X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fffmpeg_decoder.cc;h=885bd0e85e03ad11a79e1ff6da34f2173e05b184;hb=362bc777b951bdbd7e9231c4dfe5cbd183aaa64a;hp=298284512371b87c53ba21edee018434f218d009;hpb=521c09170d9e62cd72cc2da66c41816761008a4b;p=dcpomatic.git

diff --git a/src/lib/ffmpeg_decoder.cc b/src/lib/ffmpeg_decoder.cc
index 298284512..885bd0e85 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
@@ -27,25 +27,32 @@
 #include <iomanip>
 #include <iostream>
 #include <stdint.h>
-#include <boost/lexical_cast.hpp>
 #include <sndfile.h>
 extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 }
-#include "film.h"
 #include "filter.h"
 #include "exceptions.h"
 #include "image.h"
 #include "util.h"
 #include "log.h"
 #include "ffmpeg_decoder.h"
+#include "ffmpeg_audio_stream.h"
+#include "ffmpeg_subtitle_stream.h"
 #include "filter_graph.h"
 #include "audio_buffers.h"
 #include "ffmpeg_content.h"
+#include "image_proxy.h"
+#include "film.h"
+#include "timer.h"
 
 #include "i18n.h"
 
+#define LOG_GENERAL(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
+#define LOG_ERROR(...) _video_content->film()->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
+#define LOG_WARNING(...) _video_content->film()->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
+
 using std::cout;
 using std::string;
 using std::vector;
@@ -53,26 +60,19 @@ using std::stringstream;
 using std::list;
 using std::min;
 using std::pair;
+using std::make_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)
+	, SubtitleDecoder (c)
 	, FFmpeg (c)
-	, _subtitle_codec_context (0)
-	, _subtitle_codec (0)
-	, _decode_video (video)
-	, _decode_audio (audio)
-	, _video_pts_offset (0)
-	, _audio_pts_offset (0)
+	, _log (log)
 {
-	setup_subtitle ();
-
 	/* Audio and video frame PTS values may not start with 0.  We want
 	   to fiddle them so that:
 
@@ -82,45 +82,27 @@ 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 + audio_pts_offset;
-	     video_pts_to_use = video_pts_from_ffmpeg + video_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 = _decode_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 */
 
 	if (have_audio && have_video) {
-		_video_pts_offset = _audio_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
+		_pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
 	} else if (have_video) {
-		_video_pts_offset = - c->first_video().get();
+		_pts_offset = - c->first_video().get();
 	} else if (have_audio) {
-		_audio_pts_offset = - c->audio_stream()->first_audio.get();
+		_pts_offset = - c->audio_stream()->first_audio.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 ()
-{
-	boost::mutex::scoped_lock lm (_mutex);
-
-	if (_subtitle_codec_context) {
-		avcodec_close (_subtitle_codec_context);
+		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;
 	}
 }
 
@@ -134,15 +116,12 @@ 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 ();
 	}
-
-	AudioDecoder::flush ();
 }
 
 bool
@@ -155,27 +134,20 @@ 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_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
 		}
 
 		flush ();
 		return true;
 	}
 
-	avcodec_get_frame_defaults (_frame);
-
-	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() && si == _ffmpeg_content->audio_stream()->index (_format_context) && _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() && si == _ffmpeg_content->subtitle_stream()->index (_format_context) && film->with_subtitles ()) {
+	} else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si)) {
 		decode_subtitle_packet ();
 	}
 
@@ -201,6 +173,23 @@ 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]);
@@ -295,131 +284,45 @@ FFmpegDecoder::bytes_per_audio_sample () const
 	return av_get_bytes_per_sample (audio_sample_format ());
 }
 
-int
-FFmpegDecoder::minimal_run (boost::function<bool (optional<ContentTime>, optional<ContentTime>, int)> finished)
+void
+FFmpegDecoder::seek (ContentTime time, bool accurate)
 {
-	int frames_read = 0;
-	optional<ContentTime> last_video;
-	optional<ContentTime> last_audio;
-
-	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) {
+	VideoDecoder::seek (time, accurate);
+	AudioDecoder::seek (time, accurate);
+	
+	/* If we are doing an `accurate' seek, we need to use pre-roll, as
+	   we don't really know what the seek will give us.
+	*/
 
-			avcodec_get_frame_defaults (_frame);
-			
-			int finished = 0;
-			r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
-			if (r >= 0 && finished) {
-				last_video = rint (
-					(av_frame_get_best_effort_timestamp (_frame) * time_base + _video_pts_offset) * TIME_HZ
-					);
-			}
-				
-		} else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->index (_format_context)) {
-			AVPacket copy_packet = _packet;
-			while (copy_packet.size > 0) {
-
-				int finished;
-				r = avcodec_decode_audio4 (audio_codec_context(), _frame, &finished, &_packet);
-				if (r >= 0 && finished) {
-					last_audio = rint (
-						(av_frame_get_best_effort_timestamp (_frame) * time_base + _audio_pts_offset) *	TIME_HZ
-						);
-				}
-					
-				copy_packet.data += r;
-				copy_packet.size -= r;
-			}
-		}
-		
-		av_free_packet (&_packet);
+	ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
+	time -= pre_roll;
+	if (time < ContentTime (0)) {
+		time = ContentTime (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;
-}
+	ContentTime const u = time - _pts_offset;
+	int64_t s = u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base);
 
-void
-FFmpegDecoder::seek_and_flush (ContentTime t)
-{
-	int64_t s = ((double (t) / TIME_HZ) - _video_pts_offset) /
-		av_q2d (_format_context->streams[_video_stream]->time_base);
-	
 	if (_ffmpeg_content->audio_stream ()) {
 		s = min (
-			s, int64_t (
-				((double (t) / TIME_HZ) - _audio_pts_offset) /
-				av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base)
-				)
+			s, int64_t (u.seconds() / av_q2d (_ffmpeg_content->audio_stream()->stream(_format_context)->time_base))
 			);
 	}
 
-	av_seek_frame (_format_context, _video_stream, s, AVSEEK_FLAG_BACKWARD);
+	/* Ridiculous empirical hack */
+	s--;
+	if (s < 0) {
+		s = 0;
+	}
+
+	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);
-	}
-}
-
-void
-FFmpegDecoder::seek (ContentTime time, bool accurate)
-{
-	Decoder::seek (time, accurate);
-	
-	/* If we are doing an accurate seek, our initial shot will be 200ms (200 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.
-	*/
-
-	ContentTime pre_roll = accurate ? (0.2 * TIME_HZ) : 0;
-	ContentTime initial_seek = time - pre_roll;
-	if (initial_seek < 0) {
-		initial_seek = 0;
-	}
-	
-	/* Initial seek time in the video stream's timebase */
-
-	seek_and_flush (initial_seek);
-
-	if (!accurate) {
-		/* That'll do */
-		return;
-	}
-
-	int const N = minimal_run (boost::bind (&FFmpegDecoder::seek_overrun_finished, this, time, _1, _2));
-	
-	seek_and_flush (initial_seek);
-	if (N > 0) {
-		minimal_run (boost::bind (&FFmpegDecoder::seek_final_finished, this, N - 1, _3));
+	if (subtitle_codec_context ()) {
+		avcodec_flush_buffers (subtitle_codec_context ());
 	}
 }
 
@@ -438,23 +341,21 @@ FFmpegDecoder::decode_audio_packet ()
 		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_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result);
 			return;
 		}
 
 		if (frame_finished) {
-			ContentTime const t = rint (
-				(av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
-				 * av_frame_get_best_effort_timestamp(_frame) + _audio_pts_offset) * TIME_HZ
-				);
-
+			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), t);
+
+			audio (deinterleave_audio (_frame->data, data_size), ct);
 		}
 			
 		copy_packet.data += decode_result;
@@ -475,74 +376,44 @@ 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_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
 	} else {
 		graph = *i;
 	}
 
 	list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
 
-	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) {
 
 		shared_ptr<Image> image = i->first;
-		if (!post_process.empty ()) {
-			image = image->post_process (post_process, true);
-		}
 		
 		if (i->second != AV_NOPTS_VALUE) {
-			ContentTime const t = rint ((i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _video_pts_offset) * TIME_HZ);
-			video (image, false, t);
+			double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
+			video (
+				shared_ptr<ImageProxy> (new RawImageProxy (image, _video_content->film()->log())),
+				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_WARNING ("Dropping frame without PTS");
 		}
 	}
 
 	return true;
 }
-
-	
-void
-FFmpegDecoder::setup_subtitle ()
-{
-	boost::mutex::scoped_lock lm (_mutex);
 	
-	if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->index (_format_context) >= int (_format_context->nb_streams)) {
-		return;
-	}
-
-	_subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
-	_subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
-
-	if (_subtitle_codec == 0) {
-		throw DecodeError (_("could not find subtitle decoder"));
-	}
-	
-	if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
-		throw DecodeError (N_("could not open subtitle decoder"));
-	}
-}
-
 void
 FFmpegDecoder::decode_subtitle_packet ()
 {
 	int got_subtitle;
 	AVSubtitle sub;
-	if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
+	if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
 		return;
 	}
 
@@ -550,60 +421,68 @@ 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 (ContentTimePeriod (), 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;
-	
-	/* hence start time for this sub */
-	ContentTime const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
-	ContentTime const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
+	ContentTimePeriod period = subtitle_period (sub) + _pts_offset;
 
 	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;
 	}
-	
-	shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
+
+	/* 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, dcp::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 */
+	/* 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) {
-			*out_line_p++ = palette[*sub_line_p++];
+			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 ();
+	dcp::Size const vs = _ffmpeg_content->video_size ();
 
-	subtitle (
+	image_subtitle (
+		period,
 		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);
 }
+
+bool
+FFmpegDecoder::has_subtitle_during (ContentTimePeriod p) const
+{
+	return _ffmpeg_content->has_subtitle_during (p);
+}