X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;ds=sidebyside;f=src%2Flib%2Fffmpeg_decoder.cc;h=acaf149f43ade599dff78d95c109d36c86133097;hb=0b781dd65a90e617866810ecb3dee27267449f1f;hp=5eb32a0147f14e2d277474f6fe46c8fbbaf650ee;hpb=579dfce1bae018efd73c9f5f3f31770cb401d19a;p=dcpomatic.git

diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc
index 5eb32a014..acaf149f4 100644
--- a/src/lib/ffmpeg_decoder.cc
+++ b/src/lib/ffmpeg_decoder.cc
@@ -41,24 +41,30 @@ extern "C" {
 #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"
+#include "filter_graph.h"
 #include "subtitle.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)
+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)
@@ -86,8 +92,9 @@ FFmpegDecoder::~FFmpegDecoder ()
 	if (_subtitle_codec_context) {
 		avcodec_close (_subtitle_codec_context);
 	}
-	
+
 	av_free (_frame);
+	
 	avformat_close_input (&_format_context);
 }	
 
@@ -98,8 +105,8 @@ FFmpegDecoder::setup_general ()
 	
 	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) {
@@ -113,28 +120,30 @@ FFmpegDecoder::setup_general ()
 		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;
+
+			/* 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 (AudioStream (stream_name (s), i, 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) {
-			if (_subtitle_stream == -1) {
-				_subtitle_stream = i;
-			}
-			_subtitle_streams.push_back (SubtitleStream (stream_name (s), i));
+			_subtitle_streams.push_back (
+				shared_ptr<SubtitleStream> (
+					new SubtitleStream (stream_name (s), i)
+					)
+				);
 		}
 	}
 
-	/* Now override audio and subtitle streams with those from the Film, if it has any */
-
-	if (_fs->audio_stream_index() != -1) {
-		_audio_stream = _fs->audio_stream().id();
-	}
-
-	if (_fs->subtitle_stream_index() != -1) {
-		_subtitle_stream = _fs->subtitle_stream().id ();
-	}
-
 	if (_video_stream < 0) {
 		throw DecodeError ("could not find video stream");
 	}
@@ -154,7 +163,7 @@ FFmpegDecoder::setup_video ()
 	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");
 	}
@@ -163,11 +172,14 @@ FFmpegDecoder::setup_video ()
 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) {
@@ -177,24 +189,16 @@ FFmpegDecoder::setup_audio ()
 	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) {
@@ -208,13 +212,16 @@ FFmpegDecoder::setup_subtitle ()
 
 
 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 */
@@ -227,93 +234,146 @@ FFmpegDecoder::do_pass ()
 		int frame_finished;
 
 		while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-			process_video (_frame);
+			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;
 	}
 
-	double const pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base) * _packet.pts;
+	avcodec_get_frame_defaults (_frame);
+
+	shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
 	
 	if (_packet.stream_index == _video_stream) {
 
-		if (!_first_video) {
-			_first_video = pts_seconds;
-		}
-		
 		int frame_finished;
-		if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-			process_video (_frame);
-		}
+		int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
+		if (r >= 0 && frame_finished) {
 
-	} else if (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio && _first_video && _first_video.get() <= pts_seconds) {
+			if (r != _packet.size) {
+				_film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
+			}
 
-		/* Note: 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_audio) {
-			_first_audio = pts_seconds;
-			
-			/* This is our first audio packet, and if we've arrived here we must have had our
-			   first video packet.  Push some silence to make up the gap between our first
-			   video packet and our first audio.
-			*/
-			
-			/* frames of silence that we must push */
-			int const s = rint ((_first_audio.get() - _first_video.get()) * audio_sample_rate ());
-			
-			_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()
-					)
+			/* 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
 				);
-			
-			/* 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 (!_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()
+							)
+						);
+				}
+			}
+
+			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));
+			}
 		}
-		
-		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
-				);
+
+			/* 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) {
+
+				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.
+					*/
 			
-			assert (_audio_codec_context->channels == _fs->audio_channels());
-			process_audio (_frame->data[0], data_size);
+					/* 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);
+					}
+				}
+
+				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 >= 0 && _packet.stream_index == _subtitle_stream && _opt->decode_subtitles) {
+	} else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt->decode_subtitles && _first_video) {
 
 		int got_subtitle;
 		AVSubtitle sub;
 		if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
-			/* I'm not entirely sure why, but sometimes we get an AVSubtitle with
-			   no AVSubtitleRects.
+			/* 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)));
+				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);
 		}
@@ -323,10 +383,72 @@ FFmpegDecoder::do_pass ()
 	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());
+
+	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;
+			}
+		}
+	}
+
+	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;
+
+	default:
+		assert (false);
+	}
+
+	return audio;
 }
 
 float
@@ -341,26 +463,6 @@ FFmpegDecoder::frames_per_second () const
 	return av_q2d (s->r_frame_rate);
 }
 
-int
-FFmpegDecoder::audio_channels () const
-{
-	if (_audio_codec_context == 0) {
-		return 0;
-	}
-
-	return _audio_codec_context->channels;
-}
-
-int
-FFmpegDecoder::audio_sample_rate () const
-{
-	if (_audio_codec_context == 0) {
-		return 0;
-	}
-	
-	return _audio_codec_context->sample_rate;
-}
-
 AVSampleFormat
 FFmpegDecoder::audio_sample_format () const
 {
@@ -371,16 +473,6 @@ FFmpegDecoder::audio_sample_format () const
 	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
 {
@@ -417,24 +509,6 @@ FFmpegDecoder::sample_aspect_ratio_denominator () const
 	return _video_codec_context->sample_aspect_ratio.den;
 }
 
-bool
-FFmpegDecoder::has_subtitles () const
-{
-	return (_subtitle_stream != -1);
-}
-
-vector<AudioStream>
-FFmpegDecoder::audio_streams () const
-{
-	return _audio_streams;
-}
-
-vector<SubtitleStream>
-FFmpegDecoder::subtitle_streams () const
-{
-	return _subtitle_streams;
-}
-
 string
 FFmpegDecoder::stream_name (AVStream* s) const
 {
@@ -460,3 +534,106 @@ FFmpegDecoder::stream_name (AVStream* s) const
 	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();
+}