MPEG4Writer.cpp revision 985f838934510983d8a887461e98dca60a6e858f
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 // Be conservative in the estimate: do not exceed 95% of 882 // the target file limit. For small target file size limit, though, 883 // this will not help. 884 return (nTotalBytesEstimate >= (95 * mMaxFileSizeLimitBytes) / 100); 885} 886 887bool MPEG4Writer::exceedsFileDurationLimit() { 888 // No limit 889 if (mMaxFileDurationLimitUs == 0) { 890 return false; 891 } 892 893 for (List<Track *>::iterator it = mTracks.begin(); 894 it != mTracks.end(); ++it) { 895 if ((*it)->getDurationUs() >= mMaxFileDurationLimitUs) { 896 return true; 897 } 898 } 899 return false; 900} 901 902bool MPEG4Writer::reachedEOS() { 903 bool allDone = true; 904 for (List<Track *>::iterator it = mTracks.begin(); 905 it != mTracks.end(); ++it) { 906 if (!(*it)->reachedEOS()) { 907 allDone = false; 908 break; 909 } 910 } 911 912 return allDone; 913} 914 915void MPEG4Writer::setStartTimestampUs(int64_t timeUs) { 916 LOGI("setStartTimestampUs: %lld", timeUs); 917 CHECK(timeUs >= 0); 918 Mutex::Autolock autoLock(mLock); 919 if (mStartTimestampUs < 0 || mStartTimestampUs > timeUs) { 920 mStartTimestampUs = timeUs; 921 LOGI("Earliest track starting time: %lld", mStartTimestampUs); 922 } 923} 924 925int64_t MPEG4Writer::getStartTimestampUs() { 926 Mutex::Autolock autoLock(mLock); 927 return mStartTimestampUs; 928} 929 930size_t MPEG4Writer::numTracks() { 931 Mutex::Autolock autolock(mLock); 932 return mTracks.size(); 933} 934 935//////////////////////////////////////////////////////////////////////////////// 936 937MPEG4Writer::Track::Track( 938 MPEG4Writer *owner, const sp<MediaSource> &source) 939 : mOwner(owner), 940 mMeta(source->getFormat()), 941 mSource(source), 942 mDone(false), 943 mPaused(false), 944 mResumed(false), 945 mTrackDurationUs(0), 946 mEstimatedTrackSizeBytes(0), 947 mSamplesHaveSameSize(true), 948 mCodecSpecificData(NULL), 949 mCodecSpecificDataSize(0), 950 mGotAllCodecSpecificData(false), 951 mReachedEOS(false), 952 mRotation(0) { 953 getCodecSpecificDataFromInputFormatIfPossible(); 954 955 const char *mime; 956 mMeta->findCString(kKeyMIMEType, &mime); 957 mIsAvc = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC); 958 mIsAudio = !strncasecmp(mime, "audio/", 6); 959 mIsMPEG4 = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) || 960 !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC); 961 962 setTimeScale(); 963} 964 965void MPEG4Writer::Track::updateTrackSizeEstimate() { 966 967 int64_t stcoBoxSizeBytes = mOwner->use32BitFileOffset() 968 ? mNumStcoTableEntries * 4 969 : mNumStcoTableEntries * 8; 970 971 int64_t stszBoxSizeBytes = mSamplesHaveSameSize? 4: (mNumSamples * 4); 972 973 mEstimatedTrackSizeBytes = mMdatSizeBytes; // media data size 974 if (!mOwner->isFileStreamable()) { 975 // Reserved free space is not large enough to hold 976 // all meta data and thus wasted. 977 mEstimatedTrackSizeBytes += mNumStscTableEntries * 12 + // stsc box size 978 mNumStssTableEntries * 4 + // stss box size 979 mNumSttsTableEntries * 8 + // stts box size 980 stcoBoxSizeBytes + // stco box size 981 stszBoxSizeBytes; // stsz box size 982 } 983} 984 985void MPEG4Writer::Track::addOneStscTableEntry( 986 size_t chunkId, size_t sampleId) { 987 988 StscTableEntry stscEntry(chunkId, sampleId, 1); 989 mStscTableEntries.push_back(stscEntry); 990 ++mNumStscTableEntries; 991} 992 993void MPEG4Writer::Track::addOneStssTableEntry(size_t sampleId) { 994 mStssTableEntries.push_back(sampleId); 995 ++mNumStssTableEntries; 996} 997 998void MPEG4Writer::Track::addOneSttsTableEntry( 999 size_t sampleCount, int64_t durationUs) { 1000 1001 SttsTableEntry sttsEntry(sampleCount, durationUs); 1002 mSttsTableEntries.push_back(sttsEntry); 1003 ++mNumSttsTableEntries; 1004} 1005 1006void MPEG4Writer::Track::addChunkOffset(off64_t offset) { 1007 ++mNumStcoTableEntries; 1008 mChunkOffsets.push_back(offset); 1009} 1010 1011void MPEG4Writer::Track::setTimeScale() { 1012 LOGV("setTimeScale"); 1013 // Default time scale 1014 mTimeScale = 90000; 1015 1016 if (mIsAudio) { 1017 // Use the sampling rate as the default time scale for audio track. 1018 int32_t sampleRate; 1019 bool success = mMeta->findInt32(kKeySampleRate, &sampleRate); 1020 CHECK(success); 1021 mTimeScale = sampleRate; 1022 } 1023 1024 // If someone would like to overwrite the timescale, use user-supplied value. 1025 int32_t timeScale; 1026 if (mMeta->findInt32(kKeyTimeScale, &timeScale)) { 1027 mTimeScale = timeScale; 1028 } 1029 1030 CHECK(mTimeScale > 0); 1031} 1032 1033void MPEG4Writer::Track::getCodecSpecificDataFromInputFormatIfPossible() { 1034 const char *mime; 1035 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 1036 1037 if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC)) { 1038 uint32_t type; 1039 const void *data; 1040 size_t size; 1041 if (mMeta->findData(kKeyAVCC, &type, &data, &size)) { 1042 mCodecSpecificData = malloc(size); 1043 mCodecSpecificDataSize = size; 1044 memcpy(mCodecSpecificData, data, size); 1045 mGotAllCodecSpecificData = true; 1046 } 1047 } else if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) 1048 || !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC)) { 1049 uint32_t type; 1050 const void *data; 1051 size_t size; 1052 if (mMeta->findData(kKeyESDS, &type, &data, &size)) { 1053 ESDS esds(data, size); 1054 if (esds.getCodecSpecificInfo(&data, &size) == OK) { 1055 mCodecSpecificData = malloc(size); 1056 mCodecSpecificDataSize = size; 1057 memcpy(mCodecSpecificData, data, size); 1058 mGotAllCodecSpecificData = true; 1059 } 1060 } 1061 } 1062} 1063 1064MPEG4Writer::Track::~Track() { 1065 stop(); 1066 1067 if (mCodecSpecificData != NULL) { 1068 free(mCodecSpecificData); 1069 mCodecSpecificData = NULL; 1070 } 1071} 1072 1073void MPEG4Writer::Track::initTrackingProgressStatus(MetaData *params) { 1074 LOGV("initTrackingProgressStatus"); 1075 mPreviousTrackTimeUs = -1; 1076 mTrackingProgressStatus = false; 1077 mTrackEveryTimeDurationUs = 0; 1078 { 1079 int64_t timeUs; 1080 if (params && params->findInt64(kKeyTrackTimeStatus, &timeUs)) { 1081 LOGV("Receive request to track progress status for every %lld us", timeUs); 1082 mTrackEveryTimeDurationUs = timeUs; 1083 mTrackingProgressStatus = true; 1084 } 1085 } 1086} 1087 1088// static 1089void *MPEG4Writer::ThreadWrapper(void *me) { 1090 LOGV("ThreadWrapper: %p", me); 1091 MPEG4Writer *writer = static_cast<MPEG4Writer *>(me); 1092 writer->threadFunc(); 1093 return NULL; 1094} 1095 1096void MPEG4Writer::bufferChunk(const Chunk& chunk) { 1097 LOGV("bufferChunk: %p", chunk.mTrack); 1098 Mutex::Autolock autolock(mLock); 1099 CHECK_EQ(mDone, false); 1100 1101 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1102 it != mChunkInfos.end(); ++it) { 1103 1104 if (chunk.mTrack == it->mTrack) { // Found owner 1105 it->mChunks.push_back(chunk); 1106 mChunkReadyCondition.signal(); 1107 return; 1108 } 1109 } 1110 1111 CHECK("Received a chunk for a unknown track" == 0); 1112} 1113 1114void MPEG4Writer::writeFirstChunk(ChunkInfo* info) { 1115 LOGV("writeFirstChunk: %p", info->mTrack); 1116 1117 List<Chunk>::iterator chunkIt = info->mChunks.begin(); 1118 for (List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1119 it != chunkIt->mSamples.end(); ++it) { 1120 1121 off64_t offset = info->mTrack->isAvc() 1122 ? addLengthPrefixedSample_l(*it) 1123 : addSample_l(*it); 1124 if (it == chunkIt->mSamples.begin()) { 1125 info->mTrack->addChunkOffset(offset); 1126 } 1127 } 1128 1129 // Done with the current chunk. 1130 // Release all the samples in this chunk. 1131 while (!chunkIt->mSamples.empty()) { 1132 List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1133 (*it)->release(); 1134 (*it) = NULL; 1135 chunkIt->mSamples.erase(it); 1136 } 1137 chunkIt->mSamples.clear(); 1138 info->mChunks.erase(chunkIt); 1139} 1140 1141void MPEG4Writer::writeChunks() { 1142 LOGV("writeChunks"); 1143 size_t outstandingChunks = 0; 1144 while (!mChunkInfos.empty()) { 1145 List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1146 while (!it->mChunks.empty()) { 1147 CHECK_EQ(OK, writeOneChunk()); 1148 ++outstandingChunks; 1149 } 1150 it->mTrack = NULL; 1151 mChunkInfos.erase(it); 1152 } 1153 mChunkInfos.clear(); 1154 LOGD("%d chunks are written in the last batch", outstandingChunks); 1155} 1156 1157status_t MPEG4Writer::writeOneChunk() { 1158 LOGV("writeOneChunk"); 1159 1160 // Find the smallest timestamp, and write that chunk out 1161 // XXX: What if some track is just too slow? 1162 int64_t minTimestampUs = 0x7FFFFFFFFFFFFFFFLL; 1163 Track *track = NULL; 1164 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1165 it != mChunkInfos.end(); ++it) { 1166 if (!it->mChunks.empty()) { 1167 List<Chunk>::iterator chunkIt = it->mChunks.begin(); 1168 if (chunkIt->mTimeStampUs < minTimestampUs) { 1169 minTimestampUs = chunkIt->mTimeStampUs; 1170 track = it->mTrack; 1171 } 1172 } 1173 } 1174 1175 if (track == NULL) { 1176 LOGV("Nothing to be written after all"); 1177 return OK; 1178 } 1179 1180 if (mIsFirstChunk) { 1181 mIsFirstChunk = false; 1182 } 1183 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1184 it != mChunkInfos.end(); ++it) { 1185 if (it->mTrack == track) { 1186 writeFirstChunk(&(*it)); 1187 } 1188 } 1189 return OK; 1190} 1191 1192void MPEG4Writer::threadFunc() { 1193 LOGV("threadFunc"); 1194 1195 prctl(PR_SET_NAME, (unsigned long)"MPEG4Writer", 0, 0, 0); 1196 while (!mDone) { 1197 { 1198 Mutex::Autolock autolock(mLock); 1199 mChunkReadyCondition.wait(mLock); 1200 CHECK_EQ(writeOneChunk(), OK); 1201 } 1202 } 1203 1204 { 1205 // Write ALL samples 1206 Mutex::Autolock autolock(mLock); 1207 writeChunks(); 1208 } 1209} 1210 1211status_t MPEG4Writer::startWriterThread() { 1212 LOGV("startWriterThread"); 1213 1214 mDone = false; 1215 mIsFirstChunk = true; 1216 mDriftTimeUs = 0; 1217 for (List<Track *>::iterator it = mTracks.begin(); 1218 it != mTracks.end(); ++it) { 1219 ChunkInfo info; 1220 info.mTrack = *it; 1221 mChunkInfos.push_back(info); 1222 } 1223 1224 pthread_attr_t attr; 1225 pthread_attr_init(&attr); 1226 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1227 pthread_create(&mThread, &attr, ThreadWrapper, this); 1228 pthread_attr_destroy(&attr); 1229 return OK; 1230} 1231 1232 1233status_t MPEG4Writer::Track::start(MetaData *params) { 1234 if (!mDone && mPaused) { 1235 mPaused = false; 1236 mResumed = true; 1237 return OK; 1238 } 1239 1240 int64_t startTimeUs; 1241 if (params == NULL || !params->findInt64(kKeyTime, &startTimeUs)) { 1242 startTimeUs = 0; 1243 } 1244 1245 int32_t rotationDegrees; 1246 if (!mIsAudio && params && params->findInt32(kKeyRotation, &rotationDegrees)) { 1247 mRotation = rotationDegrees; 1248 } 1249 1250 mIsRealTimeRecording = true; 1251 { 1252 int32_t isNotRealTime; 1253 if (params && params->findInt32(kKeyNotRealTime, &isNotRealTime)) { 1254 mIsRealTimeRecording = (isNotRealTime == 0); 1255 } 1256 } 1257 1258 initTrackingProgressStatus(params); 1259 1260 sp<MetaData> meta = new MetaData; 1261 meta->setInt64(kKeyTime, startTimeUs); 1262 status_t err = mSource->start(meta.get()); 1263 if (err != OK) { 1264 mDone = mReachedEOS = true; 1265 return err; 1266 } 1267 1268 pthread_attr_t attr; 1269 pthread_attr_init(&attr); 1270 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1271 1272 mDone = false; 1273 mTrackDurationUs = 0; 1274 mReachedEOS = false; 1275 mEstimatedTrackSizeBytes = 0; 1276 mNumStcoTableEntries = 0; 1277 mNumStssTableEntries = 0; 1278 mNumStscTableEntries = 0; 1279 mNumSttsTableEntries = 0; 1280 mMdatSizeBytes = 0; 1281 mIsMediaTimeAdjustmentOn = false; 1282 mPrevMediaTimeAdjustTimestampUs = 0; 1283 mMediaTimeAdjustNumFrames = 0; 1284 mPrevMediaTimeAdjustSample = 0; 1285 mTotalDriftTimeToAdjustUs = 0; 1286 mPrevTotalAccumDriftTimeUs = 0; 1287 1288 pthread_create(&mThread, &attr, ThreadWrapper, this); 1289 pthread_attr_destroy(&attr); 1290 1291 return OK; 1292} 1293 1294status_t MPEG4Writer::Track::pause() { 1295 mPaused = true; 1296 return OK; 1297} 1298 1299status_t MPEG4Writer::Track::stop() { 1300 LOGD("Stopping %s track", mIsAudio? "Audio": "Video"); 1301 if (mDone) { 1302 return OK; 1303 } 1304 1305 mDone = true; 1306 1307 void *dummy; 1308 pthread_join(mThread, &dummy); 1309 1310 status_t err = (status_t) dummy; 1311 1312 LOGD("Stopping %s track source", mIsAudio? "Audio": "Video"); 1313 { 1314 status_t status = mSource->stop(); 1315 if (err == OK && status != OK && status != ERROR_END_OF_STREAM) { 1316 err = status; 1317 } 1318 } 1319 1320 LOGD("%s track stopped", mIsAudio? "Audio": "Video"); 1321 return err; 1322} 1323 1324bool MPEG4Writer::Track::reachedEOS() { 1325 return mReachedEOS; 1326} 1327 1328// static 1329void *MPEG4Writer::Track::ThreadWrapper(void *me) { 1330 Track *track = static_cast<Track *>(me); 1331 1332 status_t err = track->threadEntry(); 1333 return (void *) err; 1334} 1335 1336static void getNalUnitType(uint8_t byte, uint8_t* type) { 1337 LOGV("getNalUnitType: %d", byte); 1338 1339 // nal_unit_type: 5-bit unsigned integer 1340 *type = (byte & 0x1F); 1341} 1342 1343static const uint8_t *findNextStartCode( 1344 const uint8_t *data, size_t length) { 1345 1346 LOGV("findNextStartCode: %p %d", data, length); 1347 1348 size_t bytesLeft = length; 1349 while (bytesLeft > 4 && 1350 memcmp("\x00\x00\x00\x01", &data[length - bytesLeft], 4)) { 1351 --bytesLeft; 1352 } 1353 if (bytesLeft <= 4) { 1354 bytesLeft = 0; // Last parameter set 1355 } 1356 return &data[length - bytesLeft]; 1357} 1358 1359const uint8_t *MPEG4Writer::Track::parseParamSet( 1360 const uint8_t *data, size_t length, int type, size_t *paramSetLen) { 1361 1362 LOGV("parseParamSet"); 1363 CHECK(type == kNalUnitTypeSeqParamSet || 1364 type == kNalUnitTypePicParamSet); 1365 1366 const uint8_t *nextStartCode = findNextStartCode(data, length); 1367 *paramSetLen = nextStartCode - data; 1368 if (*paramSetLen == 0) { 1369 LOGE("Param set is malformed, since its length is 0"); 1370 return NULL; 1371 } 1372 1373 AVCParamSet paramSet(*paramSetLen, data); 1374 if (type == kNalUnitTypeSeqParamSet) { 1375 if (*paramSetLen < 4) { 1376 LOGE("Seq parameter set malformed"); 1377 return NULL; 1378 } 1379 if (mSeqParamSets.empty()) { 1380 mProfileIdc = data[1]; 1381 mProfileCompatible = data[2]; 1382 mLevelIdc = data[3]; 1383 } else { 1384 if (mProfileIdc != data[1] || 1385 mProfileCompatible != data[2] || 1386 mLevelIdc != data[3]) { 1387 LOGE("Inconsistent profile/level found in seq parameter sets"); 1388 return NULL; 1389 } 1390 } 1391 mSeqParamSets.push_back(paramSet); 1392 } else { 1393 mPicParamSets.push_back(paramSet); 1394 } 1395 return nextStartCode; 1396} 1397 1398status_t MPEG4Writer::Track::copyAVCCodecSpecificData( 1399 const uint8_t *data, size_t size) { 1400 LOGV("copyAVCCodecSpecificData"); 1401 1402 // 2 bytes for each of the parameter set length field 1403 // plus the 7 bytes for the header 1404 if (size < 4 + 7) { 1405 LOGE("Codec specific data length too short: %d", size); 1406 return ERROR_MALFORMED; 1407 } 1408 1409 mCodecSpecificDataSize = size; 1410 mCodecSpecificData = malloc(size); 1411 memcpy(mCodecSpecificData, data, size); 1412 return OK; 1413} 1414 1415status_t MPEG4Writer::Track::parseAVCCodecSpecificData( 1416 const uint8_t *data, size_t size) { 1417 1418 LOGV("parseAVCCodecSpecificData"); 1419 // Data starts with a start code. 1420 // SPS and PPS are separated with start codes. 1421 // Also, SPS must come before PPS 1422 uint8_t type = kNalUnitTypeSeqParamSet; 1423 bool gotSps = false; 1424 bool gotPps = false; 1425 const uint8_t *tmp = data; 1426 const uint8_t *nextStartCode = data; 1427 size_t bytesLeft = size; 1428 size_t paramSetLen = 0; 1429 mCodecSpecificDataSize = 0; 1430 while (bytesLeft > 4 && !memcmp("\x00\x00\x00\x01", tmp, 4)) { 1431 getNalUnitType(*(tmp + 4), &type); 1432 if (type == kNalUnitTypeSeqParamSet) { 1433 if (gotPps) { 1434 LOGE("SPS must come before PPS"); 1435 return ERROR_MALFORMED; 1436 } 1437 if (!gotSps) { 1438 gotSps = true; 1439 } 1440 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1441 } else if (type == kNalUnitTypePicParamSet) { 1442 if (!gotSps) { 1443 LOGE("SPS must come before PPS"); 1444 return ERROR_MALFORMED; 1445 } 1446 if (!gotPps) { 1447 gotPps = true; 1448 } 1449 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1450 } else { 1451 LOGE("Only SPS and PPS Nal units are expected"); 1452 return ERROR_MALFORMED; 1453 } 1454 1455 if (nextStartCode == NULL) { 1456 return ERROR_MALFORMED; 1457 } 1458 1459 // Move on to find the next parameter set 1460 bytesLeft -= nextStartCode - tmp; 1461 tmp = nextStartCode; 1462 mCodecSpecificDataSize += (2 + paramSetLen); 1463 } 1464 1465 { 1466 // Check on the number of seq parameter sets 1467 size_t nSeqParamSets = mSeqParamSets.size(); 1468 if (nSeqParamSets == 0) { 1469 LOGE("Cound not find sequence parameter set"); 1470 return ERROR_MALFORMED; 1471 } 1472 1473 if (nSeqParamSets > 0x1F) { 1474 LOGE("Too many seq parameter sets (%d) found", nSeqParamSets); 1475 return ERROR_MALFORMED; 1476 } 1477 } 1478 1479 { 1480 // Check on the number of pic parameter sets 1481 size_t nPicParamSets = mPicParamSets.size(); 1482 if (nPicParamSets == 0) { 1483 LOGE("Cound not find picture parameter set"); 1484 return ERROR_MALFORMED; 1485 } 1486 if (nPicParamSets > 0xFF) { 1487 LOGE("Too many pic parameter sets (%d) found", nPicParamSets); 1488 return ERROR_MALFORMED; 1489 } 1490 } 1491 1492 { 1493 // Check on the profiles 1494 // These profiles requires additional parameter set extensions 1495 if (mProfileIdc == 100 || mProfileIdc == 110 || 1496 mProfileIdc == 122 || mProfileIdc == 144) { 1497 LOGE("Sorry, no support for profile_idc: %d!", mProfileIdc); 1498 return BAD_VALUE; 1499 } 1500 } 1501 1502 return OK; 1503} 1504 1505status_t MPEG4Writer::Track::makeAVCCodecSpecificData( 1506 const uint8_t *data, size_t size) { 1507 1508 if (mCodecSpecificData != NULL) { 1509 LOGE("Already have codec specific data"); 1510 return ERROR_MALFORMED; 1511 } 1512 1513 if (size < 4) { 1514 LOGE("Codec specific data length too short: %d", size); 1515 return ERROR_MALFORMED; 1516 } 1517 1518 // Data is in the form of AVCCodecSpecificData 1519 if (memcmp("\x00\x00\x00\x01", data, 4)) { 1520 return copyAVCCodecSpecificData(data, size); 1521 } 1522 1523 if (parseAVCCodecSpecificData(data, size) != OK) { 1524 return ERROR_MALFORMED; 1525 } 1526 1527 // ISO 14496-15: AVC file format 1528 mCodecSpecificDataSize += 7; // 7 more bytes in the header 1529 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1530 uint8_t *header = (uint8_t *)mCodecSpecificData; 1531 header[0] = 1; // version 1532 header[1] = mProfileIdc; // profile indication 1533 header[2] = mProfileCompatible; // profile compatibility 1534 header[3] = mLevelIdc; 1535 1536 // 6-bit '111111' followed by 2-bit to lengthSizeMinuusOne 1537 if (mOwner->useNalLengthFour()) { 1538 header[4] = 0xfc | 3; // length size == 4 bytes 1539 } else { 1540 header[4] = 0xfc | 1; // length size == 2 bytes 1541 } 1542 1543 // 3-bit '111' followed by 5-bit numSequenceParameterSets 1544 int nSequenceParamSets = mSeqParamSets.size(); 1545 header[5] = 0xe0 | nSequenceParamSets; 1546 header += 6; 1547 for (List<AVCParamSet>::iterator it = mSeqParamSets.begin(); 1548 it != mSeqParamSets.end(); ++it) { 1549 // 16-bit sequence parameter set length 1550 uint16_t seqParamSetLength = it->mLength; 1551 header[0] = seqParamSetLength >> 8; 1552 header[1] = seqParamSetLength & 0xff; 1553 1554 // SPS NAL unit (sequence parameter length bytes) 1555 memcpy(&header[2], it->mData, seqParamSetLength); 1556 header += (2 + seqParamSetLength); 1557 } 1558 1559 // 8-bit nPictureParameterSets 1560 int nPictureParamSets = mPicParamSets.size(); 1561 header[0] = nPictureParamSets; 1562 header += 1; 1563 for (List<AVCParamSet>::iterator it = mPicParamSets.begin(); 1564 it != mPicParamSets.end(); ++it) { 1565 // 16-bit picture parameter set length 1566 uint16_t picParamSetLength = it->mLength; 1567 header[0] = picParamSetLength >> 8; 1568 header[1] = picParamSetLength & 0xff; 1569 1570 // PPS Nal unit (picture parameter set length bytes) 1571 memcpy(&header[2], it->mData, picParamSetLength); 1572 header += (2 + picParamSetLength); 1573 } 1574 1575 return OK; 1576} 1577 1578/* 1579* The video track's media time adjustment for real-time applications 1580* is described as follows: 1581* 1582* First, the media time adjustment is done for every period of 1583* kVideoMediaTimeAdjustPeriodTimeUs. kVideoMediaTimeAdjustPeriodTimeUs 1584* is currently a fixed value chosen heuristically. The value of 1585* kVideoMediaTimeAdjustPeriodTimeUs should not be very large or very small 1586* for two considerations: on one hand, a relatively large value 1587* helps reduce large fluctuation of drift time in the audio encoding 1588* path; while on the other hand, a relatively small value helps keep 1589* restoring synchronization in audio/video more frequently. Note for the 1590* very first period of kVideoMediaTimeAdjustPeriodTimeUs, there is 1591* no media time adjustment for the video track. 1592* 1593* Second, the total accumulated audio track time drift found 1594* in a period of kVideoMediaTimeAdjustPeriodTimeUs is distributed 1595* over a stream of incoming video frames. The number of video frames 1596* affected is determined based on the number of recorded video frames 1597* within the past kVideoMediaTimeAdjustPeriodTimeUs period. 1598* We choose to distribute the drift time over only a portion 1599* (rather than all) of the total number of recorded video frames 1600* in order to make sure that the video track media time adjustment is 1601* completed for the current period before the next video track media 1602* time adjustment period starts. Currently, the portion chosen is a 1603* half (0.5). 1604* 1605* Last, various additional checks are performed to ensure that 1606* the actual audio encoding path does not have too much drift. 1607* In particular, 1) we want to limit the average incremental time 1608* adjustment for each video frame to be less than a threshold 1609* for a single period of kVideoMediaTimeAdjustPeriodTimeUs. 1610* Currently, the threshold is set to 5 ms. If the average incremental 1611* media time adjustment for a video frame is larger than the 1612* threshold, the audio encoding path has too much time drift. 1613* 2) We also want to limit the total time drift in the audio 1614* encoding path to be less than a threshold for a period of 1615* kVideoMediaTimeAdjustPeriodTimeUs. Currently, the threshold 1616* is 0.5% of kVideoMediaTimeAdjustPeriodTimeUs. If the time drift of 1617* the audio encoding path is larger than the threshold, the audio 1618* encoding path has too much time drift. We treat the large time 1619* drift of the audio encoding path as errors, since there is no 1620* way to keep audio/video in synchronization for real-time 1621* applications if the time drift is too large unless we drop some 1622* video frames, which has its own problems that we don't want 1623* to get into for the time being. 1624*/ 1625void MPEG4Writer::Track::adjustMediaTime(int64_t *timestampUs) { 1626 if (*timestampUs - mPrevMediaTimeAdjustTimestampUs >= 1627 kVideoMediaTimeAdjustPeriodTimeUs) { 1628 1629 LOGV("New media time adjustment period at %lld us", *timestampUs); 1630 mIsMediaTimeAdjustmentOn = true; 1631 mMediaTimeAdjustNumFrames = 1632 (mNumSamples - mPrevMediaTimeAdjustSample) >> 1; 1633 1634 mPrevMediaTimeAdjustTimestampUs = *timestampUs; 1635 mPrevMediaTimeAdjustSample = mNumSamples; 1636 int64_t totalAccumDriftTimeUs = mOwner->getDriftTimeUs(); 1637 mTotalDriftTimeToAdjustUs = 1638 totalAccumDriftTimeUs - mPrevTotalAccumDriftTimeUs; 1639 1640 mPrevTotalAccumDriftTimeUs = totalAccumDriftTimeUs; 1641 1642 // Check on incremental adjusted time per frame 1643 int64_t adjustTimePerFrameUs = 1644 mTotalDriftTimeToAdjustUs / mMediaTimeAdjustNumFrames; 1645 1646 if (adjustTimePerFrameUs < 0) { 1647 adjustTimePerFrameUs = -adjustTimePerFrameUs; 1648 } 1649 if (adjustTimePerFrameUs >= 5000) { 1650 LOGE("Adjusted time per video frame is %lld us", 1651 adjustTimePerFrameUs); 1652 CHECK(!"Video frame time adjustment is too large!"); 1653 } 1654 1655 // Check on total accumulated time drift within a period of 1656 // kVideoMediaTimeAdjustPeriodTimeUs. 1657 int64_t driftPercentage = (mTotalDriftTimeToAdjustUs * 1000) 1658 / kVideoMediaTimeAdjustPeriodTimeUs; 1659 1660 if (driftPercentage < 0) { 1661 driftPercentage = -driftPercentage; 1662 } 1663 if (driftPercentage > 5) { 1664 LOGE("Audio track has time drift %lld us over %lld us", 1665 mTotalDriftTimeToAdjustUs, 1666 kVideoMediaTimeAdjustPeriodTimeUs); 1667 1668 CHECK(!"The audio track media time drifts too much!"); 1669 } 1670 1671 } 1672 1673 if (mIsMediaTimeAdjustmentOn) { 1674 if (mNumSamples - mPrevMediaTimeAdjustSample <= 1675 mMediaTimeAdjustNumFrames) { 1676 1677 // Do media time incremental adjustment 1678 int64_t incrementalAdjustTimeUs = 1679 (mTotalDriftTimeToAdjustUs * 1680 (mNumSamples - mPrevMediaTimeAdjustSample)) 1681 / mMediaTimeAdjustNumFrames; 1682 1683 *timestampUs += 1684 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs); 1685 1686 LOGV("Incremental video frame media time adjustment: %lld us", 1687 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs)); 1688 } else { 1689 // Within the remaining adjustment period, 1690 // no incremental adjustment is needed. 1691 *timestampUs += 1692 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs); 1693 1694 LOGV("Fixed video frame media time adjustment: %lld us", 1695 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs)); 1696 } 1697 } 1698} 1699 1700/* 1701 * Updates the drift time from the audio track so that 1702 * the video track can get the updated drift time information 1703 * from the file writer. The fluctuation of the drift time of the audio 1704 * encoding path is smoothed out with a simple filter by giving a larger 1705 * weight to more recently drift time. The filter coefficients, 0.5 and 0.5, 1706 * are heuristically determined. 1707 */ 1708void MPEG4Writer::Track::updateDriftTime(const sp<MetaData>& meta) { 1709 int64_t driftTimeUs = 0; 1710 if (meta->findInt64(kKeyDriftTime, &driftTimeUs)) { 1711 int64_t prevDriftTimeUs = mOwner->getDriftTimeUs(); 1712 int64_t timeUs = (driftTimeUs + prevDriftTimeUs) >> 1; 1713 mOwner->setDriftTimeUs(timeUs); 1714 } 1715} 1716 1717status_t MPEG4Writer::Track::threadEntry() { 1718 int32_t count = 0; 1719 const int64_t interleaveDurationUs = mOwner->interleaveDuration(); 1720 int64_t chunkTimestampUs = 0; 1721 int32_t nChunks = 0; 1722 int32_t nZeroLengthFrames = 0; 1723 int64_t lastTimestampUs = 0; // Previous sample time stamp in ms 1724 int64_t lastDurationUs = 0; // Between the previous two samples in ms 1725 int64_t currDurationTicks = 0; // Timescale based ticks 1726 int64_t lastDurationTicks = 0; // Timescale based ticks 1727 int32_t sampleCount = 1; // Sample count in the current stts table entry 1728 uint32_t previousSampleSize = 0; // Size of the previous sample 1729 int64_t previousPausedDurationUs = 0; 1730 int64_t timestampUs; 1731 1732 if (mIsAudio) { 1733 prctl(PR_SET_NAME, (unsigned long)"AudioTrackEncoding", 0, 0, 0); 1734 } else { 1735 prctl(PR_SET_NAME, (unsigned long)"VideoTrackEncoding", 0, 0, 0); 1736 } 1737 setpriority(PRIO_PROCESS, 0, ANDROID_PRIORITY_AUDIO); 1738 1739 sp<MetaData> meta_data; 1740 1741 mNumSamples = 0; 1742 status_t err = OK; 1743 MediaBuffer *buffer; 1744 while (!mDone && (err = mSource->read(&buffer)) == OK) { 1745 if (buffer->range_length() == 0) { 1746 buffer->release(); 1747 buffer = NULL; 1748 ++nZeroLengthFrames; 1749 continue; 1750 } 1751 1752 // If the codec specific data has not been received yet, delay pause. 1753 // After the codec specific data is received, discard what we received 1754 // when the track is to be paused. 1755 if (mPaused && !mResumed) { 1756 buffer->release(); 1757 buffer = NULL; 1758 continue; 1759 } 1760 1761 ++count; 1762 1763 int32_t isCodecConfig; 1764 if (buffer->meta_data()->findInt32(kKeyIsCodecConfig, &isCodecConfig) 1765 && isCodecConfig) { 1766 CHECK(!mGotAllCodecSpecificData); 1767 1768 if (mIsAvc) { 1769 status_t err = makeAVCCodecSpecificData( 1770 (const uint8_t *)buffer->data() 1771 + buffer->range_offset(), 1772 buffer->range_length()); 1773 CHECK_EQ(OK, err); 1774 } else if (mIsMPEG4) { 1775 mCodecSpecificDataSize = buffer->range_length(); 1776 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1777 memcpy(mCodecSpecificData, 1778 (const uint8_t *)buffer->data() 1779 + buffer->range_offset(), 1780 buffer->range_length()); 1781 } 1782 1783 buffer->release(); 1784 buffer = NULL; 1785 1786 mGotAllCodecSpecificData = true; 1787 continue; 1788 } 1789 1790 // Make a deep copy of the MediaBuffer and Metadata and release 1791 // the original as soon as we can 1792 MediaBuffer *copy = new MediaBuffer(buffer->range_length()); 1793 memcpy(copy->data(), (uint8_t *)buffer->data() + buffer->range_offset(), 1794 buffer->range_length()); 1795 copy->set_range(0, buffer->range_length()); 1796 meta_data = new MetaData(*buffer->meta_data().get()); 1797 buffer->release(); 1798 buffer = NULL; 1799 1800 if (mIsAvc) StripStartcode(copy); 1801 1802 size_t sampleSize = copy->range_length(); 1803 if (mIsAvc) { 1804 if (mOwner->useNalLengthFour()) { 1805 sampleSize += 4; 1806 } else { 1807 sampleSize += 2; 1808 } 1809 } 1810 1811 // Max file size or duration handling 1812 mMdatSizeBytes += sampleSize; 1813 updateTrackSizeEstimate(); 1814 1815 if (mOwner->exceedsFileSizeLimit()) { 1816 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_FILESIZE_REACHED, 0); 1817 break; 1818 } 1819 if (mOwner->exceedsFileDurationLimit()) { 1820 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_DURATION_REACHED, 0); 1821 break; 1822 } 1823 1824 1825 int32_t isSync = false; 1826 meta_data->findInt32(kKeyIsSyncFrame, &isSync); 1827 1828 /* 1829 * The original timestamp found in the data buffer will be modified as below: 1830 * 1831 * There is a playback offset into this track if the track's start time 1832 * is not the same as the movie start time, which will be recorded in edst 1833 * box of the output file. The playback offset is to make sure that the 1834 * starting time of the audio/video tracks are synchronized. Although the 1835 * track's media timestamp may be subject to various modifications 1836 * as outlined below, the track's playback offset time remains unchanged 1837 * once the first data buffer of the track is received. 1838 * 1839 * The media time stamp will be calculated by subtracting the playback offset 1840 * (and potential pause durations) from the original timestamp in the buffer. 1841 * 1842 * If this track is a video track for a real-time recording application with 1843 * both audio and video tracks, its media timestamp will subject to further 1844 * modification based on the media clock of the audio track. This modification 1845 * is needed for the purpose of maintaining good audio/video synchronization. 1846 * 1847 * If the recording session is paused and resumed multiple times, the track 1848 * media timestamp will be modified as if the recording session had never been 1849 * paused at all during playback of the recorded output file. In other words, 1850 * the output file will have no memory of pause/resume durations. 1851 * 1852 */ 1853 CHECK(meta_data->findInt64(kKeyTime, ×tampUs)); 1854 LOGV("%s timestampUs: %lld", mIsAudio? "Audio": "Video", timestampUs); 1855 1856//////////////////////////////////////////////////////////////////////////////// 1857 if (mSampleSizes.empty()) { 1858 mStartTimestampUs = timestampUs; 1859 mOwner->setStartTimestampUs(mStartTimestampUs); 1860 previousPausedDurationUs = mStartTimestampUs; 1861 } 1862 1863 if (mResumed) { 1864 int64_t durExcludingEarlierPausesUs = timestampUs - previousPausedDurationUs; 1865 CHECK(durExcludingEarlierPausesUs >= 0); 1866 int64_t pausedDurationUs = durExcludingEarlierPausesUs - mTrackDurationUs; 1867 CHECK(pausedDurationUs >= lastDurationUs); 1868 previousPausedDurationUs += pausedDurationUs - lastDurationUs; 1869 mResumed = false; 1870 } 1871 1872 timestampUs -= previousPausedDurationUs; 1873 CHECK(timestampUs >= 0); 1874 1875 // Media time adjustment for real-time applications 1876 if (mIsRealTimeRecording) { 1877 if (mIsAudio) { 1878 updateDriftTime(meta_data); 1879 } else { 1880 adjustMediaTime(×tampUs); 1881 } 1882 } 1883 1884 CHECK(timestampUs >= 0); 1885 if (mNumSamples > 1) { 1886 if (timestampUs <= lastTimestampUs) { 1887 LOGW("Frame arrives too late!"); 1888 // Don't drop the late frame, since dropping a frame may cause 1889 // problems later during playback 1890 1891 // The idea here is to avoid having two or more samples with the 1892 // same timestamp in the output file. 1893 if (mTimeScale >= 1000000LL) { 1894 timestampUs = lastTimestampUs + 1; 1895 } else { 1896 timestampUs = lastTimestampUs + (1000000LL + (mTimeScale >> 1)) / mTimeScale; 1897 } 1898 } 1899 } 1900 1901 LOGV("%s media time stamp: %lld and previous paused duration %lld", 1902 mIsAudio? "Audio": "Video", timestampUs, previousPausedDurationUs); 1903 if (timestampUs > mTrackDurationUs) { 1904 mTrackDurationUs = timestampUs; 1905 } 1906 1907 mSampleSizes.push_back(sampleSize); 1908 ++mNumSamples; 1909 if (mNumSamples > 2) { 1910 // We need to use the time scale based ticks, rather than the 1911 // timestamp itself to determine whether we have to use a new 1912 // stts entry, since we may have rounding errors. 1913 // The calculation is intended to reduce the accumulated 1914 // rounding errors. 1915 currDurationTicks = 1916 ((timestampUs * mTimeScale + 500000LL) / 1000000LL - 1917 (lastTimestampUs * mTimeScale + 500000LL) / 1000000LL); 1918 1919 if (currDurationTicks != lastDurationTicks) { 1920 addOneSttsTableEntry(sampleCount, lastDurationUs); 1921 sampleCount = 1; 1922 } else { 1923 ++sampleCount; 1924 } 1925 } 1926 if (mSamplesHaveSameSize) { 1927 if (mNumSamples >= 2 && previousSampleSize != sampleSize) { 1928 mSamplesHaveSameSize = false; 1929 } 1930 previousSampleSize = sampleSize; 1931 } 1932 lastDurationUs = timestampUs - lastTimestampUs; 1933 lastDurationTicks = currDurationTicks; 1934 lastTimestampUs = timestampUs; 1935 1936 if (isSync != 0) { 1937 addOneStssTableEntry(mNumSamples); 1938 } 1939 1940 if (mTrackingProgressStatus) { 1941 if (mPreviousTrackTimeUs <= 0) { 1942 mPreviousTrackTimeUs = mStartTimestampUs; 1943 } 1944 trackProgressStatus(timestampUs); 1945 } 1946 if (mOwner->numTracks() == 1) { 1947 off64_t offset = mIsAvc? mOwner->addLengthPrefixedSample_l(copy) 1948 : mOwner->addSample_l(copy); 1949 if (mChunkOffsets.empty()) { 1950 addChunkOffset(offset); 1951 } 1952 copy->release(); 1953 copy = NULL; 1954 continue; 1955 } 1956 1957 mChunkSamples.push_back(copy); 1958 if (interleaveDurationUs == 0) { 1959 addOneStscTableEntry(++nChunks, 1); 1960 bufferChunk(timestampUs); 1961 } else { 1962 if (chunkTimestampUs == 0) { 1963 chunkTimestampUs = timestampUs; 1964 } else { 1965 if (timestampUs - chunkTimestampUs > interleaveDurationUs) { 1966 ++nChunks; 1967 if (nChunks == 1 || // First chunk 1968 (--(mStscTableEntries.end()))->samplesPerChunk != 1969 mChunkSamples.size()) { 1970 addOneStscTableEntry(nChunks, mChunkSamples.size()); 1971 } 1972 bufferChunk(timestampUs); 1973 chunkTimestampUs = timestampUs; 1974 } 1975 } 1976 } 1977 1978 } 1979 1980 if (mSampleSizes.empty() || // no samples written 1981 (!mIsAudio && mNumStssTableEntries == 0) || // no sync frames for video 1982 (OK != checkCodecSpecificData())) { // no codec specific data 1983 err = ERROR_MALFORMED; 1984 } 1985 mOwner->trackProgressStatus(this, -1, err); 1986 1987 // Last chunk 1988 if (mOwner->numTracks() == 1) { 1989 addOneStscTableEntry(1, mNumSamples); 1990 } else if (!mChunkSamples.empty()) { 1991 addOneStscTableEntry(++nChunks, mChunkSamples.size()); 1992 bufferChunk(timestampUs); 1993 } 1994 1995 // We don't really know how long the last frame lasts, since 1996 // there is no frame time after it, just repeat the previous 1997 // frame's duration. 1998 if (mNumSamples == 1) { 1999 lastDurationUs = 0; // A single sample's duration 2000 } else { 2001 ++sampleCount; // Count for the last sample 2002 } 2003 addOneSttsTableEntry(sampleCount, lastDurationUs); 2004 mTrackDurationUs += lastDurationUs; 2005 mReachedEOS = true; 2006 LOGI("Received total/0-length (%d/%d) buffers and encoded %d frames. - %s", 2007 count, nZeroLengthFrames, mNumSamples, mIsAudio? "audio": "video"); 2008 if (mIsAudio) { 2009 LOGI("Audio track drift time: %lld us", mOwner->getDriftTimeUs()); 2010 } 2011 2012 if (err == ERROR_END_OF_STREAM) { 2013 return OK; 2014 } 2015 return err; 2016} 2017 2018void MPEG4Writer::Track::trackProgressStatus(int64_t timeUs, status_t err) { 2019 LOGV("trackProgressStatus: %lld us", timeUs); 2020 if (mTrackEveryTimeDurationUs > 0 && 2021 timeUs - mPreviousTrackTimeUs >= mTrackEveryTimeDurationUs) { 2022 LOGV("Fire time tracking progress status at %lld us", timeUs); 2023 mOwner->trackProgressStatus(this, timeUs - mPreviousTrackTimeUs, err); 2024 mPreviousTrackTimeUs = timeUs; 2025 } 2026} 2027 2028void MPEG4Writer::trackProgressStatus( 2029 const MPEG4Writer::Track* track, int64_t timeUs, status_t err) { 2030 Mutex::Autolock lock(mLock); 2031 int32_t nTracks = mTracks.size(); 2032 CHECK(nTracks >= 1); 2033 CHECK(nTracks < 64); // Arbitrary number 2034 2035 int32_t trackNum = 0; 2036 CHECK(trackNum < nTracks); 2037 trackNum <<= 16; 2038 2039 // Error notification 2040 // Do not consider ERROR_END_OF_STREAM an error 2041 if (err != OK && err != ERROR_END_OF_STREAM) { 2042 notify(MEDIA_RECORDER_EVENT_ERROR, 2043 trackNum | MEDIA_RECORDER_ERROR_UNKNOWN, 2044 err); 2045 return; 2046 } 2047 2048 if (timeUs == -1) { 2049 // Send completion notification 2050 notify(MEDIA_RECORDER_EVENT_INFO, 2051 trackNum | MEDIA_RECORDER_INFO_COMPLETION_STATUS, 2052 err); 2053 return; 2054 } else { 2055 // Send progress status 2056 notify(MEDIA_RECORDER_EVENT_INFO, 2057 trackNum | MEDIA_RECORDER_INFO_PROGRESS_TIME_STATUS, 2058 timeUs / 1000); 2059 } 2060} 2061 2062void MPEG4Writer::setDriftTimeUs(int64_t driftTimeUs) { 2063 LOGV("setDriftTimeUs: %lld us", driftTimeUs); 2064 Mutex::Autolock autolock(mLock); 2065 mDriftTimeUs = driftTimeUs; 2066} 2067 2068int64_t MPEG4Writer::getDriftTimeUs() { 2069 LOGV("getDriftTimeUs: %lld us", mDriftTimeUs); 2070 Mutex::Autolock autolock(mLock); 2071 return mDriftTimeUs; 2072} 2073 2074bool MPEG4Writer::useNalLengthFour() { 2075 return mUse4ByteNalLength; 2076} 2077 2078void MPEG4Writer::Track::bufferChunk(int64_t timestampUs) { 2079 LOGV("bufferChunk"); 2080 2081 Chunk chunk(this, timestampUs, mChunkSamples); 2082 mOwner->bufferChunk(chunk); 2083 mChunkSamples.clear(); 2084} 2085 2086int64_t MPEG4Writer::Track::getDurationUs() const { 2087 return mTrackDurationUs; 2088} 2089 2090int64_t MPEG4Writer::Track::getEstimatedTrackSizeBytes() const { 2091 return mEstimatedTrackSizeBytes; 2092} 2093 2094status_t MPEG4Writer::Track::checkCodecSpecificData() const { 2095 const char *mime; 2096 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 2097 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime) || 2098 !strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime) || 2099 !strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2100 if (!mCodecSpecificData || 2101 mCodecSpecificDataSize <= 0) { 2102 LOGE("Missing codec specific data"); 2103 return ERROR_MALFORMED; 2104 } 2105 } else { 2106 if (mCodecSpecificData || 2107 mCodecSpecificDataSize > 0) { 2108 LOGE("Unexepected codec specific data found"); 2109 return ERROR_MALFORMED; 2110 } 2111 } 2112 return OK; 2113} 2114 2115void MPEG4Writer::Track::writeTrackHeader( 2116 int32_t trackID, bool use32BitOffset) { 2117 const char *mime; 2118 bool success = mMeta->findCString(kKeyMIMEType, &mime); 2119 CHECK(success); 2120 2121 LOGV("%s track time scale: %d", 2122 mIsAudio? "Audio": "Video", mTimeScale); 2123 2124 time_t now = time(NULL); 2125 int32_t mvhdTimeScale = mOwner->getTimeScale(); 2126 int64_t trakDurationUs = getDurationUs(); 2127 2128 mOwner->beginBox("trak"); 2129 2130 mOwner->beginBox("tkhd"); 2131 // Flags = 7 to indicate that the track is enabled, and 2132 // part of the presentation 2133 mOwner->writeInt32(0x07); // version=0, flags=7 2134 mOwner->writeInt32(now); // creation time 2135 mOwner->writeInt32(now); // modification time 2136 mOwner->writeInt32(trackID); 2137 mOwner->writeInt32(0); // reserved 2138 int32_t tkhdDuration = 2139 (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2140 mOwner->writeInt32(tkhdDuration); // in mvhd timescale 2141 mOwner->writeInt32(0); // reserved 2142 mOwner->writeInt32(0); // reserved 2143 mOwner->writeInt16(0); // layer 2144 mOwner->writeInt16(0); // alternate group 2145 mOwner->writeInt16(mIsAudio ? 0x100 : 0); // volume 2146 mOwner->writeInt16(0); // reserved 2147 2148 mOwner->writeCompositionMatrix(mRotation); // matrix 2149 2150 if (mIsAudio) { 2151 mOwner->writeInt32(0); 2152 mOwner->writeInt32(0); 2153 } else { 2154 int32_t width, height; 2155 bool success = mMeta->findInt32(kKeyWidth, &width); 2156 success = success && mMeta->findInt32(kKeyHeight, &height); 2157 CHECK(success); 2158 2159 mOwner->writeInt32(width << 16); // 32-bit fixed-point value 2160 mOwner->writeInt32(height << 16); // 32-bit fixed-point value 2161 } 2162 mOwner->endBox(); // tkhd 2163 2164 int64_t moovStartTimeUs = mOwner->getStartTimestampUs(); 2165 if (mStartTimestampUs != moovStartTimeUs) { 2166 mOwner->beginBox("edts"); 2167 mOwner->beginBox("elst"); 2168 mOwner->writeInt32(0); // version=0, flags=0: 32-bit time 2169 mOwner->writeInt32(2); // never ends with an empty list 2170 2171 // First elst entry: specify the starting time offset 2172 int64_t offsetUs = mStartTimestampUs - moovStartTimeUs; 2173 LOGV("OffsetUs: %lld", offsetUs); 2174 int32_t seg = (offsetUs * mvhdTimeScale + 5E5) / 1E6; 2175 mOwner->writeInt32(seg); // in mvhd timecale 2176 mOwner->writeInt32(-1); // starting time offset 2177 mOwner->writeInt32(1 << 16); // rate = 1.0 2178 2179 // Second elst entry: specify the track duration 2180 seg = (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2181 mOwner->writeInt32(seg); // in mvhd timescale 2182 mOwner->writeInt32(0); 2183 mOwner->writeInt32(1 << 16); 2184 mOwner->endBox(); 2185 mOwner->endBox(); 2186 } 2187 2188 mOwner->beginBox("mdia"); 2189 2190 mOwner->beginBox("mdhd"); 2191 mOwner->writeInt32(0); // version=0, flags=0 2192 mOwner->writeInt32(now); // creation time 2193 mOwner->writeInt32(now); // modification time 2194 mOwner->writeInt32(mTimeScale); // media timescale 2195 int32_t mdhdDuration = (trakDurationUs * mTimeScale + 5E5) / 1E6; 2196 mOwner->writeInt32(mdhdDuration); // use media timescale 2197 // Language follows the three letter standard ISO-639-2/T 2198 // 'e', 'n', 'g' for "English", for instance. 2199 // Each character is packed as the difference between its ASCII value and 0x60. 2200 // For "English", these are 00101, 01110, 00111. 2201 // XXX: Where is the padding bit located: 0x15C7? 2202 mOwner->writeInt16(0); // language code 2203 mOwner->writeInt16(0); // predefined 2204 mOwner->endBox(); 2205 2206 mOwner->beginBox("hdlr"); 2207 mOwner->writeInt32(0); // version=0, flags=0 2208 mOwner->writeInt32(0); // component type: should be mhlr 2209 mOwner->writeFourcc(mIsAudio ? "soun" : "vide"); // component subtype 2210 mOwner->writeInt32(0); // reserved 2211 mOwner->writeInt32(0); // reserved 2212 mOwner->writeInt32(0); // reserved 2213 // Removing "r" for the name string just makes the string 4 byte aligned 2214 mOwner->writeCString(mIsAudio ? "SoundHandle": "VideoHandle"); // name 2215 mOwner->endBox(); 2216 2217 mOwner->beginBox("minf"); 2218 if (mIsAudio) { 2219 mOwner->beginBox("smhd"); 2220 mOwner->writeInt32(0); // version=0, flags=0 2221 mOwner->writeInt16(0); // balance 2222 mOwner->writeInt16(0); // reserved 2223 mOwner->endBox(); 2224 } else { 2225 mOwner->beginBox("vmhd"); 2226 mOwner->writeInt32(0x01); // version=0, flags=1 2227 mOwner->writeInt16(0); // graphics mode 2228 mOwner->writeInt16(0); // opcolor 2229 mOwner->writeInt16(0); 2230 mOwner->writeInt16(0); 2231 mOwner->endBox(); 2232 } 2233 2234 mOwner->beginBox("dinf"); 2235 mOwner->beginBox("dref"); 2236 mOwner->writeInt32(0); // version=0, flags=0 2237 mOwner->writeInt32(1); // entry count (either url or urn) 2238 // The table index here refers to the sample description index 2239 // in the sample table entries. 2240 mOwner->beginBox("url "); 2241 mOwner->writeInt32(1); // version=0, flags=1 (self-contained) 2242 mOwner->endBox(); // url 2243 mOwner->endBox(); // dref 2244 mOwner->endBox(); // dinf 2245 2246 mOwner->beginBox("stbl"); 2247 2248 mOwner->beginBox("stsd"); 2249 mOwner->writeInt32(0); // version=0, flags=0 2250 mOwner->writeInt32(1); // entry count 2251 if (mIsAudio) { 2252 const char *fourcc = NULL; 2253 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime)) { 2254 fourcc = "samr"; 2255 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2256 fourcc = "sawb"; 2257 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2258 fourcc = "mp4a"; 2259 } else { 2260 LOGE("Unknown mime type '%s'.", mime); 2261 CHECK(!"should not be here, unknown mime type."); 2262 } 2263 2264 mOwner->beginBox(fourcc); // audio format 2265 mOwner->writeInt32(0); // reserved 2266 mOwner->writeInt16(0); // reserved 2267 mOwner->writeInt16(0x1); // data ref index 2268 mOwner->writeInt32(0); // reserved 2269 mOwner->writeInt32(0); // reserved 2270 int32_t nChannels; 2271 CHECK_EQ(true, mMeta->findInt32(kKeyChannelCount, &nChannels)); 2272 mOwner->writeInt16(nChannels); // channel count 2273 mOwner->writeInt16(16); // sample size 2274 mOwner->writeInt16(0); // predefined 2275 mOwner->writeInt16(0); // reserved 2276 2277 int32_t samplerate; 2278 bool success = mMeta->findInt32(kKeySampleRate, &samplerate); 2279 CHECK(success); 2280 mOwner->writeInt32(samplerate << 16); 2281 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2282 mOwner->beginBox("esds"); 2283 CHECK(mCodecSpecificData); 2284 CHECK(mCodecSpecificDataSize > 0); 2285 2286 // Make sure all sizes encode to a single byte. 2287 CHECK(mCodecSpecificDataSize + 23 < 128); 2288 2289 mOwner->writeInt32(0); // version=0, flags=0 2290 mOwner->writeInt8(0x03); // ES_DescrTag 2291 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2292 mOwner->writeInt16(0x0000);// ES_ID 2293 mOwner->writeInt8(0x00); 2294 2295 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2296 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2297 mOwner->writeInt8(0x40); // objectTypeIndication ISO/IEC 14492-2 2298 mOwner->writeInt8(0x15); // streamType AudioStream 2299 2300 mOwner->writeInt16(0x03); // XXX 2301 mOwner->writeInt8(0x00); // buffer size 24-bit 2302 mOwner->writeInt32(96000); // max bit rate 2303 mOwner->writeInt32(96000); // avg bit rate 2304 2305 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2306 mOwner->writeInt8(mCodecSpecificDataSize); 2307 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2308 2309 static const uint8_t kData2[] = { 2310 0x06, // SLConfigDescriptorTag 2311 0x01, 2312 0x02 2313 }; 2314 mOwner->write(kData2, sizeof(kData2)); 2315 2316 mOwner->endBox(); // esds 2317 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime) || 2318 !strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2319 // 3gpp2 Spec AMRSampleEntry fields 2320 mOwner->beginBox("damr"); 2321 mOwner->writeCString(" "); // vendor: 4 bytes 2322 mOwner->writeInt8(0); // decoder version 2323 mOwner->writeInt16(0x83FF); // mode set: all enabled 2324 mOwner->writeInt8(0); // mode change period 2325 mOwner->writeInt8(1); // frames per sample 2326 mOwner->endBox(); 2327 } 2328 mOwner->endBox(); 2329 } else { 2330 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2331 mOwner->beginBox("mp4v"); 2332 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2333 mOwner->beginBox("s263"); 2334 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2335 mOwner->beginBox("avc1"); 2336 } else { 2337 LOGE("Unknown mime type '%s'.", mime); 2338 CHECK(!"should not be here, unknown mime type."); 2339 } 2340 2341 mOwner->writeInt32(0); // reserved 2342 mOwner->writeInt16(0); // reserved 2343 mOwner->writeInt16(1); // data ref index 2344 mOwner->writeInt16(0); // predefined 2345 mOwner->writeInt16(0); // reserved 2346 mOwner->writeInt32(0); // predefined 2347 mOwner->writeInt32(0); // predefined 2348 mOwner->writeInt32(0); // predefined 2349 2350 int32_t width, height; 2351 bool success = mMeta->findInt32(kKeyWidth, &width); 2352 success = success && mMeta->findInt32(kKeyHeight, &height); 2353 CHECK(success); 2354 2355 mOwner->writeInt16(width); 2356 mOwner->writeInt16(height); 2357 mOwner->writeInt32(0x480000); // horiz resolution 2358 mOwner->writeInt32(0x480000); // vert resolution 2359 mOwner->writeInt32(0); // reserved 2360 mOwner->writeInt16(1); // frame count 2361 mOwner->write(" ", 32); 2362 mOwner->writeInt16(0x18); // depth 2363 mOwner->writeInt16(-1); // predefined 2364 2365 CHECK(23 + mCodecSpecificDataSize < 128); 2366 2367 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2368 CHECK(mCodecSpecificData); 2369 CHECK(mCodecSpecificDataSize > 0); 2370 mOwner->beginBox("esds"); 2371 2372 mOwner->writeInt32(0); // version=0, flags=0 2373 2374 mOwner->writeInt8(0x03); // ES_DescrTag 2375 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2376 mOwner->writeInt16(0x0000); // ES_ID 2377 mOwner->writeInt8(0x1f); 2378 2379 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2380 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2381 mOwner->writeInt8(0x20); // objectTypeIndication ISO/IEC 14492-2 2382 mOwner->writeInt8(0x11); // streamType VisualStream 2383 2384 static const uint8_t kData[] = { 2385 0x01, 0x77, 0x00, 2386 0x00, 0x03, 0xe8, 0x00, 2387 0x00, 0x03, 0xe8, 0x00 2388 }; 2389 mOwner->write(kData, sizeof(kData)); 2390 2391 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2392 2393 mOwner->writeInt8(mCodecSpecificDataSize); 2394 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2395 2396 static const uint8_t kData2[] = { 2397 0x06, // SLConfigDescriptorTag 2398 0x01, 2399 0x02 2400 }; 2401 mOwner->write(kData2, sizeof(kData2)); 2402 2403 mOwner->endBox(); // esds 2404 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2405 mOwner->beginBox("d263"); 2406 2407 mOwner->writeInt32(0); // vendor 2408 mOwner->writeInt8(0); // decoder version 2409 mOwner->writeInt8(10); // level: 10 2410 mOwner->writeInt8(0); // profile: 0 2411 2412 mOwner->endBox(); // d263 2413 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2414 CHECK(mCodecSpecificData); 2415 CHECK(mCodecSpecificDataSize >= 5); 2416 2417 // Patch avcc's lengthSize field to match the number 2418 // of bytes we use to indicate the size of a nal unit. 2419 uint8_t *ptr = (uint8_t *)mCodecSpecificData; 2420 ptr[4] = 2421 (ptr[4] & 0xfc) 2422 | (mOwner->useNalLengthFour() ? 3 : 1); 2423 2424 mOwner->beginBox("avcC"); 2425 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2426 mOwner->endBox(); // avcC 2427 } 2428 2429 mOwner->beginBox("pasp"); 2430 // This is useful if the pixel is not square 2431 mOwner->writeInt32(1 << 16); // hspacing 2432 mOwner->writeInt32(1 << 16); // vspacing 2433 mOwner->endBox(); // pasp 2434 mOwner->endBox(); // mp4v, s263 or avc1 2435 } 2436 mOwner->endBox(); // stsd 2437 2438 mOwner->beginBox("stts"); 2439 mOwner->writeInt32(0); // version=0, flags=0 2440 mOwner->writeInt32(mNumSttsTableEntries); 2441 int64_t prevTimestampUs = 0; 2442 for (List<SttsTableEntry>::iterator it = mSttsTableEntries.begin(); 2443 it != mSttsTableEntries.end(); ++it) { 2444 mOwner->writeInt32(it->sampleCount); 2445 2446 // Make sure that we are calculating the sample duration the exactly 2447 // same way as we made decision on how to create stts entries. 2448 int64_t currTimestampUs = prevTimestampUs + it->sampleDurationUs; 2449 int32_t dur = ((currTimestampUs * mTimeScale + 500000LL) / 1000000LL - 2450 (prevTimestampUs * mTimeScale + 500000LL) / 1000000LL); 2451 prevTimestampUs += (it->sampleCount * it->sampleDurationUs); 2452 2453 mOwner->writeInt32(dur); 2454 } 2455 mOwner->endBox(); // stts 2456 2457 if (!mIsAudio) { 2458 mOwner->beginBox("stss"); 2459 mOwner->writeInt32(0); // version=0, flags=0 2460 mOwner->writeInt32(mNumStssTableEntries); // number of sync frames 2461 for (List<int32_t>::iterator it = mStssTableEntries.begin(); 2462 it != mStssTableEntries.end(); ++it) { 2463 mOwner->writeInt32(*it); 2464 } 2465 mOwner->endBox(); // stss 2466 } 2467 2468 mOwner->beginBox("stsz"); 2469 mOwner->writeInt32(0); // version=0, flags=0 2470 if (mSamplesHaveSameSize) { 2471 List<size_t>::iterator it = mSampleSizes.begin(); 2472 mOwner->writeInt32(*it); // default sample size 2473 } else { 2474 mOwner->writeInt32(0); 2475 } 2476 mOwner->writeInt32(mNumSamples); 2477 if (!mSamplesHaveSameSize) { 2478 for (List<size_t>::iterator it = mSampleSizes.begin(); 2479 it != mSampleSizes.end(); ++it) { 2480 mOwner->writeInt32(*it); 2481 } 2482 } 2483 mOwner->endBox(); // stsz 2484 2485 mOwner->beginBox("stsc"); 2486 mOwner->writeInt32(0); // version=0, flags=0 2487 mOwner->writeInt32(mNumStscTableEntries); 2488 for (List<StscTableEntry>::iterator it = mStscTableEntries.begin(); 2489 it != mStscTableEntries.end(); ++it) { 2490 mOwner->writeInt32(it->firstChunk); 2491 mOwner->writeInt32(it->samplesPerChunk); 2492 mOwner->writeInt32(it->sampleDescriptionId); 2493 } 2494 mOwner->endBox(); // stsc 2495 mOwner->beginBox(use32BitOffset? "stco": "co64"); 2496 mOwner->writeInt32(0); // version=0, flags=0 2497 mOwner->writeInt32(mNumStcoTableEntries); 2498 for (List<off64_t>::iterator it = mChunkOffsets.begin(); 2499 it != mChunkOffsets.end(); ++it) { 2500 if (use32BitOffset) { 2501 mOwner->writeInt32(static_cast<int32_t>(*it)); 2502 } else { 2503 mOwner->writeInt64((*it)); 2504 } 2505 } 2506 mOwner->endBox(); // stco or co64 2507 2508 mOwner->endBox(); // stbl 2509 mOwner->endBox(); // minf 2510 mOwner->endBox(); // mdia 2511 mOwner->endBox(); // trak 2512} 2513 2514} // namespace android 2515