Various fixes to PTS mangling.

diff --git a/src/lib/ffmpeg_content.cc b/src/lib/ffmpeg_content.cc
index 88fc6dbd237c3aa09a020c474e91250f24865201..f4e1b9e72b44be79d63a759d80ec00e24735e559 100644 (file)
--- a/src/lib/ffmpeg_content.cc
+++ b/src/lib/ffmpeg_content.cc
@@ -79,7 +79,7 @@ FFmpegContent::FFmpegContent (shared_ptr<const Film> f, shared_ptr<const cxml::N
                _filters.push_back (Filter::from_id ((*i)->content ()));
        }
 
-       _first_video = node->optional_number_child<Time> ("FirstVideo");
+       _first_video = node->optional_number_child<double> ("FirstVideo");
 }
 
 FFmpegContent::FFmpegContent (FFmpegContent const & o)

diff --git a/src/lib/ffmpeg_content.h b/src/lib/ffmpeg_content.h
index 69b459fb65b0960413ffb0f068da95f945124afc..dba06990b8567313421691396bf236206b578248 100644 (file)
--- a/src/lib/ffmpeg_content.h
+++ b/src/lib/ffmpeg_content.h
@@ -26,6 +26,7 @@
 #include "subtitle_content.h"
 
 class Filter;
+class ffmpeg_pts_offset_test;
 
 class FFmpegAudioStream
 {
@@ -48,6 +49,11 @@ public:
        int channels;
        AudioMapping mapping;
        boost::optional<double> first_audio;
+
+private:
+       friend class ffmpeg_pts_offset_test;
+       /* Constructor for tests */
+       FFmpegAudioStream () {}
 };
 
 extern bool operator== (FFmpegAudioStream const & a, FFmpegAudioStream const & b);
@@ -138,17 +144,19 @@ public:
        void set_subtitle_stream (boost::shared_ptr<FFmpegSubtitleStream>);
        void set_audio_stream (boost::shared_ptr<FFmpegAudioStream>);
 
-       boost::optional<Time> first_video () const {
+       boost::optional<double> first_video () const {
                boost::mutex::scoped_lock lm (_mutex);
                return _first_video;
        }
        
 private:
+       friend class ffmpeg_pts_offset_test;
+       
        std::vector<boost::shared_ptr<FFmpegSubtitleStream> > _subtitle_streams;
        boost::shared_ptr<FFmpegSubtitleStream> _subtitle_stream;
        std::vector<boost::shared_ptr<FFmpegAudioStream> > _audio_streams;
        boost::shared_ptr<FFmpegAudioStream> _audio_stream;
-       boost::optional<Time> _first_video;
+       boost::optional<double> _first_video;
        /** Video filters that should be used when generating DCPs */
        std::vector<Filter const *> _filters;
 };

diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc
index c77cb71c0ac9ca465415f17b43556c720e6c16f2..201fac8aa5a74dabd82ee4e3b577de4b9e239cd1 100644 (file)
--- a/src/lib/ffmpeg_decoder.cc
+++ b/src/lib/ffmpeg_decoder.cc
@@ -51,6 +51,7 @@ using std::vector;
 using std::stringstream;
 using std::list;
 using std::min;
+using std::pair;
 using boost::shared_ptr;
 using boost::optional;
 using boost::dynamic_pointer_cast;
@@ -66,28 +67,50 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
        , _subtitle_codec (0)
        , _decode_video (video)
        , _decode_audio (audio)
-       , _pts_offset (0)
+       , _video_pts_offset (0)
+       , _audio_pts_offset (0)
        , _just_sought (false)
 {
        setup_subtitle ();
 
-       if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) {
-               _pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate());
-       }
-}
+       /* Audio and video frame PTS values may not start with 0.  We want
+          to fiddle them so that:
 
-double
-FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
-{
-       double const old_first_video = first_video;
-       
-       /* Round the first video to a frame boundary */
-       if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) {
-               first_video = ceil (first_video * video_frame_rate) / video_frame_rate;
+          1.  One of them starts at time 0.
+          2.  The first video PTS value ends up on a frame boundary.
+
+          Then we remove big initial gaps in PTS and we allow our
+          insertion of black frames to work.
+
+          We will do:
+            audio_pts_to_use = audio_pts_from_ffmpeg + audio_pts_offset;
+            video_pts_to_use = video_pts_from_ffmpeg + video_pts_offset;
+       */
+
+       bool const have_video = video && c->first_video();
+       bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
+
+       /* First, make one of them start at 0 */
+
+       if (have_audio && have_video) {
+               _video_pts_offset = _audio_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
+       } else if (have_video) {
+               _video_pts_offset = - c->first_video().get();
        }
 
-       /* Compute the required offset (also removing any common start delay) */
-       return first_video - old_first_video - min (first_video, first_audio);
+       /* Now adjust both so that the video pts starts on a frame */
+       if (have_video && have_audio) {
+               double first_video = c->first_video().get() + _video_pts_offset;
+               double const old_first_video = first_video;
+               
+               /* Round the first video up to a frame boundary */
+               if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
+                       first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
+               }
+
+               _video_pts_offset += first_video - old_first_video;
+               _audio_pts_offset += first_video - old_first_video;
+       }
 }
 
 FFmpegDecoder::~FFmpegDecoder ()
@@ -274,11 +297,15 @@ FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
                initial -= 5;
        }
 
+       if (initial < 0) {
+               initial = 0;
+       }
+
        /* Initial seek time in the stream's timebase */
-       int64_t const initial_vt = initial / (_ffmpeg_content->video_frame_rate() * time_base);
+       int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base;
        /* Wanted final seek time in the stream's timebase */
-       int64_t const final_vt = frame / (_ffmpeg_content->video_frame_rate() * time_base);
-       
+       int64_t const final_vt = ((frame / _ffmpeg_content->video_frame_rate()) - _video_pts_offset) / time_base;
+
        av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
 
        avcodec_flush_buffers (video_codec_context());
@@ -309,7 +336,7 @@ FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
                                        int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
                                        if (bet >= final_vt) {
                                                _video_position = rint (
-                                                       (bet * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
+                                                       (bet * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate()
                                                        );
                                                av_free_packet (&_packet);
                                                break;
@@ -341,7 +368,7 @@ FFmpegDecoder::decode_audio_packet ()
                                if (_audio_position == 0) {
                                        /* Where we are in the source, in seconds */
                                        double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
-                                               * av_frame_get_best_effort_timestamp(_frame) - _pts_offset;
+                                               * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset;
 
                                        if (pts > 0) {
                                                /* Emit some silence */
@@ -393,21 +420,20 @@ FFmpegDecoder::decode_video_packet ()
                graph = *i;
        }
 
-       list<shared_ptr<Image> > images = graph->process (_frame);
+       list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
 
        string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
        
-       for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
+       for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
 
-               shared_ptr<Image> image = *i;
+               shared_ptr<Image> image = i->first;
                if (!post_process.empty ()) {
                        image = image->post_process (post_process, true);
                }
                
-               int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
-               if (bet != AV_NOPTS_VALUE) {
+               if (i->second != AV_NOPTS_VALUE) {
 
-                       double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset;
+                       double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset;
 
                        if (_just_sought) {
                                /* We just did a seek, so disable any attempts to correct for where we

diff --git a/src/lib/ffmpeg_decoder.h b/src/lib/ffmpeg_decoder.h
index 8819954db0ef06c02c810b8647b6ea6719607c36..73edcccb43ec562162825f4e82b3331d9abceaa6 100644 (file)
--- a/src/lib/ffmpeg_decoder.h
+++ b/src/lib/ffmpeg_decoder.h
@@ -84,6 +84,7 @@ private:
        bool _decode_video;
        bool _decode_audio;
 
-       double _pts_offset;
+       double _video_pts_offset;
+       double _audio_pts_offset;
        bool _just_sought;
 };

diff --git a/src/lib/filter_graph.cc b/src/lib/filter_graph.cc
index 3366f8d1b627c4dbc551763c9d64525df9c226c2..7c4df8903e18e48d11abbef510b8c5ca548f4a2d 100644 (file)
--- a/src/lib/filter_graph.cc
+++ b/src/lib/filter_graph.cc
@@ -39,6 +39,8 @@ extern "C" {
 using std::stringstream;
 using std::string;
 using std::list;
+using std::pair;
+using std::make_pair;
 using std::cout;
 using boost::shared_ptr;
 using boost::weak_ptr;
@@ -132,10 +134,10 @@ FilterGraph::~FilterGraph ()
 /** Take an AVFrame and process it using our configured filters, returning a
  *  set of Images.  Caller handles memory management of the input frame.
  */
-list<shared_ptr<Image> >
+list<pair<shared_ptr<Image>, int64_t> >
 FilterGraph::process (AVFrame* frame)
 {
-       list<shared_ptr<Image> > images;
+       list<pair<shared_ptr<Image>, int64_t> > images;
 
        if (av_buffersrc_write_frame (_buffer_src_context, frame) < 0) {
                throw DecodeError (N_("could not push buffer into filter chain."));
@@ -146,7 +148,7 @@ FilterGraph::process (AVFrame* frame)
                        break;
                }
 
-               images.push_back (shared_ptr<Image> (new Image (_frame)));
+               images.push_back (make_pair (shared_ptr<Image> (new Image (_frame)), av_frame_get_best_effort_timestamp (_frame)));
                av_frame_unref (_frame);
        }
        

diff --git a/src/lib/filter_graph.h b/src/lib/filter_graph.h
index e294812c2baa9836f5aeff99711057bae611148e..4942907959d8573d814d56e4d26e303325b67b47 100644 (file)
--- a/src/lib/filter_graph.h
+++ b/src/lib/filter_graph.h
@@ -39,7 +39,7 @@ public:
        ~FilterGraph ();
 
        bool can_process (libdcp::Size s, AVPixelFormat p) const;
-       std::list<boost::shared_ptr<Image> > process (AVFrame * frame);
+       std::list<std::pair<boost::shared_ptr<Image>, int64_t> > process (AVFrame * frame);
 
 private:
        AVFilterContext* _buffer_src_context;

diff --git a/src/lib/subtitle_content.h b/src/lib/subtitle_content.h
index 1092b7b1cc7e32440515d653bc817f9005ecb680..c29485feeb6796bf1bbe942c0949da96d1fe4501 100644 (file)
--- a/src/lib/subtitle_content.h
+++ b/src/lib/subtitle_content.h
@@ -50,7 +50,9 @@ public:
                return _subtitle_scale;
        }
        
-private:       
+private:
+       friend class ffmpeg_pts_offset_test;
+       
        /** y offset for placing subtitles, as a proportion of the container height;
            +ve is further down the frame, -ve is further up.
        */

diff --git a/src/lib/video_content.h b/src/lib/video_content.h
index e7bbad9e15c67c5b99a3b9d494819a53c9c48bbd..46c8efdeb39420bbd2439c36ad3b332a020160a4 100644 (file)
--- a/src/lib/video_content.h
+++ b/src/lib/video_content.h
@@ -87,6 +87,8 @@ protected:
        VideoContent::Frame _video_length;
 
 private:
+       friend class ffmpeg_pts_offset_test;
+       
        libdcp::Size _video_size;
        float _video_frame_rate;
        Crop _crop;

diff --git a/test/ffmpeg_pts_offset.cc b/test/ffmpeg_pts_offset.cc
index 15c2b38e1befd4ffe70f1b8ec5bcadb111d55d3d..0ed8f66d899d22753661029f31af2cd18596a3ed 100644 (file)
--- a/test/ffmpeg_pts_offset.cc
+++ b/test/ffmpeg_pts_offset.cc
@@ -19,21 +19,55 @@
 
 BOOST_AUTO_TEST_CASE (ffmpeg_pts_offset_test)
 {
-       /* Sound == video so no offset required */
-       BOOST_CHECK_EQUAL (FFmpegDecoder::compute_pts_offset (0, 0, 24), 0);
+       shared_ptr<Film> film = new_test_film ("ffmpeg_pts_offset_test");
+       shared_ptr<FFmpegContent> content (new FFmpegContent (film, "test/data/test.mp4"));
+       content->_audio_stream.reset (new FFmpegAudioStream);
+       content->_video_frame_rate = 24;
 
-       /* Common offset should be removed */
-       BOOST_CHECK_CLOSE (FFmpegDecoder::compute_pts_offset (42, 42, 24), -42, 1e-9);
+       {
+               /* Sound == video so no offset required */
+               content->_first_video = 0;
+               content->_audio_stream->first_audio = 0;
+               FFmpegDecoder decoder (film, content, true, true);
+               BOOST_CHECK_EQUAL (decoder._video_pts_offset, 0);
+               BOOST_CHECK_EQUAL (decoder._audio_pts_offset, 0);
+       }
 
-       /* Video is on a frame boundary */
-       BOOST_CHECK_EQUAL (FFmpegDecoder::compute_pts_offset (1.0 / 24.0, 0, 24), 0);
+       {
+               /* Common offset should be removed */
+               content->_first_video = 600;
+               content->_audio_stream->first_audio = 600;
+               FFmpegDecoder decoder (film, content, true, true);
+               BOOST_CHECK_EQUAL (decoder._video_pts_offset, -600);
+               BOOST_CHECK_EQUAL (decoder._audio_pts_offset, -600);
+       }
 
-       /* Again, video is on a frame boundary */
-       BOOST_CHECK_EQUAL (FFmpegDecoder::compute_pts_offset (1.0 / 23.97, 0, 23.97), 0);
+       {
+               /* Video is on a frame boundary */
+               content->_first_video = 1.0 / 24.0;
+               content->_audio_stream->first_audio = 0;
+               FFmpegDecoder decoder (film, content, true, true);
+               BOOST_CHECK_EQUAL (decoder._video_pts_offset, 0);
+               BOOST_CHECK_EQUAL (decoder._audio_pts_offset, 0);
+       }
 
-       /* And again, video is on a frame boundary */
-       BOOST_CHECK_EQUAL (FFmpegDecoder::compute_pts_offset (3.0 / 23.97, 0, 23.97), 0);
+       {
+               /* Video is off a frame boundary */
+               double const frame = 1.0 / 24.0;
+               content->_first_video = frame + 0.0215;
+               content->_audio_stream->first_audio = 0;
+               FFmpegDecoder decoder (film, content, true, true);
+               BOOST_CHECK_EQUAL (decoder._video_pts_offset, (frame - 0.0215));
+               BOOST_CHECK_EQUAL (decoder._audio_pts_offset, (frame - 0.0215));
+       }
 
-       /* Off a frame boundary */
-       BOOST_CHECK_CLOSE (FFmpegDecoder::compute_pts_offset (1.0 / 24.0 - 0.0215, 0, 24), 0.0215, 1e-9);
+       {
+               /* Video is off a frame boundary and both have a common offset */
+               double const frame = 1.0 / 24.0;
+               content->_first_video = frame + 0.0215 + 4.1;
+               content->_audio_stream->first_audio = 4.1;
+               FFmpegDecoder decoder (film, content, true, true);
+               BOOST_CHECK_EQUAL (decoder._video_pts_offset, (frame - 0.0215) - 4.1);
+               BOOST_CHECK_EQUAL (decoder._audio_pts_offset, (frame - 0.0215) - 4.1);
+       }
 }