Infobar material design refresh: bg color
[chromium-blink-merge.git] / media / formats / mp4 / track_run_iterator.cc
blob9b5a454a234cce4d364ee31c66b6501cb07689dd
1 // Copyright 2014 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #include "media/formats/mp4/track_run_iterator.h"
7 #include <algorithm>
8 #include <iomanip>
10 #include "media/base/buffers.h"
11 #include "media/formats/mp4/rcheck.h"
12 #include "media/formats/mp4/sample_to_group_iterator.h"
14 namespace media {
15 namespace mp4 {
17 struct SampleInfo {
18 int size;
19 int duration;
20 int cts_offset;
21 bool is_keyframe;
22 uint32 cenc_group_description_index;
25 struct TrackRunInfo {
26 uint32 track_id;
27 std::vector<SampleInfo> samples;
28 int64 timescale;
29 int64 start_dts;
30 int64 sample_start_offset;
32 bool is_audio;
33 const AudioSampleEntry* audio_description;
34 const VideoSampleEntry* video_description;
35 const SampleGroupDescription* track_sample_encryption_group;
37 int64 aux_info_start_offset; // Only valid if aux_info_total_size > 0.
38 int aux_info_default_size;
39 std::vector<uint8> aux_info_sizes; // Populated if default_size == 0.
40 int aux_info_total_size;
42 std::vector<CencSampleEncryptionInfoEntry> fragment_sample_encryption_info;
44 TrackRunInfo();
45 ~TrackRunInfo();
48 TrackRunInfo::TrackRunInfo()
49 : track_id(0),
50 timescale(-1),
51 start_dts(-1),
52 sample_start_offset(-1),
53 is_audio(false),
54 aux_info_start_offset(-1),
55 aux_info_default_size(-1),
56 aux_info_total_size(-1) {
58 TrackRunInfo::~TrackRunInfo() {}
60 base::TimeDelta TimeDeltaFromRational(int64 numer, int64 denom) {
61 // To avoid overflow, split the following calculation:
62 // (numer * base::Time::kMicrosecondsPerSecond) / denom
63 // into:
64 // (numer / denom) * base::Time::kMicrosecondsPerSecond +
65 // ((numer % denom) * base::Time::kMicrosecondsPerSecond) / denom
66 int64 a = numer / denom;
67 DCHECK_LE((a > 0 ? a : -a), kint64max / base::Time::kMicrosecondsPerSecond);
68 int64 timea_in_us = a * base::Time::kMicrosecondsPerSecond;
70 int64 b = numer % denom;
71 DCHECK_LE((b > 0 ? b : -b), kint64max / base::Time::kMicrosecondsPerSecond);
72 int64 timeb_in_us = (b * base::Time::kMicrosecondsPerSecond) / denom;
74 DCHECK((timeb_in_us < 0) || (timea_in_us <= kint64max - timeb_in_us));
75 DCHECK((timeb_in_us > 0) || (timea_in_us >= kint64min - timeb_in_us));
76 return base::TimeDelta::FromMicroseconds(timea_in_us + timeb_in_us);
79 DecodeTimestamp DecodeTimestampFromRational(int64 numer, int64 denom) {
80 return DecodeTimestamp::FromPresentationTime(
81 TimeDeltaFromRational(numer, denom));
84 TrackRunIterator::TrackRunIterator(const Movie* moov,
85 const scoped_refptr<MediaLog>& media_log)
86 : moov_(moov), media_log_(media_log), sample_offset_(0) {
87 CHECK(moov);
90 TrackRunIterator::~TrackRunIterator() {}
92 static std::string HexFlags(uint32 flags) {
93 std::stringstream stream;
94 stream << std::setfill('0') << std::setw(sizeof(flags)*2) << std::hex
95 << flags;
96 return stream.str();
99 static bool PopulateSampleInfo(const TrackExtends& trex,
100 const TrackFragmentHeader& tfhd,
101 const TrackFragmentRun& trun,
102 const int64 edit_list_offset,
103 const uint32 i,
104 SampleInfo* sample_info,
105 const SampleDependsOn sdtp_sample_depends_on,
106 bool is_audio,
107 const scoped_refptr<MediaLog>& media_log) {
108 if (i < trun.sample_sizes.size()) {
109 sample_info->size = trun.sample_sizes[i];
110 } else if (tfhd.default_sample_size > 0) {
111 sample_info->size = tfhd.default_sample_size;
112 } else {
113 sample_info->size = trex.default_sample_size;
116 if (i < trun.sample_durations.size()) {
117 sample_info->duration = trun.sample_durations[i];
118 } else if (tfhd.default_sample_duration > 0) {
119 sample_info->duration = tfhd.default_sample_duration;
120 } else {
121 sample_info->duration = trex.default_sample_duration;
124 if (i < trun.sample_composition_time_offsets.size()) {
125 sample_info->cts_offset = trun.sample_composition_time_offsets[i];
126 } else {
127 sample_info->cts_offset = 0;
129 sample_info->cts_offset += edit_list_offset;
131 uint32 flags;
132 if (i < trun.sample_flags.size()) {
133 flags = trun.sample_flags[i];
134 DVLOG(4) << __FUNCTION__ << " trun sample flags " << HexFlags(flags);
135 } else if (tfhd.has_default_sample_flags) {
136 flags = tfhd.default_sample_flags;
137 DVLOG(4) << __FUNCTION__ << " tfhd sample flags " << HexFlags(flags);
138 } else {
139 flags = trex.default_sample_flags;
140 DVLOG(4) << __FUNCTION__ << " trex sample flags " << HexFlags(flags);
143 SampleDependsOn sample_depends_on =
144 static_cast<SampleDependsOn>((flags >> 24) & 0x3);
145 if (sample_depends_on == kSampleDependsOnUnknown) {
146 sample_depends_on = sdtp_sample_depends_on;
148 DVLOG(4) << __FUNCTION__ << " sample_depends_on " << sample_depends_on;
149 if (sample_depends_on == kSampleDependsOnReserved) {
150 MEDIA_LOG(ERROR, media_log) << "Reserved value used in sample dependency"
151 " info.";
152 return false;
155 // Per spec (ISO 14496-12:2012), the definition for a "sync sample" is
156 // equivalent to the downstream code's "is keyframe" concept. But media exists
157 // that marks non-key video frames as sync samples (http://crbug.com/507916
158 // and http://crbug.com/310712). Hence, for video we additionally check that
159 // the sample does not depend on others (FFmpeg does too, see mov_read_trun).
160 // Sample dependency is not ignored for audio because encoded audio samples
161 // can depend on other samples and still be used for random access. Generally
162 // all audio samples are expected to be sync samples, but we prefer to check
163 // the flags to catch badly muxed audio (for now anyway ;P). History of
164 // attempts to get this right discussed in http://crrev.com/1319813002
165 bool sample_is_sync_sample = !(flags & kSampleIsNonSyncSample);
166 bool sample_depends_on_others = sample_depends_on == kSampleDependsOnOthers;
167 sample_info->is_keyframe = sample_is_sync_sample &&
168 (!sample_depends_on_others || is_audio);
170 DVLOG(4) << __FUNCTION__ << " is_kf:" << sample_info->is_keyframe
171 << " is_sync:" << sample_is_sync_sample
172 << " deps:" << sample_depends_on_others
173 << " audio:" << is_audio;
175 return true;
178 static const CencSampleEncryptionInfoEntry* GetSampleEncryptionInfoEntry(
179 const TrackRunInfo& run_info,
180 uint32 group_description_index) {
181 const std::vector<CencSampleEncryptionInfoEntry>* entries = nullptr;
183 // ISO-14496-12 Section 8.9.2.3 and 8.9.4 : group description index
184 // (1) ranges from 1 to the number of sample group entries in the track
185 // level SampleGroupDescription Box, or (2) takes the value 0 to
186 // indicate that this sample is a member of no group, in this case, the
187 // sample is associated with the default values specified in
188 // TrackEncryption Box, or (3) starts at 0x10001, i.e. the index value
189 // 1, with the value 1 in the top 16 bits, to reference fragment-local
190 // SampleGroupDescription Box.
191 // Case (2) is not supported here. The caller must handle it externally
192 // before invoking this function.
193 DCHECK_NE(group_description_index, 0u);
194 if (group_description_index >
195 SampleToGroupEntry::kFragmentGroupDescriptionIndexBase) {
196 group_description_index -=
197 SampleToGroupEntry::kFragmentGroupDescriptionIndexBase;
198 entries = &run_info.fragment_sample_encryption_info;
199 } else {
200 entries = &run_info.track_sample_encryption_group->entries;
203 // |group_description_index| is 1-based.
204 DCHECK_LE(group_description_index, entries->size());
205 return (group_description_index > entries->size())
206 ? nullptr
207 : &(*entries)[group_description_index - 1];
210 // In well-structured encrypted media, each track run will be immediately
211 // preceded by its auxiliary information; this is the only optimal storage
212 // pattern in terms of minimum number of bytes from a serial stream needed to
213 // begin playback. It also allows us to optimize caching on memory-constrained
214 // architectures, because we can cache the relatively small auxiliary
215 // information for an entire run and then discard data from the input stream,
216 // instead of retaining the entire 'mdat' box.
218 // We optimize for this situation (with no loss of generality) by sorting track
219 // runs during iteration in order of their first data offset (either sample data
220 // or auxiliary data).
221 class CompareMinTrackRunDataOffset {
222 public:
223 bool operator()(const TrackRunInfo& a, const TrackRunInfo& b) {
224 int64 a_aux = a.aux_info_total_size ? a.aux_info_start_offset : kint64max;
225 int64 b_aux = b.aux_info_total_size ? b.aux_info_start_offset : kint64max;
227 int64 a_lesser = std::min(a_aux, a.sample_start_offset);
228 int64 a_greater = std::max(a_aux, a.sample_start_offset);
229 int64 b_lesser = std::min(b_aux, b.sample_start_offset);
230 int64 b_greater = std::max(b_aux, b.sample_start_offset);
232 if (a_lesser == b_lesser) return a_greater < b_greater;
233 return a_lesser < b_lesser;
237 bool TrackRunIterator::Init(const MovieFragment& moof) {
238 runs_.clear();
240 for (size_t i = 0; i < moof.tracks.size(); i++) {
241 const TrackFragment& traf = moof.tracks[i];
243 const Track* trak = NULL;
244 for (size_t t = 0; t < moov_->tracks.size(); t++) {
245 if (moov_->tracks[t].header.track_id == traf.header.track_id)
246 trak = &moov_->tracks[t];
248 RCHECK(trak);
250 const TrackExtends* trex = NULL;
251 for (size_t t = 0; t < moov_->extends.tracks.size(); t++) {
252 if (moov_->extends.tracks[t].track_id == traf.header.track_id)
253 trex = &moov_->extends.tracks[t];
255 RCHECK(trex);
257 const SampleDescription& stsd =
258 trak->media.information.sample_table.description;
259 if (stsd.type != kAudio && stsd.type != kVideo) {
260 DVLOG(1) << "Skipping unhandled track type";
261 continue;
263 size_t desc_idx = traf.header.sample_description_index;
264 if (!desc_idx) desc_idx = trex->default_sample_description_index;
265 RCHECK(desc_idx > 0); // Descriptions are one-indexed in the file
266 desc_idx -= 1;
268 // Process edit list to remove CTS offset introduced in the presence of
269 // B-frames (those that contain a single edit with a nonnegative media
270 // time). Other uses of edit lists are not supported, as they are
271 // both uncommon and better served by higher-level protocols.
272 int64 edit_list_offset = 0;
273 const std::vector<EditListEntry>& edits = trak->edit.list.edits;
274 if (!edits.empty()) {
275 if (edits.size() > 1)
276 DVLOG(1) << "Multi-entry edit box detected; some components ignored.";
278 if (edits[0].media_time < 0) {
279 DVLOG(1) << "Empty edit list entry ignored.";
280 } else {
281 edit_list_offset = -edits[0].media_time;
285 SampleToGroupIterator sample_to_group_itr(traf.sample_to_group);
286 bool is_sample_to_group_valid = sample_to_group_itr.IsValid();
288 int64 run_start_dts = traf.decode_time.decode_time;
289 int sample_count_sum = 0;
290 for (size_t j = 0; j < traf.runs.size(); j++) {
291 const TrackFragmentRun& trun = traf.runs[j];
292 TrackRunInfo tri;
293 tri.track_id = traf.header.track_id;
294 tri.timescale = trak->media.header.timescale;
295 tri.start_dts = run_start_dts;
296 tri.sample_start_offset = trun.data_offset;
297 tri.track_sample_encryption_group =
298 &trak->media.information.sample_table.sample_group_description;
299 tri.fragment_sample_encryption_info =
300 traf.sample_group_description.entries;
302 tri.is_audio = (stsd.type == kAudio);
303 if (tri.is_audio) {
304 RCHECK(!stsd.audio_entries.empty());
305 if (desc_idx > stsd.audio_entries.size())
306 desc_idx = 0;
307 tri.audio_description = &stsd.audio_entries[desc_idx];
308 } else {
309 RCHECK(!stsd.video_entries.empty());
310 if (desc_idx > stsd.video_entries.size())
311 desc_idx = 0;
312 tri.video_description = &stsd.video_entries[desc_idx];
315 // Collect information from the auxiliary_offset entry with the same index
316 // in the 'saiz' container as the current run's index in the 'trun'
317 // container, if it is present.
318 if (traf.auxiliary_offset.offsets.size() > j) {
319 // There should be an auxiliary info entry corresponding to each sample
320 // in the auxiliary offset entry's corresponding track run.
321 RCHECK(traf.auxiliary_size.sample_count >=
322 sample_count_sum + trun.sample_count);
323 tri.aux_info_start_offset = traf.auxiliary_offset.offsets[j];
324 tri.aux_info_default_size =
325 traf.auxiliary_size.default_sample_info_size;
326 if (tri.aux_info_default_size == 0) {
327 const std::vector<uint8>& sizes =
328 traf.auxiliary_size.sample_info_sizes;
329 tri.aux_info_sizes.insert(tri.aux_info_sizes.begin(),
330 sizes.begin() + sample_count_sum,
331 sizes.begin() + sample_count_sum + trun.sample_count);
334 // If the default info size is positive, find the total size of the aux
335 // info block from it, otherwise sum over the individual sizes of each
336 // aux info entry in the aux_offset entry.
337 if (tri.aux_info_default_size) {
338 tri.aux_info_total_size =
339 tri.aux_info_default_size * trun.sample_count;
340 } else {
341 tri.aux_info_total_size = 0;
342 for (size_t k = 0; k < trun.sample_count; k++) {
343 tri.aux_info_total_size += tri.aux_info_sizes[k];
346 } else {
347 tri.aux_info_start_offset = -1;
348 tri.aux_info_total_size = 0;
351 tri.samples.resize(trun.sample_count);
352 for (size_t k = 0; k < trun.sample_count; k++) {
353 if (!PopulateSampleInfo(*trex, traf.header, trun, edit_list_offset, k,
354 &tri.samples[k], traf.sdtp.sample_depends_on(k),
355 tri.is_audio, media_log_)) {
356 return false;
359 run_start_dts += tri.samples[k].duration;
361 if (!is_sample_to_group_valid) {
362 // Set group description index to 0 to read encryption information
363 // from TrackEncryption Box.
364 tri.samples[k].cenc_group_description_index = 0;
365 continue;
368 uint32 index = sample_to_group_itr.group_description_index();
369 tri.samples[k].cenc_group_description_index = index;
370 if (index != 0)
371 RCHECK(GetSampleEncryptionInfoEntry(tri, index));
372 is_sample_to_group_valid = sample_to_group_itr.Advance();
374 runs_.push_back(tri);
375 sample_count_sum += trun.sample_count;
378 // We should have iterated through all samples in SampleToGroup Box.
379 RCHECK(!sample_to_group_itr.IsValid());
382 std::sort(runs_.begin(), runs_.end(), CompareMinTrackRunDataOffset());
383 run_itr_ = runs_.begin();
384 ResetRun();
385 return true;
388 void TrackRunIterator::AdvanceRun() {
389 ++run_itr_;
390 ResetRun();
393 void TrackRunIterator::ResetRun() {
394 if (!IsRunValid()) return;
395 sample_dts_ = run_itr_->start_dts;
396 sample_offset_ = run_itr_->sample_start_offset;
397 sample_itr_ = run_itr_->samples.begin();
398 cenc_info_.clear();
401 void TrackRunIterator::AdvanceSample() {
402 DCHECK(IsSampleValid());
403 sample_dts_ += sample_itr_->duration;
404 sample_offset_ += sample_itr_->size;
405 ++sample_itr_;
408 // This implementation only indicates a need for caching if CENC auxiliary
409 // info is available in the stream.
410 bool TrackRunIterator::AuxInfoNeedsToBeCached() {
411 DCHECK(IsRunValid());
412 return aux_info_size() > 0 && cenc_info_.size() == 0;
415 // This implementation currently only caches CENC auxiliary info.
416 bool TrackRunIterator::CacheAuxInfo(const uint8* buf, int buf_size) {
417 RCHECK(AuxInfoNeedsToBeCached() && buf_size >= aux_info_size());
419 cenc_info_.resize(run_itr_->samples.size());
420 int64 pos = 0;
421 for (size_t i = 0; i < run_itr_->samples.size(); i++) {
422 int info_size = run_itr_->aux_info_default_size;
423 if (!info_size)
424 info_size = run_itr_->aux_info_sizes[i];
426 if (IsSampleEncrypted(i)) {
427 BufferReader reader(buf + pos, info_size);
428 RCHECK(cenc_info_[i].Parse(GetIvSize(i), &reader));
430 pos += info_size;
433 return true;
436 bool TrackRunIterator::IsRunValid() const {
437 return run_itr_ != runs_.end();
440 bool TrackRunIterator::IsSampleValid() const {
441 return IsRunValid() && (sample_itr_ != run_itr_->samples.end());
444 // Because tracks are in sorted order and auxiliary information is cached when
445 // returning samples, it is guaranteed that no data will be required before the
446 // lesser of the minimum data offset of this track and the next in sequence.
447 // (The stronger condition - that no data is required before the minimum data
448 // offset of this track alone - is not guaranteed, because the BMFF spec does
449 // not have any inter-run ordering restrictions.)
450 int64 TrackRunIterator::GetMaxClearOffset() {
451 int64 offset = kint64max;
453 if (IsSampleValid()) {
454 offset = std::min(offset, sample_offset_);
455 if (AuxInfoNeedsToBeCached())
456 offset = std::min(offset, aux_info_offset());
458 if (run_itr_ != runs_.end()) {
459 std::vector<TrackRunInfo>::const_iterator next_run = run_itr_ + 1;
460 if (next_run != runs_.end()) {
461 offset = std::min(offset, next_run->sample_start_offset);
462 if (next_run->aux_info_total_size)
463 offset = std::min(offset, next_run->aux_info_start_offset);
466 if (offset == kint64max) return 0;
467 return offset;
470 uint32 TrackRunIterator::track_id() const {
471 DCHECK(IsRunValid());
472 return run_itr_->track_id;
475 bool TrackRunIterator::is_encrypted() const {
476 DCHECK(IsSampleValid());
477 return IsSampleEncrypted(sample_itr_ - run_itr_->samples.begin());
480 int64 TrackRunIterator::aux_info_offset() const {
481 return run_itr_->aux_info_start_offset;
484 int TrackRunIterator::aux_info_size() const {
485 return run_itr_->aux_info_total_size;
488 bool TrackRunIterator::is_audio() const {
489 DCHECK(IsRunValid());
490 return run_itr_->is_audio;
493 const AudioSampleEntry& TrackRunIterator::audio_description() const {
494 DCHECK(is_audio());
495 DCHECK(run_itr_->audio_description);
496 return *run_itr_->audio_description;
499 const VideoSampleEntry& TrackRunIterator::video_description() const {
500 DCHECK(!is_audio());
501 DCHECK(run_itr_->video_description);
502 return *run_itr_->video_description;
505 int64 TrackRunIterator::sample_offset() const {
506 DCHECK(IsSampleValid());
507 return sample_offset_;
510 int TrackRunIterator::sample_size() const {
511 DCHECK(IsSampleValid());
512 return sample_itr_->size;
515 DecodeTimestamp TrackRunIterator::dts() const {
516 DCHECK(IsSampleValid());
517 return DecodeTimestampFromRational(sample_dts_, run_itr_->timescale);
520 base::TimeDelta TrackRunIterator::cts() const {
521 DCHECK(IsSampleValid());
522 return TimeDeltaFromRational(sample_dts_ + sample_itr_->cts_offset,
523 run_itr_->timescale);
526 base::TimeDelta TrackRunIterator::duration() const {
527 DCHECK(IsSampleValid());
528 return TimeDeltaFromRational(sample_itr_->duration, run_itr_->timescale);
531 bool TrackRunIterator::is_keyframe() const {
532 DCHECK(IsSampleValid());
533 return sample_itr_->is_keyframe;
536 const TrackEncryption& TrackRunIterator::track_encryption() const {
537 if (is_audio())
538 return audio_description().sinf.info.track_encryption;
539 return video_description().sinf.info.track_encryption;
542 scoped_ptr<DecryptConfig> TrackRunIterator::GetDecryptConfig() {
543 DCHECK(is_encrypted());
545 if (cenc_info_.empty()) {
546 DCHECK_EQ(0, aux_info_size());
547 MEDIA_LOG(ERROR, media_log_) << "Aux Info is not available.";
548 return scoped_ptr<DecryptConfig>();
551 size_t sample_idx = sample_itr_ - run_itr_->samples.begin();
552 DCHECK_LT(sample_idx, cenc_info_.size());
553 const FrameCENCInfo& cenc_info = cenc_info_[sample_idx];
555 size_t total_size = 0;
556 if (!cenc_info.subsamples.empty() &&
557 (!cenc_info.GetTotalSizeOfSubsamples(&total_size) ||
558 total_size != static_cast<size_t>(sample_size()))) {
559 MEDIA_LOG(ERROR, media_log_) << "Incorrect CENC subsample size.";
560 return scoped_ptr<DecryptConfig>();
563 const std::vector<uint8>& kid = GetKeyId(sample_idx);
564 return scoped_ptr<DecryptConfig>(new DecryptConfig(
565 std::string(reinterpret_cast<const char*>(&kid[0]), kid.size()),
566 std::string(reinterpret_cast<const char*>(cenc_info.iv),
567 arraysize(cenc_info.iv)),
568 cenc_info.subsamples));
571 uint32 TrackRunIterator::GetGroupDescriptionIndex(uint32 sample_index) const {
572 DCHECK(IsRunValid());
573 DCHECK_LT(sample_index, run_itr_->samples.size());
574 return run_itr_->samples[sample_index].cenc_group_description_index;
577 bool TrackRunIterator::IsSampleEncrypted(size_t sample_index) const {
578 uint32 index = GetGroupDescriptionIndex(sample_index);
579 return (index == 0)
580 ? track_encryption().is_encrypted
581 : GetSampleEncryptionInfoEntry(*run_itr_, index)->is_encrypted;
584 const std::vector<uint8>& TrackRunIterator::GetKeyId(
585 size_t sample_index) const {
586 uint32 index = GetGroupDescriptionIndex(sample_index);
587 return (index == 0) ? track_encryption().default_kid
588 : GetSampleEncryptionInfoEntry(*run_itr_, index)->key_id;
591 uint8 TrackRunIterator::GetIvSize(size_t sample_index) const {
592 uint32 index = GetGroupDescriptionIndex(sample_index);
593 return (index == 0) ? track_encryption().default_iv_size
594 : GetSampleEncryptionInfoEntry(*run_itr_, index)->iv_size;
597 } // namespace mp4
598 } // namespace media