MPEG4Writer.cpp revision 674ebd0b4e1143e38392a4e3bb38b4679a4577bc
1/* 2 * Copyright (C) 2009 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17//#define LOG_NDEBUG 0 18#define LOG_TAG "MPEG4Writer" 19#include <utils/Log.h> 20 21#include <arpa/inet.h> 22 23#include <pthread.h> 24#include <sys/prctl.h> 25#include <sys/resource.h> 26 27#include <media/stagefright/MPEG4Writer.h> 28#include <media/stagefright/MediaBuffer.h> 29#include <media/stagefright/MetaData.h> 30#include <media/stagefright/MediaDebug.h> 31#include <media/stagefright/MediaDefs.h> 32#include <media/stagefright/MediaErrors.h> 33#include <media/stagefright/MediaSource.h> 34#include <media/stagefright/Utils.h> 35#include <media/mediarecorder.h> 36#include <sys/types.h> 37#include <sys/stat.h> 38#include <fcntl.h> 39#include <unistd.h> 40 41#include "include/ESDS.h" 42 43namespace android { 44 45static const int64_t kMax32BitFileSize = 0x007fffffffLL; 46static const uint8_t kNalUnitTypeSeqParamSet = 0x07; 47static const uint8_t kNalUnitTypePicParamSet = 0x08; 48 49// Using longer adjustment period to suppress fluctuations in 50// the audio encoding paths 51static const int64_t kVideoMediaTimeAdjustPeriodTimeUs = 600000000LL; // 10 minutes 52 53class MPEG4Writer::Track { 54public: 55 Track(MPEG4Writer *owner, const sp<MediaSource> &source); 56 57 ~Track(); 58 59 status_t start(MetaData *params); 60 status_t stop(); 61 status_t pause(); 62 bool reachedEOS(); 63 64 int64_t getDurationUs() const; 65 int64_t getEstimatedTrackSizeBytes() const; 66 void writeTrackHeader(int32_t trackID, bool use32BitOffset = true); 67 void bufferChunk(int64_t timestampUs); 68 bool isAvc() const { return mIsAvc; } 69 bool isAudio() const { return mIsAudio; } 70 bool isMPEG4() const { return mIsMPEG4; } 71 void addChunkOffset(off64_t offset); 72 status_t dump(int fd, const Vector<String16>& args) const; 73 74private: 75 MPEG4Writer *mOwner; 76 sp<MetaData> mMeta; 77 sp<MediaSource> mSource; 78 volatile bool mDone; 79 volatile bool mPaused; 80 volatile bool mResumed; 81 bool mIsAvc; 82 bool mIsAudio; 83 bool mIsMPEG4; 84 int64_t mTrackDurationUs; 85 86 // For realtime applications, we need to adjust the media clock 87 // for video track based on the audio media clock 88 bool mIsRealTimeRecording; 89 int64_t mMaxTimeStampUs; 90 int64_t mEstimatedTrackSizeBytes; 91 int64_t mMdatSizeBytes; 92 int32_t mTimeScale; 93 94 pthread_t mThread; 95 96 // mNumSamples is used to track how many samples in mSampleSizes List. 97 // This is to reduce the cost associated with mSampleSizes.size() call, 98 // since it is O(n). Ideally, the fix should be in List class. 99 size_t mNumSamples; 100 List<size_t> mSampleSizes; 101 bool mSamplesHaveSameSize; 102 103 List<MediaBuffer *> mChunkSamples; 104 105 size_t mNumStcoTableEntries; 106 List<off64_t> mChunkOffsets; 107 108 size_t mNumStscTableEntries; 109 struct StscTableEntry { 110 111 StscTableEntry(uint32_t chunk, uint32_t samples, uint32_t id) 112 : firstChunk(chunk), 113 samplesPerChunk(samples), 114 sampleDescriptionId(id) {} 115 116 uint32_t firstChunk; 117 uint32_t samplesPerChunk; 118 uint32_t sampleDescriptionId; 119 }; 120 List<StscTableEntry> mStscTableEntries; 121 122 size_t mNumStssTableEntries; 123 List<int32_t> mStssTableEntries; 124 125 size_t mNumSttsTableEntries; 126 struct SttsTableEntry { 127 128 SttsTableEntry(uint32_t count, uint32_t durationUs) 129 : sampleCount(count), sampleDurationUs(durationUs) {} 130 131 uint32_t sampleCount; 132 uint32_t sampleDurationUs; 133 }; 134 List<SttsTableEntry> mSttsTableEntries; 135 136 // Sequence parameter set or picture parameter set 137 struct AVCParamSet { 138 AVCParamSet(uint16_t length, const uint8_t *data) 139 : mLength(length), mData(data) {} 140 141 uint16_t mLength; 142 const uint8_t *mData; 143 }; 144 List<AVCParamSet> mSeqParamSets; 145 List<AVCParamSet> mPicParamSets; 146 uint8_t mProfileIdc; 147 uint8_t mProfileCompatible; 148 uint8_t mLevelIdc; 149 150 void *mCodecSpecificData; 151 size_t mCodecSpecificDataSize; 152 bool mGotAllCodecSpecificData; 153 bool mTrackingProgressStatus; 154 155 bool mReachedEOS; 156 int64_t mStartTimestampUs; 157 int64_t mPreviousTrackTimeUs; 158 int64_t mTrackEveryTimeDurationUs; 159 160 // Has the media time adjustment for video started? 161 bool mIsMediaTimeAdjustmentOn; 162 // The time stamp when previous media time adjustment period starts 163 int64_t mPrevMediaTimeAdjustTimestampUs; 164 // Number of vidoe frames whose time stamp may be adjusted 165 int64_t mMediaTimeAdjustNumFrames; 166 // The sample number when previous meida time adjustmnet period starts 167 int64_t mPrevMediaTimeAdjustSample; 168 // The total accumulated drift time within a period of 169 // kVideoMediaTimeAdjustPeriodTimeUs. 170 int64_t mTotalDriftTimeToAdjustUs; 171 // The total accumalated drift time since the start of the recording 172 // excluding the current time adjustment period 173 int64_t mPrevTotalAccumDriftTimeUs; 174 175 // Update the audio track's drift information. 176 void updateDriftTime(const sp<MetaData>& meta); 177 178 // Adjust the time stamp of the video track according to 179 // the drift time information from the audio track. 180 void adjustMediaTime(int64_t *timestampUs); 181 182 static void *ThreadWrapper(void *me); 183 status_t threadEntry(); 184 185 const uint8_t *parseParamSet( 186 const uint8_t *data, size_t length, int type, size_t *paramSetLen); 187 188 status_t makeAVCCodecSpecificData( 189 const uint8_t *data, size_t size); 190 status_t copyAVCCodecSpecificData( 191 const uint8_t *data, size_t size); 192 status_t parseAVCCodecSpecificData( 193 const uint8_t *data, size_t size); 194 195 // Track authoring progress status 196 void trackProgressStatus(int64_t timeUs, status_t err = OK); 197 void initTrackingProgressStatus(MetaData *params); 198 199 void getCodecSpecificDataFromInputFormatIfPossible(); 200 201 // Determine the track time scale 202 // If it is an audio track, try to use the sampling rate as 203 // the time scale; however, if user chooses the overwrite 204 // value, the user-supplied time scale will be used. 205 void setTimeScale(); 206 207 // Simple validation on the codec specific data 208 status_t checkCodecSpecificData() const; 209 int32_t mRotation; 210 211 void updateTrackSizeEstimate(); 212 void addOneStscTableEntry(size_t chunkId, size_t sampleId); 213 void addOneStssTableEntry(size_t sampleId); 214 void addOneSttsTableEntry(size_t sampleCount, int64_t durationUs); 215 216 Track(const Track &); 217 Track &operator=(const Track &); 218}; 219 220MPEG4Writer::MPEG4Writer(const char *filename) 221 : mFd(-1), 222 mInitCheck(NO_INIT), 223 mUse4ByteNalLength(true), 224 mUse32BitOffset(true), 225 mIsFileSizeLimitExplicitlyRequested(false), 226 mPaused(false), 227 mStarted(false), 228 mOffset(0), 229 mMdatOffset(0), 230 mEstimatedMoovBoxSize(0), 231 mInterleaveDurationUs(1000000) { 232 233 mFd = open(filename, O_CREAT | O_LARGEFILE | O_TRUNC); 234 if (mFd >= 0) { 235 mInitCheck = OK; 236 } 237} 238 239MPEG4Writer::MPEG4Writer(int fd) 240 : mFd(dup(fd)), 241 mInitCheck(mFd < 0? NO_INIT: OK), 242 mUse4ByteNalLength(true), 243 mUse32BitOffset(true), 244 mIsFileSizeLimitExplicitlyRequested(false), 245 mPaused(false), 246 mStarted(false), 247 mOffset(0), 248 mMdatOffset(0), 249 mEstimatedMoovBoxSize(0), 250 mInterleaveDurationUs(1000000) { 251} 252 253MPEG4Writer::~MPEG4Writer() { 254 stop(); 255 256 while (!mTracks.empty()) { 257 List<Track *>::iterator it = mTracks.begin(); 258 delete *it; 259 (*it) = NULL; 260 mTracks.erase(it); 261 } 262 mTracks.clear(); 263} 264 265status_t MPEG4Writer::dump( 266 int fd, const Vector<String16>& args) { 267 const size_t SIZE = 256; 268 char buffer[SIZE]; 269 String8 result; 270 snprintf(buffer, SIZE, " MPEG4Writer %p\n", this); 271 result.append(buffer); 272 snprintf(buffer, SIZE, " mStarted: %s\n", mStarted? "true": "false"); 273 result.append(buffer); 274 ::write(fd, result.string(), result.size()); 275 for (List<Track *>::iterator it = mTracks.begin(); 276 it != mTracks.end(); ++it) { 277 (*it)->dump(fd, args); 278 } 279 return OK; 280} 281 282status_t MPEG4Writer::Track::dump( 283 int fd, const Vector<String16>& args) const { 284 const size_t SIZE = 256; 285 char buffer[SIZE]; 286 String8 result; 287 snprintf(buffer, SIZE, " %s track\n", mIsAudio? "Audio": "Video"); 288 result.append(buffer); 289 snprintf(buffer, SIZE, " reached EOS: %s\n", 290 mReachedEOS? "true": "false"); 291 result.append(buffer); 292 ::write(fd, result.string(), result.size()); 293 return OK; 294} 295 296status_t MPEG4Writer::addSource(const sp<MediaSource> &source) { 297 Track *track = new Track(this, source); 298 mTracks.push_back(track); 299 300 return OK; 301} 302 303status_t MPEG4Writer::startTracks(MetaData *params) { 304 for (List<Track *>::iterator it = mTracks.begin(); 305 it != mTracks.end(); ++it) { 306 status_t err = (*it)->start(params); 307 308 if (err != OK) { 309 for (List<Track *>::iterator it2 = mTracks.begin(); 310 it2 != it; ++it2) { 311 (*it2)->stop(); 312 } 313 314 return err; 315 } 316 } 317 return OK; 318} 319 320int64_t MPEG4Writer::estimateMoovBoxSize(int32_t bitRate) { 321 // This implementation is highly experimental/heurisitic. 322 // 323 // Statistical analysis shows that metadata usually accounts 324 // for a small portion of the total file size, usually < 0.6%. 325 326 // The default MIN_MOOV_BOX_SIZE is set to 0.6% x 1MB / 2, 327 // where 1MB is the common file size limit for MMS application. 328 // The default MAX _MOOV_BOX_SIZE value is based on about 3 329 // minute video recording with a bit rate about 3 Mbps, because 330 // statistics also show that most of the video captured are going 331 // to be less than 3 minutes. 332 333 // If the estimation is wrong, we will pay the price of wasting 334 // some reserved space. This should not happen so often statistically. 335 static const int32_t factor = mUse32BitOffset? 1: 2; 336 static const int64_t MIN_MOOV_BOX_SIZE = 3 * 1024; // 3 KB 337 static const int64_t MAX_MOOV_BOX_SIZE = (180 * 3000000 * 6LL / 8000); 338 int64_t size = MIN_MOOV_BOX_SIZE; 339 340 // Max file size limit is set 341 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 342 size = mMaxFileSizeLimitBytes * 6 / 1000; 343 } 344 345 // Max file duration limit is set 346 if (mMaxFileDurationLimitUs != 0) { 347 if (bitRate > 0) { 348 int64_t size2 = 349 ((mMaxFileDurationLimitUs * bitRate * 6) / 1000 / 8000000); 350 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 351 // When both file size and duration limits are set, 352 // we use the smaller limit of the two. 353 if (size > size2) { 354 size = size2; 355 } 356 } else { 357 // Only max file duration limit is set 358 size = size2; 359 } 360 } 361 } 362 363 if (size < MIN_MOOV_BOX_SIZE) { 364 size = MIN_MOOV_BOX_SIZE; 365 } 366 367 // Any long duration recording will be probably end up with 368 // non-streamable mp4 file. 369 if (size > MAX_MOOV_BOX_SIZE) { 370 size = MAX_MOOV_BOX_SIZE; 371 } 372 373 LOGI("limits: %lld/%lld bytes/us, bit rate: %d bps and the estimated" 374 " moov size %lld bytes", 375 mMaxFileSizeLimitBytes, mMaxFileDurationLimitUs, bitRate, size); 376 return factor * size; 377} 378 379status_t MPEG4Writer::start(MetaData *param) { 380 if (mInitCheck != OK) { 381 return UNKNOWN_ERROR; 382 } 383 384 /* 385 * Check mMaxFileSizeLimitBytes at the beginning 386 * since mMaxFileSizeLimitBytes may be implicitly 387 * changed later for 32-bit file offset even if 388 * user does not ask to set it explicitly. 389 */ 390 if (mMaxFileSizeLimitBytes != 0) { 391 mIsFileSizeLimitExplicitlyRequested = true; 392 } 393 394 int32_t use64BitOffset; 395 if (param && 396 param->findInt32(kKey64BitFileOffset, &use64BitOffset) && 397 use64BitOffset) { 398 mUse32BitOffset = false; 399 } 400 401 if (mUse32BitOffset) { 402 // Implicit 32 bit file size limit 403 if (mMaxFileSizeLimitBytes == 0) { 404 mMaxFileSizeLimitBytes = kMax32BitFileSize; 405 } 406 407 // If file size is set to be larger than the 32 bit file 408 // size limit, treat it as an error. 409 if (mMaxFileSizeLimitBytes > kMax32BitFileSize) { 410 LOGW("32-bit file size limit (%lld bytes) too big. " 411 "It is changed to %lld bytes", 412 mMaxFileSizeLimitBytes, kMax32BitFileSize); 413 mMaxFileSizeLimitBytes = kMax32BitFileSize; 414 } 415 } 416 417 int32_t use2ByteNalLength; 418 if (param && 419 param->findInt32(kKey2ByteNalLength, &use2ByteNalLength) && 420 use2ByteNalLength) { 421 mUse4ByteNalLength = false; 422 } 423 424 mStartTimestampUs = -1; 425 426 if (mStarted) { 427 if (mPaused) { 428 mPaused = false; 429 return startTracks(param); 430 } 431 return OK; 432 } 433 434 if (!param || 435 !param->findInt32(kKeyTimeScale, &mTimeScale)) { 436 mTimeScale = 1000; 437 } 438 CHECK(mTimeScale > 0); 439 LOGV("movie time scale: %d", mTimeScale); 440 441 mStreamableFile = true; 442 mWriteMoovBoxToMemory = false; 443 mMoovBoxBuffer = NULL; 444 mMoovBoxBufferOffset = 0; 445 446 beginBox("ftyp"); 447 { 448 int32_t fileType; 449 if (param && param->findInt32(kKeyFileType, &fileType) && 450 fileType != OUTPUT_FORMAT_MPEG_4) { 451 writeFourcc("3gp4"); 452 } else { 453 writeFourcc("isom"); 454 } 455 } 456 writeInt32(0); 457 writeFourcc("isom"); 458 writeFourcc("3gp4"); 459 endBox(); 460 461 mFreeBoxOffset = mOffset; 462 463 if (mEstimatedMoovBoxSize == 0) { 464 int32_t bitRate = -1; 465 if (param) { 466 param->findInt32(kKeyBitRate, &bitRate); 467 } 468 mEstimatedMoovBoxSize = estimateMoovBoxSize(bitRate); 469 } 470 CHECK(mEstimatedMoovBoxSize >= 8); 471 lseek64(mFd, mFreeBoxOffset, SEEK_SET); 472 writeInt32(mEstimatedMoovBoxSize); 473 write("free", 4); 474 475 mMdatOffset = mFreeBoxOffset + mEstimatedMoovBoxSize; 476 mOffset = mMdatOffset; 477 lseek64(mFd, mMdatOffset, SEEK_SET); 478 if (mUse32BitOffset) { 479 write("????mdat", 8); 480 } else { 481 write("\x00\x00\x00\x01mdat????????", 16); 482 } 483 484 status_t err = startWriterThread(); 485 if (err != OK) { 486 return err; 487 } 488 489 err = startTracks(param); 490 if (err != OK) { 491 return err; 492 } 493 494 mStarted = true; 495 return OK; 496} 497 498bool MPEG4Writer::use32BitFileOffset() const { 499 return mUse32BitOffset; 500} 501 502status_t MPEG4Writer::pause() { 503 if (mInitCheck != OK) { 504 return OK; 505 } 506 mPaused = true; 507 status_t err = OK; 508 for (List<Track *>::iterator it = mTracks.begin(); 509 it != mTracks.end(); ++it) { 510 status_t status = (*it)->pause(); 511 if (status != OK) { 512 err = status; 513 } 514 } 515 return err; 516} 517 518void MPEG4Writer::stopWriterThread() { 519 LOGD("Stopping writer thread"); 520 521 { 522 Mutex::Autolock autolock(mLock); 523 524 mDone = true; 525 mChunkReadyCondition.signal(); 526 } 527 528 void *dummy; 529 pthread_join(mThread, &dummy); 530 LOGD("Writer thread stopped"); 531} 532 533/* 534 * MP4 file standard defines a composition matrix: 535 * | a b u | 536 * | c d v | 537 * | x y w | 538 * 539 * the element in the matrix is stored in the following 540 * order: {a, b, u, c, d, v, x, y, w}, 541 * where a, b, c, d, x, and y is in 16.16 format, while 542 * u, v and w is in 2.30 format. 543 */ 544void MPEG4Writer::writeCompositionMatrix(int degrees) { 545 LOGV("writeCompositionMatrix"); 546 uint32_t a = 0x00010000; 547 uint32_t b = 0; 548 uint32_t c = 0; 549 uint32_t d = 0x00010000; 550 switch (degrees) { 551 case 0: 552 break; 553 case 90: 554 a = 0; 555 b = 0x00010000; 556 c = 0xFFFF0000; 557 d = 0; 558 break; 559 case 180: 560 a = 0xFFFF0000; 561 d = 0xFFFF0000; 562 break; 563 case 270: 564 a = 0; 565 b = 0xFFFF0000; 566 c = 0x00010000; 567 d = 0; 568 break; 569 default: 570 CHECK(!"Should never reach this unknown rotation"); 571 break; 572 } 573 574 writeInt32(a); // a 575 writeInt32(b); // b 576 writeInt32(0); // u 577 writeInt32(c); // c 578 writeInt32(d); // d 579 writeInt32(0); // v 580 writeInt32(0); // x 581 writeInt32(0); // y 582 writeInt32(0x40000000); // w 583} 584 585 586status_t MPEG4Writer::stop() { 587 if (mInitCheck != OK) { 588 return OK; 589 } 590 591 status_t err = OK; 592 int64_t maxDurationUs = 0; 593 for (List<Track *>::iterator it = mTracks.begin(); 594 it != mTracks.end(); ++it) { 595 status_t status = (*it)->stop(); 596 if (err == OK && status != OK) { 597 err = status; 598 } 599 600 int64_t durationUs = (*it)->getDurationUs(); 601 if (durationUs > maxDurationUs) { 602 maxDurationUs = durationUs; 603 } 604 } 605 606 stopWriterThread(); 607 608 // Do not write out movie header on error. 609 if (err != OK) { 610 close(mFd); 611 mFd = -1; 612 mInitCheck = NO_INIT; 613 mStarted = false; 614 return err; 615 } 616 617 // Fix up the size of the 'mdat' chunk. 618 if (mUse32BitOffset) { 619 lseek64(mFd, mMdatOffset, SEEK_SET); 620 int32_t size = htonl(static_cast<int32_t>(mOffset - mMdatOffset)); 621 ::write(mFd, &size, 4); 622 } else { 623 lseek64(mFd, mMdatOffset + 8, SEEK_SET); 624 int64_t size = mOffset - mMdatOffset; 625 size = hton64(size); 626 ::write(mFd, &size, 8); 627 } 628 lseek64(mFd, mOffset, SEEK_SET); 629 630 time_t now = time(NULL); 631 const off64_t moovOffset = mOffset; 632 mWriteMoovBoxToMemory = true; 633 mMoovBoxBuffer = (uint8_t *) malloc(mEstimatedMoovBoxSize); 634 mMoovBoxBufferOffset = 0; 635 CHECK(mMoovBoxBuffer != NULL); 636 int32_t duration = (maxDurationUs * mTimeScale + 5E5) / 1E6; 637 638 beginBox("moov"); 639 640 beginBox("mvhd"); 641 writeInt32(0); // version=0, flags=0 642 writeInt32(now); // creation time 643 writeInt32(now); // modification time 644 writeInt32(mTimeScale); // mvhd timescale 645 writeInt32(duration); 646 writeInt32(0x10000); // rate: 1.0 647 writeInt16(0x100); // volume 648 writeInt16(0); // reserved 649 writeInt32(0); // reserved 650 writeInt32(0); // reserved 651 writeCompositionMatrix(0); // matrix 652 writeInt32(0); // predefined 653 writeInt32(0); // predefined 654 writeInt32(0); // predefined 655 writeInt32(0); // predefined 656 writeInt32(0); // predefined 657 writeInt32(0); // predefined 658 writeInt32(mTracks.size() + 1); // nextTrackID 659 endBox(); // mvhd 660 661 int32_t id = 1; 662 for (List<Track *>::iterator it = mTracks.begin(); 663 it != mTracks.end(); ++it, ++id) { 664 (*it)->writeTrackHeader(id, mUse32BitOffset); 665 } 666 endBox(); // moov 667 668 mWriteMoovBoxToMemory = false; 669 if (mStreamableFile) { 670 CHECK(mMoovBoxBufferOffset + 8 <= mEstimatedMoovBoxSize); 671 672 // Moov box 673 lseek64(mFd, mFreeBoxOffset, SEEK_SET); 674 mOffset = mFreeBoxOffset; 675 write(mMoovBoxBuffer, 1, mMoovBoxBufferOffset); 676 677 // Free box 678 lseek64(mFd, mOffset, SEEK_SET); 679 writeInt32(mEstimatedMoovBoxSize - mMoovBoxBufferOffset); 680 write("free", 4); 681 682 // Free temp memory 683 free(mMoovBoxBuffer); 684 mMoovBoxBuffer = NULL; 685 mMoovBoxBufferOffset = 0; 686 } else { 687 LOGI("The mp4 file will not be streamable."); 688 } 689 690 CHECK(mBoxes.empty()); 691 692 close(mFd); 693 mFd = -1; 694 mInitCheck = NO_INIT; 695 mStarted = false; 696 return err; 697} 698 699status_t MPEG4Writer::setInterleaveDuration(uint32_t durationUs) { 700 mInterleaveDurationUs = durationUs; 701 return OK; 702} 703 704void MPEG4Writer::lock() { 705 mLock.lock(); 706} 707 708void MPEG4Writer::unlock() { 709 mLock.unlock(); 710} 711 712off64_t MPEG4Writer::addSample_l(MediaBuffer *buffer) { 713 off64_t old_offset = mOffset; 714 715 ::write(mFd, 716 (const uint8_t *)buffer->data() + buffer->range_offset(), 717 buffer->range_length()); 718 719 mOffset += buffer->range_length(); 720 721 return old_offset; 722} 723 724static void StripStartcode(MediaBuffer *buffer) { 725 if (buffer->range_length() < 4) { 726 return; 727 } 728 729 const uint8_t *ptr = 730 (const uint8_t *)buffer->data() + buffer->range_offset(); 731 732 if (!memcmp(ptr, "\x00\x00\x00\x01", 4)) { 733 buffer->set_range( 734 buffer->range_offset() + 4, buffer->range_length() - 4); 735 } 736} 737 738off64_t MPEG4Writer::addLengthPrefixedSample_l(MediaBuffer *buffer) { 739 off64_t old_offset = mOffset; 740 741 size_t length = buffer->range_length(); 742 743 if (mUse4ByteNalLength) { 744 uint8_t x = length >> 24; 745 ::write(mFd, &x, 1); 746 x = (length >> 16) & 0xff; 747 ::write(mFd, &x, 1); 748 x = (length >> 8) & 0xff; 749 ::write(mFd, &x, 1); 750 x = length & 0xff; 751 ::write(mFd, &x, 1); 752 753 ::write(mFd, 754 (const uint8_t *)buffer->data() + buffer->range_offset(), 755 length); 756 757 mOffset += length + 4; 758 } else { 759 CHECK(length < 65536); 760 761 uint8_t x = length >> 8; 762 ::write(mFd, &x, 1); 763 x = length & 0xff; 764 ::write(mFd, &x, 1); 765 ::write(mFd, (const uint8_t *)buffer->data() + buffer->range_offset(), length); 766 mOffset += length + 2; 767 } 768 769 return old_offset; 770} 771 772size_t MPEG4Writer::write( 773 const void *ptr, size_t size, size_t nmemb) { 774 775 const size_t bytes = size * nmemb; 776 if (mWriteMoovBoxToMemory) { 777 // This happens only when we write the moov box at the end of 778 // recording, not for each output video/audio frame we receive. 779 off64_t moovBoxSize = 8 + mMoovBoxBufferOffset + bytes; 780 if (moovBoxSize > mEstimatedMoovBoxSize) { 781 for (List<off64_t>::iterator it = mBoxes.begin(); 782 it != mBoxes.end(); ++it) { 783 (*it) += mOffset; 784 } 785 lseek64(mFd, mOffset, SEEK_SET); 786 ::write(mFd, mMoovBoxBuffer, mMoovBoxBufferOffset); 787 ::write(mFd, ptr, size * nmemb); 788 mOffset += (bytes + mMoovBoxBufferOffset); 789 free(mMoovBoxBuffer); 790 mMoovBoxBuffer = NULL; 791 mMoovBoxBufferOffset = 0; 792 mWriteMoovBoxToMemory = false; 793 mStreamableFile = false; 794 } else { 795 memcpy(mMoovBoxBuffer + mMoovBoxBufferOffset, ptr, bytes); 796 mMoovBoxBufferOffset += bytes; 797 } 798 } else { 799 ::write(mFd, ptr, size * nmemb); 800 mOffset += bytes; 801 } 802 return bytes; 803} 804 805void MPEG4Writer::beginBox(const char *fourcc) { 806 CHECK_EQ(strlen(fourcc), 4); 807 808 mBoxes.push_back(mWriteMoovBoxToMemory? 809 mMoovBoxBufferOffset: mOffset); 810 811 writeInt32(0); 812 writeFourcc(fourcc); 813} 814 815void MPEG4Writer::endBox() { 816 CHECK(!mBoxes.empty()); 817 818 off64_t offset = *--mBoxes.end(); 819 mBoxes.erase(--mBoxes.end()); 820 821 if (mWriteMoovBoxToMemory) { 822 int32_t x = htonl(mMoovBoxBufferOffset - offset); 823 memcpy(mMoovBoxBuffer + offset, &x, 4); 824 } else { 825 lseek64(mFd, offset, SEEK_SET); 826 writeInt32(mOffset - offset); 827 mOffset -= 4; 828 lseek64(mFd, mOffset, SEEK_SET); 829 } 830} 831 832void MPEG4Writer::writeInt8(int8_t x) { 833 write(&x, 1, 1); 834} 835 836void MPEG4Writer::writeInt16(int16_t x) { 837 x = htons(x); 838 write(&x, 1, 2); 839} 840 841void MPEG4Writer::writeInt32(int32_t x) { 842 x = htonl(x); 843 write(&x, 1, 4); 844} 845 846void MPEG4Writer::writeInt64(int64_t x) { 847 x = hton64(x); 848 write(&x, 1, 8); 849} 850 851void MPEG4Writer::writeCString(const char *s) { 852 size_t n = strlen(s); 853 write(s, 1, n + 1); 854} 855 856void MPEG4Writer::writeFourcc(const char *s) { 857 CHECK_EQ(strlen(s), 4); 858 write(s, 1, 4); 859} 860 861void MPEG4Writer::write(const void *data, size_t size) { 862 write(data, 1, size); 863} 864 865bool MPEG4Writer::isFileStreamable() const { 866 return mStreamableFile; 867} 868 869bool MPEG4Writer::exceedsFileSizeLimit() { 870 // No limit 871 if (mMaxFileSizeLimitBytes == 0) { 872 return false; 873 } 874 875 int64_t nTotalBytesEstimate = static_cast<int64_t>(mEstimatedMoovBoxSize); 876 for (List<Track *>::iterator it = mTracks.begin(); 877 it != mTracks.end(); ++it) { 878 nTotalBytesEstimate += (*it)->getEstimatedTrackSizeBytes(); 879 } 880 881 return (nTotalBytesEstimate >= mMaxFileSizeLimitBytes); 882} 883 884bool MPEG4Writer::exceedsFileDurationLimit() { 885 // No limit 886 if (mMaxFileDurationLimitUs == 0) { 887 return false; 888 } 889 890 for (List<Track *>::iterator it = mTracks.begin(); 891 it != mTracks.end(); ++it) { 892 if ((*it)->getDurationUs() >= mMaxFileDurationLimitUs) { 893 return true; 894 } 895 } 896 return false; 897} 898 899bool MPEG4Writer::reachedEOS() { 900 bool allDone = true; 901 for (List<Track *>::iterator it = mTracks.begin(); 902 it != mTracks.end(); ++it) { 903 if (!(*it)->reachedEOS()) { 904 allDone = false; 905 break; 906 } 907 } 908 909 return allDone; 910} 911 912void MPEG4Writer::setStartTimestampUs(int64_t timeUs) { 913 LOGI("setStartTimestampUs: %lld", timeUs); 914 CHECK(timeUs >= 0); 915 Mutex::Autolock autoLock(mLock); 916 if (mStartTimestampUs < 0 || mStartTimestampUs > timeUs) { 917 mStartTimestampUs = timeUs; 918 LOGI("Earliest track starting time: %lld", mStartTimestampUs); 919 } 920} 921 922int64_t MPEG4Writer::getStartTimestampUs() { 923 Mutex::Autolock autoLock(mLock); 924 return mStartTimestampUs; 925} 926 927size_t MPEG4Writer::numTracks() { 928 Mutex::Autolock autolock(mLock); 929 return mTracks.size(); 930} 931 932//////////////////////////////////////////////////////////////////////////////// 933 934MPEG4Writer::Track::Track( 935 MPEG4Writer *owner, const sp<MediaSource> &source) 936 : mOwner(owner), 937 mMeta(source->getFormat()), 938 mSource(source), 939 mDone(false), 940 mPaused(false), 941 mResumed(false), 942 mTrackDurationUs(0), 943 mEstimatedTrackSizeBytes(0), 944 mSamplesHaveSameSize(true), 945 mCodecSpecificData(NULL), 946 mCodecSpecificDataSize(0), 947 mGotAllCodecSpecificData(false), 948 mReachedEOS(false), 949 mRotation(0) { 950 getCodecSpecificDataFromInputFormatIfPossible(); 951 952 const char *mime; 953 mMeta->findCString(kKeyMIMEType, &mime); 954 mIsAvc = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC); 955 mIsAudio = !strncasecmp(mime, "audio/", 6); 956 mIsMPEG4 = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) || 957 !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC); 958 959 setTimeScale(); 960} 961 962void MPEG4Writer::Track::updateTrackSizeEstimate() { 963 964 int64_t stcoBoxSizeBytes = mOwner->use32BitFileOffset() 965 ? mNumStcoTableEntries * 4 966 : mNumStcoTableEntries * 8; 967 968 int64_t stszBoxSizeBytes = mSamplesHaveSameSize? 4: (mNumSamples * 4); 969 970 mEstimatedTrackSizeBytes = mMdatSizeBytes; // media data size 971 if (!mOwner->isFileStreamable()) { 972 // Reserved free space is not large enough to hold 973 // all meta data and thus wasted. 974 mEstimatedTrackSizeBytes += mNumStscTableEntries * 12 + // stsc box size 975 mNumStssTableEntries * 4 + // stss box size 976 mNumSttsTableEntries * 8 + // stts box size 977 stcoBoxSizeBytes + // stco box size 978 stszBoxSizeBytes; // stsz box size 979 } 980} 981 982void MPEG4Writer::Track::addOneStscTableEntry( 983 size_t chunkId, size_t sampleId) { 984 985 StscTableEntry stscEntry(chunkId, sampleId, 1); 986 mStscTableEntries.push_back(stscEntry); 987 ++mNumStscTableEntries; 988} 989 990void MPEG4Writer::Track::addOneStssTableEntry(size_t sampleId) { 991 mStssTableEntries.push_back(sampleId); 992 ++mNumStssTableEntries; 993} 994 995void MPEG4Writer::Track::addOneSttsTableEntry( 996 size_t sampleCount, int64_t durationUs) { 997 998 SttsTableEntry sttsEntry(sampleCount, durationUs); 999 mSttsTableEntries.push_back(sttsEntry); 1000 ++mNumSttsTableEntries; 1001} 1002 1003void MPEG4Writer::Track::addChunkOffset(off64_t offset) { 1004 ++mNumStcoTableEntries; 1005 mChunkOffsets.push_back(offset); 1006} 1007 1008void MPEG4Writer::Track::setTimeScale() { 1009 LOGV("setTimeScale"); 1010 // Default time scale 1011 mTimeScale = 90000; 1012 1013 if (mIsAudio) { 1014 // Use the sampling rate as the default time scale for audio track. 1015 int32_t sampleRate; 1016 bool success = mMeta->findInt32(kKeySampleRate, &sampleRate); 1017 CHECK(success); 1018 mTimeScale = sampleRate; 1019 } 1020 1021 // If someone would like to overwrite the timescale, use user-supplied value. 1022 int32_t timeScale; 1023 if (mMeta->findInt32(kKeyTimeScale, &timeScale)) { 1024 mTimeScale = timeScale; 1025 } 1026 1027 CHECK(mTimeScale > 0); 1028} 1029 1030void MPEG4Writer::Track::getCodecSpecificDataFromInputFormatIfPossible() { 1031 const char *mime; 1032 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 1033 1034 if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC)) { 1035 uint32_t type; 1036 const void *data; 1037 size_t size; 1038 if (mMeta->findData(kKeyAVCC, &type, &data, &size)) { 1039 mCodecSpecificData = malloc(size); 1040 mCodecSpecificDataSize = size; 1041 memcpy(mCodecSpecificData, data, size); 1042 mGotAllCodecSpecificData = true; 1043 } 1044 } else if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) 1045 || !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC)) { 1046 uint32_t type; 1047 const void *data; 1048 size_t size; 1049 if (mMeta->findData(kKeyESDS, &type, &data, &size)) { 1050 ESDS esds(data, size); 1051 if (esds.getCodecSpecificInfo(&data, &size) == OK) { 1052 mCodecSpecificData = malloc(size); 1053 mCodecSpecificDataSize = size; 1054 memcpy(mCodecSpecificData, data, size); 1055 mGotAllCodecSpecificData = true; 1056 } 1057 } 1058 } 1059} 1060 1061MPEG4Writer::Track::~Track() { 1062 stop(); 1063 1064 if (mCodecSpecificData != NULL) { 1065 free(mCodecSpecificData); 1066 mCodecSpecificData = NULL; 1067 } 1068} 1069 1070void MPEG4Writer::Track::initTrackingProgressStatus(MetaData *params) { 1071 LOGV("initTrackingProgressStatus"); 1072 mPreviousTrackTimeUs = -1; 1073 mTrackingProgressStatus = false; 1074 mTrackEveryTimeDurationUs = 0; 1075 { 1076 int64_t timeUs; 1077 if (params && params->findInt64(kKeyTrackTimeStatus, &timeUs)) { 1078 LOGV("Receive request to track progress status for every %lld us", timeUs); 1079 mTrackEveryTimeDurationUs = timeUs; 1080 mTrackingProgressStatus = true; 1081 } 1082 } 1083} 1084 1085// static 1086void *MPEG4Writer::ThreadWrapper(void *me) { 1087 LOGV("ThreadWrapper: %p", me); 1088 MPEG4Writer *writer = static_cast<MPEG4Writer *>(me); 1089 writer->threadFunc(); 1090 return NULL; 1091} 1092 1093void MPEG4Writer::bufferChunk(const Chunk& chunk) { 1094 LOGV("bufferChunk: %p", chunk.mTrack); 1095 Mutex::Autolock autolock(mLock); 1096 CHECK_EQ(mDone, false); 1097 1098 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1099 it != mChunkInfos.end(); ++it) { 1100 1101 if (chunk.mTrack == it->mTrack) { // Found owner 1102 it->mChunks.push_back(chunk); 1103 mChunkReadyCondition.signal(); 1104 return; 1105 } 1106 } 1107 1108 CHECK("Received a chunk for a unknown track" == 0); 1109} 1110 1111void MPEG4Writer::writeFirstChunk(ChunkInfo* info) { 1112 LOGV("writeFirstChunk: %p", info->mTrack); 1113 1114 List<Chunk>::iterator chunkIt = info->mChunks.begin(); 1115 for (List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1116 it != chunkIt->mSamples.end(); ++it) { 1117 1118 off64_t offset = info->mTrack->isAvc() 1119 ? addLengthPrefixedSample_l(*it) 1120 : addSample_l(*it); 1121 if (it == chunkIt->mSamples.begin()) { 1122 info->mTrack->addChunkOffset(offset); 1123 } 1124 } 1125 1126 // Done with the current chunk. 1127 // Release all the samples in this chunk. 1128 while (!chunkIt->mSamples.empty()) { 1129 List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1130 (*it)->release(); 1131 (*it) = NULL; 1132 chunkIt->mSamples.erase(it); 1133 } 1134 chunkIt->mSamples.clear(); 1135 info->mChunks.erase(chunkIt); 1136} 1137 1138void MPEG4Writer::writeChunks() { 1139 LOGV("writeChunks"); 1140 size_t outstandingChunks = 0; 1141 while (!mChunkInfos.empty()) { 1142 List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1143 while (!it->mChunks.empty()) { 1144 CHECK_EQ(OK, writeOneChunk()); 1145 ++outstandingChunks; 1146 } 1147 it->mTrack = NULL; 1148 mChunkInfos.erase(it); 1149 } 1150 mChunkInfos.clear(); 1151 LOGD("%d chunks are written in the last batch", outstandingChunks); 1152} 1153 1154status_t MPEG4Writer::writeOneChunk() { 1155 LOGV("writeOneChunk"); 1156 1157 // Find the smallest timestamp, and write that chunk out 1158 // XXX: What if some track is just too slow? 1159 int64_t minTimestampUs = 0x7FFFFFFFFFFFFFFFLL; 1160 Track *track = NULL; 1161 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1162 it != mChunkInfos.end(); ++it) { 1163 if (!it->mChunks.empty()) { 1164 List<Chunk>::iterator chunkIt = it->mChunks.begin(); 1165 if (chunkIt->mTimeStampUs < minTimestampUs) { 1166 minTimestampUs = chunkIt->mTimeStampUs; 1167 track = it->mTrack; 1168 } 1169 } 1170 } 1171 1172 if (track == NULL) { 1173 LOGV("Nothing to be written after all"); 1174 return OK; 1175 } 1176 1177 if (mIsFirstChunk) { 1178 mIsFirstChunk = false; 1179 } 1180 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1181 it != mChunkInfos.end(); ++it) { 1182 if (it->mTrack == track) { 1183 writeFirstChunk(&(*it)); 1184 } 1185 } 1186 return OK; 1187} 1188 1189void MPEG4Writer::threadFunc() { 1190 LOGV("threadFunc"); 1191 1192 prctl(PR_SET_NAME, (unsigned long)"MPEG4Writer", 0, 0, 0); 1193 while (!mDone) { 1194 { 1195 Mutex::Autolock autolock(mLock); 1196 mChunkReadyCondition.wait(mLock); 1197 CHECK_EQ(writeOneChunk(), OK); 1198 } 1199 } 1200 1201 { 1202 // Write ALL samples 1203 Mutex::Autolock autolock(mLock); 1204 writeChunks(); 1205 } 1206} 1207 1208status_t MPEG4Writer::startWriterThread() { 1209 LOGV("startWriterThread"); 1210 1211 mDone = false; 1212 mIsFirstChunk = true; 1213 mDriftTimeUs = 0; 1214 for (List<Track *>::iterator it = mTracks.begin(); 1215 it != mTracks.end(); ++it) { 1216 ChunkInfo info; 1217 info.mTrack = *it; 1218 mChunkInfos.push_back(info); 1219 } 1220 1221 pthread_attr_t attr; 1222 pthread_attr_init(&attr); 1223 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1224 pthread_create(&mThread, &attr, ThreadWrapper, this); 1225 pthread_attr_destroy(&attr); 1226 return OK; 1227} 1228 1229 1230status_t MPEG4Writer::Track::start(MetaData *params) { 1231 if (!mDone && mPaused) { 1232 mPaused = false; 1233 mResumed = true; 1234 return OK; 1235 } 1236 1237 int64_t startTimeUs; 1238 if (params == NULL || !params->findInt64(kKeyTime, &startTimeUs)) { 1239 startTimeUs = 0; 1240 } 1241 1242 int32_t rotationDegrees; 1243 if (!mIsAudio && params && params->findInt32(kKeyRotation, &rotationDegrees)) { 1244 mRotation = rotationDegrees; 1245 } 1246 1247 mIsRealTimeRecording = true; 1248 { 1249 int32_t isNotRealTime; 1250 if (params && params->findInt32(kKeyNotRealTime, &isNotRealTime)) { 1251 mIsRealTimeRecording = (isNotRealTime == 0); 1252 } 1253 } 1254 1255 initTrackingProgressStatus(params); 1256 1257 sp<MetaData> meta = new MetaData; 1258 meta->setInt64(kKeyTime, startTimeUs); 1259 status_t err = mSource->start(meta.get()); 1260 if (err != OK) { 1261 mDone = mReachedEOS = true; 1262 return err; 1263 } 1264 1265 pthread_attr_t attr; 1266 pthread_attr_init(&attr); 1267 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1268 1269 mDone = false; 1270 mTrackDurationUs = 0; 1271 mReachedEOS = false; 1272 mEstimatedTrackSizeBytes = 0; 1273 mNumStcoTableEntries = 0; 1274 mNumStssTableEntries = 0; 1275 mNumStscTableEntries = 0; 1276 mNumSttsTableEntries = 0; 1277 mMdatSizeBytes = 0; 1278 mIsMediaTimeAdjustmentOn = false; 1279 mPrevMediaTimeAdjustTimestampUs = 0; 1280 mMediaTimeAdjustNumFrames = 0; 1281 mPrevMediaTimeAdjustSample = 0; 1282 mTotalDriftTimeToAdjustUs = 0; 1283 mPrevTotalAccumDriftTimeUs = 0; 1284 1285 pthread_create(&mThread, &attr, ThreadWrapper, this); 1286 pthread_attr_destroy(&attr); 1287 1288 return OK; 1289} 1290 1291status_t MPEG4Writer::Track::pause() { 1292 mPaused = true; 1293 return OK; 1294} 1295 1296status_t MPEG4Writer::Track::stop() { 1297 LOGD("Stopping %s track", mIsAudio? "Audio": "Video"); 1298 if (mDone) { 1299 return OK; 1300 } 1301 1302 mDone = true; 1303 1304 void *dummy; 1305 pthread_join(mThread, &dummy); 1306 1307 status_t err = (status_t) dummy; 1308 1309 LOGD("Stopping %s track source", mIsAudio? "Audio": "Video"); 1310 { 1311 status_t status = mSource->stop(); 1312 if (err == OK && status != OK && status != ERROR_END_OF_STREAM) { 1313 err = status; 1314 } 1315 } 1316 1317 LOGD("%s track stopped", mIsAudio? "Audio": "Video"); 1318 return err; 1319} 1320 1321bool MPEG4Writer::Track::reachedEOS() { 1322 return mReachedEOS; 1323} 1324 1325// static 1326void *MPEG4Writer::Track::ThreadWrapper(void *me) { 1327 Track *track = static_cast<Track *>(me); 1328 1329 status_t err = track->threadEntry(); 1330 return (void *) err; 1331} 1332 1333static void getNalUnitType(uint8_t byte, uint8_t* type) { 1334 LOGV("getNalUnitType: %d", byte); 1335 1336 // nal_unit_type: 5-bit unsigned integer 1337 *type = (byte & 0x1F); 1338} 1339 1340static const uint8_t *findNextStartCode( 1341 const uint8_t *data, size_t length) { 1342 1343 LOGV("findNextStartCode: %p %d", data, length); 1344 1345 size_t bytesLeft = length; 1346 while (bytesLeft > 4 && 1347 memcmp("\x00\x00\x00\x01", &data[length - bytesLeft], 4)) { 1348 --bytesLeft; 1349 } 1350 if (bytesLeft <= 4) { 1351 bytesLeft = 0; // Last parameter set 1352 } 1353 return &data[length - bytesLeft]; 1354} 1355 1356const uint8_t *MPEG4Writer::Track::parseParamSet( 1357 const uint8_t *data, size_t length, int type, size_t *paramSetLen) { 1358 1359 LOGV("parseParamSet"); 1360 CHECK(type == kNalUnitTypeSeqParamSet || 1361 type == kNalUnitTypePicParamSet); 1362 1363 const uint8_t *nextStartCode = findNextStartCode(data, length); 1364 *paramSetLen = nextStartCode - data; 1365 if (*paramSetLen == 0) { 1366 LOGE("Param set is malformed, since its length is 0"); 1367 return NULL; 1368 } 1369 1370 AVCParamSet paramSet(*paramSetLen, data); 1371 if (type == kNalUnitTypeSeqParamSet) { 1372 if (*paramSetLen < 4) { 1373 LOGE("Seq parameter set malformed"); 1374 return NULL; 1375 } 1376 if (mSeqParamSets.empty()) { 1377 mProfileIdc = data[1]; 1378 mProfileCompatible = data[2]; 1379 mLevelIdc = data[3]; 1380 } else { 1381 if (mProfileIdc != data[1] || 1382 mProfileCompatible != data[2] || 1383 mLevelIdc != data[3]) { 1384 LOGE("Inconsistent profile/level found in seq parameter sets"); 1385 return NULL; 1386 } 1387 } 1388 mSeqParamSets.push_back(paramSet); 1389 } else { 1390 mPicParamSets.push_back(paramSet); 1391 } 1392 return nextStartCode; 1393} 1394 1395status_t MPEG4Writer::Track::copyAVCCodecSpecificData( 1396 const uint8_t *data, size_t size) { 1397 LOGV("copyAVCCodecSpecificData"); 1398 1399 // 2 bytes for each of the parameter set length field 1400 // plus the 7 bytes for the header 1401 if (size < 4 + 7) { 1402 LOGE("Codec specific data length too short: %d", size); 1403 return ERROR_MALFORMED; 1404 } 1405 1406 mCodecSpecificDataSize = size; 1407 mCodecSpecificData = malloc(size); 1408 memcpy(mCodecSpecificData, data, size); 1409 return OK; 1410} 1411 1412status_t MPEG4Writer::Track::parseAVCCodecSpecificData( 1413 const uint8_t *data, size_t size) { 1414 1415 LOGV("parseAVCCodecSpecificData"); 1416 // Data starts with a start code. 1417 // SPS and PPS are separated with start codes. 1418 // Also, SPS must come before PPS 1419 uint8_t type = kNalUnitTypeSeqParamSet; 1420 bool gotSps = false; 1421 bool gotPps = false; 1422 const uint8_t *tmp = data; 1423 const uint8_t *nextStartCode = data; 1424 size_t bytesLeft = size; 1425 size_t paramSetLen = 0; 1426 mCodecSpecificDataSize = 0; 1427 while (bytesLeft > 4 && !memcmp("\x00\x00\x00\x01", tmp, 4)) { 1428 getNalUnitType(*(tmp + 4), &type); 1429 if (type == kNalUnitTypeSeqParamSet) { 1430 if (gotPps) { 1431 LOGE("SPS must come before PPS"); 1432 return ERROR_MALFORMED; 1433 } 1434 if (!gotSps) { 1435 gotSps = true; 1436 } 1437 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1438 } else if (type == kNalUnitTypePicParamSet) { 1439 if (!gotSps) { 1440 LOGE("SPS must come before PPS"); 1441 return ERROR_MALFORMED; 1442 } 1443 if (!gotPps) { 1444 gotPps = true; 1445 } 1446 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1447 } else { 1448 LOGE("Only SPS and PPS Nal units are expected"); 1449 return ERROR_MALFORMED; 1450 } 1451 1452 if (nextStartCode == NULL) { 1453 return ERROR_MALFORMED; 1454 } 1455 1456 // Move on to find the next parameter set 1457 bytesLeft -= nextStartCode - tmp; 1458 tmp = nextStartCode; 1459 mCodecSpecificDataSize += (2 + paramSetLen); 1460 } 1461 1462 { 1463 // Check on the number of seq parameter sets 1464 size_t nSeqParamSets = mSeqParamSets.size(); 1465 if (nSeqParamSets == 0) { 1466 LOGE("Cound not find sequence parameter set"); 1467 return ERROR_MALFORMED; 1468 } 1469 1470 if (nSeqParamSets > 0x1F) { 1471 LOGE("Too many seq parameter sets (%d) found", nSeqParamSets); 1472 return ERROR_MALFORMED; 1473 } 1474 } 1475 1476 { 1477 // Check on the number of pic parameter sets 1478 size_t nPicParamSets = mPicParamSets.size(); 1479 if (nPicParamSets == 0) { 1480 LOGE("Cound not find picture parameter set"); 1481 return ERROR_MALFORMED; 1482 } 1483 if (nPicParamSets > 0xFF) { 1484 LOGE("Too many pic parameter sets (%d) found", nPicParamSets); 1485 return ERROR_MALFORMED; 1486 } 1487 } 1488 1489 { 1490 // Check on the profiles 1491 // These profiles requires additional parameter set extensions 1492 if (mProfileIdc == 100 || mProfileIdc == 110 || 1493 mProfileIdc == 122 || mProfileIdc == 144) { 1494 LOGE("Sorry, no support for profile_idc: %d!", mProfileIdc); 1495 return BAD_VALUE; 1496 } 1497 } 1498 1499 return OK; 1500} 1501 1502status_t MPEG4Writer::Track::makeAVCCodecSpecificData( 1503 const uint8_t *data, size_t size) { 1504 1505 if (mCodecSpecificData != NULL) { 1506 LOGE("Already have codec specific data"); 1507 return ERROR_MALFORMED; 1508 } 1509 1510 if (size < 4) { 1511 LOGE("Codec specific data length too short: %d", size); 1512 return ERROR_MALFORMED; 1513 } 1514 1515 // Data is in the form of AVCCodecSpecificData 1516 if (memcmp("\x00\x00\x00\x01", data, 4)) { 1517 return copyAVCCodecSpecificData(data, size); 1518 } 1519 1520 if (parseAVCCodecSpecificData(data, size) != OK) { 1521 return ERROR_MALFORMED; 1522 } 1523 1524 // ISO 14496-15: AVC file format 1525 mCodecSpecificDataSize += 7; // 7 more bytes in the header 1526 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1527 uint8_t *header = (uint8_t *)mCodecSpecificData; 1528 header[0] = 1; // version 1529 header[1] = mProfileIdc; // profile indication 1530 header[2] = mProfileCompatible; // profile compatibility 1531 header[3] = mLevelIdc; 1532 1533 // 6-bit '111111' followed by 2-bit to lengthSizeMinuusOne 1534 if (mOwner->useNalLengthFour()) { 1535 header[4] = 0xfc | 3; // length size == 4 bytes 1536 } else { 1537 header[4] = 0xfc | 1; // length size == 2 bytes 1538 } 1539 1540 // 3-bit '111' followed by 5-bit numSequenceParameterSets 1541 int nSequenceParamSets = mSeqParamSets.size(); 1542 header[5] = 0xe0 | nSequenceParamSets; 1543 header += 6; 1544 for (List<AVCParamSet>::iterator it = mSeqParamSets.begin(); 1545 it != mSeqParamSets.end(); ++it) { 1546 // 16-bit sequence parameter set length 1547 uint16_t seqParamSetLength = it->mLength; 1548 header[0] = seqParamSetLength >> 8; 1549 header[1] = seqParamSetLength & 0xff; 1550 1551 // SPS NAL unit (sequence parameter length bytes) 1552 memcpy(&header[2], it->mData, seqParamSetLength); 1553 header += (2 + seqParamSetLength); 1554 } 1555 1556 // 8-bit nPictureParameterSets 1557 int nPictureParamSets = mPicParamSets.size(); 1558 header[0] = nPictureParamSets; 1559 header += 1; 1560 for (List<AVCParamSet>::iterator it = mPicParamSets.begin(); 1561 it != mPicParamSets.end(); ++it) { 1562 // 16-bit picture parameter set length 1563 uint16_t picParamSetLength = it->mLength; 1564 header[0] = picParamSetLength >> 8; 1565 header[1] = picParamSetLength & 0xff; 1566 1567 // PPS Nal unit (picture parameter set length bytes) 1568 memcpy(&header[2], it->mData, picParamSetLength); 1569 header += (2 + picParamSetLength); 1570 } 1571 1572 return OK; 1573} 1574 1575/* 1576* The video track's media time adjustment for real-time applications 1577* is described as follows: 1578* 1579* First, the media time adjustment is done for every period of 1580* kVideoMediaTimeAdjustPeriodTimeUs. kVideoMediaTimeAdjustPeriodTimeUs 1581* is currently a fixed value chosen heuristically. The value of 1582* kVideoMediaTimeAdjustPeriodTimeUs should not be very large or very small 1583* for two considerations: on one hand, a relatively large value 1584* helps reduce large fluctuation of drift time in the audio encoding 1585* path; while on the other hand, a relatively small value helps keep 1586* restoring synchronization in audio/video more frequently. Note for the 1587* very first period of kVideoMediaTimeAdjustPeriodTimeUs, there is 1588* no media time adjustment for the video track. 1589* 1590* Second, the total accumulated audio track time drift found 1591* in a period of kVideoMediaTimeAdjustPeriodTimeUs is distributed 1592* over a stream of incoming video frames. The number of video frames 1593* affected is determined based on the number of recorded video frames 1594* within the past kVideoMediaTimeAdjustPeriodTimeUs period. 1595* We choose to distribute the drift time over only a portion 1596* (rather than all) of the total number of recorded video frames 1597* in order to make sure that the video track media time adjustment is 1598* completed for the current period before the next video track media 1599* time adjustment period starts. Currently, the portion chosen is a 1600* half (0.5). 1601* 1602* Last, various additional checks are performed to ensure that 1603* the actual audio encoding path does not have too much drift. 1604* In particular, 1) we want to limit the average incremental time 1605* adjustment for each video frame to be less than a threshold 1606* for a single period of kVideoMediaTimeAdjustPeriodTimeUs. 1607* Currently, the threshold is set to 5 ms. If the average incremental 1608* media time adjustment for a video frame is larger than the 1609* threshold, the audio encoding path has too much time drift. 1610* 2) We also want to limit the total time drift in the audio 1611* encoding path to be less than a threshold for a period of 1612* kVideoMediaTimeAdjustPeriodTimeUs. Currently, the threshold 1613* is 0.5% of kVideoMediaTimeAdjustPeriodTimeUs. If the time drift of 1614* the audio encoding path is larger than the threshold, the audio 1615* encoding path has too much time drift. We treat the large time 1616* drift of the audio encoding path as errors, since there is no 1617* way to keep audio/video in synchronization for real-time 1618* applications if the time drift is too large unless we drop some 1619* video frames, which has its own problems that we don't want 1620* to get into for the time being. 1621*/ 1622void MPEG4Writer::Track::adjustMediaTime(int64_t *timestampUs) { 1623 if (*timestampUs - mPrevMediaTimeAdjustTimestampUs >= 1624 kVideoMediaTimeAdjustPeriodTimeUs) { 1625 1626 LOGV("New media time adjustment period at %lld us", *timestampUs); 1627 mIsMediaTimeAdjustmentOn = true; 1628 mMediaTimeAdjustNumFrames = 1629 (mNumSamples - mPrevMediaTimeAdjustSample) >> 1; 1630 1631 mPrevMediaTimeAdjustTimestampUs = *timestampUs; 1632 mPrevMediaTimeAdjustSample = mNumSamples; 1633 int64_t totalAccumDriftTimeUs = mOwner->getDriftTimeUs(); 1634 mTotalDriftTimeToAdjustUs = 1635 totalAccumDriftTimeUs - mPrevTotalAccumDriftTimeUs; 1636 1637 mPrevTotalAccumDriftTimeUs = totalAccumDriftTimeUs; 1638 1639 // Check on incremental adjusted time per frame 1640 int64_t adjustTimePerFrameUs = 1641 mTotalDriftTimeToAdjustUs / mMediaTimeAdjustNumFrames; 1642 1643 if (adjustTimePerFrameUs < 0) { 1644 adjustTimePerFrameUs = -adjustTimePerFrameUs; 1645 } 1646 if (adjustTimePerFrameUs >= 5000) { 1647 LOGE("Adjusted time per video frame is %lld us", 1648 adjustTimePerFrameUs); 1649 CHECK(!"Video frame time adjustment is too large!"); 1650 } 1651 1652 // Check on total accumulated time drift within a period of 1653 // kVideoMediaTimeAdjustPeriodTimeUs. 1654 int64_t driftPercentage = (mTotalDriftTimeToAdjustUs * 1000) 1655 / kVideoMediaTimeAdjustPeriodTimeUs; 1656 1657 if (driftPercentage < 0) { 1658 driftPercentage = -driftPercentage; 1659 } 1660 if (driftPercentage > 5) { 1661 LOGE("Audio track has time drift %lld us over %lld us", 1662 mTotalDriftTimeToAdjustUs, 1663 kVideoMediaTimeAdjustPeriodTimeUs); 1664 1665 CHECK(!"The audio track media time drifts too much!"); 1666 } 1667 1668 } 1669 1670 if (mIsMediaTimeAdjustmentOn) { 1671 if (mNumSamples - mPrevMediaTimeAdjustSample <= 1672 mMediaTimeAdjustNumFrames) { 1673 1674 // Do media time incremental adjustment 1675 int64_t incrementalAdjustTimeUs = 1676 (mTotalDriftTimeToAdjustUs * 1677 (mNumSamples - mPrevMediaTimeAdjustSample)) 1678 / mMediaTimeAdjustNumFrames; 1679 1680 *timestampUs += 1681 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs); 1682 1683 LOGV("Incremental video frame media time adjustment: %lld us", 1684 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs)); 1685 } else { 1686 // Within the remaining adjustment period, 1687 // no incremental adjustment is needed. 1688 *timestampUs += 1689 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs); 1690 1691 LOGV("Fixed video frame media time adjustment: %lld us", 1692 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs)); 1693 } 1694 } 1695} 1696 1697/* 1698 * Updates the drift time from the audio track so that 1699 * the video track can get the updated drift time information 1700 * from the file writer. The fluctuation of the drift time of the audio 1701 * encoding path is smoothed out with a simple filter by giving a larger 1702 * weight to more recently drift time. The filter coefficients, 0.5 and 0.5, 1703 * are heuristically determined. 1704 */ 1705void MPEG4Writer::Track::updateDriftTime(const sp<MetaData>& meta) { 1706 int64_t driftTimeUs = 0; 1707 if (meta->findInt64(kKeyDriftTime, &driftTimeUs)) { 1708 int64_t prevDriftTimeUs = mOwner->getDriftTimeUs(); 1709 int64_t timeUs = (driftTimeUs + prevDriftTimeUs) >> 1; 1710 mOwner->setDriftTimeUs(timeUs); 1711 } 1712} 1713 1714status_t MPEG4Writer::Track::threadEntry() { 1715 int32_t count = 0; 1716 const int64_t interleaveDurationUs = mOwner->interleaveDuration(); 1717 int64_t chunkTimestampUs = 0; 1718 int32_t nChunks = 0; 1719 int32_t nZeroLengthFrames = 0; 1720 int64_t lastTimestampUs = 0; // Previous sample time stamp in ms 1721 int64_t lastDurationUs = 0; // Between the previous two samples in ms 1722 int64_t currDurationTicks = 0; // Timescale based ticks 1723 int64_t lastDurationTicks = 0; // Timescale based ticks 1724 int32_t sampleCount = 1; // Sample count in the current stts table entry 1725 uint32_t previousSampleSize = 0; // Size of the previous sample 1726 int64_t previousPausedDurationUs = 0; 1727 int64_t timestampUs; 1728 1729 if (mIsAudio) { 1730 prctl(PR_SET_NAME, (unsigned long)"AudioTrackEncoding", 0, 0, 0); 1731 } else { 1732 prctl(PR_SET_NAME, (unsigned long)"VideoTrackEncoding", 0, 0, 0); 1733 } 1734 sp<MetaData> meta_data; 1735 1736 mNumSamples = 0; 1737 status_t err = OK; 1738 MediaBuffer *buffer; 1739 while (!mDone && (err = mSource->read(&buffer)) == OK) { 1740 if (buffer->range_length() == 0) { 1741 buffer->release(); 1742 buffer = NULL; 1743 ++nZeroLengthFrames; 1744 continue; 1745 } 1746 1747 // If the codec specific data has not been received yet, delay pause. 1748 // After the codec specific data is received, discard what we received 1749 // when the track is to be paused. 1750 if (mPaused && !mResumed) { 1751 buffer->release(); 1752 buffer = NULL; 1753 continue; 1754 } 1755 1756 ++count; 1757 1758 int32_t isCodecConfig; 1759 if (buffer->meta_data()->findInt32(kKeyIsCodecConfig, &isCodecConfig) 1760 && isCodecConfig) { 1761 CHECK(!mGotAllCodecSpecificData); 1762 1763 if (mIsAvc) { 1764 status_t err = makeAVCCodecSpecificData( 1765 (const uint8_t *)buffer->data() 1766 + buffer->range_offset(), 1767 buffer->range_length()); 1768 CHECK_EQ(OK, err); 1769 } else if (mIsMPEG4) { 1770 mCodecSpecificDataSize = buffer->range_length(); 1771 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1772 memcpy(mCodecSpecificData, 1773 (const uint8_t *)buffer->data() 1774 + buffer->range_offset(), 1775 buffer->range_length()); 1776 } 1777 1778 buffer->release(); 1779 buffer = NULL; 1780 1781 mGotAllCodecSpecificData = true; 1782 continue; 1783 } 1784 1785 // Make a deep copy of the MediaBuffer and Metadata and release 1786 // the original as soon as we can 1787 MediaBuffer *copy = new MediaBuffer(buffer->range_length()); 1788 memcpy(copy->data(), (uint8_t *)buffer->data() + buffer->range_offset(), 1789 buffer->range_length()); 1790 copy->set_range(0, buffer->range_length()); 1791 meta_data = new MetaData(*buffer->meta_data().get()); 1792 buffer->release(); 1793 buffer = NULL; 1794 1795 if (mIsAvc) StripStartcode(copy); 1796 1797 size_t sampleSize = copy->range_length(); 1798 if (mIsAvc) { 1799 if (mOwner->useNalLengthFour()) { 1800 sampleSize += 4; 1801 } else { 1802 sampleSize += 2; 1803 } 1804 } 1805 1806 // Max file size or duration handling 1807 mMdatSizeBytes += sampleSize; 1808 updateTrackSizeEstimate(); 1809 1810 if (mOwner->exceedsFileSizeLimit()) { 1811 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_FILESIZE_REACHED, 0); 1812 break; 1813 } 1814 if (mOwner->exceedsFileDurationLimit()) { 1815 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_DURATION_REACHED, 0); 1816 break; 1817 } 1818 1819 1820 int32_t isSync = false; 1821 meta_data->findInt32(kKeyIsSyncFrame, &isSync); 1822 1823 /* 1824 * The original timestamp found in the data buffer will be modified as below: 1825 * 1826 * There is a playback offset into this track if the track's start time 1827 * is not the same as the movie start time, which will be recorded in edst 1828 * box of the output file. The playback offset is to make sure that the 1829 * starting time of the audio/video tracks are synchronized. Although the 1830 * track's media timestamp may be subject to various modifications 1831 * as outlined below, the track's playback offset time remains unchanged 1832 * once the first data buffer of the track is received. 1833 * 1834 * The media time stamp will be calculated by subtracting the playback offset 1835 * (and potential pause durations) from the original timestamp in the buffer. 1836 * 1837 * If this track is a video track for a real-time recording application with 1838 * both audio and video tracks, its media timestamp will subject to further 1839 * modification based on the media clock of the audio track. This modification 1840 * is needed for the purpose of maintaining good audio/video synchronization. 1841 * 1842 * If the recording session is paused and resumed multiple times, the track 1843 * media timestamp will be modified as if the recording session had never been 1844 * paused at all during playback of the recorded output file. In other words, 1845 * the output file will have no memory of pause/resume durations. 1846 * 1847 */ 1848 CHECK(meta_data->findInt64(kKeyTime, ×tampUs)); 1849 LOGV("%s timestampUs: %lld", mIsAudio? "Audio": "Video", timestampUs); 1850 1851//////////////////////////////////////////////////////////////////////////////// 1852 if (mSampleSizes.empty()) { 1853 mStartTimestampUs = timestampUs; 1854 mOwner->setStartTimestampUs(mStartTimestampUs); 1855 previousPausedDurationUs = mStartTimestampUs; 1856 } 1857 1858 if (mResumed) { 1859 int64_t durExcludingEarlierPausesUs = timestampUs - previousPausedDurationUs; 1860 CHECK(durExcludingEarlierPausesUs >= 0); 1861 int64_t pausedDurationUs = durExcludingEarlierPausesUs - mTrackDurationUs; 1862 CHECK(pausedDurationUs >= lastDurationUs); 1863 previousPausedDurationUs += pausedDurationUs - lastDurationUs; 1864 mResumed = false; 1865 } 1866 1867 timestampUs -= previousPausedDurationUs; 1868 CHECK(timestampUs >= 0); 1869 1870 // Media time adjustment for real-time applications 1871 if (mIsRealTimeRecording) { 1872 if (mIsAudio) { 1873 updateDriftTime(meta_data); 1874 } else { 1875 adjustMediaTime(×tampUs); 1876 } 1877 } 1878 1879 CHECK(timestampUs >= 0); 1880 if (mNumSamples > 1) { 1881 if (timestampUs <= lastTimestampUs) { 1882 LOGW("Frame arrives too late!"); 1883 // Don't drop the late frame, since dropping a frame may cause 1884 // problems later during playback 1885 1886 // The idea here is to avoid having two or more samples with the 1887 // same timestamp in the output file. 1888 if (mTimeScale >= 1000000LL) { 1889 timestampUs = lastTimestampUs + 1; 1890 } else { 1891 timestampUs = lastTimestampUs + (1000000LL + (mTimeScale >> 1)) / mTimeScale; 1892 } 1893 } 1894 } 1895 1896 LOGV("%s media time stamp: %lld and previous paused duration %lld", 1897 mIsAudio? "Audio": "Video", timestampUs, previousPausedDurationUs); 1898 if (timestampUs > mTrackDurationUs) { 1899 mTrackDurationUs = timestampUs; 1900 } 1901 1902 mSampleSizes.push_back(sampleSize); 1903 ++mNumSamples; 1904 if (mNumSamples > 2) { 1905 // We need to use the time scale based ticks, rather than the 1906 // timestamp itself to determine whether we have to use a new 1907 // stts entry, since we may have rounding errors. 1908 // The calculation is intended to reduce the accumulated 1909 // rounding errors. 1910 currDurationTicks = 1911 ((timestampUs * mTimeScale + 500000LL) / 1000000LL - 1912 (lastTimestampUs * mTimeScale + 500000LL) / 1000000LL); 1913 1914 if (currDurationTicks != lastDurationTicks) { 1915 addOneSttsTableEntry(sampleCount, lastDurationUs); 1916 sampleCount = 1; 1917 } else { 1918 ++sampleCount; 1919 } 1920 } 1921 if (mSamplesHaveSameSize) { 1922 if (mNumSamples >= 2 && previousSampleSize != sampleSize) { 1923 mSamplesHaveSameSize = false; 1924 } 1925 previousSampleSize = sampleSize; 1926 } 1927 lastDurationUs = timestampUs - lastTimestampUs; 1928 lastDurationTicks = currDurationTicks; 1929 lastTimestampUs = timestampUs; 1930 1931 if (isSync != 0) { 1932 addOneStssTableEntry(mNumSamples); 1933 } 1934 1935 if (mTrackingProgressStatus) { 1936 if (mPreviousTrackTimeUs <= 0) { 1937 mPreviousTrackTimeUs = mStartTimestampUs; 1938 } 1939 trackProgressStatus(timestampUs); 1940 } 1941 if (mOwner->numTracks() == 1) { 1942 off64_t offset = mIsAvc? mOwner->addLengthPrefixedSample_l(copy) 1943 : mOwner->addSample_l(copy); 1944 if (mChunkOffsets.empty()) { 1945 addChunkOffset(offset); 1946 } 1947 copy->release(); 1948 copy = NULL; 1949 continue; 1950 } 1951 1952 mChunkSamples.push_back(copy); 1953 if (interleaveDurationUs == 0) { 1954 addOneStscTableEntry(++nChunks, 1); 1955 bufferChunk(timestampUs); 1956 } else { 1957 if (chunkTimestampUs == 0) { 1958 chunkTimestampUs = timestampUs; 1959 } else { 1960 if (timestampUs - chunkTimestampUs > interleaveDurationUs) { 1961 ++nChunks; 1962 if (nChunks == 1 || // First chunk 1963 (--(mStscTableEntries.end()))->samplesPerChunk != 1964 mChunkSamples.size()) { 1965 addOneStscTableEntry(nChunks, mChunkSamples.size()); 1966 } 1967 bufferChunk(timestampUs); 1968 chunkTimestampUs = timestampUs; 1969 } 1970 } 1971 } 1972 1973 } 1974 1975 if (mSampleSizes.empty() || // no samples written 1976 (!mIsAudio && mNumStssTableEntries == 0) || // no sync frames for video 1977 (OK != checkCodecSpecificData())) { // no codec specific data 1978 err = ERROR_MALFORMED; 1979 } 1980 mOwner->trackProgressStatus(this, -1, err); 1981 1982 // Last chunk 1983 if (mOwner->numTracks() == 1) { 1984 addOneStscTableEntry(1, mNumSamples); 1985 } else if (!mChunkSamples.empty()) { 1986 addOneStscTableEntry(++nChunks, mChunkSamples.size()); 1987 bufferChunk(timestampUs); 1988 } 1989 1990 // We don't really know how long the last frame lasts, since 1991 // there is no frame time after it, just repeat the previous 1992 // frame's duration. 1993 if (mNumSamples == 1) { 1994 lastDurationUs = 0; // A single sample's duration 1995 } else { 1996 ++sampleCount; // Count for the last sample 1997 } 1998 addOneSttsTableEntry(sampleCount, lastDurationUs); 1999 mTrackDurationUs += lastDurationUs; 2000 mReachedEOS = true; 2001 LOGI("Received total/0-length (%d/%d) buffers and encoded %d frames. - %s", 2002 count, nZeroLengthFrames, mNumSamples, mIsAudio? "audio": "video"); 2003 if (mIsAudio) { 2004 LOGI("Audio track drift time: %lld us", mOwner->getDriftTimeUs()); 2005 } 2006 2007 if (err == ERROR_END_OF_STREAM) { 2008 return OK; 2009 } 2010 return err; 2011} 2012 2013void MPEG4Writer::Track::trackProgressStatus(int64_t timeUs, status_t err) { 2014 LOGV("trackProgressStatus: %lld us", timeUs); 2015 if (mTrackEveryTimeDurationUs > 0 && 2016 timeUs - mPreviousTrackTimeUs >= mTrackEveryTimeDurationUs) { 2017 LOGV("Fire time tracking progress status at %lld us", timeUs); 2018 mOwner->trackProgressStatus(this, timeUs - mPreviousTrackTimeUs, err); 2019 mPreviousTrackTimeUs = timeUs; 2020 } 2021} 2022 2023void MPEG4Writer::trackProgressStatus( 2024 const MPEG4Writer::Track* track, int64_t timeUs, status_t err) { 2025 Mutex::Autolock lock(mLock); 2026 int32_t nTracks = mTracks.size(); 2027 CHECK(nTracks >= 1); 2028 CHECK(nTracks < 64); // Arbitrary number 2029 2030 int32_t trackNum = 0; 2031 CHECK(trackNum < nTracks); 2032 trackNum <<= 16; 2033 2034 // Error notification 2035 // Do not consider ERROR_END_OF_STREAM an error 2036 if (err != OK && err != ERROR_END_OF_STREAM) { 2037 notify(MEDIA_RECORDER_EVENT_ERROR, 2038 trackNum | MEDIA_RECORDER_ERROR_UNKNOWN, 2039 err); 2040 return; 2041 } 2042 2043 if (timeUs == -1) { 2044 // Send completion notification 2045 notify(MEDIA_RECORDER_EVENT_INFO, 2046 trackNum | MEDIA_RECORDER_INFO_COMPLETION_STATUS, 2047 err); 2048 return; 2049 } else { 2050 // Send progress status 2051 notify(MEDIA_RECORDER_EVENT_INFO, 2052 trackNum | MEDIA_RECORDER_INFO_PROGRESS_TIME_STATUS, 2053 timeUs / 1000); 2054 } 2055} 2056 2057void MPEG4Writer::setDriftTimeUs(int64_t driftTimeUs) { 2058 LOGV("setDriftTimeUs: %lld us", driftTimeUs); 2059 Mutex::Autolock autolock(mLock); 2060 mDriftTimeUs = driftTimeUs; 2061} 2062 2063int64_t MPEG4Writer::getDriftTimeUs() { 2064 LOGV("getDriftTimeUs: %lld us", mDriftTimeUs); 2065 Mutex::Autolock autolock(mLock); 2066 return mDriftTimeUs; 2067} 2068 2069bool MPEG4Writer::useNalLengthFour() { 2070 return mUse4ByteNalLength; 2071} 2072 2073void MPEG4Writer::Track::bufferChunk(int64_t timestampUs) { 2074 LOGV("bufferChunk"); 2075 2076 Chunk chunk(this, timestampUs, mChunkSamples); 2077 mOwner->bufferChunk(chunk); 2078 mChunkSamples.clear(); 2079} 2080 2081int64_t MPEG4Writer::Track::getDurationUs() const { 2082 return mTrackDurationUs; 2083} 2084 2085int64_t MPEG4Writer::Track::getEstimatedTrackSizeBytes() const { 2086 return mEstimatedTrackSizeBytes; 2087} 2088 2089status_t MPEG4Writer::Track::checkCodecSpecificData() const { 2090 const char *mime; 2091 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 2092 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime) || 2093 !strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime) || 2094 !strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2095 if (!mCodecSpecificData || 2096 mCodecSpecificDataSize <= 0) { 2097 LOGE("Missing codec specific data"); 2098 return ERROR_MALFORMED; 2099 } 2100 } else { 2101 if (mCodecSpecificData || 2102 mCodecSpecificDataSize > 0) { 2103 LOGE("Unexepected codec specific data found"); 2104 return ERROR_MALFORMED; 2105 } 2106 } 2107 return OK; 2108} 2109 2110void MPEG4Writer::Track::writeTrackHeader( 2111 int32_t trackID, bool use32BitOffset) { 2112 const char *mime; 2113 bool success = mMeta->findCString(kKeyMIMEType, &mime); 2114 CHECK(success); 2115 2116 LOGV("%s track time scale: %d", 2117 mIsAudio? "Audio": "Video", mTimeScale); 2118 2119 time_t now = time(NULL); 2120 int32_t mvhdTimeScale = mOwner->getTimeScale(); 2121 int64_t trakDurationUs = getDurationUs(); 2122 2123 mOwner->beginBox("trak"); 2124 2125 mOwner->beginBox("tkhd"); 2126 // Flags = 7 to indicate that the track is enabled, and 2127 // part of the presentation 2128 mOwner->writeInt32(0x07); // version=0, flags=7 2129 mOwner->writeInt32(now); // creation time 2130 mOwner->writeInt32(now); // modification time 2131 mOwner->writeInt32(trackID); 2132 mOwner->writeInt32(0); // reserved 2133 int32_t tkhdDuration = 2134 (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2135 mOwner->writeInt32(tkhdDuration); // in mvhd timescale 2136 mOwner->writeInt32(0); // reserved 2137 mOwner->writeInt32(0); // reserved 2138 mOwner->writeInt16(0); // layer 2139 mOwner->writeInt16(0); // alternate group 2140 mOwner->writeInt16(mIsAudio ? 0x100 : 0); // volume 2141 mOwner->writeInt16(0); // reserved 2142 2143 mOwner->writeCompositionMatrix(mRotation); // matrix 2144 2145 if (mIsAudio) { 2146 mOwner->writeInt32(0); 2147 mOwner->writeInt32(0); 2148 } else { 2149 int32_t width, height; 2150 bool success = mMeta->findInt32(kKeyWidth, &width); 2151 success = success && mMeta->findInt32(kKeyHeight, &height); 2152 CHECK(success); 2153 2154 mOwner->writeInt32(width << 16); // 32-bit fixed-point value 2155 mOwner->writeInt32(height << 16); // 32-bit fixed-point value 2156 } 2157 mOwner->endBox(); // tkhd 2158 2159 int64_t moovStartTimeUs = mOwner->getStartTimestampUs(); 2160 if (mStartTimestampUs != moovStartTimeUs) { 2161 mOwner->beginBox("edts"); 2162 mOwner->beginBox("elst"); 2163 mOwner->writeInt32(0); // version=0, flags=0: 32-bit time 2164 mOwner->writeInt32(2); // never ends with an empty list 2165 2166 // First elst entry: specify the starting time offset 2167 int64_t offsetUs = mStartTimestampUs - moovStartTimeUs; 2168 LOGV("OffsetUs: %lld", offsetUs); 2169 int32_t seg = (offsetUs * mvhdTimeScale + 5E5) / 1E6; 2170 mOwner->writeInt32(seg); // in mvhd timecale 2171 mOwner->writeInt32(-1); // starting time offset 2172 mOwner->writeInt32(1 << 16); // rate = 1.0 2173 2174 // Second elst entry: specify the track duration 2175 seg = (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2176 mOwner->writeInt32(seg); // in mvhd timescale 2177 mOwner->writeInt32(0); 2178 mOwner->writeInt32(1 << 16); 2179 mOwner->endBox(); 2180 mOwner->endBox(); 2181 } 2182 2183 mOwner->beginBox("mdia"); 2184 2185 mOwner->beginBox("mdhd"); 2186 mOwner->writeInt32(0); // version=0, flags=0 2187 mOwner->writeInt32(now); // creation time 2188 mOwner->writeInt32(now); // modification time 2189 mOwner->writeInt32(mTimeScale); // media timescale 2190 int32_t mdhdDuration = (trakDurationUs * mTimeScale + 5E5) / 1E6; 2191 mOwner->writeInt32(mdhdDuration); // use media timescale 2192 // Language follows the three letter standard ISO-639-2/T 2193 // 'e', 'n', 'g' for "English", for instance. 2194 // Each character is packed as the difference between its ASCII value and 0x60. 2195 // For "English", these are 00101, 01110, 00111. 2196 // XXX: Where is the padding bit located: 0x15C7? 2197 mOwner->writeInt16(0); // language code 2198 mOwner->writeInt16(0); // predefined 2199 mOwner->endBox(); 2200 2201 mOwner->beginBox("hdlr"); 2202 mOwner->writeInt32(0); // version=0, flags=0 2203 mOwner->writeInt32(0); // component type: should be mhlr 2204 mOwner->writeFourcc(mIsAudio ? "soun" : "vide"); // component subtype 2205 mOwner->writeInt32(0); // reserved 2206 mOwner->writeInt32(0); // reserved 2207 mOwner->writeInt32(0); // reserved 2208 // Removing "r" for the name string just makes the string 4 byte aligned 2209 mOwner->writeCString(mIsAudio ? "SoundHandle": "VideoHandle"); // name 2210 mOwner->endBox(); 2211 2212 mOwner->beginBox("minf"); 2213 if (mIsAudio) { 2214 mOwner->beginBox("smhd"); 2215 mOwner->writeInt32(0); // version=0, flags=0 2216 mOwner->writeInt16(0); // balance 2217 mOwner->writeInt16(0); // reserved 2218 mOwner->endBox(); 2219 } else { 2220 mOwner->beginBox("vmhd"); 2221 mOwner->writeInt32(0x01); // version=0, flags=1 2222 mOwner->writeInt16(0); // graphics mode 2223 mOwner->writeInt16(0); // opcolor 2224 mOwner->writeInt16(0); 2225 mOwner->writeInt16(0); 2226 mOwner->endBox(); 2227 } 2228 2229 mOwner->beginBox("dinf"); 2230 mOwner->beginBox("dref"); 2231 mOwner->writeInt32(0); // version=0, flags=0 2232 mOwner->writeInt32(1); // entry count (either url or urn) 2233 // The table index here refers to the sample description index 2234 // in the sample table entries. 2235 mOwner->beginBox("url "); 2236 mOwner->writeInt32(1); // version=0, flags=1 (self-contained) 2237 mOwner->endBox(); // url 2238 mOwner->endBox(); // dref 2239 mOwner->endBox(); // dinf 2240 2241 mOwner->beginBox("stbl"); 2242 2243 mOwner->beginBox("stsd"); 2244 mOwner->writeInt32(0); // version=0, flags=0 2245 mOwner->writeInt32(1); // entry count 2246 if (mIsAudio) { 2247 const char *fourcc = NULL; 2248 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime)) { 2249 fourcc = "samr"; 2250 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2251 fourcc = "sawb"; 2252 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2253 fourcc = "mp4a"; 2254 } else { 2255 LOGE("Unknown mime type '%s'.", mime); 2256 CHECK(!"should not be here, unknown mime type."); 2257 } 2258 2259 mOwner->beginBox(fourcc); // audio format 2260 mOwner->writeInt32(0); // reserved 2261 mOwner->writeInt16(0); // reserved 2262 mOwner->writeInt16(0x1); // data ref index 2263 mOwner->writeInt32(0); // reserved 2264 mOwner->writeInt32(0); // reserved 2265 int32_t nChannels; 2266 CHECK_EQ(true, mMeta->findInt32(kKeyChannelCount, &nChannels)); 2267 mOwner->writeInt16(nChannels); // channel count 2268 mOwner->writeInt16(16); // sample size 2269 mOwner->writeInt16(0); // predefined 2270 mOwner->writeInt16(0); // reserved 2271 2272 int32_t samplerate; 2273 bool success = mMeta->findInt32(kKeySampleRate, &samplerate); 2274 CHECK(success); 2275 mOwner->writeInt32(samplerate << 16); 2276 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2277 mOwner->beginBox("esds"); 2278 CHECK(mCodecSpecificData); 2279 CHECK(mCodecSpecificDataSize > 0); 2280 2281 // Make sure all sizes encode to a single byte. 2282 CHECK(mCodecSpecificDataSize + 23 < 128); 2283 2284 mOwner->writeInt32(0); // version=0, flags=0 2285 mOwner->writeInt8(0x03); // ES_DescrTag 2286 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2287 mOwner->writeInt16(0x0000);// ES_ID 2288 mOwner->writeInt8(0x00); 2289 2290 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2291 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2292 mOwner->writeInt8(0x40); // objectTypeIndication ISO/IEC 14492-2 2293 mOwner->writeInt8(0x15); // streamType AudioStream 2294 2295 mOwner->writeInt16(0x03); // XXX 2296 mOwner->writeInt8(0x00); // buffer size 24-bit 2297 mOwner->writeInt32(96000); // max bit rate 2298 mOwner->writeInt32(96000); // avg bit rate 2299 2300 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2301 mOwner->writeInt8(mCodecSpecificDataSize); 2302 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2303 2304 static const uint8_t kData2[] = { 2305 0x06, // SLConfigDescriptorTag 2306 0x01, 2307 0x02 2308 }; 2309 mOwner->write(kData2, sizeof(kData2)); 2310 2311 mOwner->endBox(); // esds 2312 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime) || 2313 !strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2314 // 3gpp2 Spec AMRSampleEntry fields 2315 mOwner->beginBox("damr"); 2316 mOwner->writeCString(" "); // vendor: 4 bytes 2317 mOwner->writeInt8(0); // decoder version 2318 mOwner->writeInt16(0x83FF); // mode set: all enabled 2319 mOwner->writeInt8(0); // mode change period 2320 mOwner->writeInt8(1); // frames per sample 2321 mOwner->endBox(); 2322 } 2323 mOwner->endBox(); 2324 } else { 2325 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2326 mOwner->beginBox("mp4v"); 2327 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2328 mOwner->beginBox("s263"); 2329 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2330 mOwner->beginBox("avc1"); 2331 } else { 2332 LOGE("Unknown mime type '%s'.", mime); 2333 CHECK(!"should not be here, unknown mime type."); 2334 } 2335 2336 mOwner->writeInt32(0); // reserved 2337 mOwner->writeInt16(0); // reserved 2338 mOwner->writeInt16(1); // data ref index 2339 mOwner->writeInt16(0); // predefined 2340 mOwner->writeInt16(0); // reserved 2341 mOwner->writeInt32(0); // predefined 2342 mOwner->writeInt32(0); // predefined 2343 mOwner->writeInt32(0); // predefined 2344 2345 int32_t width, height; 2346 bool success = mMeta->findInt32(kKeyWidth, &width); 2347 success = success && mMeta->findInt32(kKeyHeight, &height); 2348 CHECK(success); 2349 2350 mOwner->writeInt16(width); 2351 mOwner->writeInt16(height); 2352 mOwner->writeInt32(0x480000); // horiz resolution 2353 mOwner->writeInt32(0x480000); // vert resolution 2354 mOwner->writeInt32(0); // reserved 2355 mOwner->writeInt16(1); // frame count 2356 mOwner->write(" ", 32); 2357 mOwner->writeInt16(0x18); // depth 2358 mOwner->writeInt16(-1); // predefined 2359 2360 CHECK(23 + mCodecSpecificDataSize < 128); 2361 2362 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2363 CHECK(mCodecSpecificData); 2364 CHECK(mCodecSpecificDataSize > 0); 2365 mOwner->beginBox("esds"); 2366 2367 mOwner->writeInt32(0); // version=0, flags=0 2368 2369 mOwner->writeInt8(0x03); // ES_DescrTag 2370 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2371 mOwner->writeInt16(0x0000); // ES_ID 2372 mOwner->writeInt8(0x1f); 2373 2374 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2375 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2376 mOwner->writeInt8(0x20); // objectTypeIndication ISO/IEC 14492-2 2377 mOwner->writeInt8(0x11); // streamType VisualStream 2378 2379 static const uint8_t kData[] = { 2380 0x01, 0x77, 0x00, 2381 0x00, 0x03, 0xe8, 0x00, 2382 0x00, 0x03, 0xe8, 0x00 2383 }; 2384 mOwner->write(kData, sizeof(kData)); 2385 2386 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2387 2388 mOwner->writeInt8(mCodecSpecificDataSize); 2389 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2390 2391 static const uint8_t kData2[] = { 2392 0x06, // SLConfigDescriptorTag 2393 0x01, 2394 0x02 2395 }; 2396 mOwner->write(kData2, sizeof(kData2)); 2397 2398 mOwner->endBox(); // esds 2399 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2400 mOwner->beginBox("d263"); 2401 2402 mOwner->writeInt32(0); // vendor 2403 mOwner->writeInt8(0); // decoder version 2404 mOwner->writeInt8(10); // level: 10 2405 mOwner->writeInt8(0); // profile: 0 2406 2407 mOwner->endBox(); // d263 2408 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2409 CHECK(mCodecSpecificData); 2410 CHECK(mCodecSpecificDataSize >= 5); 2411 2412 // Patch avcc's lengthSize field to match the number 2413 // of bytes we use to indicate the size of a nal unit. 2414 uint8_t *ptr = (uint8_t *)mCodecSpecificData; 2415 ptr[4] = 2416 (ptr[4] & 0xfc) 2417 | (mOwner->useNalLengthFour() ? 3 : 1); 2418 2419 mOwner->beginBox("avcC"); 2420 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2421 mOwner->endBox(); // avcC 2422 } 2423 2424 mOwner->beginBox("pasp"); 2425 // This is useful if the pixel is not square 2426 mOwner->writeInt32(1 << 16); // hspacing 2427 mOwner->writeInt32(1 << 16); // vspacing 2428 mOwner->endBox(); // pasp 2429 mOwner->endBox(); // mp4v, s263 or avc1 2430 } 2431 mOwner->endBox(); // stsd 2432 2433 mOwner->beginBox("stts"); 2434 mOwner->writeInt32(0); // version=0, flags=0 2435 mOwner->writeInt32(mNumSttsTableEntries); 2436 int64_t prevTimestampUs = 0; 2437 for (List<SttsTableEntry>::iterator it = mSttsTableEntries.begin(); 2438 it != mSttsTableEntries.end(); ++it) { 2439 mOwner->writeInt32(it->sampleCount); 2440 2441 // Make sure that we are calculating the sample duration the exactly 2442 // same way as we made decision on how to create stts entries. 2443 int64_t currTimestampUs = prevTimestampUs + it->sampleDurationUs; 2444 int32_t dur = ((currTimestampUs * mTimeScale + 500000LL) / 1000000LL - 2445 (prevTimestampUs * mTimeScale + 500000LL) / 1000000LL); 2446 prevTimestampUs += (it->sampleCount * it->sampleDurationUs); 2447 2448 mOwner->writeInt32(dur); 2449 } 2450 mOwner->endBox(); // stts 2451 2452 if (!mIsAudio) { 2453 mOwner->beginBox("stss"); 2454 mOwner->writeInt32(0); // version=0, flags=0 2455 mOwner->writeInt32(mNumStssTableEntries); // number of sync frames 2456 for (List<int32_t>::iterator it = mStssTableEntries.begin(); 2457 it != mStssTableEntries.end(); ++it) { 2458 mOwner->writeInt32(*it); 2459 } 2460 mOwner->endBox(); // stss 2461 } 2462 2463 mOwner->beginBox("stsz"); 2464 mOwner->writeInt32(0); // version=0, flags=0 2465 if (mSamplesHaveSameSize) { 2466 List<size_t>::iterator it = mSampleSizes.begin(); 2467 mOwner->writeInt32(*it); // default sample size 2468 } else { 2469 mOwner->writeInt32(0); 2470 } 2471 mOwner->writeInt32(mNumSamples); 2472 if (!mSamplesHaveSameSize) { 2473 for (List<size_t>::iterator it = mSampleSizes.begin(); 2474 it != mSampleSizes.end(); ++it) { 2475 mOwner->writeInt32(*it); 2476 } 2477 } 2478 mOwner->endBox(); // stsz 2479 2480 mOwner->beginBox("stsc"); 2481 mOwner->writeInt32(0); // version=0, flags=0 2482 mOwner->writeInt32(mNumStscTableEntries); 2483 for (List<StscTableEntry>::iterator it = mStscTableEntries.begin(); 2484 it != mStscTableEntries.end(); ++it) { 2485 mOwner->writeInt32(it->firstChunk); 2486 mOwner->writeInt32(it->samplesPerChunk); 2487 mOwner->writeInt32(it->sampleDescriptionId); 2488 } 2489 mOwner->endBox(); // stsc 2490 mOwner->beginBox(use32BitOffset? "stco": "co64"); 2491 mOwner->writeInt32(0); // version=0, flags=0 2492 mOwner->writeInt32(mNumStcoTableEntries); 2493 for (List<off64_t>::iterator it = mChunkOffsets.begin(); 2494 it != mChunkOffsets.end(); ++it) { 2495 if (use32BitOffset) { 2496 mOwner->writeInt32(static_cast<int32_t>(*it)); 2497 } else { 2498 mOwner->writeInt64((*it)); 2499 } 2500 } 2501 mOwner->endBox(); // stco or co64 2502 2503 mOwner->endBox(); // stbl 2504 mOwner->endBox(); // minf 2505 mOwner->endBox(); // mdia 2506 mOwner->endBox(); // trak 2507} 2508 2509} // namespace android 2510