#include <iomanip>
#include <iostream>
#include <stdint.h>
+#include <boost/lexical_cast.hpp>
extern "C" {
#include <tiffio.h>
#include <libavcodec/avcodec.h>
#include "transcoder.h"
#include "job.h"
#include "filter.h"
-#include "film_state.h"
#include "options.h"
#include "exceptions.h"
#include "image.h"
#include "util.h"
#include "log.h"
#include "ffmpeg_decoder.h"
-
-using namespace std;
-using namespace boost;
-
-FFmpegDecoder::FFmpegDecoder (boost::shared_ptr<const FilmState> s, boost::shared_ptr<const Options> o, Job* j, Log* l, bool minimal, bool ignore_length)
- : Decoder (s, o, j, l, minimal, ignore_length)
+#include "filter_graph.h"
+#include "subtitle.h"
+
+using std::cout;
+using std::string;
+using std::vector;
+using std::stringstream;
+using std::list;
+using boost::shared_ptr;
+using boost::optional;
+using boost::dynamic_pointer_cast;
+
+FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, Job* j)
+ : Decoder (f, o, j)
+ , VideoDecoder (f, o, j)
+ , AudioDecoder (f, o, j)
, _format_context (0)
, _video_stream (-1)
- , _audio_stream (-1)
- , _subtitle_stream (-1)
, _frame (0)
, _video_codec_context (0)
, _video_codec (0)
, _audio_codec (0)
, _subtitle_codec_context (0)
, _subtitle_codec (0)
- , _have_subtitle (false)
{
setup_general ();
setup_video ();
avcodec_close (_video_codec_context);
}
- if (_have_subtitle) {
- avsubtitle_free (&_subtitle);
- }
-
if (_subtitle_codec_context) {
avcodec_close (_subtitle_codec_context);
}
-
+
av_free (_frame);
+
avformat_close_input (&_format_context);
}
av_register_all ();
- if ((r = avformat_open_input (&_format_context, _fs->content_path().c_str(), 0, 0)) != 0) {
- throw OpenFileError (_fs->content_path ());
+ if ((r = avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0)) != 0) {
+ throw OpenFileError (_film->content_path ());
}
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) {
- if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
+ AVStream* s = _format_context->streams[i];
+ if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
_video_stream = i;
- } else if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
- _audio_stream = i;
- } else if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
- _subtitle_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.
+ */
+
+ if (s->codec->channel_layout == 0) {
+ s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
+ }
+
+ _audio_streams.push_back (
+ shared_ptr<AudioStream> (
+ new 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 (
+ shared_ptr<SubtitleStream> (
+ new SubtitleStream (stream_name (s), i)
+ )
+ );
}
}
if (_video_codec == 0) {
throw DecodeError ("could not find video decoder");
}
-
+
if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
throw DecodeError ("could not open video decoder");
}
void
FFmpegDecoder::setup_audio ()
{
- if (_audio_stream < 0) {
+ if (!_audio_stream) {
return;
}
+
+ shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
+ assert (ffa);
- _audio_codec_context = _format_context->streams[_audio_stream]->codec;
+ _audio_codec_context = _format_context->streams[ffa->id()]->codec;
_audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
if (_audio_codec == 0) {
if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
throw DecodeError ("could not open audio decoder");
}
-
- /* 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.
- */
-
- if (_audio_codec_context->channel_layout == 0) {
- _audio_codec_context->channel_layout = av_get_default_channel_layout (audio_channels ());
- }
}
void
FFmpegDecoder::setup_subtitle ()
{
- if (_subtitle_stream < 0) {
+ if (!_subtitle_stream) {
return;
}
- _subtitle_codec_context = _format_context->streams[_subtitle_stream]->codec;
+ _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec;
_subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
if (_subtitle_codec == 0) {
bool
-FFmpegDecoder::do_pass ()
+FFmpegDecoder::pass ()
{
int r = av_read_frame (_format_context, &_packet);
+
if (r < 0) {
if (r != AVERROR_EOF) {
- throw DecodeError ("error on av_read_frame");
+ /* 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;
- int frame_finished;
+ /* XXX: should we reset _packet.data and size after each *_decode_* call? */
- if (_opt->decode_video) {
- while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
+ int frame_finished;
- process_video (_frame);
- }
+ while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
+ filter_and_emit_video (_frame);
}
- if (_audio_stream >= 0 && _opt->decode_audio) {
+ if (_audio_stream && _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 == _fs->audio_channels);
- process_audio (_frame->data[0], data_size);
+
+ assert (_audio_codec_context->channels == _film->audio_channels());
+ Audio (deinterleave_audio (_frame->data[0], data_size));
}
}
return true;
}
- if (_packet.stream_index == _video_stream && _opt->decode_video) {
+ avcodec_get_frame_defaults (_frame);
+
+ shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
+
+ if (_packet.stream_index == _video_stream) {
int frame_finished;
- if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-
- if (_have_subtitle) {
- for (unsigned int i = 0; i < _subtitle.num_rects; ++i) {
- AVSubtitleRect* rect = _subtitle.rects[i];
- if (rect->type != SUBTITLE_BITMAP) {
- throw DecodeError ("non-bitmap subtitles not yet supported");
- }
+ int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
+ if (r >= 0 && frame_finished) {
- /* XXX: all this assumes YUV420 in _frame */
-
- assert (rect->nb_colors == 4);
- assert (rect->pict.data[0]);
+ if (r != _packet.size) {
+ _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
+ }
- /* Start of the first line in the target frame */
- uint8_t* frame_y_p = _frame->data[0] + rect->y * _frame->linesize[0];
- uint8_t* frame_u_p = _frame->data[1] + (rect->y / 2) * _frame->linesize[1];
- uint8_t* frame_v_p = _frame->data[2] + (rect->y / 2) * _frame->linesize[2];
-
- /* 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];
-
- for (int sub_y = 0; sub_y < rect->h; ++sub_y) {
- /* Pointers to the start of this line */
- uint8_t* sub_line_p = sub_p;
- uint8_t* frame_line_y_p = frame_y_p + rect->x;
- uint8_t* frame_line_u_p = frame_u_p + (rect->x / 2);
- uint8_t* frame_line_v_p = frame_v_p + (rect->x / 2);
-
- /* U and V are subsampled */
- uint8_t current_u = 0;
- uint8_t current_v = 0;
- int subsample_step = 0;
-
- for (int sub_x = 0; sub_x < rect->w; ++sub_x) {
-
- /* RGB value for this subtitle pixel */
- uint32_t const val = palette[*sub_line_p++];
-
- int const red = (val & 0xff);
- int const green = (val & 0xff00) >> 8;
- int const blue = (val & 0xff0000) >> 16;
- float const alpha = ((val & 0xff000000) >> 24) / 255.0;
-
- /* Alpha-blend Y */
- int const cy = *frame_line_y_p;
- *frame_line_y_p++ = int (cy * (1 - alpha)) + int (RGB_TO_Y_CCIR (red, green, blue) * alpha);
-
- /* Store up U and V */
- current_u |= ((RGB_TO_U_CCIR (red, green, blue, 0) & 0xf0) >> 4) << (4 * subsample_step);
- current_v |= ((RGB_TO_V_CCIR (red, green, blue, 0) & 0xf0) >> 4) << (4 * subsample_step);
-
- if (subsample_step == 1 && (sub_y % 2) == 0) {
- /* We have complete U and V bytes, so alpha-blend them into the frame */
- int const cu = *frame_line_u_p;
- int const cv = *frame_line_v_p;
- *frame_line_u_p++ = int (cu * (1 - alpha)) + int (current_u * alpha);
- *frame_line_v_p++ = int (cv * (1 - alpha)) + int (current_v * alpha);
- current_u = current_v = 0;
- }
-
- subsample_step = (subsample_step + 1) % 2;
- }
-
- sub_p += rect->pict.linesize[0];
- frame_y_p += _frame->linesize[0];
- if ((sub_y % 2) == 0) {
- frame_u_p += _frame->linesize[1];
- frame_v_p += _frame->linesize[2];
- }
- }
+ /* Where we are in the output, in seconds */
+ double const out_pts_seconds = video_frame() / frames_per_second();
+
+ /* 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 ("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;
+ }
+
+ /* 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 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()
+ )
+ );
}
}
-
- process_video (_frame);
+
+ 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 ("Frame removed at %1s", out_pts_seconds));
+ }
}
- } else if (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio) {
-
- avcodec_get_frame_defaults (_frame);
-
+ } else if (ffa && _packet.stream_index == ffa->id() && _opt->decode_audio) {
+
int frame_finished;
if (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 == _fs->audio_channels);
- process_audio (_frame->data[0], data_size);
- }
+ /* 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) {
- } else if (_subtitle_stream >= 0 && _packet.stream_index == _subtitle_stream) {
+ 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()) * ffa->sample_rate ());
+
+ _film->log()->log (
+ String::compose (
+ "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, ffa->channels(), bytes_per_audio_sample()
+ )
+ );
+
+ if (s) {
+ shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), s));
+ audio->make_silent ();
+ Audio (audio);
+ }
+ }
- if (_have_subtitle) {
- avsubtitle_free (&_subtitle);
- _have_subtitle = false;
+ 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());
+ Audio (deinterleave_audio (_frame->data[0], data_size));
+ }
}
+
+ } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt->decode_subtitles && _first_video) {
int got_subtitle;
- if (avcodec_decode_subtitle2 (_subtitle_codec_context, &_subtitle, &got_subtitle, &_packet) && got_subtitle) {
- _have_subtitle = true;
+ 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, _first_video.get())));
+ } catch (...) {
+ /* some problem with the subtitle; we probably didn't understand it */
+ }
+ } else {
+ emit_subtitle (shared_ptr<TimedSubtitle> ());
+ }
+ avsubtitle_free (&sub);
}
}
return false;
}
-int
-FFmpegDecoder::length_in_frames () const
+shared_ptr<AudioBuffers>
+FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
{
- return (_format_context->duration / AV_TIME_BASE) * frames_per_second ();
-}
+ assert (_film->audio_channels());
+ assert (bytes_per_audio_sample());
-float
-FFmpegDecoder::frames_per_second () const
-{
- return av_q2d (_format_context->streams[_video_stream]->avg_frame_rate);
-}
+ shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
+ assert (ffa);
+
+ /* Deinterleave and convert to float */
+
+ assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0);
+
+ int const total_samples = size / bytes_per_audio_sample();
+ int const frames = total_samples / _film->audio_channels();
+ shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames));
+
+ switch (audio_sample_format()) {
+ case AV_SAMPLE_FMT_S16:
+ {
+ int16_t* p = (int16_t *) data;
+ 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 == _film->audio_channels()) {
+ channel = 0;
+ ++sample;
+ }
+ }
+ }
+ break;
+
+ case AV_SAMPLE_FMT_S32:
+ {
+ int32_t* p = (int32_t *) data;
+ int sample = 0;
+ int channel = 0;
+ for (int i = 0; i < total_samples; ++i) {
+ audio->data(channel)[sample] = float(*p++) / (1 << 31);
+
+ ++channel;
+ if (channel == _film->audio_channels()) {
+ channel = 0;
+ ++sample;
+ }
+ }
+ }
-int
-FFmpegDecoder::audio_channels () const
-{
- if (_audio_codec_context == 0) {
- return 0;
+ case AV_SAMPLE_FMT_FLTP:
+ {
+ float* p = reinterpret_cast<float*> (data);
+ for (int i = 0; i < _film->audio_channels(); ++i) {
+ memcpy (audio->data(i), p, frames * sizeof(float));
+ p += frames;
+ }
}
+ break;
- return _audio_codec_context->channels;
+ default:
+ assert (false);
+ }
+
+ return audio;
}
-int
-FFmpegDecoder::audio_sample_rate () const
+float
+FFmpegDecoder::frames_per_second () const
{
- if (_audio_codec_context == 0) {
- return 0;
+ 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 _audio_codec_context->sample_rate;
+
+ return av_q2d (s->r_frame_rate);
}
AVSampleFormat
return _audio_codec_context->sample_fmt;
}
-int64_t
-FFmpegDecoder::audio_channel_layout () const
-{
- if (_audio_codec_context == 0) {
- return 0;
- }
-
- return _audio_codec_context->channel_layout;
-}
-
Size
FFmpegDecoder::native_size () const
{
return _video_codec_context->sample_aspect_ratio.den;
}
+string
+FFmpegDecoder::stream_name (AVStream* s) const
+{
+ stringstream n;
+
+ AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
+ if (lang) {
+ n << lang->value;
+ }
+
+ AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
+ if (title) {
+ if (!n.str().empty()) {
+ n << " ";
+ }
+ n << title->value;
+ }
+
+ if (n.str().empty()) {
+ n << "unknown";
+ }
+
+ return n.str ();
+}
+
+int
+FFmpegDecoder::bytes_per_audio_sample () const
+{
+ return av_get_bytes_per_sample (audio_sample_format ());
+}
+
+void
+FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s)
+{
+ AudioDecoder::set_audio_stream (s);
+ setup_audio ();
+}
+
+void
+FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
+{
+ VideoDecoder::set_subtitle_stream (s);
+ setup_subtitle ();
+}
+
+void
+FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
+{
+ shared_ptr<FilterGraph> graph;
+
+ list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
+ while (i != _filter_graphs.end() && !(*i)->can_process (Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
+ ++i;
+ }
+
+ if (i == _filter_graphs.end ()) {
+ graph.reset (new FilterGraph (_film, this, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format));
+ _filter_graphs.push_back (graph);
+ _film->log()->log (String::compose ("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);
+
+ for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
+ emit_video (*i);
+ }
+}
+
+shared_ptr<FFmpegAudioStream>
+FFmpegAudioStream::create (string t, optional<int> v)
+{
+ if (!v) {
+ /* version < 1; no type in the string, and there's only FFmpeg streams anyway */
+ return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
+ }
+
+ stringstream s (t);
+ string type;
+ s >> type;
+ if (type != "ffmpeg") {
+ return shared_ptr<FFmpegAudioStream> ();
+ }
+
+ return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
+}
+
+FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
+{
+ stringstream n (t);
+
+ int name_index = 4;
+ if (!version) {
+ name_index = 2;
+ int channels;
+ n >> _id >> channels;
+ _channel_layout = av_get_default_channel_layout (channels);
+ _sample_rate = 0;
+ } else {
+ string type;
+ /* Current (marked version 1) */
+ n >> type >> _id >> _sample_rate >> _channel_layout;
+ assert (type == "ffmpeg");
+ }
+
+ for (int i = 0; i < name_index; ++i) {
+ size_t const s = t.find (' ');
+ if (s != string::npos) {
+ t = t.substr (s + 1);
+ }
+ }
+
+ _name = t;
+}
+
+string
+FFmpegAudioStream::to_string () const
+{
+ return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name);
+}
+
+/** @return Length (in video frames) according to our content's header */
+SourceFrame
+FFmpegDecoder::length () const
+{
+ return (double(_format_context->duration) / AV_TIME_BASE) * frames_per_second();
+}
projects / dcpomatic.git / blobdiff