#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 "options.h"
#include "exceptions.h"
#include "image.h"
#include "util.h"
#include "log.h"
#include "ffmpeg_decoder.h"
-#include "subtitle.h"
+#include "filter_graph.h"
+#include "audio_buffers.h"
+
+#include "i18n.h"
using std::cout;
using std::string;
using std::vector;
using std::stringstream;
+using std::list;
+using std::min;
+using std::pair;
using boost::shared_ptr;
-
-FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, Job* j, bool minimal)
- : Decoder (f, o, j, minimal)
- , _format_context (0)
- , _video_stream (-1)
- , _audio_stream (-1)
- , _subtitle_stream (-1)
- , _frame (0)
- , _video_codec_context (0)
- , _video_codec (0)
- , _audio_codec_context (0)
- , _audio_codec (0)
+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)
+ : Decoder (f)
+ , VideoDecoder (f)
+ , AudioDecoder (f)
+ , SubtitleDecoder (f)
+ , FFmpeg (c)
, _subtitle_codec_context (0)
, _subtitle_codec (0)
+ , _decode_video (video)
+ , _decode_audio (audio)
+ , _video_pts_offset (0)
+ , _audio_pts_offset (0)
+ , _just_sought (false)
{
- setup_general ();
- setup_video ();
- setup_audio ();
setup_subtitle ();
+
+ /* Audio and video frame PTS values may not start with 0. We want
+ to fiddle them so that:
+
+ 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();
+ }
+
+ /* 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 ()
{
- if (_audio_codec_context) {
- avcodec_close (_audio_codec_context);
- }
-
- if (_video_codec_context) {
- avcodec_close (_video_codec_context);
- }
+ boost::mutex::scoped_lock lm (_mutex);
if (_subtitle_codec_context) {
avcodec_close (_subtitle_codec_context);
}
-
- av_free (_frame);
-
- avformat_close_input (&_format_context);
}
void
-FFmpegDecoder::setup_general ()
+FFmpegDecoder::pass ()
{
- int r;
-
- av_register_all ();
+ int r = av_read_frame (_format_context, &_packet);
- if ((r = avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0)) != 0) {
- throw OpenFileError (_film->content_path ());
+ 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));
+ shared_ptr<const Film> film = _film.lock ();
+ assert (film);
+ 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? */
+
+ if (_decode_video) {
+ while (decode_video_packet ()) {}
+ }
+
+ if (_ffmpeg_content->audio_stream() && _decode_audio) {
+ decode_audio_packet ();
+ }
+
+ /* Stop us being asked for any more data */
+ _video_position = _ffmpeg_content->video_length ();
+ _audio_position = _ffmpeg_content->audio_length ();
+ return;
}
- if (avformat_find_stream_info (_format_context, 0) < 0) {
- throw DecodeError ("could not find stream information");
+ avcodec_get_frame_defaults (_frame);
+
+ if (_packet.stream_index == _video_stream && _decode_video) {
+ decode_video_packet ();
+ } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+ decode_audio_packet ();
+ } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id) {
+ decode_subtitle_packet ();
}
- /* Find video, audio and subtitle streams and choose the first of each */
+ av_free_packet (&_packet);
+}
- 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) {
- if (_audio_stream == -1) {
- _audio_stream = i;
+/** @param data pointer to array of pointers to buffers.
+ * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
+ */
+shared_ptr<AudioBuffers>
+FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
+{
+ assert (_ffmpeg_content->audio_channels());
+ assert (bytes_per_audio_sample());
+
+ /* Deinterleave and convert to float */
+
+ assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
+
+ int const total_samples = size / bytes_per_audio_sample();
+ int const frames = total_samples / _ffmpeg_content->audio_channels();
+ shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
+
+ switch (audio_sample_format()) {
+ case AV_SAMPLE_FMT_S16:
+ {
+ int16_t* p = reinterpret_cast<int16_t *> (data[0]);
+ int sample = 0;
+ int channel = 0;
+ for (int i = 0; i < total_samples; ++i) {
+ audio->data(channel)[sample] = float(*p++) / (1 << 15);
+
+ ++channel;
+ if (channel == _ffmpeg_content->audio_channels()) {
+ channel = 0;
+ ++sample;
}
- _audio_streams.push_back (AudioStream (stream_name (s), i, s->codec->channels));
- } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
- if (_subtitle_stream == -1) {
- _subtitle_stream = i;
+ }
+ }
+ break;
+
+ case AV_SAMPLE_FMT_S16P:
+ {
+ int16_t** p = reinterpret_cast<int16_t **> (data);
+ for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
+ for (int j = 0; j < frames; ++j) {
+ audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
}
- _subtitle_streams.push_back (SubtitleStream (stream_name (s), i));
}
}
-
- /* Now override audio and subtitle streams with those from the Film, if it has any */
-
- if (_film->audio_stream_index() != -1) {
- _audio_stream = _film->audio_stream().id();
+ break;
+
+ case AV_SAMPLE_FMT_S32:
+ {
+ int32_t* p = reinterpret_cast<int32_t *> (data[0]);
+ int sample = 0;
+ int channel = 0;
+ for (int i = 0; i < total_samples; ++i) {
+ audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
+
+ ++channel;
+ if (channel == _ffmpeg_content->audio_channels()) {
+ channel = 0;
+ ++sample;
+ }
+ }
}
-
- if (_film->subtitle_stream_index() != -1) {
- _subtitle_stream = _film->subtitle_stream().id ();
+ break;
+
+ case AV_SAMPLE_FMT_FLT:
+ {
+ float* p = reinterpret_cast<float*> (data[0]);
+ int sample = 0;
+ int channel = 0;
+ for (int i = 0; i < total_samples; ++i) {
+ audio->data(channel)[sample] = *p++;
+
+ ++channel;
+ if (channel == _ffmpeg_content->audio_channels()) {
+ channel = 0;
+ ++sample;
+ }
+ }
}
-
- if (_video_stream < 0) {
- throw DecodeError ("could not find video stream");
+ break;
+
+ case AV_SAMPLE_FMT_FLTP:
+ {
+ float** p = reinterpret_cast<float**> (data);
+ for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
+ memcpy (audio->data(i), p[i], frames * sizeof(float));
+ }
}
+ break;
- _frame = avcodec_alloc_frame ();
- if (_frame == 0) {
- throw DecodeError ("could not allocate frame");
+ default:
+ throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
}
+
+ return audio;
}
-void
-FFmpegDecoder::setup_video ()
+AVSampleFormat
+FFmpegDecoder::audio_sample_format () const
{
- _video_codec_context = _format_context->streams[_video_stream]->codec;
- _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
-
- if (_video_codec == 0) {
- throw DecodeError ("could not find video decoder");
+ if (!_ffmpeg_content->audio_stream()) {
+ return (AVSampleFormat) 0;
}
-
- /* I think this prevents problems with green hash on decodes and
- "changing frame properties on the fly is not supported by all filters"
- messages with some content. Although I'm not sure; needs checking.
- */
- AVDictionary* opts = 0;
- av_dict_set (&opts, "threads", "1", 0);
- if (avcodec_open2 (_video_codec_context, _video_codec, &opts) < 0) {
- throw DecodeError ("could not open video decoder");
- }
+ return audio_codec_context()->sample_fmt;
+}
+
+int
+FFmpegDecoder::bytes_per_audio_sample () const
+{
+ return av_get_bytes_per_sample (audio_sample_format ());
}
void
-FFmpegDecoder::setup_audio ()
+FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
{
- if (_audio_stream < 0) {
- return;
- }
-
- _audio_codec_context = _format_context->streams[_audio_stream]->codec;
- _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
+ double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
- if (_audio_codec == 0) {
- throw DecodeError ("could not find audio decoder");
+ /* If we are doing an accurate seek, our initial shot will be 5 frames (5 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.
+ */
+ int initial = frame;
+ if (accurate) {
+ initial -= 5;
}
- if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
- throw DecodeError ("could not open audio decoder");
+ if (initial < 0) {
+ initial = 0;
}
- /* 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.
- */
+ /* Initial seek time in the stream's timebase */
+ 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()) - _video_pts_offset) / time_base;
- if (_audio_codec_context->channel_layout == 0) {
- _audio_codec_context->channel_layout = av_get_default_channel_layout (audio_channels ());
- }
-}
+ av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
-void
-FFmpegDecoder::setup_subtitle ()
-{
- if (_subtitle_stream < 0) {
- return;
+ avcodec_flush_buffers (video_codec_context());
+ if (_subtitle_codec_context) {
+ avcodec_flush_buffers (_subtitle_codec_context);
}
- _subtitle_codec_context = _format_context->streams[_subtitle_stream]->codec;
- _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
+ _just_sought = true;
- if (_subtitle_codec == 0) {
- throw DecodeError ("could not find subtitle decoder");
+ if (frame == 0) {
+ /* We're already there; from here on we can only seek non-zero amounts */
+ return;
}
-
- if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
- throw DecodeError ("could not open subtitle decoder");
+
+ if (accurate) {
+ while (1) {
+ int r = av_read_frame (_format_context, &_packet);
+ if (r < 0) {
+ return;
+ }
+
+ avcodec_get_frame_defaults (_frame);
+
+ if (_packet.stream_index == _video_stream) {
+ int finished = 0;
+ int const r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
+ if (r >= 0 && finished) {
+ int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
+ if (bet >= final_vt) {
+ _video_position = rint (
+ (bet * time_base + _video_pts_offset) * _ffmpeg_content->video_frame_rate()
+ );
+ av_free_packet (&_packet);
+ break;
+ }
+ }
+ }
+
+ av_free_packet (&_packet);
+ }
}
}
-
-bool
-FFmpegDecoder::pass ()
+void
+FFmpegDecoder::decode_audio_packet ()
{
- int r = av_read_frame (_format_context, &_packet);
+ /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
+ several times.
+ */
- 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 ("error on av_read_frame (%1) (%2)", buf, r));
- }
-
- /* Get any remaining frames */
-
- _packet.data = 0;
- _packet.size = 0;
+ AVPacket copy_packet = _packet;
- /* XXX: should we reset _packet.data and size after each *_decode_* call? */
+ while (copy_packet.size > 0) {
int frame_finished;
-
- while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
- process_video (_frame);
- }
-
- if (_audio_stream >= 0 && _opt->decode_audio) {
- while (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
- 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 == _film->audio_channels());
- process_audio (_frame->data[0], data_size);
+ int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
+ if (decode_result >= 0) {
+ if (frame_finished) {
+
+ 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) + _audio_pts_offset;
+
+ if (pts > 0) {
+ /* Emit some silence */
+ shared_ptr<AudioBuffers> silence (
+ new AudioBuffers (
+ _ffmpeg_content->audio_channels(),
+ pts * _ffmpeg_content->content_audio_frame_rate()
+ )
+ );
+
+ silence->make_silent ();
+ audio (silence, _audio_position);
+ }
+ }
+
+ copy_packet.data += decode_result;
+ copy_packet.size -= decode_result;
}
}
+ }
+}
- return true;
+bool
+FFmpegDecoder::decode_video_packet ()
+{
+ int frame_finished;
+ if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
+ return false;
}
- avcodec_get_frame_defaults (_frame);
-
- if (_packet.stream_index == _video_stream) {
+ boost::mutex::scoped_lock lm (_filter_graphs_mutex);
- int frame_finished;
- int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
- if (r >= 0 && frame_finished) {
+ 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)) {
+ ++i;
+ }
- if (r != _packet.size) {
- _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
- }
+ if (i == _filter_graphs.end ()) {
+ shared_ptr<const Film> film = _film.lock ();
+ assert (film);
- /* Where we are in the output, in seconds */
- double const out_pts_seconds = video_frames_in() / frames_per_second();
+ graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
+ _filter_graphs.push_back (graph);
- /* 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_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
+ } else {
+ graph = *i;
+ }
- if (!_first_video) {
- _first_video = source_pts_seconds;
- }
+ list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
- /* 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 (
- "Extra frame inserted at %1s; source frame %2, source PTS %3",
- out_pts_seconds, video_frames_in(), source_pts_seconds
- )
- );
- }
- }
+ string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
+
+ for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
- if (delta > -one_frame) {
- /* Process this frame */
- process_video (_frame);
- } else {
- /* Otherwise we are omitting a frame to keep things right */
- _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds));
- }
+ shared_ptr<Image> image = i->first;
+ if (!post_process.empty ()) {
+ image = image->post_process (post_process, true);
}
+
+ if (i->second != AV_NOPTS_VALUE) {
- } else if (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio) {
-
- int frame_finished;
- if (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-
- /* 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);
-
- /* 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) {
+ double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset;
- if (!_first_audio) {
- _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()) * audio_sample_rate ());
-
- _film->log()->log (
- String::compose (
- "First video at %1, first audio at %2, pushing %3 frames of silence for %4 channels (%5 bytes per sample)",
- _first_video.get(), _first_audio.get(), s, audio_channels(), bytes_per_audio_sample()
- )
- );
-
- if (s) {
- /* hence bytes */
- int const b = s * audio_channels() * bytes_per_audio_sample();
-
- /* XXX: this assumes that it won't be too much, and there are shaky assumptions
- that all sound representations are silent with memset()ed zero data.
- */
- uint8_t silence[b];
- memset (silence, 0, b);
- process_audio (silence, b);
- }
- }
+ if (_just_sought) {
+ /* We just did a seek, so disable any attempts to correct for where we
+ are / should be.
+ */
+ _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
+ _just_sought = false;
+ }
- int const data_size = av_samples_get_buffer_size (
- 0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
+ double const next = _video_position / _ffmpeg_content->video_frame_rate();
+ double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
+ double delta = pts - next;
+
+ while (delta > one_frame) {
+ /* This PTS is more than one frame forward in time of where we think we should be; emit
+ a black frame.
+ */
+ boost::shared_ptr<Image> black (
+ new Image (
+ static_cast<AVPixelFormat> (_frame->format),
+ libdcp::Size (video_codec_context()->width, video_codec_context()->height),
+ true
+ )
);
- assert (_audio_codec_context->channels == _film->audio_channels());
- process_audio (_frame->data[0], data_size);
+ black->make_black ();
+ video (image, false, _video_position);
+ delta -= one_frame;
}
- }
-
- } else if (_subtitle_stream >= 0 && _packet.stream_index == _subtitle_stream && _opt->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) {
- process_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub, _first_video.get())));
- } else {
- process_subtitle (shared_ptr<TimedSubtitle> ());
+
+ if (delta > -one_frame) {
+ /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
+ video (image, false, _video_position);
}
- avsubtitle_free (&sub);
+
+ } else {
+ shared_ptr<const Film> film = _film.lock ();
+ assert (film);
+ film->log()->log ("Dropping frame without PTS");
}
}
-
- av_free_packet (&_packet);
- return false;
-}
-
-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);
+ return true;
}
-int
-FFmpegDecoder::audio_channels () const
+
+void
+FFmpegDecoder::setup_subtitle ()
{
- if (_audio_codec_context == 0) {
- return 0;
+ boost::mutex::scoped_lock lm (_mutex);
+
+ if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
+ return;
}
- return _audio_codec_context->channels;
-}
-
-int
-FFmpegDecoder::audio_sample_rate () const
-{
- if (_audio_codec_context == 0) {
- return 0;
- }
-
- return _audio_codec_context->sample_rate;
-}
+ _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
+ _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
-AVSampleFormat
-FFmpegDecoder::audio_sample_format () const
-{
- if (_audio_codec_context == 0) {
- return (AVSampleFormat) 0;
+ if (_subtitle_codec == 0) {
+ throw DecodeError (_("could not find subtitle decoder"));
}
- return _audio_codec_context->sample_fmt;
-}
-
-int64_t
-FFmpegDecoder::audio_channel_layout () const
-{
- if (_audio_codec_context == 0) {
- return 0;
+ if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
+ throw DecodeError (N_("could not open subtitle decoder"));
}
-
- return _audio_codec_context->channel_layout;
-}
-
-Size
-FFmpegDecoder::native_size () const
-{
- return Size (_video_codec_context->width, _video_codec_context->height);
-}
-
-PixelFormat
-FFmpegDecoder::pixel_format () const
-{
- return _video_codec_context->pix_fmt;
-}
-
-int
-FFmpegDecoder::time_base_numerator () const
-{
- return _video_codec_context->time_base.num;
-}
-
-int
-FFmpegDecoder::time_base_denominator () const
-{
- return _video_codec_context->time_base.den;
-}
-
-int
-FFmpegDecoder::sample_aspect_ratio_numerator () const
-{
- return _video_codec_context->sample_aspect_ratio.num;
-}
-
-int
-FFmpegDecoder::sample_aspect_ratio_denominator () const
-{
- return _video_codec_context->sample_aspect_ratio.den;
}
bool
-FFmpegDecoder::has_subtitles () const
+FFmpegDecoder::done () const
{
- return (_subtitle_stream != -1);
+ bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
+ bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
+ return vd && ad;
}
-
-vector<AudioStream>
-FFmpegDecoder::audio_streams () const
-{
- return _audio_streams;
-}
-
-vector<SubtitleStream>
-FFmpegDecoder::subtitle_streams () const
+
+void
+FFmpegDecoder::decode_subtitle_packet ()
{
- return _subtitle_streams;
-}
+ int got_subtitle;
+ AVSubtitle sub;
+ if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
+ return;
+ }
-string
-FFmpegDecoder::stream_name (AVStream* s) const
-{
- stringstream n;
+ /* 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) {
+ subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
+ return;
+ } else if (sub.num_rects > 1) {
+ throw DecodeError (_("multi-part subtitles not yet supported"));
+ }
+
+ /* Subtitle PTS in seconds (within the source, not taking into account any of the
+ source that we may have chopped off for the DCP)
+ */
+ double const packet_time = static_cast<double> (sub.pts) / AV_TIME_BASE;
- AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
- if (lang) {
- n << lang->value;
+ /* hence start time for this sub */
+ Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
+ Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
+
+ AVSubtitleRect const * rect = sub.rects[0];
+
+ if (rect->type != SUBTITLE_BITMAP) {
+ throw DecodeError (_("non-bitmap subtitles not yet supported"));
}
- AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
- if (title) {
- if (!n.str().empty()) {
- n << " ";
+ shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
+
+ /* Start of the first line in the subtitle */
+ uint8_t* sub_p = rect->pict.data[0];
+ /* sub_p looks up into a RGB palette which is here */
+ 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) {
+ *out_line_p++ = palette[*sub_line_p++];
}
- n << title->value;
+ sub_p += rect->pict.linesize[0];
+ out_p += image->stride()[0] / sizeof (uint32_t);
}
- if (n.str().empty()) {
- n << "unknown";
- }
-
- return n.str ();
+ libdcp::Size const vs = _ffmpeg_content->video_size ();
+
+ 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