !_video_streams.empty();
 }
 
 
@@ -481,3 +514,13 @@ FFmpegExaminer::range () const
                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 };
+}
+