X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;fp=src%2Flib%2Fffmpeg_decoder.cc;h=d251a37446ff96e2e49d514575b7719e53f44b6c;hp=7a5bf8ba832f51a538ea57775147ba20296a3c06;hb=39bc73fe192f932ed6695eb87b19de446e8b4f55;hpb=884093edac5f63d26c02d411c73dfb52376ccf8f

diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc
index 7a5bf8ba8..d251a3744 100644
--- a/src/lib/ffmpeg_decoder.cc
+++ b/src/lib/ffmpeg_decoder.cc
@@ -1,5 +1,5 @@
 /*
-    Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2012-2014 Carl Hetherington <cth@carlh.net>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
@@ -32,7 +32,6 @@ extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 }
-#include "film.h"
 #include "filter.h"
 #include "exceptions.h"
 #include "image.h"
@@ -56,20 +55,15 @@ using std::pair;
 using boost::shared_ptr;
 using boost::optional;
 using boost::dynamic_pointer_cast;
-using libdcp::Size;
+using dcp::Size;
 
-FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
-	: Decoder (f)
-	, VideoDecoder (f, c)
-	, AudioDecoder (f, c)
-	, SubtitleDecoder (f)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log)
+	: VideoDecoder (c)
+	, AudioDecoder (c)
 	, FFmpeg (c)
+	, _log (log)
 	, _subtitle_codec_context (0)
 	, _subtitle_codec (0)
-	, _decode_video (video)
-	, _decode_audio (audio)
-	, _pts_offset (0)
-	, _just_sought (false)
 {
 	setup_subtitle ();
 
@@ -82,13 +76,11 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
 	   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;
+	   We will do pts_to_use = 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;
+	bool const have_video = c->first_video();
+	bool const have_audio = c->audio_stream () && c->audio_stream()->first_audio;
 
 	/* First, make one of them start at 0 */
 
@@ -102,15 +94,9 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegC
 
 	/* 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;
+		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;
 	}
 }
 
@@ -133,20 +119,15 @@ FFmpegDecoder::flush ()
 	
 	/* XXX: should we reset _packet.data and size after each *_decode_* call? */
 	
-	if (_decode_video) {
-		while (decode_video_packet ()) {}
-	}
+	while (decode_video_packet ()) {}
 	
-	if (_ffmpeg_content->audio_stream() && _decode_audio) {
+	if (_ffmpeg_content->audio_stream()) {
 		decode_audio_packet ();
+		AudioDecoder::flush ();
 	}
-
-	/* Stop us being asked for any more data */
-	_video_position = _ffmpeg_content->video_length_after_3d_combine ();
-	_audio_position = _ffmpeg_content->audio_length ();
 }
 
-void
+bool
 FFmpegDecoder::pass ()
 {
 	int r = av_read_frame (_format_context, &_packet);
@@ -156,29 +137,25 @@ FFmpegDecoder::pass ()
 			/* 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));
+			_log->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
 		}
 
 		flush ();
-		return;
+		return true;
 	}
 
-	shared_ptr<const Film> film = _film.lock ();
-	assert (film);
-
 	int const si = _packet.stream_index;
 	
-	if (si == _video_stream && _decode_video) {
+	if (si == _video_stream) {
 		decode_video_packet ();
-	} else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
+	} else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si)) {
 		decode_audio_packet ();
-	} else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
+	} else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) {
 		decode_subtitle_packet ();
 	}
 
 	av_free_packet (&_packet);
+	return false;
 }
 
 /** @param data pointer to array of pointers to buffers.
@@ -310,77 +287,131 @@ FFmpegDecoder::bytes_per_audio_sample () const
 	return av_get_bytes_per_sample (audio_sample_format ());
 }
 
-void
-FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
+int
+FFmpegDecoder::minimal_run (boost::function<bool (optional<ContentTime>, optional<ContentTime>, int)> finished)
 {
-	double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
+	int frames_read = 0;
+	optional<ContentTime> last_video;
+	optional<ContentTime> last_audio;
 
-	/* 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;
+	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 got_picture = 0;
+			r = avcodec_decode_video2 (video_codec_context(), _frame, &got_picture, &_packet);
+			if (r >= 0 && got_picture) {
+				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) {
 
-	if (accurate) {
-		initial -= 5;
+				int got_frame;
+				r = avcodec_decode_audio4 (audio_codec_context(), _frame, &got_frame, &_packet);
+				if (r >= 0 && got_frame) {
+					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);
 	}
 
-	if (initial < 0) {
-		initial = 0;
+	return frames_read;
+}
+
+bool
+FFmpegDecoder::seek_overrun_finished (ContentTime seek, optional<ContentTime> last_video, optional<ContentTime> last_audio) const
+{
+	return (last_video && last_video.get() >= seek) || (last_audio && last_audio.get() >= seek);
+}
+
+bool
+FFmpegDecoder::seek_final_finished (int n, int done) const
+{
+	return n == done;
+}
+
+void
+FFmpegDecoder::seek_and_flush (ContentTime t)
+{
+	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))
+			);
 	}
 
-	/* Initial seek time in the stream's timebase */
-	int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base;
+	/* Ridiculous empirical hack */
+	s--;
+	if (s < 0) {
+		s = 0;
+	}
 
-	av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
+	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);
 	}
+}
 
-	/* 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.
+void
+FFmpegDecoder::seek (ContentTime time, bool accurate)
+{
+	VideoDecoder::seek (time, accurate);
+	AudioDecoder::seek (time, accurate);
+	
+	/* If we are doing an accurate seek, our initial shot will be 2s (2 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.
 	*/
-	if (!accurate) {
-		_just_sought = true;
+
+	ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
+	ContentTime initial_seek = time - pre_roll;
+	if (initial_seek < ContentTime (0)) {
+		initial_seek = ContentTime (0);
 	}
-	
-	_video_position = frame;
-	
-	if (frame == 0 || !accurate) {
-		/* We're already there, or we're as close as we need to be */
+
+	/* Initial seek time in the video stream's timebase */
+
+	seek_and_flush (initial_seek);
+
+	if (!accurate) {
+		/* That'll do */
 		return;
 	}
 
-	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()
-				);
+	int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2));
 
-			if (_video_position >= (frame - 1)) {
-				av_free_packet (&_packet);
-				break;
-			}
-		}
-		
-		av_free_packet (&_packet);
+	seek_and_flush (initial_seek);
+	if (N > 0) {
+		minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3));
 	}
 }
 
@@ -397,39 +428,23 @@ FFmpegDecoder::decode_audio_packet ()
 
 		int frame_finished;
 		int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, &copy_packet);
+
 		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));
+			_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()
-							)
-						);
-					
-					silence->make_silent ();
-					audio (silence, _audio_position);
-				}
-			}
+			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), _audio_position);
+
+			audio (deinterleave_audio (_frame->data, data_size), ct);
 		}
 			
 		copy_packet.data += decode_result;
@@ -450,18 +465,14 @@ FFmpegDecoder::decode_video_packet ()
 	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 ()) {
-		shared_ptr<const Film> film = _film.lock ();
-		assert (film);
-
-		graph.reset (new FilterGraph (_ffmpeg_content, 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;
 	}
@@ -473,49 +484,10 @@ FFmpegDecoder::decode_video_packet ()
 		shared_ptr<Image> image = i->first;
 		
 		if (i->second != AV_NOPTS_VALUE) {
-
-			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 = 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;
-
-			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.
-				*/
-
-				/* XXX: I think this should be a copy of the last frame... */
-				boost::shared_ptr<Image> black (
-					new Image (
-						static_cast<AVPixelFormat> (_frame->format),
-						libdcp::Size (video_codec_context()->width, video_codec_context()->height),
-						true
-						)
-					);
-				
-				black->make_black ();
-				video (shared_ptr<ImageProxy> (new RawImageProxy (image)), false, _video_position);
-				delta -= one_frame;
-			}
-
-			if (delta > -one_frame) {
-				/* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
-				video (shared_ptr<ImageProxy> (new RawImageProxy (image)), false, _video_position);
-			}
-				
+			double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
+			video (shared_ptr<ImageProxy> (new RawImageProxy (image)), rint (pts * _ffmpeg_content->video_frame_rate ()));
 		} else {
-			shared_ptr<const Film> film = _film.lock ();
-			assert (film);
-			film->log()->log ("Dropping frame without PTS");
+			_log->log ("Dropping frame without PTS");
 		}
 	}
 
@@ -548,14 +520,6 @@ FFmpegDecoder::setup_subtitle ()
 	}
 }
 
-bool
-FFmpegDecoder::done () const
-{
-	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;
-}
-	
 void
 FFmpegDecoder::decode_subtitle_packet ()
 {
@@ -569,31 +533,33 @@ FFmpegDecoder::decode_subtitle_packet ()
 	   indicate that the previous subtitle should stop.
 	*/
 	if (sub.num_rects <= 0) {
-		subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
+		image_subtitle (ContentTime (), ContentTime (), shared_ptr<Image> (), dcpomatic::Rect<double> ());
 		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
+	/* Subtitle PTS (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;
-
+	ContentTime packet_time = ContentTime::from_seconds (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;
+	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);
 
 	AVSubtitleRect const * rect = sub.rects[0];
 
 	if (rect->type != SUBTITLE_BITMAP) {
-		throw DecodeError (_("non-bitmap subtitles not yet supported"));
+		/* XXX */
+		// throw DecodeError (_("non-bitmap subtitles not yet supported"));
+		return;
 	}
 
 	/* 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));
+	shared_ptr<Image> image (new Image (PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true));
 
 	/* Start of the first line in the subtitle */
 	uint8_t* sub_p = rect->pict.data[0];
@@ -615,20 +581,19 @@ FFmpegDecoder::decode_subtitle_packet ()
 		out_p += image->stride()[0] / sizeof (uint32_t);
 	}
 
-	libdcp::Size const vs = _ffmpeg_content->video_size ();
+	dcp::Size const vs = _ffmpeg_content->video_size ();
 
-	subtitle (
+	image_subtitle (
+		from,
+		to,
 		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);
 }