2 Copyright (C) 2012-2015 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 "filter_graph.h"
33 #include "audio_buffers.h"
34 #include "ffmpeg_content.h"
35 #include "raw_image_proxy.h"
37 #include "compose.hpp"
39 #include <libavcodec/avcodec.h>
40 #include <libavformat/avformat.h>
42 #include <boost/foreach.hpp>
50 #define LOG_GENERAL(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_GENERAL);
51 #define LOG_ERROR(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_ERROR);
52 #define LOG_WARNING_NC(...) _log->log (__VA_ARGS__, LogEntry::TYPE_WARNING);
53 #define LOG_WARNING(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_WARNING);
61 using boost::shared_ptr;
64 FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log, bool fast)
66 , AudioDecoder (c, fast)
70 , _pts_offset (pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->video_frame_rate()))
76 FFmpegDecoder::flush ()
78 /* Get any remaining frames */
83 /* XXX: should we reset _packet.data and size after each *_decode_* call? */
85 while (decode_video_packet ()) {}
87 decode_audio_packet ();
88 AudioDecoder::flush ();
92 FFmpegDecoder::pass (PassReason reason, bool accurate)
94 int r = av_read_frame (_format_context, &_packet);
96 /* AVERROR_INVALIDDATA can apparently be returned sometimes even when av_read_frame
97 has pretty-much succeeded (and hence generated data which should be processed).
98 Hence it makes sense to continue here in that case.
100 if (r < 0 && r != AVERROR_INVALIDDATA) {
101 if (r != AVERROR_EOF) {
102 /* Maybe we should fail here, but for now we'll just finish off instead */
104 av_strerror (r, buf, sizeof(buf));
105 LOG_ERROR (N_("error on av_read_frame (%1) (%2)"), buf, r);
112 int const si = _packet.stream_index;
113 shared_ptr<const FFmpegContent> fc = _ffmpeg_content;
115 if (si == _video_stream && !_ignore_video && (accurate || reason != PASS_REASON_SUBTITLE)) {
116 decode_video_packet ();
117 } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index (_format_context, si)) {
118 decode_subtitle_packet ();
119 } else if (accurate || reason != PASS_REASON_SUBTITLE) {
120 decode_audio_packet ();
123 av_free_packet (&_packet);
127 /** @param data pointer to array of pointers to buffers.
128 * Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
130 shared_ptr<AudioBuffers>
131 FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream, uint8_t** data, int size)
133 DCPOMATIC_ASSERT (bytes_per_audio_sample (stream));
135 /* Deinterleave and convert to float */
137 /* total_samples and frames will be rounded down here, so if there are stray samples at the end
138 of the block that do not form a complete sample or frame they will be dropped.
140 int const total_samples = size / bytes_per_audio_sample (stream);
141 int const frames = total_samples / stream->channels();
142 shared_ptr<AudioBuffers> audio (new AudioBuffers (stream->channels(), frames));
144 switch (audio_sample_format (stream)) {
145 case AV_SAMPLE_FMT_U8:
147 uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
150 for (int i = 0; i < total_samples; ++i) {
151 audio->data(channel)[sample] = float(*p++) / (1 << 23);
154 if (channel == stream->channels()) {
162 case AV_SAMPLE_FMT_S16:
164 int16_t* p = reinterpret_cast<int16_t *> (data[0]);
167 for (int i = 0; i < total_samples; ++i) {
168 audio->data(channel)[sample] = float(*p++) / (1 << 15);
171 if (channel == stream->channels()) {
179 case AV_SAMPLE_FMT_S16P:
181 int16_t** p = reinterpret_cast<int16_t **> (data);
182 for (int i = 0; i < stream->channels(); ++i) {
183 for (int j = 0; j < frames; ++j) {
184 audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
190 case AV_SAMPLE_FMT_S32:
192 int32_t* p = reinterpret_cast<int32_t *> (data[0]);
195 for (int i = 0; i < total_samples; ++i) {
196 audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
199 if (channel == stream->channels()) {
207 case AV_SAMPLE_FMT_FLT:
209 float* p = reinterpret_cast<float*> (data[0]);
212 for (int i = 0; i < total_samples; ++i) {
213 audio->data(channel)[sample] = *p++;
216 if (channel == stream->channels()) {
224 case AV_SAMPLE_FMT_FLTP:
226 float** p = reinterpret_cast<float**> (data);
227 for (int i = 0; i < stream->channels(); ++i) {
228 memcpy (audio->data(i), p[i], frames * sizeof(float));
234 throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format (stream))));
241 FFmpegDecoder::audio_sample_format (shared_ptr<FFmpegAudioStream> stream) const
243 return stream->stream (_format_context)->codec->sample_fmt;
247 FFmpegDecoder::bytes_per_audio_sample (shared_ptr<FFmpegAudioStream> stream) const
249 return av_get_bytes_per_sample (audio_sample_format (stream));
253 FFmpegDecoder::seek (ContentTime time, bool accurate)
255 VideoDecoder::seek (time, accurate);
256 AudioDecoder::seek (time, accurate);
257 SubtitleDecoder::seek (time, accurate);
259 /* If we are doing an `accurate' seek, we need to use pre-roll, as
260 we don't really know what the seek will give us.
263 ContentTime pre_roll = accurate ? ContentTime::from_seconds (2) : ContentTime (0);
266 /* XXX: it seems debatable whether PTS should be used here...
267 http://www.mjbshaw.com/2012/04/seeking-in-ffmpeg-know-your-timestamp.html
270 ContentTime u = time - _pts_offset;
271 if (u < ContentTime ()) {
274 av_seek_frame (_format_context, _video_stream, u.seconds() / av_q2d (_format_context->streams[_video_stream]->time_base), AVSEEK_FLAG_BACKWARD);
276 avcodec_flush_buffers (video_codec_context());
278 /* XXX: should be flushing audio buffers? */
280 if (subtitle_codec_context ()) {
281 avcodec_flush_buffers (subtitle_codec_context ());
286 FFmpegDecoder::decode_audio_packet ()
288 /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
292 AVPacket copy_packet = _packet;
294 /* XXX: inefficient */
295 vector<shared_ptr<FFmpegAudioStream> > streams = ffmpeg_content()->ffmpeg_audio_streams ();
296 vector<shared_ptr<FFmpegAudioStream> >::const_iterator stream = streams.begin ();
297 while (stream != streams.end () && !(*stream)->uses_index (_format_context, copy_packet.stream_index)) {
301 if (stream == streams.end ()) {
302 /* The packet's stream may not be an audio one; just ignore it in this method if so */
306 while (copy_packet.size > 0) {
309 int decode_result = avcodec_decode_audio4 ((*stream)->stream (_format_context)->codec, _frame, &frame_finished, ©_packet);
310 if (decode_result < 0) {
311 /* avcodec_decode_audio4 can sometimes return an error even though it has decoded
312 some valid data; for example dca_subframe_footer can return AVERROR_INVALIDDATA
313 if it overreads the auxiliary data. ffplay carries on if frame_finished is true,
314 even in the face of such an error, so I think we should too.
316 Returning from the method here caused mantis #352.
318 LOG_WARNING ("avcodec_decode_audio4 failed (%1)", decode_result);
320 /* Fudge decode_result so that we come out of the while loop when
321 we've processed this data.
323 decode_result = copy_packet.size;
326 if (frame_finished) {
327 ContentTime ct = ContentTime::from_seconds (
328 av_frame_get_best_effort_timestamp (_frame) *
329 av_q2d ((*stream)->stream (_format_context)->time_base))
332 int const data_size = av_samples_get_buffer_size (
333 0, (*stream)->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (*stream), 1
336 shared_ptr<AudioBuffers> data = deinterleave_audio (*stream, _frame->data, data_size);
338 if (ct < ContentTime ()) {
339 /* Discard audio data that comes before time 0 */
340 Frame const remove = min (int64_t (data->frames()), (-ct).frames_ceil(double((*stream)->frame_rate ())));
341 data->move (remove, 0, data->frames() - remove);
342 data->set_frames (data->frames() - remove);
343 ct += ContentTime::from_frames (remove, (*stream)->frame_rate ());
346 if (data->frames() > 0) {
347 audio (*stream, data, ct);
351 copy_packet.data += decode_result;
352 copy_packet.size -= decode_result;
357 FFmpegDecoder::decode_video_packet ()
360 if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
364 boost::mutex::scoped_lock lm (_filter_graphs_mutex);
366 shared_ptr<FilterGraph> graph;
368 list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
369 while (i != _filter_graphs.end() && !(*i)->can_process (dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
373 if (i == _filter_graphs.end ()) {
374 graph.reset (new FilterGraph (_ffmpeg_content, dcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
375 _filter_graphs.push_back (graph);
376 LOG_GENERAL (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format);
381 list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
383 for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
385 shared_ptr<Image> image = i->first;
387 if (i->second != AV_NOPTS_VALUE) {
388 double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset.seconds ();
390 shared_ptr<ImageProxy> (new RawImageProxy (image)),
391 llrint (pts * _ffmpeg_content->video_frame_rate ())
394 LOG_WARNING_NC ("Dropping frame without PTS");
402 FFmpegDecoder::decode_subtitle_packet ()
406 if (avcodec_decode_subtitle2 (subtitle_codec_context(), &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
410 if (sub.num_rects <= 0) {
411 /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
412 indicate that the previous subtitle should stop. We can ignore it here.
417 /* Subtitle PTS (within the source, not taking into account any of the
418 source that we may have chopped off for the DCP).
420 FFmpegSubtitlePeriod sub_period = subtitle_period (sub);
421 ContentTimePeriod period;
422 period.from = sub_period.from + _pts_offset;
424 /* We already know the subtitle period `to' time */
425 period.to = sub_period.to.get() + _pts_offset;
427 /* We have to look up the `to' time in the stream's records */
428 period.to = ffmpeg_content()->subtitle_stream()->find_subtitle_to (period.from);
431 for (unsigned int i = 0; i < sub.num_rects; ++i) {
432 AVSubtitleRect const * rect = sub.rects[i];
434 switch (rect->type) {
437 case SUBTITLE_BITMAP:
438 decode_bitmap_subtitle (rect, period);
441 cout << "XXX: SUBTITLE_TEXT " << rect->text << "\n";
444 cout << "XXX: SUBTITLE_ASS " << rect->ass << "\n";
449 avsubtitle_free (&sub);
452 list<ContentTimePeriod>
453 FFmpegDecoder::image_subtitles_during (ContentTimePeriod p, bool starting) const
455 return _ffmpeg_content->subtitles_during (p, starting);
458 list<ContentTimePeriod>
459 FFmpegDecoder::text_subtitles_during (ContentTimePeriod, bool) const
461 return list<ContentTimePeriod> ();
465 FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTimePeriod period)
467 /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
468 G, third B, fourth A.
470 shared_ptr<Image> image (new Image (AV_PIX_FMT_RGBA, dcp::Size (rect->w, rect->h), true));
472 /* Start of the first line in the subtitle */
473 uint8_t* sub_p = rect->pict.data[0];
474 /* sub_p looks up into a BGRA palette which is here
475 (i.e. first byte B, second G, third R, fourth A)
477 uint32_t const * palette = (uint32_t *) rect->pict.data[1];
478 /* Start of the output data */
479 uint32_t* out_p = (uint32_t *) image->data()[0];
481 for (int y = 0; y < rect->h; ++y) {
482 uint8_t* sub_line_p = sub_p;
483 uint32_t* out_line_p = out_p;
484 for (int x = 0; x < rect->w; ++x) {
485 uint32_t const p = palette[*sub_line_p++];
486 *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
488 sub_p += rect->pict.linesize[0];
489 out_p += image->stride()[0] / sizeof (uint32_t);
492 dcp::Size const vs = _ffmpeg_content->video_size ();
493 dcpomatic::Rect<double> const scaled_rect (
494 static_cast<double> (rect->x) / vs.width,
495 static_cast<double> (rect->y) / vs.height,
496 static_cast<double> (rect->w) / vs.width,
497 static_cast<double> (rect->h) / vs.height
500 image_subtitle (period, image, scaled_rect);