2 Copyright (C) 2012 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.
30 #include <boost/lexical_cast.hpp>
33 #include <libavcodec/avcodec.h>
34 #include <libavformat/avformat.h>
38 #include "exceptions.h"
42 #include "ffmpeg_decoder.h"
43 #include "filter_graph.h"
44 #include "audio_buffers.h"
45 #include "ffmpeg_content.h"
52 using std::stringstream;
56 using boost::shared_ptr;
57 using boost::optional;
58 using boost::dynamic_pointer_cast;
61 FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
67 , _subtitle_codec_context (0)
69 , _decode_video (video)
70 , _decode_audio (audio)
72 , _just_sought (false)
76 /* Audio and video frame PTS values may not start with 0. We want
77 to fiddle them so that:
79 1. One of them starts at time 0.
80 2. The first video PTS value ends up on a frame boundary.
82 Then we remove big initial gaps in PTS and we allow our
83 insertion of black frames to work.
86 audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
87 video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
90 bool const have_video = video && c->first_video();
91 bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
93 /* First, make one of them start at 0 */
95 if (have_audio && have_video) {
96 _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
97 } else if (have_video) {
98 _pts_offset = - c->first_video().get();
99 } else if (have_audio) {
100 _pts_offset = - c->audio_stream()->first_audio.get();
103 /* Now adjust both so that the video pts starts on a frame */
104 if (have_video && have_audio) {
105 double first_video = c->first_video().get() + _pts_offset;
106 double const old_first_video = first_video;
108 /* Round the first video up to a frame boundary */
109 if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
110 first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
113 _pts_offset += first_video - old_first_video;
117 FFmpegDecoder::~FFmpegDecoder ()
119 boost::mutex::scoped_lock lm (_mutex);
121 if (_subtitle_codec_context) {
122 avcodec_close (_subtitle_codec_context);
127 FFmpegDecoder::flush ()
129 /* Get any remaining frames */
134 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
137 while (decode_video_packet ()) {}
140 if (_ffmpeg_content->audio_stream() && _decode_audio) {
141 decode_audio_packet ();
144 /* Stop us being asked for any more data */
145 _video_position = _ffmpeg_content->video_length ();
146 _audio_position = _ffmpeg_content->audio_length ();
150 FFmpegDecoder::pass ()
152 int r = av_read_frame (_format_context, &_packet);
155 if (r != AVERROR_EOF) {
156 /* Maybe we should fail here, but for now we'll just finish off instead */
158 av_strerror (r, buf, sizeof(buf));
159 shared_ptr<const Film> film = _film.lock ();
161 film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
168 shared_ptr<const Film> film = _film.lock ();
171 int const si = _packet.stream_index;
173 if (si == _video_stream && _decode_video) {
174 decode_video_packet ();
175 } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _decode_audio) {
176 decode_audio_packet ();
177 } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
178 decode_subtitle_packet ();
181 av_free_packet (&_packet);
184 /** @param data pointer to array of pointers to buffers.
185 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
187 shared_ptr<AudioBuffers>
188 FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
190 assert (_ffmpeg_content->audio_channels());
191 assert (bytes_per_audio_sample());
193 /* Deinterleave and convert to float */
195 assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
197 int const total_samples = size / bytes_per_audio_sample();
198 int const frames = total_samples / _ffmpeg_content->audio_channels();
199 shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
201 switch (audio_sample_format()) {
202 case AV_SAMPLE_FMT_U8:
204 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
207 for (int i = 0; i < total_samples; ++i) {
208 audio->data(channel)[sample] = float(*p++) / (1 << 23);
211 if (channel == _ffmpeg_content->audio_channels()) {
219 case AV_SAMPLE_FMT_S16:
221 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
224 for (int i = 0; i < total_samples; ++i) {
225 audio->data(channel)[sample] = float(*p++) / (1 << 15);
228 if (channel == _ffmpeg_content->audio_channels()) {
236 case AV_SAMPLE_FMT_S16P:
238 int16_t** p = reinterpret_cast<int16_t **> (data);
239 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
240 for (int j = 0; j < frames; ++j) {
241 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
247 case AV_SAMPLE_FMT_S32:
249 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
252 for (int i = 0; i < total_samples; ++i) {
253 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
256 if (channel == _ffmpeg_content->audio_channels()) {
264 case AV_SAMPLE_FMT_FLT:
266 float* p = reinterpret_cast<float*> (data[0]);
269 for (int i = 0; i < total_samples; ++i) {
270 audio->data(channel)[sample] = *p++;
273 if (channel == _ffmpeg_content->audio_channels()) {
281 case AV_SAMPLE_FMT_FLTP:
283 float** p = reinterpret_cast<float**> (data);
284 for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
285 memcpy (audio->data(i), p[i], frames * sizeof(float));
291 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
298 FFmpegDecoder::audio_sample_format () const
300 if (!_ffmpeg_content->audio_stream()) {
301 return (AVSampleFormat) 0;
304 return audio_codec_context()->sample_fmt;
308 FFmpegDecoder::bytes_per_audio_sample () const
310 return av_get_bytes_per_sample (audio_sample_format ());
314 FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
316 double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
318 /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
319 a number plucked from the air) earlier than we want to end up. The loop below
320 will hopefully then step through to where we want to be.
332 /* Initial seek time in the stream's timebase */
333 int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base;
335 av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
337 avcodec_flush_buffers (video_codec_context());
338 if (_subtitle_codec_context) {
339 avcodec_flush_buffers (_subtitle_codec_context);
342 /* This !accurate is piling hack upon hack; setting _just_sought to true
343 even with accurate == true defeats our attempt to align the start
344 of the video and audio. Here we disable that defeat when accurate == true
345 i.e. when we are making a DCP rather than just previewing one.
346 Ewww. This should be gone in 2.0.
352 _video_position = frame;
354 if (frame == 0 || !accurate) {
355 /* We're already there, or we're as close as we need to be */
360 int r = av_read_frame (_format_context, &_packet);
365 if (_packet.stream_index != _video_stream) {
366 av_free_packet (&_packet);
371 r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
372 if (r >= 0 && finished) {
373 _video_position = rint (
374 (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
377 if (_video_position >= (frame - 1)) {
378 av_free_packet (&_packet);
383 av_free_packet (&_packet);
388 FFmpegDecoder::decode_audio_packet ()
390 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
394 AVPacket copy_packet = _packet;
396 while (copy_packet.size > 0) {
399 int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, ©_packet);
400 if (decode_result < 0) {
401 shared_ptr<const Film> film = _film.lock ();
403 film->log()->log (String::compose ("avcodec_decode_audio4 failed (%1)", decode_result));
407 if (frame_finished) {
409 if (_audio_position == 0) {
410 /* Where we are in the source, in seconds */
411 double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
412 * av_frame_get_best_effort_timestamp(_frame) + _pts_offset;
415 /* Emit some silence */
416 shared_ptr<AudioBuffers> silence (
418 _ffmpeg_content->audio_channels(),
419 pts * _ffmpeg_content->content_audio_frame_rate()
423 silence->make_silent ();
424 audio (silence, _audio_position);
428 int const data_size = av_samples_get_buffer_size (
429 0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
432 audio (deinterleave_audio (_frame->data, data_size), _audio_position);
435 copy_packet.data += decode_result;
436 copy_packet.size -= decode_result;
441 FFmpegDecoder::decode_video_packet ()
444 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
448 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
450 shared_ptr<FilterGraph> graph;
452 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
453 while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
457 if (i == _filter_graphs.end ()) {
458 shared_ptr<const Film> film = _film.lock ();
461 graph.reset (new FilterGraph (_ffmpeg_content, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
462 _filter_graphs.push_back (graph);
464 film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
469 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
471 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
473 shared_ptr<Image> image = i->first;
475 if (i->second != AV_NOPTS_VALUE) {
477 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
480 /* We just did a seek, so disable any attempts to correct for where we
483 _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
484 _just_sought = false;
487 double const next = _video_position / _ffmpeg_content->video_frame_rate();
488 double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
489 double delta = pts - next;
491 while (delta > one_frame) {
492 /* This PTS is more than one frame forward in time of where we think we should be; emit
496 /* XXX: I think this should be a copy of the last frame... */
497 boost::shared_ptr<Image> black (
499 static_cast<AVPixelFormat> (_frame->format),
500 libdcp::Size (video_codec_context()->width, video_codec_context()->height),
505 black->make_black ();
506 video (image, false, _video_position);
510 if (delta > -one_frame) {
511 /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
512 video (image, false, _video_position);
516 shared_ptr<const Film> film = _film.lock ();
518 film->log()->log ("Dropping frame without PTS");
527 FFmpegDecoder::setup_subtitle ()
529 boost::mutex::scoped_lock lm (_mutex);
531 if (!_ffmpeg_content->subtitle_stream()) {
535 _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
536 if (_subtitle_codec_context == 0) {
537 throw DecodeError (N_("could not find subtitle stream"));
540 _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
542 if (_subtitle_codec == 0) {
543 throw DecodeError (N_("could not find subtitle decoder"));
546 if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
547 throw DecodeError (N_("could not open subtitle decoder"));
552 FFmpegDecoder::done () const
554 bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
555 bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
560 FFmpegDecoder::decode_subtitle_packet ()
564 if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
568 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
569 indicate that the previous subtitle should stop.
571 if (sub.num_rects <= 0) {
572 subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
574 } else if (sub.num_rects > 1) {
575 throw DecodeError (_("multi-part subtitles not yet supported"));
578 /* Subtitle PTS in seconds (within the source, not taking into account any of the
579 source that we may have chopped off for the DCP)
581 double const packet_time = (static_cast<double> (sub.pts ) / AV_TIME_BASE) + _pts_offset;
583 /* hence start time for this sub */
584 Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
585 Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
587 AVSubtitleRect const * rect = sub.rects[0];
589 if (rect->type != SUBTITLE_BITMAP) {
590 throw DecodeError (_("non-bitmap subtitles not yet supported"));
593 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
594 G, third B, fourth A.
596 shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
598 /* Start of the first line in the subtitle */
599 uint8_t* sub_p = rect->pict.data[0];
600 /* sub_p looks up into a BGRA palette which is here
601 (i.e. first byte B, second G, third R, fourth A)
603 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
604 /* Start of the output data */
605 uint32_t* out_p = (uint32_t *) image->data()[0];
607 for (int y = 0; y < rect->h; ++y) {
608 uint8_t* sub_line_p = sub_p;
609 uint32_t* out_line_p = out_p;
610 for (int x = 0; x < rect->w; ++x) {
611 uint32_t const p = palette[*sub_line_p++];
612 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
614 sub_p += rect->pict.linesize[0];
615 out_p += image->stride()[0] / sizeof (uint32_t);
618 libdcp::Size const vs = _ffmpeg_content->video_size ();
622 dcpomatic::Rect<double> (
623 static_cast<double> (rect->x) / vs.width,
624 static_cast<double> (rect->y) / vs.height,
625 static_cast<double> (rect->w) / vs.width,
626 static_cast<double> (rect->h) / vs.height
633 avsubtitle_free (&sub);