MPEG4Writer.cpp revision eaae38445a340c4857c1c5569475879a728e63b7
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 volatile bool mStarted; 82 bool mIsAvc; 83 bool mIsAudio; 84 bool mIsMPEG4; 85 int64_t mTrackDurationUs; 86 87 // For realtime applications, we need to adjust the media clock 88 // for video track based on the audio media clock 89 bool mIsRealTimeRecording; 90 int64_t mMaxTimeStampUs; 91 int64_t mEstimatedTrackSizeBytes; 92 int64_t mMdatSizeBytes; 93 int32_t mTimeScale; 94 95 pthread_t mThread; 96 97 // mNumSamples is used to track how many samples in mSampleSizes List. 98 // This is to reduce the cost associated with mSampleSizes.size() call, 99 // since it is O(n). Ideally, the fix should be in List class. 100 size_t mNumSamples; 101 List<size_t> mSampleSizes; 102 bool mSamplesHaveSameSize; 103 104 List<MediaBuffer *> mChunkSamples; 105 106 size_t mNumStcoTableEntries; 107 List<off64_t> mChunkOffsets; 108 109 size_t mNumStscTableEntries; 110 struct StscTableEntry { 111 112 StscTableEntry(uint32_t chunk, uint32_t samples, uint32_t id) 113 : firstChunk(chunk), 114 samplesPerChunk(samples), 115 sampleDescriptionId(id) {} 116 117 uint32_t firstChunk; 118 uint32_t samplesPerChunk; 119 uint32_t sampleDescriptionId; 120 }; 121 List<StscTableEntry> mStscTableEntries; 122 123 size_t mNumStssTableEntries; 124 List<int32_t> mStssTableEntries; 125 126 size_t mNumSttsTableEntries; 127 struct SttsTableEntry { 128 129 SttsTableEntry(uint32_t count, uint32_t durationUs) 130 : sampleCount(count), sampleDurationUs(durationUs) {} 131 132 uint32_t sampleCount; 133 uint32_t sampleDurationUs; 134 }; 135 List<SttsTableEntry> mSttsTableEntries; 136 137 // Sequence parameter set or picture parameter set 138 struct AVCParamSet { 139 AVCParamSet(uint16_t length, const uint8_t *data) 140 : mLength(length), mData(data) {} 141 142 uint16_t mLength; 143 const uint8_t *mData; 144 }; 145 List<AVCParamSet> mSeqParamSets; 146 List<AVCParamSet> mPicParamSets; 147 uint8_t mProfileIdc; 148 uint8_t mProfileCompatible; 149 uint8_t mLevelIdc; 150 151 void *mCodecSpecificData; 152 size_t mCodecSpecificDataSize; 153 bool mGotAllCodecSpecificData; 154 bool mTrackingProgressStatus; 155 156 bool mReachedEOS; 157 int64_t mStartTimestampUs; 158 int64_t mPreviousTrackTimeUs; 159 int64_t mTrackEveryTimeDurationUs; 160 161 // Has the media time adjustment for video started? 162 bool mIsMediaTimeAdjustmentOn; 163 // The time stamp when previous media time adjustment period starts 164 int64_t mPrevMediaTimeAdjustTimestampUs; 165 // Number of vidoe frames whose time stamp may be adjusted 166 int64_t mMediaTimeAdjustNumFrames; 167 // The sample number when previous meida time adjustmnet period starts 168 int64_t mPrevMediaTimeAdjustSample; 169 // The total accumulated drift time within a period of 170 // kVideoMediaTimeAdjustPeriodTimeUs. 171 int64_t mTotalDriftTimeToAdjustUs; 172 // The total accumalated drift time since the start of the recording 173 // excluding the current time adjustment period 174 int64_t mPrevTotalAccumDriftTimeUs; 175 176 // Update the audio track's drift information. 177 void updateDriftTime(const sp<MetaData>& meta); 178 179 // Adjust the time stamp of the video track according to 180 // the drift time information from the audio track. 181 void adjustMediaTime(int64_t *timestampUs); 182 183 static void *ThreadWrapper(void *me); 184 status_t threadEntry(); 185 186 const uint8_t *parseParamSet( 187 const uint8_t *data, size_t length, int type, size_t *paramSetLen); 188 189 status_t makeAVCCodecSpecificData( 190 const uint8_t *data, size_t size); 191 status_t copyAVCCodecSpecificData( 192 const uint8_t *data, size_t size); 193 status_t parseAVCCodecSpecificData( 194 const uint8_t *data, size_t size); 195 196 // Track authoring progress status 197 void trackProgressStatus(int64_t timeUs, status_t err = OK); 198 void initTrackingProgressStatus(MetaData *params); 199 200 void getCodecSpecificDataFromInputFormatIfPossible(); 201 202 // Determine the track time scale 203 // If it is an audio track, try to use the sampling rate as 204 // the time scale; however, if user chooses the overwrite 205 // value, the user-supplied time scale will be used. 206 void setTimeScale(); 207 208 // Simple validation on the codec specific data 209 status_t checkCodecSpecificData() const; 210 int32_t mRotation; 211 212 void updateTrackSizeEstimate(); 213 void addOneStscTableEntry(size_t chunkId, size_t sampleId); 214 void addOneStssTableEntry(size_t sampleId); 215 void addOneSttsTableEntry(size_t sampleCount, int64_t durationUs); 216 217 Track(const Track &); 218 Track &operator=(const Track &); 219}; 220 221MPEG4Writer::MPEG4Writer(const char *filename) 222 : mFd(-1), 223 mInitCheck(NO_INIT), 224 mUse4ByteNalLength(true), 225 mUse32BitOffset(true), 226 mIsFileSizeLimitExplicitlyRequested(false), 227 mPaused(false), 228 mStarted(false), 229 mOffset(0), 230 mMdatOffset(0), 231 mEstimatedMoovBoxSize(0), 232 mInterleaveDurationUs(1000000) { 233 234 mFd = open(filename, O_CREAT | O_LARGEFILE | O_TRUNC); 235 if (mFd >= 0) { 236 mInitCheck = OK; 237 } 238} 239 240MPEG4Writer::MPEG4Writer(int fd) 241 : mFd(dup(fd)), 242 mInitCheck(mFd < 0? NO_INIT: OK), 243 mUse4ByteNalLength(true), 244 mUse32BitOffset(true), 245 mIsFileSizeLimitExplicitlyRequested(false), 246 mPaused(false), 247 mStarted(false), 248 mOffset(0), 249 mMdatOffset(0), 250 mEstimatedMoovBoxSize(0), 251 mInterleaveDurationUs(1000000) { 252} 253 254MPEG4Writer::~MPEG4Writer() { 255 stop(); 256 257 while (!mTracks.empty()) { 258 List<Track *>::iterator it = mTracks.begin(); 259 delete *it; 260 (*it) = NULL; 261 mTracks.erase(it); 262 } 263 mTracks.clear(); 264} 265 266status_t MPEG4Writer::dump( 267 int fd, const Vector<String16>& args) { 268 const size_t SIZE = 256; 269 char buffer[SIZE]; 270 String8 result; 271 snprintf(buffer, SIZE, " MPEG4Writer %p\n", this); 272 result.append(buffer); 273 snprintf(buffer, SIZE, " mStarted: %s\n", mStarted? "true": "false"); 274 result.append(buffer); 275 ::write(fd, result.string(), result.size()); 276 for (List<Track *>::iterator it = mTracks.begin(); 277 it != mTracks.end(); ++it) { 278 (*it)->dump(fd, args); 279 } 280 return OK; 281} 282 283status_t MPEG4Writer::Track::dump( 284 int fd, const Vector<String16>& args) const { 285 const size_t SIZE = 256; 286 char buffer[SIZE]; 287 String8 result; 288 snprintf(buffer, SIZE, " %s track\n", mIsAudio? "Audio": "Video"); 289 result.append(buffer); 290 snprintf(buffer, SIZE, " reached EOS: %s\n", 291 mReachedEOS? "true": "false"); 292 result.append(buffer); 293 ::write(fd, result.string(), result.size()); 294 return OK; 295} 296 297status_t MPEG4Writer::addSource(const sp<MediaSource> &source) { 298 Track *track = new Track(this, source); 299 mTracks.push_back(track); 300 301 return OK; 302} 303 304status_t MPEG4Writer::startTracks(MetaData *params) { 305 for (List<Track *>::iterator it = mTracks.begin(); 306 it != mTracks.end(); ++it) { 307 status_t err = (*it)->start(params); 308 309 if (err != OK) { 310 for (List<Track *>::iterator it2 = mTracks.begin(); 311 it2 != it; ++it2) { 312 (*it2)->stop(); 313 } 314 315 return err; 316 } 317 } 318 return OK; 319} 320 321int64_t MPEG4Writer::estimateMoovBoxSize(int32_t bitRate) { 322 // This implementation is highly experimental/heurisitic. 323 // 324 // Statistical analysis shows that metadata usually accounts 325 // for a small portion of the total file size, usually < 0.6%. 326 327 // The default MIN_MOOV_BOX_SIZE is set to 0.6% x 1MB / 2, 328 // where 1MB is the common file size limit for MMS application. 329 // The default MAX _MOOV_BOX_SIZE value is based on about 3 330 // minute video recording with a bit rate about 3 Mbps, because 331 // statistics also show that most of the video captured are going 332 // to be less than 3 minutes. 333 334 // If the estimation is wrong, we will pay the price of wasting 335 // some reserved space. This should not happen so often statistically. 336 static const int32_t factor = mUse32BitOffset? 1: 2; 337 static const int64_t MIN_MOOV_BOX_SIZE = 3 * 1024; // 3 KB 338 static const int64_t MAX_MOOV_BOX_SIZE = (180 * 3000000 * 6LL / 8000); 339 int64_t size = MIN_MOOV_BOX_SIZE; 340 341 // Max file size limit is set 342 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 343 size = mMaxFileSizeLimitBytes * 6 / 1000; 344 } 345 346 // Max file duration limit is set 347 if (mMaxFileDurationLimitUs != 0) { 348 if (bitRate > 0) { 349 int64_t size2 = 350 ((mMaxFileDurationLimitUs * bitRate * 6) / 1000 / 8000000); 351 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 352 // When both file size and duration limits are set, 353 // we use the smaller limit of the two. 354 if (size > size2) { 355 size = size2; 356 } 357 } else { 358 // Only max file duration limit is set 359 size = size2; 360 } 361 } 362 } 363 364 if (size < MIN_MOOV_BOX_SIZE) { 365 size = MIN_MOOV_BOX_SIZE; 366 } 367 368 // Any long duration recording will be probably end up with 369 // non-streamable mp4 file. 370 if (size > MAX_MOOV_BOX_SIZE) { 371 size = MAX_MOOV_BOX_SIZE; 372 } 373 374 LOGI("limits: %lld/%lld bytes/us, bit rate: %d bps and the estimated" 375 " moov size %lld bytes", 376 mMaxFileSizeLimitBytes, mMaxFileDurationLimitUs, bitRate, size); 377 return factor * size; 378} 379 380status_t MPEG4Writer::start(MetaData *param) { 381 if (mInitCheck != OK) { 382 return UNKNOWN_ERROR; 383 } 384 385 /* 386 * Check mMaxFileSizeLimitBytes at the beginning 387 * since mMaxFileSizeLimitBytes may be implicitly 388 * changed later for 32-bit file offset even if 389 * user does not ask to set it explicitly. 390 */ 391 if (mMaxFileSizeLimitBytes != 0) { 392 mIsFileSizeLimitExplicitlyRequested = true; 393 } 394 395 int32_t use64BitOffset; 396 if (param && 397 param->findInt32(kKey64BitFileOffset, &use64BitOffset) && 398 use64BitOffset) { 399 mUse32BitOffset = false; 400 } 401 402 if (mUse32BitOffset) { 403 // Implicit 32 bit file size limit 404 if (mMaxFileSizeLimitBytes == 0) { 405 mMaxFileSizeLimitBytes = kMax32BitFileSize; 406 } 407 408 // If file size is set to be larger than the 32 bit file 409 // size limit, treat it as an error. 410 if (mMaxFileSizeLimitBytes > kMax32BitFileSize) { 411 LOGW("32-bit file size limit (%lld bytes) too big. " 412 "It is changed to %lld bytes", 413 mMaxFileSizeLimitBytes, kMax32BitFileSize); 414 mMaxFileSizeLimitBytes = kMax32BitFileSize; 415 } 416 } 417 418 int32_t use2ByteNalLength; 419 if (param && 420 param->findInt32(kKey2ByteNalLength, &use2ByteNalLength) && 421 use2ByteNalLength) { 422 mUse4ByteNalLength = false; 423 } 424 425 mStartTimestampUs = -1; 426 427 if (mStarted) { 428 if (mPaused) { 429 mPaused = false; 430 return startTracks(param); 431 } 432 return OK; 433 } 434 435 if (!param || 436 !param->findInt32(kKeyTimeScale, &mTimeScale)) { 437 mTimeScale = 1000; 438 } 439 CHECK(mTimeScale > 0); 440 LOGV("movie time scale: %d", mTimeScale); 441 442 mStreamableFile = true; 443 mWriteMoovBoxToMemory = false; 444 mMoovBoxBuffer = NULL; 445 mMoovBoxBufferOffset = 0; 446 447 beginBox("ftyp"); 448 { 449 int32_t fileType; 450 if (param && param->findInt32(kKeyFileType, &fileType) && 451 fileType != OUTPUT_FORMAT_MPEG_4) { 452 writeFourcc("3gp4"); 453 } else { 454 writeFourcc("isom"); 455 } 456 } 457 writeInt32(0); 458 writeFourcc("isom"); 459 writeFourcc("3gp4"); 460 endBox(); 461 462 mFreeBoxOffset = mOffset; 463 464 if (mEstimatedMoovBoxSize == 0) { 465 int32_t bitRate = -1; 466 if (param) { 467 param->findInt32(kKeyBitRate, &bitRate); 468 } 469 mEstimatedMoovBoxSize = estimateMoovBoxSize(bitRate); 470 } 471 CHECK(mEstimatedMoovBoxSize >= 8); 472 lseek64(mFd, mFreeBoxOffset, SEEK_SET); 473 writeInt32(mEstimatedMoovBoxSize); 474 write("free", 4); 475 476 mMdatOffset = mFreeBoxOffset + mEstimatedMoovBoxSize; 477 mOffset = mMdatOffset; 478 lseek64(mFd, mMdatOffset, SEEK_SET); 479 if (mUse32BitOffset) { 480 write("????mdat", 8); 481 } else { 482 write("\x00\x00\x00\x01mdat????????", 16); 483 } 484 485 status_t err = startWriterThread(); 486 if (err != OK) { 487 return err; 488 } 489 490 err = startTracks(param); 491 if (err != OK) { 492 return err; 493 } 494 495 mStarted = true; 496 return OK; 497} 498 499bool MPEG4Writer::use32BitFileOffset() const { 500 return mUse32BitOffset; 501} 502 503status_t MPEG4Writer::pause() { 504 if (mInitCheck != OK) { 505 return OK; 506 } 507 mPaused = true; 508 status_t err = OK; 509 for (List<Track *>::iterator it = mTracks.begin(); 510 it != mTracks.end(); ++it) { 511 status_t status = (*it)->pause(); 512 if (status != OK) { 513 err = status; 514 } 515 } 516 return err; 517} 518 519void MPEG4Writer::stopWriterThread() { 520 LOGD("Stopping writer thread"); 521 522 { 523 Mutex::Autolock autolock(mLock); 524 525 mDone = true; 526 mChunkReadyCondition.signal(); 527 } 528 529 void *dummy; 530 pthread_join(mThread, &dummy); 531 LOGD("Writer thread stopped"); 532} 533 534/* 535 * MP4 file standard defines a composition matrix: 536 * | a b u | 537 * | c d v | 538 * | x y w | 539 * 540 * the element in the matrix is stored in the following 541 * order: {a, b, u, c, d, v, x, y, w}, 542 * where a, b, c, d, x, and y is in 16.16 format, while 543 * u, v and w is in 2.30 format. 544 */ 545void MPEG4Writer::writeCompositionMatrix(int degrees) { 546 LOGV("writeCompositionMatrix"); 547 uint32_t a = 0x00010000; 548 uint32_t b = 0; 549 uint32_t c = 0; 550 uint32_t d = 0x00010000; 551 switch (degrees) { 552 case 0: 553 break; 554 case 90: 555 a = 0; 556 b = 0x00010000; 557 c = 0xFFFF0000; 558 d = 0; 559 break; 560 case 180: 561 a = 0xFFFF0000; 562 d = 0xFFFF0000; 563 break; 564 case 270: 565 a = 0; 566 b = 0xFFFF0000; 567 c = 0x00010000; 568 d = 0; 569 break; 570 default: 571 CHECK(!"Should never reach this unknown rotation"); 572 break; 573 } 574 575 writeInt32(a); // a 576 writeInt32(b); // b 577 writeInt32(0); // u 578 writeInt32(c); // c 579 writeInt32(d); // d 580 writeInt32(0); // v 581 writeInt32(0); // x 582 writeInt32(0); // y 583 writeInt32(0x40000000); // w 584} 585 586 587status_t MPEG4Writer::stop() { 588 if (mInitCheck != OK) { 589 return OK; 590 } 591 592 status_t err = OK; 593 int64_t maxDurationUs = 0; 594 int64_t minDurationUs = 0x7fffffffffffffffLL; 595 for (List<Track *>::iterator it = mTracks.begin(); 596 it != mTracks.end(); ++it) { 597 status_t status = (*it)->stop(); 598 if (err == OK && status != OK) { 599 err = status; 600 } 601 602 int64_t durationUs = (*it)->getDurationUs(); 603 if (durationUs > maxDurationUs) { 604 maxDurationUs = durationUs; 605 } 606 if (durationUs < minDurationUs) { 607 minDurationUs = durationUs; 608 } 609 } 610 611 if (mTracks.size() > 1) { 612 LOGD("Duration from tracks range is [%lld, %lld] us", 613 minDurationUs, maxDurationUs); 614 } 615 616 stopWriterThread(); 617 618 // Do not write out movie header on error. 619 if (err != OK) { 620 close(mFd); 621 mFd = -1; 622 mInitCheck = NO_INIT; 623 mStarted = false; 624 return err; 625 } 626 627 // Fix up the size of the 'mdat' chunk. 628 if (mUse32BitOffset) { 629 lseek64(mFd, mMdatOffset, SEEK_SET); 630 int32_t size = htonl(static_cast<int32_t>(mOffset - mMdatOffset)); 631 ::write(mFd, &size, 4); 632 } else { 633 lseek64(mFd, mMdatOffset + 8, SEEK_SET); 634 int64_t size = mOffset - mMdatOffset; 635 size = hton64(size); 636 ::write(mFd, &size, 8); 637 } 638 lseek64(mFd, mOffset, SEEK_SET); 639 640 time_t now = time(NULL); 641 const off64_t moovOffset = mOffset; 642 mWriteMoovBoxToMemory = true; 643 mMoovBoxBuffer = (uint8_t *) malloc(mEstimatedMoovBoxSize); 644 mMoovBoxBufferOffset = 0; 645 CHECK(mMoovBoxBuffer != NULL); 646 int32_t duration = (maxDurationUs * mTimeScale + 5E5) / 1E6; 647 648 beginBox("moov"); 649 650 beginBox("mvhd"); 651 writeInt32(0); // version=0, flags=0 652 writeInt32(now); // creation time 653 writeInt32(now); // modification time 654 writeInt32(mTimeScale); // mvhd timescale 655 writeInt32(duration); 656 writeInt32(0x10000); // rate: 1.0 657 writeInt16(0x100); // volume 658 writeInt16(0); // reserved 659 writeInt32(0); // reserved 660 writeInt32(0); // reserved 661 writeCompositionMatrix(0); // matrix 662 writeInt32(0); // predefined 663 writeInt32(0); // predefined 664 writeInt32(0); // predefined 665 writeInt32(0); // predefined 666 writeInt32(0); // predefined 667 writeInt32(0); // predefined 668 writeInt32(mTracks.size() + 1); // nextTrackID 669 endBox(); // mvhd 670 671 int32_t id = 1; 672 for (List<Track *>::iterator it = mTracks.begin(); 673 it != mTracks.end(); ++it, ++id) { 674 (*it)->writeTrackHeader(id, mUse32BitOffset); 675 } 676 endBox(); // moov 677 678 mWriteMoovBoxToMemory = false; 679 if (mStreamableFile) { 680 CHECK(mMoovBoxBufferOffset + 8 <= mEstimatedMoovBoxSize); 681 682 // Moov box 683 lseek64(mFd, mFreeBoxOffset, SEEK_SET); 684 mOffset = mFreeBoxOffset; 685 write(mMoovBoxBuffer, 1, mMoovBoxBufferOffset); 686 687 // Free box 688 lseek64(mFd, mOffset, SEEK_SET); 689 writeInt32(mEstimatedMoovBoxSize - mMoovBoxBufferOffset); 690 write("free", 4); 691 692 // Free temp memory 693 free(mMoovBoxBuffer); 694 mMoovBoxBuffer = NULL; 695 mMoovBoxBufferOffset = 0; 696 } else { 697 LOGI("The mp4 file will not be streamable."); 698 } 699 700 CHECK(mBoxes.empty()); 701 702 close(mFd); 703 mFd = -1; 704 mInitCheck = NO_INIT; 705 mStarted = false; 706 return err; 707} 708 709status_t MPEG4Writer::setInterleaveDuration(uint32_t durationUs) { 710 mInterleaveDurationUs = durationUs; 711 return OK; 712} 713 714void MPEG4Writer::lock() { 715 mLock.lock(); 716} 717 718void MPEG4Writer::unlock() { 719 mLock.unlock(); 720} 721 722off64_t MPEG4Writer::addSample_l(MediaBuffer *buffer) { 723 off64_t old_offset = mOffset; 724 725 ::write(mFd, 726 (const uint8_t *)buffer->data() + buffer->range_offset(), 727 buffer->range_length()); 728 729 mOffset += buffer->range_length(); 730 731 return old_offset; 732} 733 734static void StripStartcode(MediaBuffer *buffer) { 735 if (buffer->range_length() < 4) { 736 return; 737 } 738 739 const uint8_t *ptr = 740 (const uint8_t *)buffer->data() + buffer->range_offset(); 741 742 if (!memcmp(ptr, "\x00\x00\x00\x01", 4)) { 743 buffer->set_range( 744 buffer->range_offset() + 4, buffer->range_length() - 4); 745 } 746} 747 748off64_t MPEG4Writer::addLengthPrefixedSample_l(MediaBuffer *buffer) { 749 off64_t old_offset = mOffset; 750 751 size_t length = buffer->range_length(); 752 753 if (mUse4ByteNalLength) { 754 uint8_t x = length >> 24; 755 ::write(mFd, &x, 1); 756 x = (length >> 16) & 0xff; 757 ::write(mFd, &x, 1); 758 x = (length >> 8) & 0xff; 759 ::write(mFd, &x, 1); 760 x = length & 0xff; 761 ::write(mFd, &x, 1); 762 763 ::write(mFd, 764 (const uint8_t *)buffer->data() + buffer->range_offset(), 765 length); 766 767 mOffset += length + 4; 768 } else { 769 CHECK(length < 65536); 770 771 uint8_t x = length >> 8; 772 ::write(mFd, &x, 1); 773 x = length & 0xff; 774 ::write(mFd, &x, 1); 775 ::write(mFd, (const uint8_t *)buffer->data() + buffer->range_offset(), length); 776 mOffset += length + 2; 777 } 778 779 return old_offset; 780} 781 782size_t MPEG4Writer::write( 783 const void *ptr, size_t size, size_t nmemb) { 784 785 const size_t bytes = size * nmemb; 786 if (mWriteMoovBoxToMemory) { 787 // This happens only when we write the moov box at the end of 788 // recording, not for each output video/audio frame we receive. 789 off64_t moovBoxSize = 8 + mMoovBoxBufferOffset + bytes; 790 if (moovBoxSize > mEstimatedMoovBoxSize) { 791 for (List<off64_t>::iterator it = mBoxes.begin(); 792 it != mBoxes.end(); ++it) { 793 (*it) += mOffset; 794 } 795 lseek64(mFd, mOffset, SEEK_SET); 796 ::write(mFd, mMoovBoxBuffer, mMoovBoxBufferOffset); 797 ::write(mFd, ptr, size * nmemb); 798 mOffset += (bytes + mMoovBoxBufferOffset); 799 free(mMoovBoxBuffer); 800 mMoovBoxBuffer = NULL; 801 mMoovBoxBufferOffset = 0; 802 mWriteMoovBoxToMemory = false; 803 mStreamableFile = false; 804 } else { 805 memcpy(mMoovBoxBuffer + mMoovBoxBufferOffset, ptr, bytes); 806 mMoovBoxBufferOffset += bytes; 807 } 808 } else { 809 ::write(mFd, ptr, size * nmemb); 810 mOffset += bytes; 811 } 812 return bytes; 813} 814 815void MPEG4Writer::beginBox(const char *fourcc) { 816 CHECK_EQ(strlen(fourcc), 4); 817 818 mBoxes.push_back(mWriteMoovBoxToMemory? 819 mMoovBoxBufferOffset: mOffset); 820 821 writeInt32(0); 822 writeFourcc(fourcc); 823} 824 825void MPEG4Writer::endBox() { 826 CHECK(!mBoxes.empty()); 827 828 off64_t offset = *--mBoxes.end(); 829 mBoxes.erase(--mBoxes.end()); 830 831 if (mWriteMoovBoxToMemory) { 832 int32_t x = htonl(mMoovBoxBufferOffset - offset); 833 memcpy(mMoovBoxBuffer + offset, &x, 4); 834 } else { 835 lseek64(mFd, offset, SEEK_SET); 836 writeInt32(mOffset - offset); 837 mOffset -= 4; 838 lseek64(mFd, mOffset, SEEK_SET); 839 } 840} 841 842void MPEG4Writer::writeInt8(int8_t x) { 843 write(&x, 1, 1); 844} 845 846void MPEG4Writer::writeInt16(int16_t x) { 847 x = htons(x); 848 write(&x, 1, 2); 849} 850 851void MPEG4Writer::writeInt32(int32_t x) { 852 x = htonl(x); 853 write(&x, 1, 4); 854} 855 856void MPEG4Writer::writeInt64(int64_t x) { 857 x = hton64(x); 858 write(&x, 1, 8); 859} 860 861void MPEG4Writer::writeCString(const char *s) { 862 size_t n = strlen(s); 863 write(s, 1, n + 1); 864} 865 866void MPEG4Writer::writeFourcc(const char *s) { 867 CHECK_EQ(strlen(s), 4); 868 write(s, 1, 4); 869} 870 871void MPEG4Writer::write(const void *data, size_t size) { 872 write(data, 1, size); 873} 874 875bool MPEG4Writer::isFileStreamable() const { 876 return mStreamableFile; 877} 878 879bool MPEG4Writer::exceedsFileSizeLimit() { 880 // No limit 881 if (mMaxFileSizeLimitBytes == 0) { 882 return false; 883 } 884 885 int64_t nTotalBytesEstimate = static_cast<int64_t>(mEstimatedMoovBoxSize); 886 for (List<Track *>::iterator it = mTracks.begin(); 887 it != mTracks.end(); ++it) { 888 nTotalBytesEstimate += (*it)->getEstimatedTrackSizeBytes(); 889 } 890 891 // Be conservative in the estimate: do not exceed 95% of 892 // the target file limit. For small target file size limit, though, 893 // this will not help. 894 return (nTotalBytesEstimate >= (95 * mMaxFileSizeLimitBytes) / 100); 895} 896 897bool MPEG4Writer::exceedsFileDurationLimit() { 898 // No limit 899 if (mMaxFileDurationLimitUs == 0) { 900 return false; 901 } 902 903 for (List<Track *>::iterator it = mTracks.begin(); 904 it != mTracks.end(); ++it) { 905 if ((*it)->getDurationUs() >= mMaxFileDurationLimitUs) { 906 return true; 907 } 908 } 909 return false; 910} 911 912bool MPEG4Writer::reachedEOS() { 913 bool allDone = true; 914 for (List<Track *>::iterator it = mTracks.begin(); 915 it != mTracks.end(); ++it) { 916 if (!(*it)->reachedEOS()) { 917 allDone = false; 918 break; 919 } 920 } 921 922 return allDone; 923} 924 925void MPEG4Writer::setStartTimestampUs(int64_t timeUs) { 926 LOGI("setStartTimestampUs: %lld", timeUs); 927 CHECK(timeUs >= 0); 928 Mutex::Autolock autoLock(mLock); 929 if (mStartTimestampUs < 0 || mStartTimestampUs > timeUs) { 930 mStartTimestampUs = timeUs; 931 LOGI("Earliest track starting time: %lld", mStartTimestampUs); 932 } 933} 934 935int64_t MPEG4Writer::getStartTimestampUs() { 936 Mutex::Autolock autoLock(mLock); 937 return mStartTimestampUs; 938} 939 940size_t MPEG4Writer::numTracks() { 941 Mutex::Autolock autolock(mLock); 942 return mTracks.size(); 943} 944 945//////////////////////////////////////////////////////////////////////////////// 946 947MPEG4Writer::Track::Track( 948 MPEG4Writer *owner, const sp<MediaSource> &source) 949 : mOwner(owner), 950 mMeta(source->getFormat()), 951 mSource(source), 952 mDone(false), 953 mPaused(false), 954 mResumed(false), 955 mStarted(false), 956 mTrackDurationUs(0), 957 mEstimatedTrackSizeBytes(0), 958 mSamplesHaveSameSize(true), 959 mCodecSpecificData(NULL), 960 mCodecSpecificDataSize(0), 961 mGotAllCodecSpecificData(false), 962 mReachedEOS(false), 963 mRotation(0) { 964 getCodecSpecificDataFromInputFormatIfPossible(); 965 966 const char *mime; 967 mMeta->findCString(kKeyMIMEType, &mime); 968 mIsAvc = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC); 969 mIsAudio = !strncasecmp(mime, "audio/", 6); 970 mIsMPEG4 = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) || 971 !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC); 972 973 setTimeScale(); 974} 975 976void MPEG4Writer::Track::updateTrackSizeEstimate() { 977 978 int64_t stcoBoxSizeBytes = mOwner->use32BitFileOffset() 979 ? mNumStcoTableEntries * 4 980 : mNumStcoTableEntries * 8; 981 982 int64_t stszBoxSizeBytes = mSamplesHaveSameSize? 4: (mNumSamples * 4); 983 984 mEstimatedTrackSizeBytes = mMdatSizeBytes; // media data size 985 if (!mOwner->isFileStreamable()) { 986 // Reserved free space is not large enough to hold 987 // all meta data and thus wasted. 988 mEstimatedTrackSizeBytes += mNumStscTableEntries * 12 + // stsc box size 989 mNumStssTableEntries * 4 + // stss box size 990 mNumSttsTableEntries * 8 + // stts box size 991 stcoBoxSizeBytes + // stco box size 992 stszBoxSizeBytes; // stsz box size 993 } 994} 995 996void MPEG4Writer::Track::addOneStscTableEntry( 997 size_t chunkId, size_t sampleId) { 998 999 StscTableEntry stscEntry(chunkId, sampleId, 1); 1000 mStscTableEntries.push_back(stscEntry); 1001 ++mNumStscTableEntries; 1002} 1003 1004void MPEG4Writer::Track::addOneStssTableEntry(size_t sampleId) { 1005 mStssTableEntries.push_back(sampleId); 1006 ++mNumStssTableEntries; 1007} 1008 1009void MPEG4Writer::Track::addOneSttsTableEntry( 1010 size_t sampleCount, int64_t durationUs) { 1011 1012 SttsTableEntry sttsEntry(sampleCount, durationUs); 1013 mSttsTableEntries.push_back(sttsEntry); 1014 ++mNumSttsTableEntries; 1015} 1016 1017void MPEG4Writer::Track::addChunkOffset(off64_t offset) { 1018 ++mNumStcoTableEntries; 1019 mChunkOffsets.push_back(offset); 1020} 1021 1022void MPEG4Writer::Track::setTimeScale() { 1023 LOGV("setTimeScale"); 1024 // Default time scale 1025 mTimeScale = 90000; 1026 1027 if (mIsAudio) { 1028 // Use the sampling rate as the default time scale for audio track. 1029 int32_t sampleRate; 1030 bool success = mMeta->findInt32(kKeySampleRate, &sampleRate); 1031 CHECK(success); 1032 mTimeScale = sampleRate; 1033 } 1034 1035 // If someone would like to overwrite the timescale, use user-supplied value. 1036 int32_t timeScale; 1037 if (mMeta->findInt32(kKeyTimeScale, &timeScale)) { 1038 mTimeScale = timeScale; 1039 } 1040 1041 CHECK(mTimeScale > 0); 1042} 1043 1044void MPEG4Writer::Track::getCodecSpecificDataFromInputFormatIfPossible() { 1045 const char *mime; 1046 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 1047 1048 if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC)) { 1049 uint32_t type; 1050 const void *data; 1051 size_t size; 1052 if (mMeta->findData(kKeyAVCC, &type, &data, &size)) { 1053 mCodecSpecificData = malloc(size); 1054 mCodecSpecificDataSize = size; 1055 memcpy(mCodecSpecificData, data, size); 1056 mGotAllCodecSpecificData = true; 1057 } 1058 } else if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) 1059 || !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC)) { 1060 uint32_t type; 1061 const void *data; 1062 size_t size; 1063 if (mMeta->findData(kKeyESDS, &type, &data, &size)) { 1064 ESDS esds(data, size); 1065 if (esds.getCodecSpecificInfo(&data, &size) == OK) { 1066 mCodecSpecificData = malloc(size); 1067 mCodecSpecificDataSize = size; 1068 memcpy(mCodecSpecificData, data, size); 1069 mGotAllCodecSpecificData = true; 1070 } 1071 } 1072 } 1073} 1074 1075MPEG4Writer::Track::~Track() { 1076 stop(); 1077 1078 if (mCodecSpecificData != NULL) { 1079 free(mCodecSpecificData); 1080 mCodecSpecificData = NULL; 1081 } 1082} 1083 1084void MPEG4Writer::Track::initTrackingProgressStatus(MetaData *params) { 1085 LOGV("initTrackingProgressStatus"); 1086 mPreviousTrackTimeUs = -1; 1087 mTrackingProgressStatus = false; 1088 mTrackEveryTimeDurationUs = 0; 1089 { 1090 int64_t timeUs; 1091 if (params && params->findInt64(kKeyTrackTimeStatus, &timeUs)) { 1092 LOGV("Receive request to track progress status for every %lld us", timeUs); 1093 mTrackEveryTimeDurationUs = timeUs; 1094 mTrackingProgressStatus = true; 1095 } 1096 } 1097} 1098 1099// static 1100void *MPEG4Writer::ThreadWrapper(void *me) { 1101 LOGV("ThreadWrapper: %p", me); 1102 MPEG4Writer *writer = static_cast<MPEG4Writer *>(me); 1103 writer->threadFunc(); 1104 return NULL; 1105} 1106 1107void MPEG4Writer::bufferChunk(const Chunk& chunk) { 1108 LOGV("bufferChunk: %p", chunk.mTrack); 1109 Mutex::Autolock autolock(mLock); 1110 CHECK_EQ(mDone, false); 1111 1112 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1113 it != mChunkInfos.end(); ++it) { 1114 1115 if (chunk.mTrack == it->mTrack) { // Found owner 1116 it->mChunks.push_back(chunk); 1117 mChunkReadyCondition.signal(); 1118 return; 1119 } 1120 } 1121 1122 CHECK("Received a chunk for a unknown track" == 0); 1123} 1124 1125void MPEG4Writer::writeFirstChunk(ChunkInfo* info) { 1126 LOGV("writeFirstChunk: %p", info->mTrack); 1127 1128 List<Chunk>::iterator chunkIt = info->mChunks.begin(); 1129 for (List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1130 it != chunkIt->mSamples.end(); ++it) { 1131 1132 off64_t offset = info->mTrack->isAvc() 1133 ? addLengthPrefixedSample_l(*it) 1134 : addSample_l(*it); 1135 if (it == chunkIt->mSamples.begin()) { 1136 info->mTrack->addChunkOffset(offset); 1137 } 1138 } 1139 1140 // Done with the current chunk. 1141 // Release all the samples in this chunk. 1142 while (!chunkIt->mSamples.empty()) { 1143 List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1144 (*it)->release(); 1145 (*it) = NULL; 1146 chunkIt->mSamples.erase(it); 1147 } 1148 chunkIt->mSamples.clear(); 1149 info->mChunks.erase(chunkIt); 1150} 1151 1152void MPEG4Writer::writeChunks() { 1153 LOGV("writeChunks"); 1154 size_t outstandingChunks = 0; 1155 while (!mChunkInfos.empty()) { 1156 List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1157 while (!it->mChunks.empty()) { 1158 CHECK_EQ(OK, writeOneChunk()); 1159 ++outstandingChunks; 1160 } 1161 it->mTrack = NULL; 1162 mChunkInfos.erase(it); 1163 } 1164 mChunkInfos.clear(); 1165 LOGD("%d chunks are written in the last batch", outstandingChunks); 1166} 1167 1168status_t MPEG4Writer::writeOneChunk() { 1169 LOGV("writeOneChunk"); 1170 1171 // Find the smallest timestamp, and write that chunk out 1172 // XXX: What if some track is just too slow? 1173 int64_t minTimestampUs = 0x7FFFFFFFFFFFFFFFLL; 1174 Track *track = NULL; 1175 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1176 it != mChunkInfos.end(); ++it) { 1177 if (!it->mChunks.empty()) { 1178 List<Chunk>::iterator chunkIt = it->mChunks.begin(); 1179 if (chunkIt->mTimeStampUs < minTimestampUs) { 1180 minTimestampUs = chunkIt->mTimeStampUs; 1181 track = it->mTrack; 1182 } 1183 } 1184 } 1185 1186 if (track == NULL) { 1187 LOGV("Nothing to be written after all"); 1188 return OK; 1189 } 1190 1191 if (mIsFirstChunk) { 1192 mIsFirstChunk = false; 1193 } 1194 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1195 it != mChunkInfos.end(); ++it) { 1196 if (it->mTrack == track) { 1197 writeFirstChunk(&(*it)); 1198 } 1199 } 1200 return OK; 1201} 1202 1203void MPEG4Writer::threadFunc() { 1204 LOGV("threadFunc"); 1205 1206 prctl(PR_SET_NAME, (unsigned long)"MPEG4Writer", 0, 0, 0); 1207 while (!mDone) { 1208 { 1209 Mutex::Autolock autolock(mLock); 1210 mChunkReadyCondition.wait(mLock); 1211 CHECK_EQ(writeOneChunk(), OK); 1212 } 1213 } 1214 1215 { 1216 // Write ALL samples 1217 Mutex::Autolock autolock(mLock); 1218 writeChunks(); 1219 } 1220} 1221 1222status_t MPEG4Writer::startWriterThread() { 1223 LOGV("startWriterThread"); 1224 1225 mDone = false; 1226 mIsFirstChunk = true; 1227 mDriftTimeUs = 0; 1228 for (List<Track *>::iterator it = mTracks.begin(); 1229 it != mTracks.end(); ++it) { 1230 ChunkInfo info; 1231 info.mTrack = *it; 1232 mChunkInfos.push_back(info); 1233 } 1234 1235 pthread_attr_t attr; 1236 pthread_attr_init(&attr); 1237 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1238 pthread_create(&mThread, &attr, ThreadWrapper, this); 1239 pthread_attr_destroy(&attr); 1240 return OK; 1241} 1242 1243 1244status_t MPEG4Writer::Track::start(MetaData *params) { 1245 if (!mDone && mPaused) { 1246 mPaused = false; 1247 mResumed = true; 1248 return OK; 1249 } 1250 1251 int64_t startTimeUs; 1252 if (params == NULL || !params->findInt64(kKeyTime, &startTimeUs)) { 1253 startTimeUs = 0; 1254 } 1255 1256 int32_t rotationDegrees; 1257 if (!mIsAudio && params && params->findInt32(kKeyRotation, &rotationDegrees)) { 1258 mRotation = rotationDegrees; 1259 } 1260 1261 mIsRealTimeRecording = true; 1262 { 1263 int32_t isNotRealTime; 1264 if (params && params->findInt32(kKeyNotRealTime, &isNotRealTime)) { 1265 mIsRealTimeRecording = (isNotRealTime == 0); 1266 } 1267 } 1268 1269 initTrackingProgressStatus(params); 1270 1271 sp<MetaData> meta = new MetaData; 1272 meta->setInt64(kKeyTime, startTimeUs); 1273 status_t err = mSource->start(meta.get()); 1274 if (err != OK) { 1275 mDone = mReachedEOS = true; 1276 return err; 1277 } 1278 1279 pthread_attr_t attr; 1280 pthread_attr_init(&attr); 1281 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1282 1283 mDone = false; 1284 mStarted = true; 1285 mTrackDurationUs = 0; 1286 mReachedEOS = false; 1287 mEstimatedTrackSizeBytes = 0; 1288 mNumStcoTableEntries = 0; 1289 mNumStssTableEntries = 0; 1290 mNumStscTableEntries = 0; 1291 mNumSttsTableEntries = 0; 1292 mMdatSizeBytes = 0; 1293 mIsMediaTimeAdjustmentOn = false; 1294 mPrevMediaTimeAdjustTimestampUs = 0; 1295 mMediaTimeAdjustNumFrames = 0; 1296 mPrevMediaTimeAdjustSample = 0; 1297 mTotalDriftTimeToAdjustUs = 0; 1298 mPrevTotalAccumDriftTimeUs = 0; 1299 1300 pthread_create(&mThread, &attr, ThreadWrapper, this); 1301 pthread_attr_destroy(&attr); 1302 1303 return OK; 1304} 1305 1306status_t MPEG4Writer::Track::pause() { 1307 mPaused = true; 1308 return OK; 1309} 1310 1311status_t MPEG4Writer::Track::stop() { 1312 LOGD("Stopping %s track", mIsAudio? "Audio": "Video"); 1313 if (!mStarted) { 1314 LOGE("Stop() called but track is not started"); 1315 return ERROR_END_OF_STREAM; 1316 } 1317 1318 if (mDone) { 1319 return OK; 1320 } 1321 mDone = true; 1322 1323 void *dummy; 1324 pthread_join(mThread, &dummy); 1325 1326 status_t err = (status_t) dummy; 1327 1328 LOGD("Stopping %s track source", mIsAudio? "Audio": "Video"); 1329 { 1330 status_t status = mSource->stop(); 1331 if (err == OK && status != OK && status != ERROR_END_OF_STREAM) { 1332 err = status; 1333 } 1334 } 1335 1336 LOGD("%s track stopped", mIsAudio? "Audio": "Video"); 1337 return err; 1338} 1339 1340bool MPEG4Writer::Track::reachedEOS() { 1341 return mReachedEOS; 1342} 1343 1344// static 1345void *MPEG4Writer::Track::ThreadWrapper(void *me) { 1346 Track *track = static_cast<Track *>(me); 1347 1348 status_t err = track->threadEntry(); 1349 return (void *) err; 1350} 1351 1352static void getNalUnitType(uint8_t byte, uint8_t* type) { 1353 LOGV("getNalUnitType: %d", byte); 1354 1355 // nal_unit_type: 5-bit unsigned integer 1356 *type = (byte & 0x1F); 1357} 1358 1359static const uint8_t *findNextStartCode( 1360 const uint8_t *data, size_t length) { 1361 1362 LOGV("findNextStartCode: %p %d", data, length); 1363 1364 size_t bytesLeft = length; 1365 while (bytesLeft > 4 && 1366 memcmp("\x00\x00\x00\x01", &data[length - bytesLeft], 4)) { 1367 --bytesLeft; 1368 } 1369 if (bytesLeft <= 4) { 1370 bytesLeft = 0; // Last parameter set 1371 } 1372 return &data[length - bytesLeft]; 1373} 1374 1375const uint8_t *MPEG4Writer::Track::parseParamSet( 1376 const uint8_t *data, size_t length, int type, size_t *paramSetLen) { 1377 1378 LOGV("parseParamSet"); 1379 CHECK(type == kNalUnitTypeSeqParamSet || 1380 type == kNalUnitTypePicParamSet); 1381 1382 const uint8_t *nextStartCode = findNextStartCode(data, length); 1383 *paramSetLen = nextStartCode - data; 1384 if (*paramSetLen == 0) { 1385 LOGE("Param set is malformed, since its length is 0"); 1386 return NULL; 1387 } 1388 1389 AVCParamSet paramSet(*paramSetLen, data); 1390 if (type == kNalUnitTypeSeqParamSet) { 1391 if (*paramSetLen < 4) { 1392 LOGE("Seq parameter set malformed"); 1393 return NULL; 1394 } 1395 if (mSeqParamSets.empty()) { 1396 mProfileIdc = data[1]; 1397 mProfileCompatible = data[2]; 1398 mLevelIdc = data[3]; 1399 } else { 1400 if (mProfileIdc != data[1] || 1401 mProfileCompatible != data[2] || 1402 mLevelIdc != data[3]) { 1403 LOGE("Inconsistent profile/level found in seq parameter sets"); 1404 return NULL; 1405 } 1406 } 1407 mSeqParamSets.push_back(paramSet); 1408 } else { 1409 mPicParamSets.push_back(paramSet); 1410 } 1411 return nextStartCode; 1412} 1413 1414status_t MPEG4Writer::Track::copyAVCCodecSpecificData( 1415 const uint8_t *data, size_t size) { 1416 LOGV("copyAVCCodecSpecificData"); 1417 1418 // 2 bytes for each of the parameter set length field 1419 // plus the 7 bytes for the header 1420 if (size < 4 + 7) { 1421 LOGE("Codec specific data length too short: %d", size); 1422 return ERROR_MALFORMED; 1423 } 1424 1425 mCodecSpecificDataSize = size; 1426 mCodecSpecificData = malloc(size); 1427 memcpy(mCodecSpecificData, data, size); 1428 return OK; 1429} 1430 1431status_t MPEG4Writer::Track::parseAVCCodecSpecificData( 1432 const uint8_t *data, size_t size) { 1433 1434 LOGV("parseAVCCodecSpecificData"); 1435 // Data starts with a start code. 1436 // SPS and PPS are separated with start codes. 1437 // Also, SPS must come before PPS 1438 uint8_t type = kNalUnitTypeSeqParamSet; 1439 bool gotSps = false; 1440 bool gotPps = false; 1441 const uint8_t *tmp = data; 1442 const uint8_t *nextStartCode = data; 1443 size_t bytesLeft = size; 1444 size_t paramSetLen = 0; 1445 mCodecSpecificDataSize = 0; 1446 while (bytesLeft > 4 && !memcmp("\x00\x00\x00\x01", tmp, 4)) { 1447 getNalUnitType(*(tmp + 4), &type); 1448 if (type == kNalUnitTypeSeqParamSet) { 1449 if (gotPps) { 1450 LOGE("SPS must come before PPS"); 1451 return ERROR_MALFORMED; 1452 } 1453 if (!gotSps) { 1454 gotSps = true; 1455 } 1456 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1457 } else if (type == kNalUnitTypePicParamSet) { 1458 if (!gotSps) { 1459 LOGE("SPS must come before PPS"); 1460 return ERROR_MALFORMED; 1461 } 1462 if (!gotPps) { 1463 gotPps = true; 1464 } 1465 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1466 } else { 1467 LOGE("Only SPS and PPS Nal units are expected"); 1468 return ERROR_MALFORMED; 1469 } 1470 1471 if (nextStartCode == NULL) { 1472 return ERROR_MALFORMED; 1473 } 1474 1475 // Move on to find the next parameter set 1476 bytesLeft -= nextStartCode - tmp; 1477 tmp = nextStartCode; 1478 mCodecSpecificDataSize += (2 + paramSetLen); 1479 } 1480 1481 { 1482 // Check on the number of seq parameter sets 1483 size_t nSeqParamSets = mSeqParamSets.size(); 1484 if (nSeqParamSets == 0) { 1485 LOGE("Cound not find sequence parameter set"); 1486 return ERROR_MALFORMED; 1487 } 1488 1489 if (nSeqParamSets > 0x1F) { 1490 LOGE("Too many seq parameter sets (%d) found", nSeqParamSets); 1491 return ERROR_MALFORMED; 1492 } 1493 } 1494 1495 { 1496 // Check on the number of pic parameter sets 1497 size_t nPicParamSets = mPicParamSets.size(); 1498 if (nPicParamSets == 0) { 1499 LOGE("Cound not find picture parameter set"); 1500 return ERROR_MALFORMED; 1501 } 1502 if (nPicParamSets > 0xFF) { 1503 LOGE("Too many pic parameter sets (%d) found", nPicParamSets); 1504 return ERROR_MALFORMED; 1505 } 1506 } 1507 1508 { 1509 // Check on the profiles 1510 // These profiles requires additional parameter set extensions 1511 if (mProfileIdc == 100 || mProfileIdc == 110 || 1512 mProfileIdc == 122 || mProfileIdc == 144) { 1513 LOGE("Sorry, no support for profile_idc: %d!", mProfileIdc); 1514 return BAD_VALUE; 1515 } 1516 } 1517 1518 return OK; 1519} 1520 1521status_t MPEG4Writer::Track::makeAVCCodecSpecificData( 1522 const uint8_t *data, size_t size) { 1523 1524 if (mCodecSpecificData != NULL) { 1525 LOGE("Already have codec specific data"); 1526 return ERROR_MALFORMED; 1527 } 1528 1529 if (size < 4) { 1530 LOGE("Codec specific data length too short: %d", size); 1531 return ERROR_MALFORMED; 1532 } 1533 1534 // Data is in the form of AVCCodecSpecificData 1535 if (memcmp("\x00\x00\x00\x01", data, 4)) { 1536 return copyAVCCodecSpecificData(data, size); 1537 } 1538 1539 if (parseAVCCodecSpecificData(data, size) != OK) { 1540 return ERROR_MALFORMED; 1541 } 1542 1543 // ISO 14496-15: AVC file format 1544 mCodecSpecificDataSize += 7; // 7 more bytes in the header 1545 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1546 uint8_t *header = (uint8_t *)mCodecSpecificData; 1547 header[0] = 1; // version 1548 header[1] = mProfileIdc; // profile indication 1549 header[2] = mProfileCompatible; // profile compatibility 1550 header[3] = mLevelIdc; 1551 1552 // 6-bit '111111' followed by 2-bit to lengthSizeMinuusOne 1553 if (mOwner->useNalLengthFour()) { 1554 header[4] = 0xfc | 3; // length size == 4 bytes 1555 } else { 1556 header[4] = 0xfc | 1; // length size == 2 bytes 1557 } 1558 1559 // 3-bit '111' followed by 5-bit numSequenceParameterSets 1560 int nSequenceParamSets = mSeqParamSets.size(); 1561 header[5] = 0xe0 | nSequenceParamSets; 1562 header += 6; 1563 for (List<AVCParamSet>::iterator it = mSeqParamSets.begin(); 1564 it != mSeqParamSets.end(); ++it) { 1565 // 16-bit sequence parameter set length 1566 uint16_t seqParamSetLength = it->mLength; 1567 header[0] = seqParamSetLength >> 8; 1568 header[1] = seqParamSetLength & 0xff; 1569 1570 // SPS NAL unit (sequence parameter length bytes) 1571 memcpy(&header[2], it->mData, seqParamSetLength); 1572 header += (2 + seqParamSetLength); 1573 } 1574 1575 // 8-bit nPictureParameterSets 1576 int nPictureParamSets = mPicParamSets.size(); 1577 header[0] = nPictureParamSets; 1578 header += 1; 1579 for (List<AVCParamSet>::iterator it = mPicParamSets.begin(); 1580 it != mPicParamSets.end(); ++it) { 1581 // 16-bit picture parameter set length 1582 uint16_t picParamSetLength = it->mLength; 1583 header[0] = picParamSetLength >> 8; 1584 header[1] = picParamSetLength & 0xff; 1585 1586 // PPS Nal unit (picture parameter set length bytes) 1587 memcpy(&header[2], it->mData, picParamSetLength); 1588 header += (2 + picParamSetLength); 1589 } 1590 1591 return OK; 1592} 1593 1594/* 1595* The video track's media time adjustment for real-time applications 1596* is described as follows: 1597* 1598* First, the media time adjustment is done for every period of 1599* kVideoMediaTimeAdjustPeriodTimeUs. kVideoMediaTimeAdjustPeriodTimeUs 1600* is currently a fixed value chosen heuristically. The value of 1601* kVideoMediaTimeAdjustPeriodTimeUs should not be very large or very small 1602* for two considerations: on one hand, a relatively large value 1603* helps reduce large fluctuation of drift time in the audio encoding 1604* path; while on the other hand, a relatively small value helps keep 1605* restoring synchronization in audio/video more frequently. Note for the 1606* very first period of kVideoMediaTimeAdjustPeriodTimeUs, there is 1607* no media time adjustment for the video track. 1608* 1609* Second, the total accumulated audio track time drift found 1610* in a period of kVideoMediaTimeAdjustPeriodTimeUs is distributed 1611* over a stream of incoming video frames. The number of video frames 1612* affected is determined based on the number of recorded video frames 1613* within the past kVideoMediaTimeAdjustPeriodTimeUs period. 1614* We choose to distribute the drift time over only a portion 1615* (rather than all) of the total number of recorded video frames 1616* in order to make sure that the video track media time adjustment is 1617* completed for the current period before the next video track media 1618* time adjustment period starts. Currently, the portion chosen is a 1619* half (0.5). 1620* 1621* Last, various additional checks are performed to ensure that 1622* the actual audio encoding path does not have too much drift. 1623* In particular, 1) we want to limit the average incremental time 1624* adjustment for each video frame to be less than a threshold 1625* for a single period of kVideoMediaTimeAdjustPeriodTimeUs. 1626* Currently, the threshold is set to 5 ms. If the average incremental 1627* media time adjustment for a video frame is larger than the 1628* threshold, the audio encoding path has too much time drift. 1629* 2) We also want to limit the total time drift in the audio 1630* encoding path to be less than a threshold for a period of 1631* kVideoMediaTimeAdjustPeriodTimeUs. Currently, the threshold 1632* is 0.5% of kVideoMediaTimeAdjustPeriodTimeUs. If the time drift of 1633* the audio encoding path is larger than the threshold, the audio 1634* encoding path has too much time drift. We treat the large time 1635* drift of the audio encoding path as errors, since there is no 1636* way to keep audio/video in synchronization for real-time 1637* applications if the time drift is too large unless we drop some 1638* video frames, which has its own problems that we don't want 1639* to get into for the time being. 1640*/ 1641void MPEG4Writer::Track::adjustMediaTime(int64_t *timestampUs) { 1642 if (*timestampUs - mPrevMediaTimeAdjustTimestampUs >= 1643 kVideoMediaTimeAdjustPeriodTimeUs) { 1644 1645 LOGV("New media time adjustment period at %lld us", *timestampUs); 1646 mIsMediaTimeAdjustmentOn = true; 1647 mMediaTimeAdjustNumFrames = 1648 (mNumSamples - mPrevMediaTimeAdjustSample) >> 1; 1649 1650 mPrevMediaTimeAdjustTimestampUs = *timestampUs; 1651 mPrevMediaTimeAdjustSample = mNumSamples; 1652 int64_t totalAccumDriftTimeUs = mOwner->getDriftTimeUs(); 1653 mTotalDriftTimeToAdjustUs = 1654 totalAccumDriftTimeUs - mPrevTotalAccumDriftTimeUs; 1655 1656 mPrevTotalAccumDriftTimeUs = totalAccumDriftTimeUs; 1657 1658 // Check on incremental adjusted time per frame 1659 int64_t adjustTimePerFrameUs = 1660 mTotalDriftTimeToAdjustUs / mMediaTimeAdjustNumFrames; 1661 1662 if (adjustTimePerFrameUs < 0) { 1663 adjustTimePerFrameUs = -adjustTimePerFrameUs; 1664 } 1665 if (adjustTimePerFrameUs >= 5000) { 1666 LOGE("Adjusted time per video frame is %lld us", 1667 adjustTimePerFrameUs); 1668 CHECK(!"Video frame time adjustment is too large!"); 1669 } 1670 1671 // Check on total accumulated time drift within a period of 1672 // kVideoMediaTimeAdjustPeriodTimeUs. 1673 int64_t driftPercentage = (mTotalDriftTimeToAdjustUs * 1000) 1674 / kVideoMediaTimeAdjustPeriodTimeUs; 1675 1676 if (driftPercentage < 0) { 1677 driftPercentage = -driftPercentage; 1678 } 1679 if (driftPercentage > 5) { 1680 LOGE("Audio track has time drift %lld us over %lld us", 1681 mTotalDriftTimeToAdjustUs, 1682 kVideoMediaTimeAdjustPeriodTimeUs); 1683 1684 CHECK(!"The audio track media time drifts too much!"); 1685 } 1686 1687 } 1688 1689 if (mIsMediaTimeAdjustmentOn) { 1690 if (mNumSamples - mPrevMediaTimeAdjustSample <= 1691 mMediaTimeAdjustNumFrames) { 1692 1693 // Do media time incremental adjustment 1694 int64_t incrementalAdjustTimeUs = 1695 (mTotalDriftTimeToAdjustUs * 1696 (mNumSamples - mPrevMediaTimeAdjustSample)) 1697 / mMediaTimeAdjustNumFrames; 1698 1699 *timestampUs += 1700 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs); 1701 1702 LOGV("Incremental video frame media time adjustment: %lld us", 1703 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs)); 1704 } else { 1705 // Within the remaining adjustment period, 1706 // no incremental adjustment is needed. 1707 *timestampUs += 1708 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs); 1709 1710 LOGV("Fixed video frame media time adjustment: %lld us", 1711 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs)); 1712 } 1713 } 1714} 1715 1716/* 1717 * Updates the drift time from the audio track so that 1718 * the video track can get the updated drift time information 1719 * from the file writer. The fluctuation of the drift time of the audio 1720 * encoding path is smoothed out with a simple filter by giving a larger 1721 * weight to more recently drift time. The filter coefficients, 0.5 and 0.5, 1722 * are heuristically determined. 1723 */ 1724void MPEG4Writer::Track::updateDriftTime(const sp<MetaData>& meta) { 1725 int64_t driftTimeUs = 0; 1726 if (meta->findInt64(kKeyDriftTime, &driftTimeUs)) { 1727 int64_t prevDriftTimeUs = mOwner->getDriftTimeUs(); 1728 int64_t timeUs = (driftTimeUs + prevDriftTimeUs) >> 1; 1729 mOwner->setDriftTimeUs(timeUs); 1730 } 1731} 1732 1733status_t MPEG4Writer::Track::threadEntry() { 1734 int32_t count = 0; 1735 const int64_t interleaveDurationUs = mOwner->interleaveDuration(); 1736 int64_t chunkTimestampUs = 0; 1737 int32_t nChunks = 0; 1738 int32_t nZeroLengthFrames = 0; 1739 int64_t lastTimestampUs = 0; // Previous sample time stamp in ms 1740 int64_t lastDurationUs = 0; // Between the previous two samples in ms 1741 int64_t currDurationTicks = 0; // Timescale based ticks 1742 int64_t lastDurationTicks = 0; // Timescale based ticks 1743 int32_t sampleCount = 1; // Sample count in the current stts table entry 1744 uint32_t previousSampleSize = 0; // Size of the previous sample 1745 int64_t previousPausedDurationUs = 0; 1746 int64_t timestampUs; 1747 1748 if (mIsAudio) { 1749 prctl(PR_SET_NAME, (unsigned long)"AudioTrackEncoding", 0, 0, 0); 1750 } else { 1751 prctl(PR_SET_NAME, (unsigned long)"VideoTrackEncoding", 0, 0, 0); 1752 } 1753 setpriority(PRIO_PROCESS, 0, ANDROID_PRIORITY_AUDIO); 1754 1755 sp<MetaData> meta_data; 1756 1757 mNumSamples = 0; 1758 status_t err = OK; 1759 MediaBuffer *buffer; 1760 while (!mDone && (err = mSource->read(&buffer)) == OK) { 1761 if (buffer->range_length() == 0) { 1762 buffer->release(); 1763 buffer = NULL; 1764 ++nZeroLengthFrames; 1765 continue; 1766 } 1767 1768 // If the codec specific data has not been received yet, delay pause. 1769 // After the codec specific data is received, discard what we received 1770 // when the track is to be paused. 1771 if (mPaused && !mResumed) { 1772 buffer->release(); 1773 buffer = NULL; 1774 continue; 1775 } 1776 1777 ++count; 1778 1779 int32_t isCodecConfig; 1780 if (buffer->meta_data()->findInt32(kKeyIsCodecConfig, &isCodecConfig) 1781 && isCodecConfig) { 1782 CHECK(!mGotAllCodecSpecificData); 1783 1784 if (mIsAvc) { 1785 status_t err = makeAVCCodecSpecificData( 1786 (const uint8_t *)buffer->data() 1787 + buffer->range_offset(), 1788 buffer->range_length()); 1789 CHECK_EQ(OK, err); 1790 } else if (mIsMPEG4) { 1791 mCodecSpecificDataSize = buffer->range_length(); 1792 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1793 memcpy(mCodecSpecificData, 1794 (const uint8_t *)buffer->data() 1795 + buffer->range_offset(), 1796 buffer->range_length()); 1797 } 1798 1799 buffer->release(); 1800 buffer = NULL; 1801 1802 mGotAllCodecSpecificData = true; 1803 continue; 1804 } 1805 1806 // Make a deep copy of the MediaBuffer and Metadata and release 1807 // the original as soon as we can 1808 MediaBuffer *copy = new MediaBuffer(buffer->range_length()); 1809 memcpy(copy->data(), (uint8_t *)buffer->data() + buffer->range_offset(), 1810 buffer->range_length()); 1811 copy->set_range(0, buffer->range_length()); 1812 meta_data = new MetaData(*buffer->meta_data().get()); 1813 buffer->release(); 1814 buffer = NULL; 1815 1816 if (mIsAvc) StripStartcode(copy); 1817 1818 size_t sampleSize = copy->range_length(); 1819 if (mIsAvc) { 1820 if (mOwner->useNalLengthFour()) { 1821 sampleSize += 4; 1822 } else { 1823 sampleSize += 2; 1824 } 1825 } 1826 1827 // Max file size or duration handling 1828 mMdatSizeBytes += sampleSize; 1829 updateTrackSizeEstimate(); 1830 1831 if (mOwner->exceedsFileSizeLimit()) { 1832 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_FILESIZE_REACHED, 0); 1833 break; 1834 } 1835 if (mOwner->exceedsFileDurationLimit()) { 1836 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_DURATION_REACHED, 0); 1837 break; 1838 } 1839 1840 1841 int32_t isSync = false; 1842 meta_data->findInt32(kKeyIsSyncFrame, &isSync); 1843 1844 /* 1845 * The original timestamp found in the data buffer will be modified as below: 1846 * 1847 * There is a playback offset into this track if the track's start time 1848 * is not the same as the movie start time, which will be recorded in edst 1849 * box of the output file. The playback offset is to make sure that the 1850 * starting time of the audio/video tracks are synchronized. Although the 1851 * track's media timestamp may be subject to various modifications 1852 * as outlined below, the track's playback offset time remains unchanged 1853 * once the first data buffer of the track is received. 1854 * 1855 * The media time stamp will be calculated by subtracting the playback offset 1856 * (and potential pause durations) from the original timestamp in the buffer. 1857 * 1858 * If this track is a video track for a real-time recording application with 1859 * both audio and video tracks, its media timestamp will subject to further 1860 * modification based on the media clock of the audio track. This modification 1861 * is needed for the purpose of maintaining good audio/video synchronization. 1862 * 1863 * If the recording session is paused and resumed multiple times, the track 1864 * media timestamp will be modified as if the recording session had never been 1865 * paused at all during playback of the recorded output file. In other words, 1866 * the output file will have no memory of pause/resume durations. 1867 * 1868 */ 1869 CHECK(meta_data->findInt64(kKeyTime, ×tampUs)); 1870 LOGV("%s timestampUs: %lld", mIsAudio? "Audio": "Video", timestampUs); 1871 1872//////////////////////////////////////////////////////////////////////////////// 1873 if (mSampleSizes.empty()) { 1874 mStartTimestampUs = timestampUs; 1875 mOwner->setStartTimestampUs(mStartTimestampUs); 1876 previousPausedDurationUs = mStartTimestampUs; 1877 } 1878 1879 if (mResumed) { 1880 int64_t durExcludingEarlierPausesUs = timestampUs - previousPausedDurationUs; 1881 CHECK(durExcludingEarlierPausesUs >= 0); 1882 int64_t pausedDurationUs = durExcludingEarlierPausesUs - mTrackDurationUs; 1883 CHECK(pausedDurationUs >= lastDurationUs); 1884 previousPausedDurationUs += pausedDurationUs - lastDurationUs; 1885 mResumed = false; 1886 } 1887 1888 timestampUs -= previousPausedDurationUs; 1889 CHECK(timestampUs >= 0); 1890 1891 // Media time adjustment for real-time applications 1892 if (mIsRealTimeRecording) { 1893 if (mIsAudio) { 1894 updateDriftTime(meta_data); 1895 } else { 1896 adjustMediaTime(×tampUs); 1897 } 1898 } 1899 1900 CHECK(timestampUs >= 0); 1901 if (mNumSamples > 1) { 1902 if (timestampUs <= lastTimestampUs) { 1903 LOGW("Frame arrives too late!"); 1904 // Don't drop the late frame, since dropping a frame may cause 1905 // problems later during playback 1906 1907 // The idea here is to avoid having two or more samples with the 1908 // same timestamp in the output file. 1909 if (mTimeScale >= 1000000LL) { 1910 timestampUs = lastTimestampUs + 1; 1911 } else { 1912 timestampUs = lastTimestampUs + (1000000LL + (mTimeScale >> 1)) / mTimeScale; 1913 } 1914 } 1915 } 1916 1917 LOGV("%s media time stamp: %lld and previous paused duration %lld", 1918 mIsAudio? "Audio": "Video", timestampUs, previousPausedDurationUs); 1919 if (timestampUs > mTrackDurationUs) { 1920 mTrackDurationUs = timestampUs; 1921 } 1922 1923 mSampleSizes.push_back(sampleSize); 1924 ++mNumSamples; 1925 if (mNumSamples > 2) { 1926 // We need to use the time scale based ticks, rather than the 1927 // timestamp itself to determine whether we have to use a new 1928 // stts entry, since we may have rounding errors. 1929 // The calculation is intended to reduce the accumulated 1930 // rounding errors. 1931 currDurationTicks = 1932 ((timestampUs * mTimeScale + 500000LL) / 1000000LL - 1933 (lastTimestampUs * mTimeScale + 500000LL) / 1000000LL); 1934 1935 if (currDurationTicks != lastDurationTicks) { 1936 addOneSttsTableEntry(sampleCount, lastDurationUs); 1937 sampleCount = 1; 1938 } else { 1939 ++sampleCount; 1940 } 1941 } 1942 if (mSamplesHaveSameSize) { 1943 if (mNumSamples >= 2 && previousSampleSize != sampleSize) { 1944 mSamplesHaveSameSize = false; 1945 } 1946 previousSampleSize = sampleSize; 1947 } 1948 lastDurationUs = timestampUs - lastTimestampUs; 1949 lastDurationTicks = currDurationTicks; 1950 lastTimestampUs = timestampUs; 1951 1952 if (isSync != 0) { 1953 addOneStssTableEntry(mNumSamples); 1954 } 1955 1956 if (mTrackingProgressStatus) { 1957 if (mPreviousTrackTimeUs <= 0) { 1958 mPreviousTrackTimeUs = mStartTimestampUs; 1959 } 1960 trackProgressStatus(timestampUs); 1961 } 1962 if (mOwner->numTracks() == 1) { 1963 off64_t offset = mIsAvc? mOwner->addLengthPrefixedSample_l(copy) 1964 : mOwner->addSample_l(copy); 1965 if (mChunkOffsets.empty()) { 1966 addChunkOffset(offset); 1967 } 1968 copy->release(); 1969 copy = NULL; 1970 continue; 1971 } 1972 1973 mChunkSamples.push_back(copy); 1974 if (interleaveDurationUs == 0) { 1975 addOneStscTableEntry(++nChunks, 1); 1976 bufferChunk(timestampUs); 1977 } else { 1978 if (chunkTimestampUs == 0) { 1979 chunkTimestampUs = timestampUs; 1980 } else { 1981 if (timestampUs - chunkTimestampUs > interleaveDurationUs) { 1982 ++nChunks; 1983 if (nChunks == 1 || // First chunk 1984 (--(mStscTableEntries.end()))->samplesPerChunk != 1985 mChunkSamples.size()) { 1986 addOneStscTableEntry(nChunks, mChunkSamples.size()); 1987 } 1988 bufferChunk(timestampUs); 1989 chunkTimestampUs = timestampUs; 1990 } 1991 } 1992 } 1993 1994 } 1995 1996 if (mSampleSizes.empty() || // no samples written 1997 (!mIsAudio && mNumStssTableEntries == 0) || // no sync frames for video 1998 (OK != checkCodecSpecificData())) { // no codec specific data 1999 err = ERROR_MALFORMED; 2000 } 2001 mOwner->trackProgressStatus(this, -1, err); 2002 2003 // Last chunk 2004 if (mOwner->numTracks() == 1) { 2005 addOneStscTableEntry(1, mNumSamples); 2006 } else if (!mChunkSamples.empty()) { 2007 addOneStscTableEntry(++nChunks, mChunkSamples.size()); 2008 bufferChunk(timestampUs); 2009 } 2010 2011 // We don't really know how long the last frame lasts, since 2012 // there is no frame time after it, just repeat the previous 2013 // frame's duration. 2014 if (mNumSamples == 1) { 2015 lastDurationUs = 0; // A single sample's duration 2016 } else { 2017 ++sampleCount; // Count for the last sample 2018 } 2019 addOneSttsTableEntry(sampleCount, lastDurationUs); 2020 mTrackDurationUs += lastDurationUs; 2021 mReachedEOS = true; 2022 LOGI("Received total/0-length (%d/%d) buffers and encoded %d frames. - %s", 2023 count, nZeroLengthFrames, mNumSamples, mIsAudio? "audio": "video"); 2024 if (mIsAudio) { 2025 LOGI("Audio track drift time: %lld us", mOwner->getDriftTimeUs()); 2026 } 2027 2028 if (err == ERROR_END_OF_STREAM) { 2029 return OK; 2030 } 2031 return err; 2032} 2033 2034void MPEG4Writer::Track::trackProgressStatus(int64_t timeUs, status_t err) { 2035 LOGV("trackProgressStatus: %lld us", timeUs); 2036 if (mTrackEveryTimeDurationUs > 0 && 2037 timeUs - mPreviousTrackTimeUs >= mTrackEveryTimeDurationUs) { 2038 LOGV("Fire time tracking progress status at %lld us", timeUs); 2039 mOwner->trackProgressStatus(this, timeUs - mPreviousTrackTimeUs, err); 2040 mPreviousTrackTimeUs = timeUs; 2041 } 2042} 2043 2044void MPEG4Writer::trackProgressStatus( 2045 const MPEG4Writer::Track* track, int64_t timeUs, status_t err) { 2046 Mutex::Autolock lock(mLock); 2047 int32_t nTracks = mTracks.size(); 2048 CHECK(nTracks >= 1); 2049 CHECK(nTracks < 64); // Arbitrary number 2050 2051 int32_t trackNum = 0; 2052 CHECK(trackNum < nTracks); 2053 trackNum <<= 16; 2054 2055 // Error notification 2056 // Do not consider ERROR_END_OF_STREAM an error 2057 if (err != OK && err != ERROR_END_OF_STREAM) { 2058 notify(MEDIA_RECORDER_EVENT_ERROR, 2059 trackNum | MEDIA_RECORDER_ERROR_UNKNOWN, 2060 err); 2061 return; 2062 } 2063 2064 if (timeUs == -1) { 2065 // Send completion notification 2066 notify(MEDIA_RECORDER_EVENT_INFO, 2067 trackNum | MEDIA_RECORDER_INFO_COMPLETION_STATUS, 2068 err); 2069 return; 2070 } else { 2071 // Send progress status 2072 notify(MEDIA_RECORDER_EVENT_INFO, 2073 trackNum | MEDIA_RECORDER_INFO_PROGRESS_TIME_STATUS, 2074 timeUs / 1000); 2075 } 2076} 2077 2078void MPEG4Writer::setDriftTimeUs(int64_t driftTimeUs) { 2079 LOGV("setDriftTimeUs: %lld us", driftTimeUs); 2080 Mutex::Autolock autolock(mLock); 2081 mDriftTimeUs = driftTimeUs; 2082} 2083 2084int64_t MPEG4Writer::getDriftTimeUs() { 2085 LOGV("getDriftTimeUs: %lld us", mDriftTimeUs); 2086 Mutex::Autolock autolock(mLock); 2087 return mDriftTimeUs; 2088} 2089 2090bool MPEG4Writer::useNalLengthFour() { 2091 return mUse4ByteNalLength; 2092} 2093 2094void MPEG4Writer::Track::bufferChunk(int64_t timestampUs) { 2095 LOGV("bufferChunk"); 2096 2097 Chunk chunk(this, timestampUs, mChunkSamples); 2098 mOwner->bufferChunk(chunk); 2099 mChunkSamples.clear(); 2100} 2101 2102int64_t MPEG4Writer::Track::getDurationUs() const { 2103 return mTrackDurationUs; 2104} 2105 2106int64_t MPEG4Writer::Track::getEstimatedTrackSizeBytes() const { 2107 return mEstimatedTrackSizeBytes; 2108} 2109 2110status_t MPEG4Writer::Track::checkCodecSpecificData() const { 2111 const char *mime; 2112 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 2113 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime) || 2114 !strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime) || 2115 !strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2116 if (!mCodecSpecificData || 2117 mCodecSpecificDataSize <= 0) { 2118 LOGE("Missing codec specific data"); 2119 return ERROR_MALFORMED; 2120 } 2121 } else { 2122 if (mCodecSpecificData || 2123 mCodecSpecificDataSize > 0) { 2124 LOGE("Unexepected codec specific data found"); 2125 return ERROR_MALFORMED; 2126 } 2127 } 2128 return OK; 2129} 2130 2131void MPEG4Writer::Track::writeTrackHeader( 2132 int32_t trackID, bool use32BitOffset) { 2133 const char *mime; 2134 bool success = mMeta->findCString(kKeyMIMEType, &mime); 2135 CHECK(success); 2136 2137 LOGV("%s track time scale: %d", 2138 mIsAudio? "Audio": "Video", mTimeScale); 2139 2140 time_t now = time(NULL); 2141 int32_t mvhdTimeScale = mOwner->getTimeScale(); 2142 int64_t trakDurationUs = getDurationUs(); 2143 2144 mOwner->beginBox("trak"); 2145 2146 mOwner->beginBox("tkhd"); 2147 // Flags = 7 to indicate that the track is enabled, and 2148 // part of the presentation 2149 mOwner->writeInt32(0x07); // version=0, flags=7 2150 mOwner->writeInt32(now); // creation time 2151 mOwner->writeInt32(now); // modification time 2152 mOwner->writeInt32(trackID); 2153 mOwner->writeInt32(0); // reserved 2154 int32_t tkhdDuration = 2155 (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2156 mOwner->writeInt32(tkhdDuration); // in mvhd timescale 2157 mOwner->writeInt32(0); // reserved 2158 mOwner->writeInt32(0); // reserved 2159 mOwner->writeInt16(0); // layer 2160 mOwner->writeInt16(0); // alternate group 2161 mOwner->writeInt16(mIsAudio ? 0x100 : 0); // volume 2162 mOwner->writeInt16(0); // reserved 2163 2164 mOwner->writeCompositionMatrix(mRotation); // matrix 2165 2166 if (mIsAudio) { 2167 mOwner->writeInt32(0); 2168 mOwner->writeInt32(0); 2169 } else { 2170 int32_t width, height; 2171 bool success = mMeta->findInt32(kKeyWidth, &width); 2172 success = success && mMeta->findInt32(kKeyHeight, &height); 2173 CHECK(success); 2174 2175 mOwner->writeInt32(width << 16); // 32-bit fixed-point value 2176 mOwner->writeInt32(height << 16); // 32-bit fixed-point value 2177 } 2178 mOwner->endBox(); // tkhd 2179 2180 int64_t moovStartTimeUs = mOwner->getStartTimestampUs(); 2181 if (mStartTimestampUs != moovStartTimeUs) { 2182 mOwner->beginBox("edts"); 2183 mOwner->beginBox("elst"); 2184 mOwner->writeInt32(0); // version=0, flags=0: 32-bit time 2185 mOwner->writeInt32(2); // never ends with an empty list 2186 2187 // First elst entry: specify the starting time offset 2188 int64_t offsetUs = mStartTimestampUs - moovStartTimeUs; 2189 LOGV("OffsetUs: %lld", offsetUs); 2190 int32_t seg = (offsetUs * mvhdTimeScale + 5E5) / 1E6; 2191 mOwner->writeInt32(seg); // in mvhd timecale 2192 mOwner->writeInt32(-1); // starting time offset 2193 mOwner->writeInt32(1 << 16); // rate = 1.0 2194 2195 // Second elst entry: specify the track duration 2196 seg = (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2197 mOwner->writeInt32(seg); // in mvhd timescale 2198 mOwner->writeInt32(0); 2199 mOwner->writeInt32(1 << 16); 2200 mOwner->endBox(); 2201 mOwner->endBox(); 2202 } 2203 2204 mOwner->beginBox("mdia"); 2205 2206 mOwner->beginBox("mdhd"); 2207 mOwner->writeInt32(0); // version=0, flags=0 2208 mOwner->writeInt32(now); // creation time 2209 mOwner->writeInt32(now); // modification time 2210 mOwner->writeInt32(mTimeScale); // media timescale 2211 int32_t mdhdDuration = (trakDurationUs * mTimeScale + 5E5) / 1E6; 2212 mOwner->writeInt32(mdhdDuration); // use media timescale 2213 // Language follows the three letter standard ISO-639-2/T 2214 // 'e', 'n', 'g' for "English", for instance. 2215 // Each character is packed as the difference between its ASCII value and 0x60. 2216 // For "English", these are 00101, 01110, 00111. 2217 // XXX: Where is the padding bit located: 0x15C7? 2218 mOwner->writeInt16(0); // language code 2219 mOwner->writeInt16(0); // predefined 2220 mOwner->endBox(); 2221 2222 mOwner->beginBox("hdlr"); 2223 mOwner->writeInt32(0); // version=0, flags=0 2224 mOwner->writeInt32(0); // component type: should be mhlr 2225 mOwner->writeFourcc(mIsAudio ? "soun" : "vide"); // component subtype 2226 mOwner->writeInt32(0); // reserved 2227 mOwner->writeInt32(0); // reserved 2228 mOwner->writeInt32(0); // reserved 2229 // Removing "r" for the name string just makes the string 4 byte aligned 2230 mOwner->writeCString(mIsAudio ? "SoundHandle": "VideoHandle"); // name 2231 mOwner->endBox(); 2232 2233 mOwner->beginBox("minf"); 2234 if (mIsAudio) { 2235 mOwner->beginBox("smhd"); 2236 mOwner->writeInt32(0); // version=0, flags=0 2237 mOwner->writeInt16(0); // balance 2238 mOwner->writeInt16(0); // reserved 2239 mOwner->endBox(); 2240 } else { 2241 mOwner->beginBox("vmhd"); 2242 mOwner->writeInt32(0x01); // version=0, flags=1 2243 mOwner->writeInt16(0); // graphics mode 2244 mOwner->writeInt16(0); // opcolor 2245 mOwner->writeInt16(0); 2246 mOwner->writeInt16(0); 2247 mOwner->endBox(); 2248 } 2249 2250 mOwner->beginBox("dinf"); 2251 mOwner->beginBox("dref"); 2252 mOwner->writeInt32(0); // version=0, flags=0 2253 mOwner->writeInt32(1); // entry count (either url or urn) 2254 // The table index here refers to the sample description index 2255 // in the sample table entries. 2256 mOwner->beginBox("url "); 2257 mOwner->writeInt32(1); // version=0, flags=1 (self-contained) 2258 mOwner->endBox(); // url 2259 mOwner->endBox(); // dref 2260 mOwner->endBox(); // dinf 2261 2262 mOwner->beginBox("stbl"); 2263 2264 mOwner->beginBox("stsd"); 2265 mOwner->writeInt32(0); // version=0, flags=0 2266 mOwner->writeInt32(1); // entry count 2267 if (mIsAudio) { 2268 const char *fourcc = NULL; 2269 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime)) { 2270 fourcc = "samr"; 2271 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2272 fourcc = "sawb"; 2273 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2274 fourcc = "mp4a"; 2275 } else { 2276 LOGE("Unknown mime type '%s'.", mime); 2277 CHECK(!"should not be here, unknown mime type."); 2278 } 2279 2280 mOwner->beginBox(fourcc); // audio format 2281 mOwner->writeInt32(0); // reserved 2282 mOwner->writeInt16(0); // reserved 2283 mOwner->writeInt16(0x1); // data ref index 2284 mOwner->writeInt32(0); // reserved 2285 mOwner->writeInt32(0); // reserved 2286 int32_t nChannels; 2287 CHECK_EQ(true, mMeta->findInt32(kKeyChannelCount, &nChannels)); 2288 mOwner->writeInt16(nChannels); // channel count 2289 mOwner->writeInt16(16); // sample size 2290 mOwner->writeInt16(0); // predefined 2291 mOwner->writeInt16(0); // reserved 2292 2293 int32_t samplerate; 2294 bool success = mMeta->findInt32(kKeySampleRate, &samplerate); 2295 CHECK(success); 2296 mOwner->writeInt32(samplerate << 16); 2297 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2298 mOwner->beginBox("esds"); 2299 CHECK(mCodecSpecificData); 2300 CHECK(mCodecSpecificDataSize > 0); 2301 2302 // Make sure all sizes encode to a single byte. 2303 CHECK(mCodecSpecificDataSize + 23 < 128); 2304 2305 mOwner->writeInt32(0); // version=0, flags=0 2306 mOwner->writeInt8(0x03); // ES_DescrTag 2307 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2308 mOwner->writeInt16(0x0000);// ES_ID 2309 mOwner->writeInt8(0x00); 2310 2311 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2312 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2313 mOwner->writeInt8(0x40); // objectTypeIndication ISO/IEC 14492-2 2314 mOwner->writeInt8(0x15); // streamType AudioStream 2315 2316 mOwner->writeInt16(0x03); // XXX 2317 mOwner->writeInt8(0x00); // buffer size 24-bit 2318 mOwner->writeInt32(96000); // max bit rate 2319 mOwner->writeInt32(96000); // avg bit rate 2320 2321 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2322 mOwner->writeInt8(mCodecSpecificDataSize); 2323 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2324 2325 static const uint8_t kData2[] = { 2326 0x06, // SLConfigDescriptorTag 2327 0x01, 2328 0x02 2329 }; 2330 mOwner->write(kData2, sizeof(kData2)); 2331 2332 mOwner->endBox(); // esds 2333 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime) || 2334 !strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2335 // 3gpp2 Spec AMRSampleEntry fields 2336 mOwner->beginBox("damr"); 2337 mOwner->writeCString(" "); // vendor: 4 bytes 2338 mOwner->writeInt8(0); // decoder version 2339 mOwner->writeInt16(0x83FF); // mode set: all enabled 2340 mOwner->writeInt8(0); // mode change period 2341 mOwner->writeInt8(1); // frames per sample 2342 mOwner->endBox(); 2343 } 2344 mOwner->endBox(); 2345 } else { 2346 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2347 mOwner->beginBox("mp4v"); 2348 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2349 mOwner->beginBox("s263"); 2350 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2351 mOwner->beginBox("avc1"); 2352 } else { 2353 LOGE("Unknown mime type '%s'.", mime); 2354 CHECK(!"should not be here, unknown mime type."); 2355 } 2356 2357 mOwner->writeInt32(0); // reserved 2358 mOwner->writeInt16(0); // reserved 2359 mOwner->writeInt16(1); // data ref index 2360 mOwner->writeInt16(0); // predefined 2361 mOwner->writeInt16(0); // reserved 2362 mOwner->writeInt32(0); // predefined 2363 mOwner->writeInt32(0); // predefined 2364 mOwner->writeInt32(0); // predefined 2365 2366 int32_t width, height; 2367 bool success = mMeta->findInt32(kKeyWidth, &width); 2368 success = success && mMeta->findInt32(kKeyHeight, &height); 2369 CHECK(success); 2370 2371 mOwner->writeInt16(width); 2372 mOwner->writeInt16(height); 2373 mOwner->writeInt32(0x480000); // horiz resolution 2374 mOwner->writeInt32(0x480000); // vert resolution 2375 mOwner->writeInt32(0); // reserved 2376 mOwner->writeInt16(1); // frame count 2377 mOwner->write(" ", 32); 2378 mOwner->writeInt16(0x18); // depth 2379 mOwner->writeInt16(-1); // predefined 2380 2381 CHECK(23 + mCodecSpecificDataSize < 128); 2382 2383 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2384 CHECK(mCodecSpecificData); 2385 CHECK(mCodecSpecificDataSize > 0); 2386 mOwner->beginBox("esds"); 2387 2388 mOwner->writeInt32(0); // version=0, flags=0 2389 2390 mOwner->writeInt8(0x03); // ES_DescrTag 2391 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2392 mOwner->writeInt16(0x0000); // ES_ID 2393 mOwner->writeInt8(0x1f); 2394 2395 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2396 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2397 mOwner->writeInt8(0x20); // objectTypeIndication ISO/IEC 14492-2 2398 mOwner->writeInt8(0x11); // streamType VisualStream 2399 2400 static const uint8_t kData[] = { 2401 0x01, 0x77, 0x00, 2402 0x00, 0x03, 0xe8, 0x00, 2403 0x00, 0x03, 0xe8, 0x00 2404 }; 2405 mOwner->write(kData, sizeof(kData)); 2406 2407 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2408 2409 mOwner->writeInt8(mCodecSpecificDataSize); 2410 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2411 2412 static const uint8_t kData2[] = { 2413 0x06, // SLConfigDescriptorTag 2414 0x01, 2415 0x02 2416 }; 2417 mOwner->write(kData2, sizeof(kData2)); 2418 2419 mOwner->endBox(); // esds 2420 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2421 mOwner->beginBox("d263"); 2422 2423 mOwner->writeInt32(0); // vendor 2424 mOwner->writeInt8(0); // decoder version 2425 mOwner->writeInt8(10); // level: 10 2426 mOwner->writeInt8(0); // profile: 0 2427 2428 mOwner->endBox(); // d263 2429 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2430 CHECK(mCodecSpecificData); 2431 CHECK(mCodecSpecificDataSize >= 5); 2432 2433 // Patch avcc's lengthSize field to match the number 2434 // of bytes we use to indicate the size of a nal unit. 2435 uint8_t *ptr = (uint8_t *)mCodecSpecificData; 2436 ptr[4] = 2437 (ptr[4] & 0xfc) 2438 | (mOwner->useNalLengthFour() ? 3 : 1); 2439 2440 mOwner->beginBox("avcC"); 2441 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2442 mOwner->endBox(); // avcC 2443 } 2444 2445 mOwner->beginBox("pasp"); 2446 // This is useful if the pixel is not square 2447 mOwner->writeInt32(1 << 16); // hspacing 2448 mOwner->writeInt32(1 << 16); // vspacing 2449 mOwner->endBox(); // pasp 2450 mOwner->endBox(); // mp4v, s263 or avc1 2451 } 2452 mOwner->endBox(); // stsd 2453 2454 mOwner->beginBox("stts"); 2455 mOwner->writeInt32(0); // version=0, flags=0 2456 mOwner->writeInt32(mNumSttsTableEntries); 2457 int64_t prevTimestampUs = 0; 2458 for (List<SttsTableEntry>::iterator it = mSttsTableEntries.begin(); 2459 it != mSttsTableEntries.end(); ++it) { 2460 mOwner->writeInt32(it->sampleCount); 2461 2462 // Make sure that we are calculating the sample duration the exactly 2463 // same way as we made decision on how to create stts entries. 2464 int64_t currTimestampUs = prevTimestampUs + it->sampleDurationUs; 2465 int32_t dur = ((currTimestampUs * mTimeScale + 500000LL) / 1000000LL - 2466 (prevTimestampUs * mTimeScale + 500000LL) / 1000000LL); 2467 prevTimestampUs += (it->sampleCount * it->sampleDurationUs); 2468 2469 mOwner->writeInt32(dur); 2470 } 2471 mOwner->endBox(); // stts 2472 2473 if (!mIsAudio) { 2474 mOwner->beginBox("stss"); 2475 mOwner->writeInt32(0); // version=0, flags=0 2476 mOwner->writeInt32(mNumStssTableEntries); // number of sync frames 2477 for (List<int32_t>::iterator it = mStssTableEntries.begin(); 2478 it != mStssTableEntries.end(); ++it) { 2479 mOwner->writeInt32(*it); 2480 } 2481 mOwner->endBox(); // stss 2482 } 2483 2484 mOwner->beginBox("stsz"); 2485 mOwner->writeInt32(0); // version=0, flags=0 2486 if (mSamplesHaveSameSize) { 2487 List<size_t>::iterator it = mSampleSizes.begin(); 2488 mOwner->writeInt32(*it); // default sample size 2489 } else { 2490 mOwner->writeInt32(0); 2491 } 2492 mOwner->writeInt32(mNumSamples); 2493 if (!mSamplesHaveSameSize) { 2494 for (List<size_t>::iterator it = mSampleSizes.begin(); 2495 it != mSampleSizes.end(); ++it) { 2496 mOwner->writeInt32(*it); 2497 } 2498 } 2499 mOwner->endBox(); // stsz 2500 2501 mOwner->beginBox("stsc"); 2502 mOwner->writeInt32(0); // version=0, flags=0 2503 mOwner->writeInt32(mNumStscTableEntries); 2504 for (List<StscTableEntry>::iterator it = mStscTableEntries.begin(); 2505 it != mStscTableEntries.end(); ++it) { 2506 mOwner->writeInt32(it->firstChunk); 2507 mOwner->writeInt32(it->samplesPerChunk); 2508 mOwner->writeInt32(it->sampleDescriptionId); 2509 } 2510 mOwner->endBox(); // stsc 2511 mOwner->beginBox(use32BitOffset? "stco": "co64"); 2512 mOwner->writeInt32(0); // version=0, flags=0 2513 mOwner->writeInt32(mNumStcoTableEntries); 2514 for (List<off64_t>::iterator it = mChunkOffsets.begin(); 2515 it != mChunkOffsets.end(); ++it) { 2516 if (use32BitOffset) { 2517 mOwner->writeInt32(static_cast<int32_t>(*it)); 2518 } else { 2519 mOwner->writeInt64((*it)); 2520 } 2521 } 2522 mOwner->endBox(); // stco or co64 2523 2524 mOwner->endBox(); // stbl 2525 mOwner->endBox(); // minf 2526 mOwner->endBox(); // mdia 2527 mOwner->endBox(); // trak 2528} 2529 2530} // namespace android 2531