2 Copyright (C) 2012-2016 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.
25 #include "exceptions.h"
29 #include "ffmpeg_decoder.h"
30 #include "ffmpeg_audio_stream.h"
31 #include "ffmpeg_subtitle_stream.h"
32 #include "video_filter_graph.h"
33 #include "audio_buffers.h"
34 #include "ffmpeg_content.h"
35 #include "raw_image_proxy.h"
37 #include "md5_digester.h"
38 #include "compose.hpp"
40 #include <libavcodec/avcodec.h>
41 #include <libavformat/avformat.h>
43 #include <boost/foreach.hpp>
51 #define LOG_GENERAL(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_GENERAL);
52 #define LOG_ERROR(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_ERROR);
53 #define LOG_WARNING_NC(...) _log->log (__VA_ARGS__, LogEntry::TYPE_WARNING);
54 #define LOG_WARNING(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_WARNING);
62 using boost::shared_ptr;
65 FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log, bool fast)
67 , AudioDecoder (c, fast)
71 , _pts_offset (pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->video_frame_rate()))
77 FFmpegDecoder::flush ()
79 /* Get any remaining frames */
84 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
86 while (decode_video_packet ()) {}
88 decode_audio_packet ();
89 AudioDecoder::flush ();
93 FFmpegDecoder::pass (PassReason reason, bool accurate)
95 int r = av_read_frame (_format_context, &_packet);
97 /* AVERROR_INVALIDDATA can apparently be returned sometimes even when av_read_frame
98 has pretty-much succeeded (and hence generated data which should be processed).
99 Hence it makes sense to continue here in that case.
101 if (r < 0 && r != AVERROR_INVALIDDATA) {
102 if (r != AVERROR_EOF) {
103 /* Maybe we should fail here, but for now we'll just finish off instead */
105 av_strerror (r, buf, sizeof(buf));
106 LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
113 int const si = _packet.stream_index;
114 shared_ptr<const FFmpegContent> fc = _ffmpeg_content;
116 if (si == _video_stream && !_ignore_video && (accurate || reason != PASS_REASON_SUBTITLE)) {
117 decode_video_packet ();
118 } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index (_format_context, si)) {
119 decode_subtitle_packet ();
120 } else if (accurate || reason != PASS_REASON_SUBTITLE) {
121 decode_audio_packet ();
124 av_free_packet (&_packet);
128 /** @param data pointer to array of pointers to buffers.
129 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
131 shared_ptr<AudioBuffers>
132 FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream, uint8_t** data, int size)
134 DCPOMATIC_ASSERT (bytes_per_audio_sample (stream));
136 /* Deinterleave and convert to float */
138 /* total_samples and frames will be rounded down here, so if there are stray samples at the end
139 of the block that do not form a complete sample or frame they will be dropped.
141 int const total_samples = size / bytes_per_audio_sample (stream);
142 int const frames = total_samples / stream->channels();
143 shared_ptr<AudioBuffers> audio (new AudioBuffers (stream->channels(), frames));
145 switch (audio_sample_format (stream)) {
146 case AV_SAMPLE_FMT_U8:
148 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
151 for (int i = 0; i < total_samples; ++i) {
152 audio->data(channel)[sample] = float(*p++) / (1 << 23);
155 if (channel == stream->channels()) {
163 case AV_SAMPLE_FMT_S16:
165 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
168 for (int i = 0; i < total_samples; ++i) {
169 audio->data(channel)[sample] = float(*p++) / (1 << 15);
172 if (channel == stream->channels()) {
180 case AV_SAMPLE_FMT_S16P:
182 int16_t** p = reinterpret_cast<int16_t **> (data);
183 for (int i = 0; i < stream->channels(); ++i) {
184 for (int j = 0; j < frames; ++j) {
185 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
191 case AV_SAMPLE_FMT_S32:
193 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
196 for (int i = 0; i < total_samples; ++i) {
197 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
200 if (channel == stream->channels()) {
208 case AV_SAMPLE_FMT_FLT:
210 float* p = reinterpret_cast<float*> (data[0]);
213 for (int i = 0; i < total_samples; ++i) {
214 audio->data(channel)[sample] = *p++;
217 if (channel == stream->channels()) {
225 case AV_SAMPLE_FMT_FLTP:
227 float** p = reinterpret_cast<float**> (data);
228 /* Sometimes there aren't as many channels in the _frame as in the stream */
229 for (int i = 0; i < _frame->channels; ++i) {
230 memcpy (audio->data(i), p[i], frames * sizeof(float));
232 for (int i = _frame->channels; i < stream->channels(); ++i) {
233 audio->make_silent (i);
239 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format (stream))));
246 FFmpegDecoder::audio_sample_format (shared_ptr<FFmpegAudioStream> stream) const
248 return stream->stream (_format_context)->codec->sample_fmt;
252 FFmpegDecoder::bytes_per_audio_sample (shared_ptr<FFmpegAudioStream> stream) const
254 return av_get_bytes_per_sample (audio_sample_format (stream));
258 FFmpegDecoder::seek (ContentTime time, bool accurate)
260 VideoDecoder::seek (time, accurate);
261 AudioDecoder::seek (time, accurate);
262 SubtitleDecoder::seek (time, accurate);
264 /* If we are doing an `accurate' seek, we need to use pre-roll, as
265 we don't really know what the seek will give us.
268 ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
271 /* XXX: it seems debatable whether PTS should be used here...
272 http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html
275 ContentTime u = time - _pts_offset;
276 if (u < ContentTime ()) {
279 av_seek_frame (_format_context, _video_stream, u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base), AVSEEK_FLAG_BACKWARD);
281 avcodec_flush_buffers (video_codec_context());
283 /* XXX: should be flushing audio buffers? */
285 if (subtitle_codec_context ()) {
286 avcodec_flush_buffers (subtitle_codec_context ());
291 FFmpegDecoder::decode_audio_packet ()
293 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
297 AVPacket copy_packet = _packet;
299 /* XXX: inefficient */
300 vector<shared_ptr<FFmpegAudioStream> > streams = ffmpeg_content()->ffmpeg_audio_streams ();
301 vector<shared_ptr<FFmpegAudioStream> >::const_iterator stream = streams.begin ();
302 while (stream != streams.end () && !(*stream)->uses_index (_format_context, copy_packet.stream_index)) {
306 if (stream == streams.end ()) {
307 /* The packet's stream may not be an audio one; just ignore it in this method if so */
311 while (copy_packet.size > 0) {
314 int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet);
315 if (decode_result < 0) {
316 /* avcodec_decode_audio4 can sometimes return an error even though it has decoded
317 some valid data; for example dca_subframe_footer can return AVERROR_INVALIDDATA
318 if it overreads the auxiliary data. ffplay carries on if frame_finished is true,
319 even in the face of such an error, so I think we should too.
321 Returning from the method here caused mantis #352.
323 LOG_WARNING ("avcodec_decode_audio4 failed (%1)", decode_result);
325 /* Fudge decode_result so that we come out of the while loop when
326 we've processed this data.
328 decode_result = copy_packet.size;
331 if (frame_finished) {
332 ContentTime ct = ContentTime::from_seconds (
333 av_frame_get_best_effort_timestamp (_frame) *
334 av_q2d ((*stream)->stream (_format_context)->time_base))
337 int const data_size = av_samples_get_buffer_size (
338 0, (*stream)->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (*stream), 1
341 shared_ptr<AudioBuffers> data = deinterleave_audio (*stream, _frame->data, data_size);
343 if (ct < ContentTime ()) {
344 /* Discard audio data that comes before time 0 */
345 Frame const remove = min (int64_t (data->frames()), (-ct).frames_ceil(double((*stream)->frame_rate ())));
346 data->move (remove, 0, data->frames() - remove);
347 data->set_frames (data->frames() - remove);
348 ct += ContentTime::from_frames (remove, (*stream)->frame_rate ());
351 if (data->frames() > 0) {
352 audio (*stream, data, ct);
356 copy_packet.data += decode_result;
357 copy_packet.size -= decode_result;
362 FFmpegDecoder::decode_video_packet ()
365 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
369 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
371 shared_ptr<VideoFilterGraph> graph;
373 list<shared_ptr<VideoFilterGraph> >::iterator i = _filter_graphs.begin();
374 while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
378 if (i == _filter_graphs.end ()) {
379 graph.reset (new VideoFilterGraph (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
380 graph->setup (_ffmpeg_content->filters ());
381 _filter_graphs.push_back (graph);
382 LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
387 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
389 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
391 shared_ptr<Image> image = i->first;
393 if (i->second != AV_NOPTS_VALUE) {
394 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
396 shared_ptr<ImageProxy> (new RawImageProxy (image)),
397 llrint (pts * _ffmpeg_content->video_frame_rate ())
400 LOG_WARNING_NC ("Dropping frame without PTS");
408 FFmpegDecoder::decode_subtitle_packet ()
412 if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
416 if (sub.num_rects <= 0) {
417 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
418 indicate that the previous subtitle should stop. We can ignore it here.
423 /* Subtitle PTS (within the source, not taking into account any of the
424 source that we may have chopped off for the DCP).
426 FFmpegSubtitlePeriod sub_period = subtitle_period (sub);
427 ContentTimePeriod period;
428 period.from = sub_period.from + _pts_offset;
430 /* We already know the subtitle period `to' time */
431 period.to = sub_period.to.get() + _pts_offset;
433 /* We have to look up the `to' time in the stream's records */
434 period.to = ffmpeg_content()->subtitle_stream()->find_subtitle_to (subtitle_id (sub));
437 for (unsigned int i = 0; i < sub.num_rects; ++i) {
438 AVSubtitleRect const * rect = sub.rects[i];
440 switch (rect->type) {
443 case SUBTITLE_BITMAP:
444 decode_bitmap_subtitle (rect, period);
447 cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n";
450 cout << "XXX: SUBTITLE_ASS " << rect->ass << "\n";
455 avsubtitle_free (&sub);
458 list<ContentTimePeriod>
459 FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const
461 return _ffmpeg_content->subtitles_during (p, starting);
464 list<ContentTimePeriod>
465 FFmpegDecoder::text_subtitles_during (ContentTimePeriod, bool) const
467 return list<ContentTimePeriod> ();
471 FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period)
473 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
474 G, third B, fourth A.
476 shared_ptr<Image> image (new Image (AV_PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true));
478 /* Start of the first line in the subtitle */
479 uint8_t* sub_p = rect->pict.data[0];
480 /* sub_p looks up into a BGRA palette which is here
481 (i.e. first byte B, second G, third R, fourth A)
483 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
484 /* Start of the output data */
485 uint32_t* out_p = (uint32_t *) image->data()[0];
487 for (int y = 0; y < rect->h; ++y) {
488 uint8_t* sub_line_p = sub_p;
489 uint32_t* out_line_p = out_p;
490 for (int x = 0; x < rect->w; ++x) {
491 uint32_t const p = palette[*sub_line_p++];
492 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
494 sub_p += rect->pict.linesize[0];
495 out_p += image->stride()[0] / sizeof (uint32_t);
498 dcp::Size const vs = _ffmpeg_content->video_size ();
499 dcpomatic::Rect<double> const scaled_rect (
500 static_cast<double> (rect->x) / vs.width,
501 static_cast<double> (rect->y) / vs.height,
502 static_cast<double> (rect->w) / vs.width,
503 static_cast<double> (rect->h) / vs.height
506 image_subtitle (period, image, scaled_rect);