X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=851c64606b705af1974c1c953fa1d14863c6af89;hb=16e8c1c7222796246f74153bf294909c3efe57dc;hp=3b7d727b88eec115b508da254d806da5f90b9f27;hpb=df52b97f307605aad15ab5f01c8fdcf93afc9d15;p=dcpomatic.git

diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc
index 3b7d727b8..851c64606 100644
--- a/src/lib/ffmpeg_decoder.cc
+++ b/src/lib/ffmpeg_decoder.cc
@@ -41,8 +41,8 @@ extern "C" {
 #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"
 
@@ -52,6 +52,7 @@ 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;
@@ -59,8 +60,9 @@ 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)
+	, VideoDecoder (f, c)
+	, AudioDecoder (f, c)
+	, SubtitleDecoder (f)
 	, FFmpeg (c)
 	, _subtitle_codec_context (0)
 	, _subtitle_codec (0)
@@ -71,23 +73,45 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
 {
 	setup_subtitle ();
 
-	if (video && audio && c->audio_stream() && c->first_video() && c->audio_stream()->first_audio) {
-		_pts_offset = compute_pts_offset (c->first_video().get(), c->audio_stream()->first_audio.get(), c->video_frame_rate());
-	}
-}
+	/* Audio and video frame PTS values may not start with 0.  We want
+	   to fiddle them so that:
 
-double
-FFmpegDecoder::compute_pts_offset (double first_video, double first_audio, float video_frame_rate)
-{
-	double const old_first_video = first_video;
-	
-	/* Round the first video to a frame boundary */
-	if (fabs (rint (first_video * video_frame_rate) - first_video * video_frame_rate) > 1e-6) {
-		first_video = ceil (first_video * video_frame_rate) / video_frame_rate;
+	   1.  One of them starts at time 0.
+	   2.  The first video PTS value ends up on a frame boundary.
+
+	   Then we remove big initial gaps in PTS and we allow our
+	   insertion of black frames to work.
+
+	   We will do:
+	     audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
+	     video_pts_to_use = video_pts_from_ffmpeg + 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) {
+		_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();
 	}
 
-	/* Compute the required offset (also removing any common start delay) */
-	return first_video - old_first_video - min (first_video, first_audio);
+	/* Now adjust both so that the video pts starts on a frame */
+	if (have_video && have_audio) {
+		double first_video = c->first_video().get() + _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 ();
+		}
+
+		_pts_offset += first_video - old_first_video;
+	}
 }
 
 FFmpegDecoder::~FFmpegDecoder ()
@@ -97,7 +121,30 @@ FFmpegDecoder::~FFmpegDecoder ()
 	if (_subtitle_codec_context) {
 		avcodec_close (_subtitle_codec_context);
 	}
-}	
+}
+
+void
+FFmpegDecoder::flush ()
+{
+	/* 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 ();
+}
 
 void
 FFmpegDecoder::pass ()
@@ -114,55 +161,21 @@ FFmpegDecoder::pass ()
 			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 ();
+		flush ();
 		return;
 	}
 
-	avcodec_get_frame_defaults (_frame);
+	shared_ptr<const Film> film = _film.lock ();
+	assert (film);
 
-	if (_packet.stream_index == _video_stream && _decode_video) {
+	int const si = _packet.stream_index;
+	
+	if (si == _video_stream && _decode_video) {
 		decode_video_packet ();
-	} else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+	} 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) {
-#if 0		
-
-		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 {
-					subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
-				} catch (...) {
-					/* some problem with the subtitle; we probably didn't understand it */
-				}
-			} else {
-				subtitle (shared_ptr<TimedSubtitle> ());
-			}
-			avsubtitle_free (&sub);
-		}
-#endif		
+	} else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
+		decode_subtitle_packet ();
 	}
 
 	av_free_packet (&_packet);
@@ -186,6 +199,23 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
 	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]);
@@ -281,58 +311,76 @@ FFmpegDecoder::bytes_per_audio_sample () const
 }
 
 void
-FFmpegDecoder::seek (VideoContent::Frame frame)
+FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
 {
-	do_seek (frame, false, false);
-}
+	double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
 
-void
-FFmpegDecoder::seek_back ()
-{
-	if (_video_position == 0) {
-		return;
+	/* 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;
 	}
-	
-	do_seek (_video_position - 1, true, true);
-}
 
-void
-FFmpegDecoder::do_seek (VideoContent::Frame frame, bool backwards, bool accurate)
-{
-	int64_t const vt = frame / (_ffmpeg_content->video_frame_rate() * av_q2d (_format_context->streams[_video_stream]->time_base));
-	av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
+	if (initial < 0) {
+		initial = 0;
+	}
+
+	/* Initial seek time in the stream's timebase */
+	int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base;
+
+	av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
 
 	avcodec_flush_buffers (video_codec_context());
 	if (_subtitle_codec_context) {
 		avcodec_flush_buffers (_subtitle_codec_context);
 	}
 
-	_just_sought = true;
+	/* This !accurate is piling hack upon hack; setting _just_sought to true
+	   even with accurate == true defeats our attempt to align the start
+	   of the video and audio.  Here we disable that defeat when accurate == true
+	   i.e. when we are making a DCP rather than just previewing one.
+	   Ewww.  This should be gone in 2.0.
+	*/
+	if (!accurate) {
+		_just_sought = true;
+	}
+	
+	_video_position = frame;
+	
+	if (frame == 0 || !accurate) {
+		/* We're already there, or we're as close as we need to be */
+		return;
+	}
 
-	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 > vt) {
-						_video_position = (bet * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset)
-							* _ffmpeg_content->video_frame_rate();
-						break;
-					}
-				}
-			}
-			
+	while (1) {
+		int r = av_read_frame (_format_context, &_packet);
+		if (r < 0) {
+			return;
+		}
+
+		if (_packet.stream_index != _video_stream) {
 			av_free_packet (&_packet);
+			continue;
 		}
+		
+		int finished = 0;
+		r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
+		if (r >= 0 && finished) {
+			_video_position = rint (
+				(av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
+				);
+
+			if (_video_position >= (frame - 1)) {
+				av_free_packet (&_packet);
+				break;
+			}
+		}
+		
+		av_free_packet (&_packet);
 	}
 }
 
@@ -344,37 +392,48 @@ FFmpegDecoder::decode_audio_packet ()
 	*/
 	
 	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, &copy_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) - _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);
-					}
-				}
+		if (decode_result < 0) {
+			shared_ptr<const Film> film = _film.lock ();
+			assert (film);
+			film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
+			return;
+		}
+
+		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) + _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()
+							)
+						);
 					
-				copy_packet.data += decode_result;
-				copy_packet.size -= decode_result;
+					silence->make_silent ();
+					audio (silence, _audio_position);
+				}
 			}
+			
+			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), _audio_position);
 		}
+			
+		copy_packet.data += decode_result;
+		copy_packet.size -= decode_result;
 	}
 }
 
@@ -385,7 +444,7 @@ FFmpegDecoder::decode_video_packet ()
 	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;
@@ -407,30 +466,24 @@ FFmpegDecoder::decode_video_packet ()
 		graph = *i;
 	}
 
-	list<shared_ptr<Image> > images = graph->process (_frame);
+	list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
 
-	string post_process = Filter::ffmpeg_strings (_ffmpeg_content->filters()).second;
-	
-	for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
+	for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
 
-		shared_ptr<Image> image = *i;
-		if (!post_process.empty ()) {
-			image = image->post_process (post_process, true);
-		}
+		shared_ptr<Image> image = i->first;
 		
-		int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
-		if (bet != AV_NOPTS_VALUE) {
+		if (i->second != AV_NOPTS_VALUE) {
 
-			double const pts = bet * av_q2d (_format_context->streams[_video_stream]->time_base) - _pts_offset;
+			double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
 
 			if (_just_sought) {
 				/* We just did a seek, so disable any attempts to correct for where we
 				   are / should be.
 				*/
-				_video_position = pts * _ffmpeg_content->video_frame_rate ();
+				_video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
 				_just_sought = false;
 			}
-			
+
 			double const next = _video_position / _ffmpeg_content->video_frame_rate();
 			double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
 			double delta = pts - next;
@@ -439,8 +492,10 @@ FFmpegDecoder::decode_video_packet ()
 				/* This PTS is more than one frame forward in time of where we think we should be; emit
 				   a black frame.
 				*/
+
+				/* XXX: I think this should be a copy of the last frame... */
 				boost::shared_ptr<Image> black (
-					new SimpleImage (
+					new Image (
 						static_cast<AVPixelFormat> (_frame->format),
 						libdcp::Size (video_codec_context()->width, video_codec_context()->height),
 						true
@@ -473,15 +528,19 @@ FFmpegDecoder::setup_subtitle ()
 {
 	boost::mutex::scoped_lock lm (_mutex);
 	
-	if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
+	if (!_ffmpeg_content->subtitle_stream()) {
 		return;
 	}
 
-	_subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
+	_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 (_("could not find subtitle decoder"));
+		throw DecodeError (N_("could not find subtitle decoder"));
 	}
 	
 	if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
@@ -497,3 +556,79 @@ FFmpegDecoder::done () const
 	return vd && ad;
 }
 	
+void
+FFmpegDecoder::decode_subtitle_packet ()
+{
+	int got_subtitle;
+	AVSubtitle sub;
+	if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
+		return;
+	}
+
+	/* 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) + _pts_offset;
+
+	/* 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"));
+	}
+
+	/* 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, 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 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);
+	}
+
+	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);
+}