#include <iostream>
#include <stdint.h>
#include <boost/lexical_cast.hpp>
+#include <sndfile.h>
extern "C" {
-#include <tiffio.h>
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
-#include <libswscale/swscale.h>
-#include <libpostproc/postprocess.h>
}
-#include <sndfile.h>
-#include "film.h"
-#include "format.h"
-#include "transcoder.h"
-#include "job.h"
#include "filter.h"
#include "exceptions.h"
#include "image.h"
#include "log.h"
#include "ffmpeg_decoder.h"
#include "filter_graph.h"
-#include "subtitle.h"
+#include "audio_buffers.h"
+#include "ffmpeg_content.h"
#include "i18n.h"
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;
-using libdcp::Size;
-
-FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles, bool video_sync)
- : Decoder (f)
- , VideoDecoder (f)
- , AudioDecoder (f)
- , _ffmpeg_content (c)
- , _format_context (0)
- , _video_stream (-1)
- , _frame (0)
- , _video_codec_context (0)
- , _video_codec (0)
- , _audio_codec_context (0)
- , _audio_codec (0)
+using dcp::Size;
+
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log, bool video, bool audio, bool subtitles)
+ : VideoDecoder (c)
+ , AudioDecoder (c)
+ , FFmpeg (c)
+ , _log (log)
, _subtitle_codec_context (0)
, _subtitle_codec (0)
, _decode_video (video)
, _decode_audio (audio)
, _decode_subtitles (subtitles)
- , _video_sync (video_sync)
+ , _pts_offset (0)
{
- setup_general ();
- setup_video ();
- setup_audio ();
setup_subtitle ();
- if (!video_sync) {
- _first_video = 0;
- }
-}
-
-FFmpegDecoder::~FFmpegDecoder ()
-{
- if (_audio_codec_context) {
- avcodec_close (_audio_codec_context);
- }
-
- if (_video_codec_context) {
- avcodec_close (_video_codec_context);
- }
-
- if (_subtitle_codec_context) {
- avcodec_close (_subtitle_codec_context);
- }
-
- av_free (_frame);
-
- avformat_close_input (&_format_context);
-}
-
-void
-FFmpegDecoder::setup_general ()
-{
- av_register_all ();
+ /* Audio and video frame PTS values may not start with 0. We want
+ to fiddle them so that:
- if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
- throw OpenFileError (_ffmpeg_content->file().string ());
- }
+ 1. One of them starts at time 0.
+ 2. The first video PTS value ends up on a frame boundary.
- if (avformat_find_stream_info (_format_context, 0) < 0) {
- throw DecodeError (_("could not find stream information"));
- }
-
- /* Find video, audio and subtitle streams and choose the first of each */
-
- for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
- AVStream* s = _format_context->streams[i];
- if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
- _video_stream = i;
- } else 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,
- so bodge it here. No idea why we should have to do this.
- */
+ Then we remove big initial gaps in PTS and we allow our
+ insertion of black frames to work.
- if (s->codec->channel_layout == 0) {
- s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
- }
-
- _audio_streams.push_back (
- FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
- );
-
- } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
- _subtitle_streams.push_back (FFmpegSubtitleStream (stream_name (s), i));
- }
- }
-
- if (_video_stream < 0) {
- throw DecodeError (N_("could not find video stream"));
- }
+ We will do pts_to_use = pts_from_ffmpeg + pts_offset;
+ */
- _frame = avcodec_alloc_frame ();
- if (_frame == 0) {
- throw DecodeError (N_("could not allocate frame"));
- }
-}
+ bool const have_video = video && c->first_video();
+ bool const have_audio = _decode_audio && c->audio_stream () && c->audio_stream()->first_audio;
-void
-FFmpegDecoder::setup_video ()
-{
- _video_codec_context = _format_context->streams[_video_stream]->codec;
- _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
+ /* First, make one of them start at 0 */
- if (_video_codec == 0) {
- throw DecodeError (_("could not find video decoder"));
+ if (have_audio && have_video) {
+ _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
+ } else if (have_video) {
+ _pts_offset = - c->first_video().get();
+ } else if (have_audio) {
+ _pts_offset = - c->audio_stream()->first_audio.get();
}
- if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
- throw DecodeError (N_("could not open video decoder"));
+ /* Now adjust both so that the video pts starts on a frame */
+ if (have_video && have_audio) {
+ ContentTime first_video = c->first_video().get() + _pts_offset;
+ ContentTime const old_first_video = first_video;
+ _pts_offset += first_video.round_up (c->video_frame_rate ()) - old_first_video;
}
}
-void
-FFmpegDecoder::setup_audio ()
+FFmpegDecoder::~FFmpegDecoder ()
{
- if (!_ffmpeg_content->audio_stream ()) {
- return;
- }
+ boost::mutex::scoped_lock lm (_mutex);
- _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
- _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
-
- if (_audio_codec == 0) {
- throw DecodeError (_("could not find audio decoder"));
- }
-
- if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
- throw DecodeError (N_("could not open audio decoder"));
+ if (_subtitle_codec_context) {
+ avcodec_close (_subtitle_codec_context);
}
}
void
-FFmpegDecoder::setup_subtitle ()
+FFmpegDecoder::flush ()
{
- if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
- return;
- }
-
- _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
- _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
-
- if (_subtitle_codec == 0) {
- throw DecodeError (_("could not find subtitle decoder"));
+ /* Get any remaining frames */
+
+ _packet.data = 0;
+ _packet.size = 0;
+
+ /* XXX: should we reset _packet.data and size after each *_decode_* call? */
+
+ if (_decode_video) {
+ while (decode_video_packet ()) {}
}
- if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
- throw DecodeError (N_("could not open subtitle decoder"));
+ if (_ffmpeg_content->audio_stream() && _decode_audio) {
+ decode_audio_packet ();
+ AudioDecoder::flush ();
}
}
-
bool
FFmpegDecoder::pass ()
{
int r = av_read_frame (_format_context, &_packet);
-
+
if (r < 0) {
if (r != AVERROR_EOF) {
/* Maybe we should fail here, but for now we'll just finish off instead */
char buf[256];
av_strerror (r, buf, sizeof(buf));
- _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
- }
-
- /* Get any remaining frames */
-
- _packet.data = 0;
- _packet.size = 0;
-
- /* XXX: should we reset _packet.data and size after each *_decode_* call? */
-
- int frame_finished;
-
- if (_decode_video) {
- while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
- filter_and_emit_video (_frame);
- }
- }
-
- if (_ffmpeg_content->audio_stream() && _decode_audio) {
- decode_audio_packet ();
+ _log->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
}
+ flush ();
return true;
}
- avcodec_get_frame_defaults (_frame);
-
- if (_packet.stream_index == _video_stream && _decode_video) {
-
- int frame_finished;
- int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
- if (r >= 0 && frame_finished) {
-
- if (r != _packet.size) {
- _film->log()->log (String::compose (N_("Used only %1 bytes of %2 in packet"), r, _packet.size));
- }
-
- if (_video_sync) {
- out_with_sync ();
- } else {
- filter_and_emit_video (_frame);
- }
- }
-
- } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+ int const si = _packet.stream_index;
+
+ if (si == _video_stream && _decode_video) {
+ decode_video_packet ();
+ } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
decode_audio_packet ();
- } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles && _first_video) {
-
- int got_subtitle;
- AVSubtitle sub;
- if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
- /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
- indicate that the previous subtitle should stop.
- */
- if (sub.num_rects > 0) {
- shared_ptr<TimedSubtitle> ts;
- try {
- emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
- } catch (...) {
- /* some problem with the subtitle; we probably didn't understand it */
- }
- } else {
- emit_subtitle (shared_ptr<TimedSubtitle> ());
- }
- avsubtitle_free (&sub);
- }
+ } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && _decode_subtitles) {
+ decode_subtitle_packet ();
}
-
+
av_free_packet (&_packet);
return false;
}
shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
switch (audio_sample_format()) {
+ case AV_SAMPLE_FMT_U8:
+ {
+ uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
+ int sample = 0;
+ int channel = 0;
+ for (int i = 0; i < total_samples; ++i) {
+ audio->data(channel)[sample] = float(*p++) / (1 << 23);
+
+ ++channel;
+ if (channel == _ffmpeg_content->audio_channels()) {
+ channel = 0;
+ ++sample;
+ }
+ }
+ }
+ break;
+
case AV_SAMPLE_FMT_S16:
{
int16_t* p = reinterpret_cast<int16_t *> (data[0]);
return audio;
}
-float
-FFmpegDecoder::frames_per_second () const
-{
- AVStream* s = _format_context->streams[_video_stream];
-
- if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
- return av_q2d (s->avg_frame_rate);
- }
-
- return av_q2d (s->r_frame_rate);
-}
-
AVSampleFormat
FFmpegDecoder::audio_sample_format () const
{
- if (_audio_codec_context == 0) {
+ if (!_ffmpeg_content->audio_stream()) {
return (AVSampleFormat) 0;
}
- return _audio_codec_context->sample_fmt;
+ return audio_codec_context()->sample_fmt;
}
-libdcp::Size
-FFmpegDecoder::native_size () const
-{
- return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
-}
-
-PixelFormat
-FFmpegDecoder::pixel_format () const
+int
+FFmpegDecoder::bytes_per_audio_sample () const
{
- return _video_codec_context->pix_fmt;
+ return av_get_bytes_per_sample (audio_sample_format ());
}
int
-FFmpegDecoder::time_base_numerator () const
+FFmpegDecoder::minimal_run (boost::function<bool (optional<ContentTime>, optional<ContentTime>, int)> finished)
{
- return _video_codec_context->time_base.num;
+ int frames_read = 0;
+ optional<ContentTime> last_video;
+ optional<ContentTime> last_audio;
+
+ while (!finished (last_video, last_audio, frames_read)) {
+ int r = av_read_frame (_format_context, &_packet);
+ if (r < 0) {
+ /* We should flush our decoders here, possibly yielding a few more frames,
+ but the consequence of having to do that is too hideous to contemplate.
+ Instead we give up and say that you can't seek too close to the end
+ of a file.
+ */
+ return frames_read;
+ }
+
+ ++frames_read;
+
+ double const time_base = av_q2d (_format_context->streams[_packet.stream_index]->time_base);
+
+ if (_packet.stream_index == _video_stream) {
+
+ avcodec_get_frame_defaults (_frame);
+
+ int finished = 0;
+ r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
+ if (r >= 0 && finished) {
+ last_video = ContentTime::from_seconds (av_frame_get_best_effort_timestamp (_frame) * time_base) + _pts_offset;
+ }
+
+ } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, _packet.stream_index)) {
+ AVPacket copy_packet = _packet;
+ while (copy_packet.size > 0) {
+
+ int finished;
+ r = avcodec_decode_audio4 (audio_codec_context(), _frame, &finished, &_packet);
+ if (r >= 0 && finished) {
+ last_audio = ContentTime::from_seconds (av_frame_get_best_effort_timestamp (_frame) * time_base) + _pts_offset;
+ }
+
+ copy_packet.data += r;
+ copy_packet.size -= r;
+ }
+ }
+
+ av_free_packet (&_packet);
+ }
+
+ return frames_read;
}
-int
-FFmpegDecoder::time_base_denominator () const
+bool
+FFmpegDecoder::seek_overrun_finished (ContentTime seek, optional<ContentTime> last_video, optional<ContentTime> last_audio) const
{
- return _video_codec_context->time_base.den;
+ return (last_video && last_video.get() >= seek) || (last_audio && last_audio.get() >= seek);
}
-int
-FFmpegDecoder::sample_aspect_ratio_numerator () const
+bool
+FFmpegDecoder::seek_final_finished (int n, int done) const
{
- return _video_codec_context->sample_aspect_ratio.num;
+ return n == done;
}
-int
-FFmpegDecoder::sample_aspect_ratio_denominator () const
+void
+FFmpegDecoder::seek_and_flush (ContentTime t)
{
- return _video_codec_context->sample_aspect_ratio.den;
+ ContentTime const u = t - _pts_offset;
+ int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base);
+
+ if (_ffmpeg_content->audio_stream ()) {
+ s = min (
+ s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base))
+ );
+ }
+
+ /* Ridiculous empirical hack */
+ s--;
+ if (s < 0) {
+ s = 0;
+ }
+
+ av_seek_frame (_format_context, _video_stream, s, 0);
+
+ avcodec_flush_buffers (video_codec_context());
+ if (audio_codec_context ()) {
+ avcodec_flush_buffers (audio_codec_context ());
+ }
+ if (_subtitle_codec_context) {
+ avcodec_flush_buffers (_subtitle_codec_context);
+ }
}
-string
-FFmpegDecoder::stream_name (AVStream* s) const
+void
+FFmpegDecoder::seek (ContentTime time, bool accurate)
{
- stringstream n;
-
- AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
- if (lang) {
- n << lang->value;
+ Decoder::seek (time, accurate);
+ if (_decode_audio) {
+ AudioDecoder::seek (time, accurate);
}
- AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
- if (title) {
- if (!n.str().empty()) {
- n << N_(" ");
- }
- n << title->value;
+ /* If we are doing an accurate seek, our initial shot will be 200ms (200 being
+ a number plucked from the air) earlier than we want to end up. The loop below
+ will hopefully then step through to where we want to be.
+ */
+
+ ContentTime pre_roll = accurate ? ContentTime::from_seconds (0.2) : ContentTime (0);
+ ContentTime initial_seek = time - pre_roll;
+ if (initial_seek < ContentTime (0)) {
+ initial_seek = ContentTime (0);
}
- if (n.str().empty()) {
- n << N_("unknown");
+ /* Initial seek time in the video stream's timebase */
+
+ seek_and_flush (initial_seek);
+
+ if (!accurate) {
+ /* That'll do */
+ return;
}
- return n.str ();
+ int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2));
+
+ seek_and_flush (initial_seek);
+ if (N > 0) {
+ minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3));
+ }
}
-int
-FFmpegDecoder::bytes_per_audio_sample () const
+void
+FFmpegDecoder::decode_audio_packet ()
{
- return av_get_bytes_per_sample (audio_sample_format ());
+ /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
+ several times.
+ */
+
+ AVPacket copy_packet = _packet;
+
+ while (copy_packet.size > 0) {
+
+ int frame_finished;
+ int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
+
+ if (decode_result < 0) {
+ _log->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
+ return;
+ }
+
+ if (frame_finished) {
+ ContentTime const ct = ContentTime::from_seconds (
+ av_frame_get_best_effort_timestamp (_frame) *
+ av_q2d (_ffmpeg_content->audio_stream()->stream (_format_context)->time_base))
+ + _pts_offset;
+
+ int const data_size = av_samples_get_buffer_size (
+ 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
+ );
+
+ audio (deinterleave_audio (_frame->data, data_size), ct);
+ }
+
+ copy_packet.data += decode_result;
+ copy_packet.size -= decode_result;
+ }
}
-void
-FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
+bool
+FFmpegDecoder::decode_video_packet ()
{
+ int frame_finished;
+ if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
+ return false;
+ }
+
boost::mutex::scoped_lock lm (_filter_graphs_mutex);
-
- shared_ptr<FilterGraph> graph;
+ shared_ptr<FilterGraph> graph;
+
list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
- while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
+ while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
++i;
}
if (i == _filter_graphs.end ()) {
- graph.reset (new FilterGraph (_film, this, libdcp::Size (frame->width, frame->height), (AVPixelFormat) frame->format));
+ graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
_filter_graphs.push_back (graph);
- _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), frame->width, frame->height, frame->format));
+ _log->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
} else {
graph = *i;
}
- list<shared_ptr<Image> > images = graph->process (frame);
+ list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
- for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
- emit_video (*i, frame_time ());
- }
-}
+ for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
-bool
-FFmpegDecoder::seek (double p)
-{
- return do_seek (p, false);
-}
+ shared_ptr<Image> image = i->first;
+
+ if (i->second != AV_NOPTS_VALUE) {
+ video (image, false, ContentTime::from_seconds (i->second * av_q2d (_format_context->streams[_video_stream]->time_base)) + _pts_offset);
+ } else {
+ _log->log ("Dropping frame without PTS");
+ }
+ }
-bool
-FFmpegDecoder::seek_to_last ()
-{
- /* This AVSEEK_FLAG_BACKWARD in do_seek is a bit of a hack; without it, if we ask for a seek to the same place as last time
- (used when we change decoder parameters and want to re-fetch the frame) we end up going forwards rather than
- staying in the same place.
- */
- return do_seek (last_source_time(), true);
+ return true;
}
-bool
-FFmpegDecoder::do_seek (double p, bool backwards)
-{
- int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
-
- int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
- avcodec_flush_buffers (_video_codec_context);
- if (_subtitle_codec_context) {
- avcodec_flush_buffers (_subtitle_codec_context);
- }
-
- return r < 0;
-}
-
void
-FFmpegDecoder::out_with_sync ()
+FFmpegDecoder::setup_subtitle ()
{
- /* Where we are in the output, in seconds */
- double const out_pts_seconds = video_frame() / frames_per_second();
+ boost::mutex::scoped_lock lm (_mutex);
- /* Where we are in the source, in seconds */
- double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
- * av_frame_get_best_effort_timestamp(_frame);
-
- _film->log()->log (
- String::compose (N_("Source video frame ready; source at %1, output at %2"), source_pts_seconds, out_pts_seconds),
- Log::VERBOSE
- );
-
- if (!_first_video) {
- _first_video = source_pts_seconds;
+ if (!_ffmpeg_content->subtitle_stream()) {
+ return;
}
-
- /* Difference between where we are and where we should be */
- double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds;
- double const one_frame = 1 / frames_per_second();
-
- /* Insert frames if required to get out_pts_seconds up to pts_seconds */
- if (delta > one_frame) {
- int const extra = rint (delta / one_frame);
- for (int i = 0; i < extra; ++i) {
- repeat_last_video ();
- _film->log()->log (
- String::compose (
- N_("Extra video frame inserted at %1s; source frame %2, source PTS %3 (at %4 fps)"),
- out_pts_seconds, video_frame(), source_pts_seconds, frames_per_second()
- )
- );
- }
+
+ _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
+ if (_subtitle_codec_context == 0) {
+ throw DecodeError (N_("could not find subtitle stream"));
+ }
+
+ _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
+
+ if (_subtitle_codec == 0) {
+ throw DecodeError (N_("could not find subtitle decoder"));
}
- if (delta > -one_frame) {
- /* Process this frame */
- filter_and_emit_video (_frame);
- } else {
- /* Otherwise we are omitting a frame to keep things right */
- _film->log()->log (String::compose (N_("Frame removed at %1s"), out_pts_seconds));
+ if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
+ throw DecodeError (N_("could not open subtitle decoder"));
}
}
void
-FFmpegDecoder::film_changed (Film::Property p)
+FFmpegDecoder::decode_subtitle_packet ()
{
- switch (p) {
- case Film::CROP:
- case Film::FILTERS:
- {
- boost::mutex::scoped_lock lm (_filter_graphs_mutex);
- _filter_graphs.clear ();
+ int got_subtitle;
+ AVSubtitle sub;
+ if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
+ return;
}
- OutputChanged ();
- break;
- default:
- break;
+ /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
+ indicate that the previous subtitle should stop.
+ */
+ if (sub.num_rects <= 0) {
+ image_subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), ContentTime (), ContentTime ());
+ return;
+ } else if (sub.num_rects > 1) {
+ throw DecodeError (_("multi-part subtitles not yet supported"));
}
-}
-
-/** @return Length (in video frames) according to our content's header */
-ContentVideoFrame
-FFmpegDecoder::video_length () const
-{
- return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
-}
-
-double
-FFmpegDecoder::frame_time () const
-{
- return av_frame_get_best_effort_timestamp(_frame) * av_q2d (_format_context->streams[_video_stream]->time_base);
-}
-
-void
-FFmpegDecoder::decode_audio_packet ()
-{
- /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
- several times.
+
+ /* Subtitle PTS (within the source, not taking into account any of the
+ source that we may have chopped off for the DCP)
*/
+ ContentTime packet_time = ContentTime::from_seconds (static_cast<double> (sub.pts) / AV_TIME_BASE) + _pts_offset;
- AVPacket copy_packet = _packet;
+ /* hence start time for this sub */
+ ContentTime const from = packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3);
+ ContentTime const to = packet_time + ContentTime::from_seconds (sub.end_display_time / 1e3);
- while (copy_packet.size > 0) {
+ AVSubtitleRect const * rect = sub.rects[0];
- int frame_finished;
- int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, ©_packet);
- if (decode_result >= 0 && frame_finished) {
-
- /* Where we are in the source, in seconds */
- double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
- * av_frame_get_best_effort_timestamp(_frame);
-
- /* We only decode audio if we've had our first video packet through, and if it
- was before this packet. Until then audio is thrown away.
- */
-
- if ((_first_video && _first_video.get() <= source_pts_seconds) || !_decode_video) {
-
- if (!_first_audio && _decode_video) {
- _first_audio = source_pts_seconds;
-
- /* This is our first audio frame, and if we've arrived here we must have had our
- first video frame. Push some silence to make up any gap between our first
- video frame and our first audio.
- */
-
- /* frames of silence that we must push */
- int const s = rint ((_first_audio.get() - _first_video.get()) * _ffmpeg_content->audio_frame_rate ());
-
- _film->log()->log (
- String::compose (
- N_("First video at %1, first audio at %2, pushing %3 audio frames of silence for %4 channels (%5 bytes per sample)"),
- _first_video.get(), _first_audio.get(), s, _ffmpeg_content->audio_channels(), bytes_per_audio_sample()
- )
- );
-
- if (s) {
- shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), s));
- audio->make_silent ();
- Audio (audio);
- }
- }
-
- int const data_size = av_samples_get_buffer_size (
- 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
- );
-
- assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
- Audio (deinterleave_audio (_frame->data, data_size));
- }
- }
+ if (rect->type != SUBTITLE_BITMAP) {
+ throw DecodeError (_("non-bitmap subtitles not yet supported"));
+ }
+
+ /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
+ G, third B, fourth A.
+ */
+ shared_ptr<Image> image (new Image (PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true));
- if (decode_result >= 0) {
- copy_packet.data += decode_result;
- copy_packet.size -= decode_result;
+ /* Start of the first line in the subtitle */
+ uint8_t* sub_p = rect->pict.data[0];
+ /* sub_p looks up into a BGRA palette which is here
+ (i.e. first byte B, second G, third R, fourth A)
+ */
+ uint32_t const * palette = (uint32_t *) rect->pict.data[1];
+ /* Start of the output data */
+ uint32_t* out_p = (uint32_t *) image->data()[0];
+
+ for (int y = 0; y < rect->h; ++y) {
+ uint8_t* sub_line_p = sub_p;
+ uint32_t* out_line_p = out_p;
+ for (int x = 0; x < rect->w; ++x) {
+ uint32_t const p = palette[*sub_line_p++];
+ *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
}
- }
+ sub_p += rect->pict.linesize[0];
+ out_p += image->stride()[0] / sizeof (uint32_t);
+ }
+
+ dcp::Size const vs = _ffmpeg_content->video_size ();
+
+ image_subtitle (
+ image,
+ dcpomatic::Rect<double> (
+ static_cast<double> (rect->x) / vs.width,
+ static_cast<double> (rect->y) / vs.height,
+ static_cast<double> (rect->w) / vs.width,
+ static_cast<double> (rect->h) / vs.height
+ ),
+ from,
+ to
+ );
+
+
+ avsubtitle_free (&sub);
}
projects / dcpomatic.git / blobdiff