2 Copyright (C) 2012-2014 Carl Hetherington <cth@carlh.net>
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 /** @file src/ffmpeg_decoder.cc
21 * @brief A decoder using FFmpeg to decode content.
32 #include <libavcodec/avcodec.h>
33 #include <libavformat/avformat.h>
37 #include "exceptions.h"
41 #include "ffmpeg_decoder.h"
42 #include "filter_graph.h"
43 #include "audio_buffers.h"
44 #include "ffmpeg_content.h"
45 #include "image_proxy.h"
49 #define LOG_GENERAL(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
50 #define LOG_ERROR(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
51 #define LOG_WARNING(...) film->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
59 using boost::shared_ptr;
60 using boost::optional;
61 using boost::dynamic_pointer_cast;
64 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
70 , _subtitle_codec_context (0)
72 , _decode_video (video)
73 , _decode_audio (audio)
75 , _just_sought (false)
79 /* Audio and video frame PTS values may not start with 0. We want
80 to fiddle them so that:
82 1. One of them starts at time 0.
83 2. The first video PTS value ends up on a frame boundary.
85 Then we remove big initial gaps in PTS and we allow our
86 insertion of black frames to work.
89 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
90 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
93 bool const have_video = video && c->first_video();
94 bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
96 /* First, make one of them start at 0 */
98 if (have_audio && have_video) {
99 _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
100 } else if (have_video) {
101 _pts_offset = - c->first_video().get();
102 } else if (have_audio) {
103 _pts_offset = - c->audio_stream()->first_audio.get();
106 /* Now adjust both so that the video pts starts on a frame */
107 if (have_video && have_audio) {
108 double first_video = c->first_video().get() + _pts_offset;
109 double const old_first_video = first_video;
111 /* Round the first video up to a frame boundary */
112 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
113 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
116 _pts_offset += first_video - old_first_video;
120 FFmpegDecoder::~FFmpegDecoder ()
122 boost::mutex::scoped_lock lm (_mutex);
124 if (_subtitle_codec_context) {
125 avcodec_close (_subtitle_codec_context);
130 FFmpegDecoder::flush ()
132 /* Get any remaining frames */
137 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
140 while (decode_video_packet ()) {}
143 if (_ffmpeg_content->audio_stream() && _decode_audio) {
144 decode_audio_packet ();
147 /* Stop us being asked for any more data */
148 _video_position = _ffmpeg_content->video_length_after_3d_combine ();
149 _audio_position = _ffmpeg_content->audio_length ();
153 FFmpegDecoder::pass ()
155 int r = av_read_frame (_format_context, &_packet);
158 if (r != AVERROR_EOF) {
159 /* Maybe we should fail here, but for now we'll just finish off instead */
161 av_strerror (r, buf, sizeof(buf));
162 shared_ptr<const Film> film = _film.lock ();
164 LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
171 shared_ptr<const Film> film = _film.lock ();
174 int const si = _packet.stream_index;
176 if (si == _video_stream && _decode_video) {
177 decode_video_packet ();
178 } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
179 decode_audio_packet ();
180 } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
181 decode_subtitle_packet ();
184 av_free_packet (&_packet);
187 /** @param data pointer to array of pointers to buffers.
188 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
190 shared_ptr<AudioBuffers>
191 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
193 assert (_ffmpeg_content->audio_channels());
194 assert (bytes_per_audio_sample());
196 /* Deinterleave and convert to float */
198 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
200 int const total_samples = size / bytes_per_audio_sample();
201 int const frames = total_samples / _ffmpeg_content->audio_channels();
202 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
204 switch (audio_sample_format()) {
205 case AV_SAMPLE_FMT_U8:
207 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
210 for (int i = 0; i < total_samples; ++i) {
211 audio->data(channel)[sample] = float(*p++) / (1 << 23);
214 if (channel == _ffmpeg_content->audio_channels()) {
222 case AV_SAMPLE_FMT_S16:
224 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
227 for (int i = 0; i < total_samples; ++i) {
228 audio->data(channel)[sample] = float(*p++) / (1 << 15);
231 if (channel == _ffmpeg_content->audio_channels()) {
239 case AV_SAMPLE_FMT_S16P:
241 int16_t** p = reinterpret_cast<int16_t **> (data);
242 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
243 for (int j = 0; j < frames; ++j) {
244 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
250 case AV_SAMPLE_FMT_S32:
252 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
255 for (int i = 0; i < total_samples; ++i) {
256 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
259 if (channel == _ffmpeg_content->audio_channels()) {
267 case AV_SAMPLE_FMT_FLT:
269 float* p = reinterpret_cast<float*> (data[0]);
272 for (int i = 0; i < total_samples; ++i) {
273 audio->data(channel)[sample] = *p++;
276 if (channel == _ffmpeg_content->audio_channels()) {
284 case AV_SAMPLE_FMT_FLTP:
286 float** p = reinterpret_cast<float**> (data);
287 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
288 memcpy (audio->data(i), p[i], frames * sizeof(float));
294 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
301 FFmpegDecoder::audio_sample_format () const
303 if (!_ffmpeg_content->audio_stream()) {
304 return (AVSampleFormat) 0;
307 return audio_codec_context()->sample_fmt;
311 FFmpegDecoder::bytes_per_audio_sample () const
313 return av_get_bytes_per_sample (audio_sample_format ());
317 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
319 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
321 /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
322 a number plucked from the air) earlier than we want to end up. The loop below
323 will hopefully then step through to where we want to be.
335 /* Initial seek time in the stream's timebase */
336 int64_t const initial_vt = ((initial / _ffmpeg_content->original_video_frame_rate()) - _pts_offset) / time_base;
338 av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
340 avcodec_flush_buffers (video_codec_context());
341 if (_subtitle_codec_context) {
342 avcodec_flush_buffers (_subtitle_codec_context);
345 /* This !accurate is piling hack upon hack; setting _just_sought to true
346 even with accurate == true defeats our attempt to align the start
347 of the video and audio. Here we disable that defeat when accurate == true
348 i.e. when we are making a DCP rather than just previewing one.
349 Ewww. This should be gone in 2.0.
355 _video_position = frame;
357 if (frame == 0 || !accurate) {
358 /* We're already there, or we're as close as we need to be */
363 int r = av_read_frame (_format_context, &_packet);
368 if (_packet.stream_index != _video_stream) {
369 av_free_packet (&_packet);
374 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
375 if (r >= 0 && finished) {
376 _video_position = rint (
377 (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->original_video_frame_rate()
380 if (_video_position >= (frame - 1)) {
381 av_free_packet (&_packet);
386 av_free_packet (&_packet);
389 /* _video_position should be the next thing to be emitted, which will the one after the thing
396 FFmpegDecoder::decode_audio_packet ()
398 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
402 AVPacket copy_packet = _packet;
404 while (copy_packet.size > 0) {
407 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
408 if (decode_result < 0) {
409 shared_ptr<const Film> film = _film.lock ();
411 LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result);
415 if (frame_finished) {
417 if (_audio_position == 0) {
418 /* Where we are in the source, in seconds */
419 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
420 * av_frame_get_best_effort_timestamp(_frame) + _pts_offset;
423 /* Emit some silence */
424 int64_t frames = pts * _ffmpeg_content->content_audio_frame_rate ();
426 int64_t const this_time = min (frames, (int64_t) _ffmpeg_content->content_audio_frame_rate() / 2);
428 shared_ptr<AudioBuffers> silence (
429 new AudioBuffers (_ffmpeg_content->audio_channels(), this_time)
432 silence->make_silent ();
433 audio (silence, _audio_position);
439 int const data_size = av_samples_get_buffer_size (
440 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
443 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
446 copy_packet.data += decode_result;
447 copy_packet.size -= decode_result;
452 FFmpegDecoder::decode_video_packet ()
455 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
459 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
461 shared_ptr<FilterGraph> graph;
463 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
464 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
468 if (i == _filter_graphs.end ()) {
469 shared_ptr<const Film> film = _film.lock ();
472 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
473 _filter_graphs.push_back (graph);
475 LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
480 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
482 shared_ptr<const Film> film = _film.lock ();
485 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
487 shared_ptr<Image> image = i->first;
489 if (i->second != AV_NOPTS_VALUE) {
491 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
494 /* We just did a seek, so disable any attempts to correct for where we
497 _video_position = rint (pts * _ffmpeg_content->original_video_frame_rate ());
498 _just_sought = false;
501 double const next = _video_position / _ffmpeg_content->original_video_frame_rate();
502 double const one_frame = 1 / _ffmpeg_content->original_video_frame_rate ();
503 double delta = pts - next;
505 while (delta > one_frame) {
506 /* This PTS is more than one frame forward in time of where we think we should be; emit
510 /* XXX: I think this should be a copy of the last frame... */
511 boost::shared_ptr<Image> black (
513 static_cast<AVPixelFormat> (_frame->format),
514 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
519 shared_ptr<const Film> film = _film.lock ();
522 black->make_black ();
523 video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
527 if (delta > -one_frame) {
528 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
529 video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
533 LOG_WARNING ("Dropping frame without PTS");
542 FFmpegDecoder::setup_subtitle ()
544 boost::mutex::scoped_lock lm (_mutex);
546 if (!_ffmpeg_content->subtitle_stream()) {
550 _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
551 if (_subtitle_codec_context == 0) {
552 throw DecodeError (N_("could not find subtitle stream"));
555 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
557 if (_subtitle_codec == 0) {
558 throw DecodeError (N_("could not find subtitle decoder"));
561 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
562 throw DecodeError (N_("could not open subtitle decoder"));
567 FFmpegDecoder::done () const
569 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
570 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
575 FFmpegDecoder::decode_subtitle_packet ()
579 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
583 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
584 indicate that the previous subtitle should stop.
586 if (sub.num_rects <= 0) {
587 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
589 } else if (sub.num_rects > 1) {
590 throw DecodeError (_("multi-part subtitles not yet supported"));
593 /* Subtitle PTS in seconds (within the source, not taking into account any of the
594 source that we may have chopped off for the DCP)
596 double const packet_time = (static_cast<double> (sub.pts ) / AV_TIME_BASE) + _pts_offset;
598 /* hence start time for this sub */
599 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
600 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
602 AVSubtitleRect const * rect = sub.rects[0];
604 if (rect->type != SUBTITLE_BITMAP) {
605 throw DecodeError (_("non-bitmap subtitles not yet supported"));
608 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
609 G, third B, fourth A.
611 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
613 /* Start of the first line in the subtitle */
614 uint8_t* sub_p = rect->pict.data[0];
615 /* sub_p looks up into a BGRA palette which is here
616 (i.e. first byte B, second G, third R, fourth A)
618 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
619 /* Start of the output data */
620 uint32_t* out_p = (uint32_t *) image->data()[0];
622 for (int y = 0; y < rect->h; ++y) {
623 uint8_t* sub_line_p = sub_p;
624 uint32_t* out_line_p = out_p;
625 for (int x = 0; x < rect->w; ++x) {
626 uint32_t const p = palette[*sub_line_p++];
627 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
629 sub_p += rect->pict.linesize[0];
630 out_p += image->stride()[0] / sizeof (uint32_t);
633 libdcp::Size const vs = _ffmpeg_content->video_size ();
637 dcpomatic::Rect<double> (
638 static_cast<double> (rect->x) / vs.width,
639 static_cast<double> (rect->y) / vs.height,
640 static_cast<double> (rect->w) / vs.width,
641 static_cast<double> (rect->h) / vs.height
648 avsubtitle_free (&sub);