Use enum class for VideoRange.
[dcpomatic.git] / src / lib / ffmpeg_examiner.cc
index a145ae25c8f2c6fe6ab1d3fde72c8217c37577e4..384c21e9c27288cbfc26b92c7c640985cfcf5a0f 100644 (file)
 
 */
 
-extern "C" {
-#include <libavcodec/avcodec.h>
-#include <libavformat/avformat.h>
-#include <libavutil/pixfmt.h>
-#include <libavutil/pixdesc.h>
-}
+#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 <boost/foreach.hpp>
+#include "warnings.h"
+DCPOMATIC_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/display.h>
+}
+DCPOMATIC_ENABLE_WARNINGS
 #include <iostream>
 
 #include "i18n.h"
@@ -38,19 +43,30 @@ extern "C" {
 using std::string;
 using std::cout;
 using std::max;
-using boost::shared_ptr;
+using std::vector;
+using std::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<const FFmpegContent> c, shared_ptr<Job> 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,
@@ -72,7 +88,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                                                s->codec->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
                                                )
                                        )
@@ -87,7 +103,7 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                /* See if the header has duration information in it */
                _need_video_length = _format_context->duration == 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());
                }
        }
 
@@ -98,9 +114,15 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
        /* 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<bool> 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) {
@@ -116,9 +138,10 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
                }
 
                AVCodecContext* context = _format_context->streams[_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;
@@ -134,24 +157,60 @@ FFmpegExaminer::FFmpegExaminer (shared_ptr<const FFmpegContent> c, shared_ptr<Jo
 
                av_packet_unref (&_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) {
+               /* 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;
+                       }
+               }
+
+               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)");
+       }
 }
 
+
+/** @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()]);
                }
@@ -160,9 +219,14 @@ 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<FFmpegAudioStream> stream)
 {
@@ -171,17 +235,22 @@ FFmpegExaminer::audio_packet (AVCodecContext* context, shared_ptr<FFmpegAudioStr
        }
 
        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));
        }
 }
 
+
 optional<ContentTime>
 FFmpegExaminer::frame_time (AVStream* s) const
 {
        optional<ContentTime> 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));
        }
@@ -189,15 +258,12 @@ FFmpegExaminer::frame_time (AVStream* s) const
        return t;
 }
 
+
 optional<double>
 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
@@ -260,11 +326,11 @@ FFmpegExaminer::stream_name (AVStream* s) const
        return n;
 }
 
-int
+optional<int>
 FFmpegExaminer::bits_per_pixel () const
 {
        if (video_codec_context()->pix_fmt == -1) {
-               throw DecodeError (_("Could not find pixel format for video."));
+               return optional<int>();
        }
 
        AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (video_codec_context()->pix_fmt);
@@ -358,3 +424,16 @@ FFmpegExaminer::has_video () const
 {
        return static_cast<bool> (_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;
+       }
+}