_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)
#include "subtitle_content.h"
class Filter;
+class ffmpeg_pts_offset_test;
class FFmpegAudioStream
{
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);
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;
};
using std::stringstream;
using std::list;
using std::min;
+using std::pair;
using boost::shared_ptr;
using boost::optional;
using boost::dynamic_pointer_cast;
, _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 ()
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());
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;
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 */
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
bool _decode_video;
bool _decode_audio;
- double _pts_offset;
+ double _video_pts_offset;
+ double _audio_pts_offset;
bool _just_sought;
};
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;
/** 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."));
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);
}
~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;
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.
*/
VideoContent::Frame _video_length;
private:
+ friend class ffmpeg_pts_offset_test;
+
libdcp::Size _video_size;
float _video_frame_rate;
Crop _crop;
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);
+ }
}