MPEG4Writer.cpp revision c7fc37a3dab9bd1f96713649f351b5990e6316ff
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 37#include "include/ESDS.h" 38 39namespace android { 40 41static const int64_t kMax32BitFileSize = 0x007fffffffLL; 42static const uint8_t kNalUnitTypeSeqParamSet = 0x07; 43static const uint8_t kNalUnitTypePicParamSet = 0x08; 44 45// Using longer adjustment period to suppress fluctuations in 46// the audio encoding paths 47static const int64_t kVideoMediaTimeAdjustPeriodTimeUs = 600000000LL; // 10 minutes 48 49class MPEG4Writer::Track { 50public: 51 Track(MPEG4Writer *owner, const sp<MediaSource> &source); 52 53 ~Track(); 54 55 status_t start(MetaData *params); 56 status_t stop(); 57 status_t pause(); 58 bool reachedEOS(); 59 60 int64_t getDurationUs() const; 61 int64_t getEstimatedTrackSizeBytes() const; 62 void writeTrackHeader(int32_t trackID, bool use32BitOffset = true); 63 void bufferChunk(int64_t timestampUs); 64 bool isAvc() const { return mIsAvc; } 65 bool isAudio() const { return mIsAudio; } 66 bool isMPEG4() const { return mIsMPEG4; } 67 void addChunkOffset(off64_t offset); 68 status_t dump(int fd, const Vector<String16>& args) const; 69 70private: 71 MPEG4Writer *mOwner; 72 sp<MetaData> mMeta; 73 sp<MediaSource> mSource; 74 volatile bool mDone; 75 volatile bool mPaused; 76 volatile bool mResumed; 77 bool mIsAvc; 78 bool mIsAudio; 79 bool mIsMPEG4; 80 int64_t mTrackDurationUs; 81 82 // For realtime applications, we need to adjust the media clock 83 // for video track based on the audio media clock 84 bool mIsRealTimeRecording; 85 int64_t mMaxTimeStampUs; 86 int64_t mEstimatedTrackSizeBytes; 87 int64_t mMdatSizeBytes; 88 int32_t mTimeScale; 89 90 pthread_t mThread; 91 92 // mNumSamples is used to track how many samples in mSampleSizes List. 93 // This is to reduce the cost associated with mSampleSizes.size() call, 94 // since it is O(n). Ideally, the fix should be in List class. 95 size_t mNumSamples; 96 List<size_t> mSampleSizes; 97 bool mSamplesHaveSameSize; 98 99 List<MediaBuffer *> mChunkSamples; 100 101 size_t mNumStcoTableEntries; 102 List<off64_t> mChunkOffsets; 103 104 size_t mNumStscTableEntries; 105 struct StscTableEntry { 106 107 StscTableEntry(uint32_t chunk, uint32_t samples, uint32_t id) 108 : firstChunk(chunk), 109 samplesPerChunk(samples), 110 sampleDescriptionId(id) {} 111 112 uint32_t firstChunk; 113 uint32_t samplesPerChunk; 114 uint32_t sampleDescriptionId; 115 }; 116 List<StscTableEntry> mStscTableEntries; 117 118 size_t mNumStssTableEntries; 119 List<int32_t> mStssTableEntries; 120 121 size_t mNumSttsTableEntries; 122 struct SttsTableEntry { 123 124 SttsTableEntry(uint32_t count, uint32_t durationUs) 125 : sampleCount(count), sampleDurationUs(durationUs) {} 126 127 uint32_t sampleCount; 128 uint32_t sampleDurationUs; 129 }; 130 List<SttsTableEntry> mSttsTableEntries; 131 132 // Sequence parameter set or picture parameter set 133 struct AVCParamSet { 134 AVCParamSet(uint16_t length, const uint8_t *data) 135 : mLength(length), mData(data) {} 136 137 uint16_t mLength; 138 const uint8_t *mData; 139 }; 140 List<AVCParamSet> mSeqParamSets; 141 List<AVCParamSet> mPicParamSets; 142 uint8_t mProfileIdc; 143 uint8_t mProfileCompatible; 144 uint8_t mLevelIdc; 145 146 void *mCodecSpecificData; 147 size_t mCodecSpecificDataSize; 148 bool mGotAllCodecSpecificData; 149 bool mTrackingProgressStatus; 150 151 bool mReachedEOS; 152 int64_t mStartTimestampUs; 153 int64_t mPreviousTrackTimeUs; 154 int64_t mTrackEveryTimeDurationUs; 155 156 // Has the media time adjustment for video started? 157 bool mIsMediaTimeAdjustmentOn; 158 // The time stamp when previous media time adjustment period starts 159 int64_t mPrevMediaTimeAdjustTimestampUs; 160 // Number of vidoe frames whose time stamp may be adjusted 161 int64_t mMediaTimeAdjustNumFrames; 162 // The sample number when previous meida time adjustmnet period starts 163 int64_t mPrevMediaTimeAdjustSample; 164 // The total accumulated drift time within a period of 165 // kVideoMediaTimeAdjustPeriodTimeUs. 166 int64_t mTotalDriftTimeToAdjustUs; 167 // The total accumalated drift time since the start of the recording 168 // excluding the current time adjustment period 169 int64_t mPrevTotalAccumDriftTimeUs; 170 171 // Update the audio track's drift information. 172 void updateDriftTime(const sp<MetaData>& meta); 173 174 // Adjust the time stamp of the video track according to 175 // the drift time information from the audio track. 176 void adjustMediaTime(int64_t *timestampUs); 177 178 static void *ThreadWrapper(void *me); 179 status_t threadEntry(); 180 181 const uint8_t *parseParamSet( 182 const uint8_t *data, size_t length, int type, size_t *paramSetLen); 183 184 status_t makeAVCCodecSpecificData( 185 const uint8_t *data, size_t size); 186 status_t copyAVCCodecSpecificData( 187 const uint8_t *data, size_t size); 188 status_t parseAVCCodecSpecificData( 189 const uint8_t *data, size_t size); 190 191 // Track authoring progress status 192 void trackProgressStatus(int64_t timeUs, status_t err = OK); 193 void initTrackingProgressStatus(MetaData *params); 194 195 void getCodecSpecificDataFromInputFormatIfPossible(); 196 197 // Determine the track time scale 198 // If it is an audio track, try to use the sampling rate as 199 // the time scale; however, if user chooses the overwrite 200 // value, the user-supplied time scale will be used. 201 void setTimeScale(); 202 203 // Simple validation on the codec specific data 204 status_t checkCodecSpecificData() const; 205 int32_t mRotation; 206 207 void updateTrackSizeEstimate(); 208 void addOneStscTableEntry(size_t chunkId, size_t sampleId); 209 void addOneStssTableEntry(size_t sampleId); 210 void addOneSttsTableEntry(size_t sampleCount, int64_t durationUs); 211 212 Track(const Track &); 213 Track &operator=(const Track &); 214}; 215 216MPEG4Writer::MPEG4Writer(const char *filename) 217 : mFile(fopen(filename, "wb")), 218 mFd(mFile == NULL? -1: fileno(mFile)), 219 mUse4ByteNalLength(true), 220 mUse32BitOffset(true), 221 mIsFileSizeLimitExplicitlyRequested(false), 222 mPaused(false), 223 mStarted(false), 224 mOffset(0), 225 mMdatOffset(0), 226 mEstimatedMoovBoxSize(0), 227 mInterleaveDurationUs(1000000) { 228 CHECK(mFile != NULL); 229} 230 231MPEG4Writer::MPEG4Writer(int fd) 232 : mFile(fdopen(fd, "wb")), 233 mFd(mFile == NULL? -1: fileno(mFile)), 234 mUse4ByteNalLength(true), 235 mUse32BitOffset(true), 236 mIsFileSizeLimitExplicitlyRequested(false), 237 mPaused(false), 238 mStarted(false), 239 mOffset(0), 240 mMdatOffset(0), 241 mEstimatedMoovBoxSize(0), 242 mInterleaveDurationUs(1000000) { 243 CHECK(mFile != NULL); 244} 245 246MPEG4Writer::~MPEG4Writer() { 247 stop(); 248 249 while (!mTracks.empty()) { 250 List<Track *>::iterator it = mTracks.begin(); 251 delete *it; 252 (*it) = NULL; 253 mTracks.erase(it); 254 } 255 mTracks.clear(); 256} 257 258status_t MPEG4Writer::dump( 259 int fd, const Vector<String16>& args) { 260 const size_t SIZE = 256; 261 char buffer[SIZE]; 262 String8 result; 263 snprintf(buffer, SIZE, " MPEG4Writer %p\n", this); 264 result.append(buffer); 265 snprintf(buffer, SIZE, " mStarted: %s\n", mStarted? "true": "false"); 266 result.append(buffer); 267 ::write(fd, result.string(), result.size()); 268 for (List<Track *>::iterator it = mTracks.begin(); 269 it != mTracks.end(); ++it) { 270 (*it)->dump(fd, args); 271 } 272 return OK; 273} 274 275status_t MPEG4Writer::Track::dump( 276 int fd, const Vector<String16>& args) const { 277 const size_t SIZE = 256; 278 char buffer[SIZE]; 279 String8 result; 280 snprintf(buffer, SIZE, " %s track\n", mIsAudio? "Audio": "Video"); 281 result.append(buffer); 282 snprintf(buffer, SIZE, " reached EOS: %s\n", 283 mReachedEOS? "true": "false"); 284 result.append(buffer); 285 ::write(fd, result.string(), result.size()); 286 return OK; 287} 288 289status_t MPEG4Writer::addSource(const sp<MediaSource> &source) { 290 Track *track = new Track(this, source); 291 mTracks.push_back(track); 292 293 return OK; 294} 295 296status_t MPEG4Writer::startTracks(MetaData *params) { 297 for (List<Track *>::iterator it = mTracks.begin(); 298 it != mTracks.end(); ++it) { 299 status_t err = (*it)->start(params); 300 301 if (err != OK) { 302 for (List<Track *>::iterator it2 = mTracks.begin(); 303 it2 != it; ++it2) { 304 (*it2)->stop(); 305 } 306 307 return err; 308 } 309 } 310 return OK; 311} 312 313int64_t MPEG4Writer::estimateMoovBoxSize(int32_t bitRate) { 314 // This implementation is highly experimental/heurisitic. 315 // 316 // Statistical analysis shows that metadata usually accounts 317 // for a small portion of the total file size, usually < 0.6%. 318 319 // The default MIN_MOOV_BOX_SIZE is set to 0.6% x 1MB / 2, 320 // where 1MB is the common file size limit for MMS application. 321 // The default MAX _MOOV_BOX_SIZE value is based on about 3 322 // minute video recording with a bit rate about 3 Mbps, because 323 // statistics also show that most of the video captured are going 324 // to be less than 3 minutes. 325 326 // If the estimation is wrong, we will pay the price of wasting 327 // some reserved space. This should not happen so often statistically. 328 static const int32_t factor = mUse32BitOffset? 1: 2; 329 static const int64_t MIN_MOOV_BOX_SIZE = 3 * 1024; // 3 KB 330 static const int64_t MAX_MOOV_BOX_SIZE = (180 * 3000000 * 6LL / 8000); 331 int64_t size = MIN_MOOV_BOX_SIZE; 332 333 // Max file size limit is set 334 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 335 size = mMaxFileSizeLimitBytes * 6 / 1000; 336 } 337 338 // Max file duration limit is set 339 if (mMaxFileDurationLimitUs != 0) { 340 if (bitRate > 0) { 341 int64_t size2 = 342 ((mMaxFileDurationLimitUs * bitRate * 6) / 1000 / 8000000); 343 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 344 // When both file size and duration limits are set, 345 // we use the smaller limit of the two. 346 if (size > size2) { 347 size = size2; 348 } 349 } else { 350 // Only max file duration limit is set 351 size = size2; 352 } 353 } 354 } 355 356 if (size < MIN_MOOV_BOX_SIZE) { 357 size = MIN_MOOV_BOX_SIZE; 358 } 359 360 // Any long duration recording will be probably end up with 361 // non-streamable mp4 file. 362 if (size > MAX_MOOV_BOX_SIZE) { 363 size = MAX_MOOV_BOX_SIZE; 364 } 365 366 LOGI("limits: %lld/%lld bytes/us, bit rate: %d bps and the estimated" 367 " moov size %lld bytes", 368 mMaxFileSizeLimitBytes, mMaxFileDurationLimitUs, bitRate, size); 369 return factor * size; 370} 371 372status_t MPEG4Writer::start(MetaData *param) { 373 if (mFile == NULL) { 374 return UNKNOWN_ERROR; 375 } 376 377 /* 378 * Check mMaxFileSizeLimitBytes at the beginning 379 * since mMaxFileSizeLimitBytes may be implicitly 380 * changed later for 32-bit file offset even if 381 * user does not ask to set it explicitly. 382 */ 383 if (mMaxFileSizeLimitBytes != 0) { 384 mIsFileSizeLimitExplicitlyRequested = true; 385 } 386 387 int32_t use64BitOffset; 388 if (param && 389 param->findInt32(kKey64BitFileOffset, &use64BitOffset) && 390 use64BitOffset) { 391 mUse32BitOffset = false; 392 } 393 394 if (mUse32BitOffset) { 395 // Implicit 32 bit file size limit 396 if (mMaxFileSizeLimitBytes == 0) { 397 mMaxFileSizeLimitBytes = kMax32BitFileSize; 398 } 399 400 // If file size is set to be larger than the 32 bit file 401 // size limit, treat it as an error. 402 if (mMaxFileSizeLimitBytes > kMax32BitFileSize) { 403 LOGW("32-bit file size limit (%lld bytes) too big. " 404 "It is changed to %lld bytes", 405 mMaxFileSizeLimitBytes, kMax32BitFileSize); 406 mMaxFileSizeLimitBytes = kMax32BitFileSize; 407 } 408 } 409 410 int32_t use2ByteNalLength; 411 if (param && 412 param->findInt32(kKey2ByteNalLength, &use2ByteNalLength) && 413 use2ByteNalLength) { 414 mUse4ByteNalLength = false; 415 } 416 417 mStartTimestampUs = -1; 418 419 if (mStarted) { 420 if (mPaused) { 421 mPaused = false; 422 return startTracks(param); 423 } 424 return OK; 425 } 426 427 if (!param || 428 !param->findInt32(kKeyTimeScale, &mTimeScale)) { 429 mTimeScale = 1000; 430 } 431 CHECK(mTimeScale > 0); 432 LOGV("movie time scale: %d", mTimeScale); 433 434 mStreamableFile = true; 435 mWriteMoovBoxToMemory = false; 436 mMoovBoxBuffer = NULL; 437 mMoovBoxBufferOffset = 0; 438 439 beginBox("ftyp"); 440 { 441 int32_t fileType; 442 if (param && param->findInt32(kKeyFileType, &fileType) && 443 fileType != OUTPUT_FORMAT_MPEG_4) { 444 writeFourcc("3gp4"); 445 } else { 446 writeFourcc("isom"); 447 } 448 } 449 writeInt32(0); 450 writeFourcc("isom"); 451 writeFourcc("3gp4"); 452 endBox(); 453 454 mFreeBoxOffset = mOffset; 455 456 if (mEstimatedMoovBoxSize == 0) { 457 int32_t bitRate = -1; 458 if (param) { 459 param->findInt32(kKeyBitRate, &bitRate); 460 } 461 mEstimatedMoovBoxSize = estimateMoovBoxSize(bitRate); 462 } 463 CHECK(mEstimatedMoovBoxSize >= 8); 464 lseek64(mFd, mFreeBoxOffset, SEEK_SET); 465 writeInt32(mEstimatedMoovBoxSize); 466 write("free", 4); 467 468 mMdatOffset = mFreeBoxOffset + mEstimatedMoovBoxSize; 469 mOffset = mMdatOffset; 470 lseek64(mFd, mMdatOffset, SEEK_SET); 471 if (mUse32BitOffset) { 472 write("????mdat", 8); 473 } else { 474 write("\x00\x00\x00\x01mdat????????", 16); 475 } 476 477 status_t err = startWriterThread(); 478 if (err != OK) { 479 return err; 480 } 481 482 err = startTracks(param); 483 if (err != OK) { 484 return err; 485 } 486 487 mStarted = true; 488 return OK; 489} 490 491bool MPEG4Writer::use32BitFileOffset() const { 492 return mUse32BitOffset; 493} 494 495status_t MPEG4Writer::pause() { 496 if (mFile == NULL) { 497 return OK; 498 } 499 mPaused = true; 500 status_t err = OK; 501 for (List<Track *>::iterator it = mTracks.begin(); 502 it != mTracks.end(); ++it) { 503 status_t status = (*it)->pause(); 504 if (status != OK) { 505 err = status; 506 } 507 } 508 return err; 509} 510 511void MPEG4Writer::stopWriterThread() { 512 LOGD("Stopping writer thread"); 513 514 { 515 Mutex::Autolock autolock(mLock); 516 517 mDone = true; 518 mChunkReadyCondition.signal(); 519 } 520 521 void *dummy; 522 pthread_join(mThread, &dummy); 523 LOGD("Writer thread stopped"); 524} 525 526/* 527 * MP4 file standard defines a composition matrix: 528 * | a b u | 529 * | c d v | 530 * | x y w | 531 * 532 * the element in the matrix is stored in the following 533 * order: {a, b, u, c, d, v, x, y, w}, 534 * where a, b, c, d, x, and y is in 16.16 format, while 535 * u, v and w is in 2.30 format. 536 */ 537void MPEG4Writer::writeCompositionMatrix(int degrees) { 538 LOGV("writeCompositionMatrix"); 539 uint32_t a = 0x00010000; 540 uint32_t b = 0; 541 uint32_t c = 0; 542 uint32_t d = 0x00010000; 543 switch (degrees) { 544 case 0: 545 break; 546 case 90: 547 a = 0; 548 b = 0x00010000; 549 c = 0xFFFF0000; 550 d = 0; 551 break; 552 case 180: 553 a = 0xFFFF0000; 554 d = 0xFFFF0000; 555 break; 556 case 270: 557 a = 0; 558 b = 0xFFFF0000; 559 c = 0x00010000; 560 d = 0; 561 break; 562 default: 563 CHECK(!"Should never reach this unknown rotation"); 564 break; 565 } 566 567 writeInt32(a); // a 568 writeInt32(b); // b 569 writeInt32(0); // u 570 writeInt32(c); // c 571 writeInt32(d); // d 572 writeInt32(0); // v 573 writeInt32(0); // x 574 writeInt32(0); // y 575 writeInt32(0x40000000); // w 576} 577 578 579status_t MPEG4Writer::stop() { 580 if (mFile == NULL) { 581 return OK; 582 } 583 584 status_t err = OK; 585 int64_t maxDurationUs = 0; 586 for (List<Track *>::iterator it = mTracks.begin(); 587 it != mTracks.end(); ++it) { 588 status_t status = (*it)->stop(); 589 if (err == OK && status != OK) { 590 err = status; 591 } 592 593 int64_t durationUs = (*it)->getDurationUs(); 594 if (durationUs > maxDurationUs) { 595 maxDurationUs = durationUs; 596 } 597 } 598 599 stopWriterThread(); 600 601 // Do not write out movie header on error. 602 if (err != OK) { 603 fflush(mFile); 604 fclose(mFile); 605 mFile = NULL; 606 mStarted = false; 607 return err; 608 } 609 610 // Fix up the size of the 'mdat' chunk. 611 if (mUse32BitOffset) { 612 lseek64(mFd, mMdatOffset, SEEK_SET); 613 int32_t size = htonl(static_cast<int32_t>(mOffset - mMdatOffset)); 614 ::write(mFd, &size, 4); 615 } else { 616 lseek64(mFd, mMdatOffset + 8, SEEK_SET); 617 int64_t size = mOffset - mMdatOffset; 618 size = hton64(size); 619 ::write(mFd, &size, 8); 620 } 621 lseek64(mFd, mOffset, SEEK_SET); 622 623 time_t now = time(NULL); 624 const off64_t moovOffset = mOffset; 625 mWriteMoovBoxToMemory = true; 626 mMoovBoxBuffer = (uint8_t *) malloc(mEstimatedMoovBoxSize); 627 mMoovBoxBufferOffset = 0; 628 CHECK(mMoovBoxBuffer != NULL); 629 int32_t duration = (maxDurationUs * mTimeScale + 5E5) / 1E6; 630 631 beginBox("moov"); 632 633 beginBox("mvhd"); 634 writeInt32(0); // version=0, flags=0 635 writeInt32(now); // creation time 636 writeInt32(now); // modification time 637 writeInt32(mTimeScale); // mvhd timescale 638 writeInt32(duration); 639 writeInt32(0x10000); // rate: 1.0 640 writeInt16(0x100); // volume 641 writeInt16(0); // reserved 642 writeInt32(0); // reserved 643 writeInt32(0); // reserved 644 writeCompositionMatrix(0); // matrix 645 writeInt32(0); // predefined 646 writeInt32(0); // predefined 647 writeInt32(0); // predefined 648 writeInt32(0); // predefined 649 writeInt32(0); // predefined 650 writeInt32(0); // predefined 651 writeInt32(mTracks.size() + 1); // nextTrackID 652 endBox(); // mvhd 653 654 int32_t id = 1; 655 for (List<Track *>::iterator it = mTracks.begin(); 656 it != mTracks.end(); ++it, ++id) { 657 (*it)->writeTrackHeader(id, mUse32BitOffset); 658 } 659 endBox(); // moov 660 661 mWriteMoovBoxToMemory = false; 662 if (mStreamableFile) { 663 CHECK(mMoovBoxBufferOffset + 8 <= mEstimatedMoovBoxSize); 664 665 // Moov box 666 lseek64(mFd, mFreeBoxOffset, SEEK_SET); 667 mOffset = mFreeBoxOffset; 668 write(mMoovBoxBuffer, 1, mMoovBoxBufferOffset, mFile); 669 670 // Free box 671 lseek64(mFd, mOffset, SEEK_SET); 672 writeInt32(mEstimatedMoovBoxSize - mMoovBoxBufferOffset); 673 write("free", 4); 674 675 // Free temp memory 676 free(mMoovBoxBuffer); 677 mMoovBoxBuffer = NULL; 678 mMoovBoxBufferOffset = 0; 679 } else { 680 LOGI("The mp4 file will not be streamable."); 681 } 682 683 CHECK(mBoxes.empty()); 684 685 fflush(mFile); 686 fclose(mFile); 687 mFile = NULL; 688 mStarted = false; 689 return err; 690} 691 692status_t MPEG4Writer::setInterleaveDuration(uint32_t durationUs) { 693 mInterleaveDurationUs = durationUs; 694 return OK; 695} 696 697void MPEG4Writer::lock() { 698 mLock.lock(); 699} 700 701void MPEG4Writer::unlock() { 702 mLock.unlock(); 703} 704 705off64_t MPEG4Writer::addSample_l(MediaBuffer *buffer) { 706 off64_t old_offset = mOffset; 707 708 ::write(mFd, 709 (const uint8_t *)buffer->data() + buffer->range_offset(), 710 buffer->range_length()); 711 712 mOffset += buffer->range_length(); 713 714 return old_offset; 715} 716 717static void StripStartcode(MediaBuffer *buffer) { 718 if (buffer->range_length() < 4) { 719 return; 720 } 721 722 const uint8_t *ptr = 723 (const uint8_t *)buffer->data() + buffer->range_offset(); 724 725 if (!memcmp(ptr, "\x00\x00\x00\x01", 4)) { 726 buffer->set_range( 727 buffer->range_offset() + 4, buffer->range_length() - 4); 728 } 729} 730 731off64_t MPEG4Writer::addLengthPrefixedSample_l(MediaBuffer *buffer) { 732 off64_t old_offset = mOffset; 733 734 size_t length = buffer->range_length(); 735 736 if (mUse4ByteNalLength) { 737 uint8_t x = length >> 24; 738 ::write(mFd, &x, 1); 739 x = (length >> 16) & 0xff; 740 ::write(mFd, &x, 1); 741 x = (length >> 8) & 0xff; 742 ::write(mFd, &x, 1); 743 x = length & 0xff; 744 ::write(mFd, &x, 1); 745 746 ::write(mFd, 747 (const uint8_t *)buffer->data() + buffer->range_offset(), 748 length); 749 750 mOffset += length + 4; 751 } else { 752 CHECK(length < 65536); 753 754 uint8_t x = length >> 8; 755 ::write(mFd, &x, 1); 756 x = length & 0xff; 757 ::write(mFd, &x, 1); 758 ::write(mFd, (const uint8_t *)buffer->data() + buffer->range_offset(), length); 759 mOffset += length + 2; 760 } 761 762 return old_offset; 763} 764 765size_t MPEG4Writer::write( 766 const void *ptr, size_t size, size_t nmemb, FILE *stream) { 767 768 const size_t bytes = size * nmemb; 769 int fd = fileno(stream); 770 if (mWriteMoovBoxToMemory) { 771 off64_t moovBoxSize = 8 + mMoovBoxBufferOffset + bytes; 772 if (moovBoxSize > mEstimatedMoovBoxSize) { 773 for (List<off64_t>::iterator it = mBoxes.begin(); 774 it != mBoxes.end(); ++it) { 775 (*it) += mOffset; 776 } 777 lseek64(fd, mOffset, SEEK_SET); 778 ::write(fd, mMoovBoxBuffer, mMoovBoxBufferOffset); 779 ::write(fd, ptr, size * nmemb); 780 mOffset += (bytes + mMoovBoxBufferOffset); 781 free(mMoovBoxBuffer); 782 mMoovBoxBuffer = NULL; 783 mMoovBoxBufferOffset = 0; 784 mWriteMoovBoxToMemory = false; 785 mStreamableFile = false; 786 } else { 787 memcpy(mMoovBoxBuffer + mMoovBoxBufferOffset, ptr, bytes); 788 mMoovBoxBufferOffset += bytes; 789 } 790 } else { 791 ::write(fd, ptr, size * nmemb); 792 mOffset += bytes; 793 } 794 return bytes; 795} 796 797void MPEG4Writer::beginBox(const char *fourcc) { 798 CHECK_EQ(strlen(fourcc), 4); 799 800 mBoxes.push_back(mWriteMoovBoxToMemory? 801 mMoovBoxBufferOffset: mOffset); 802 803 writeInt32(0); 804 writeFourcc(fourcc); 805} 806 807void MPEG4Writer::endBox() { 808 CHECK(!mBoxes.empty()); 809 810 off64_t offset = *--mBoxes.end(); 811 mBoxes.erase(--mBoxes.end()); 812 813 if (mWriteMoovBoxToMemory) { 814 int32_t x = htonl(mMoovBoxBufferOffset - offset); 815 memcpy(mMoovBoxBuffer + offset, &x, 4); 816 } else { 817 lseek64(mFd, offset, SEEK_SET); 818 writeInt32(mOffset - offset); 819 mOffset -= 4; 820 lseek64(mFd, mOffset, SEEK_SET); 821 } 822} 823 824void MPEG4Writer::writeInt8(int8_t x) { 825 write(&x, 1, 1, mFile); 826} 827 828void MPEG4Writer::writeInt16(int16_t x) { 829 x = htons(x); 830 write(&x, 1, 2, mFile); 831} 832 833void MPEG4Writer::writeInt32(int32_t x) { 834 x = htonl(x); 835 write(&x, 1, 4, mFile); 836} 837 838void MPEG4Writer::writeInt64(int64_t x) { 839 x = hton64(x); 840 write(&x, 1, 8, mFile); 841} 842 843void MPEG4Writer::writeCString(const char *s) { 844 size_t n = strlen(s); 845 write(s, 1, n + 1, mFile); 846} 847 848void MPEG4Writer::writeFourcc(const char *s) { 849 CHECK_EQ(strlen(s), 4); 850 write(s, 1, 4, mFile); 851} 852 853void MPEG4Writer::write(const void *data, size_t size) { 854 write(data, 1, size, mFile); 855} 856 857bool MPEG4Writer::isFileStreamable() const { 858 return mStreamableFile; 859} 860 861bool MPEG4Writer::exceedsFileSizeLimit() { 862 // No limit 863 if (mMaxFileSizeLimitBytes == 0) { 864 return false; 865 } 866 867 int64_t nTotalBytesEstimate = static_cast<int64_t>(mEstimatedMoovBoxSize); 868 for (List<Track *>::iterator it = mTracks.begin(); 869 it != mTracks.end(); ++it) { 870 nTotalBytesEstimate += (*it)->getEstimatedTrackSizeBytes(); 871 } 872 873 return (nTotalBytesEstimate >= mMaxFileSizeLimitBytes); 874} 875 876bool MPEG4Writer::exceedsFileDurationLimit() { 877 // No limit 878 if (mMaxFileDurationLimitUs == 0) { 879 return false; 880 } 881 882 for (List<Track *>::iterator it = mTracks.begin(); 883 it != mTracks.end(); ++it) { 884 if ((*it)->getDurationUs() >= mMaxFileDurationLimitUs) { 885 return true; 886 } 887 } 888 return false; 889} 890 891bool MPEG4Writer::reachedEOS() { 892 bool allDone = true; 893 for (List<Track *>::iterator it = mTracks.begin(); 894 it != mTracks.end(); ++it) { 895 if (!(*it)->reachedEOS()) { 896 allDone = false; 897 break; 898 } 899 } 900 901 return allDone; 902} 903 904void MPEG4Writer::setStartTimestampUs(int64_t timeUs) { 905 LOGI("setStartTimestampUs: %lld", timeUs); 906 CHECK(timeUs >= 0); 907 Mutex::Autolock autoLock(mLock); 908 if (mStartTimestampUs < 0 || mStartTimestampUs > timeUs) { 909 mStartTimestampUs = timeUs; 910 LOGI("Earliest track starting time: %lld", mStartTimestampUs); 911 } 912} 913 914int64_t MPEG4Writer::getStartTimestampUs() { 915 Mutex::Autolock autoLock(mLock); 916 return mStartTimestampUs; 917} 918 919size_t MPEG4Writer::numTracks() { 920 Mutex::Autolock autolock(mLock); 921 return mTracks.size(); 922} 923 924//////////////////////////////////////////////////////////////////////////////// 925 926MPEG4Writer::Track::Track( 927 MPEG4Writer *owner, const sp<MediaSource> &source) 928 : mOwner(owner), 929 mMeta(source->getFormat()), 930 mSource(source), 931 mDone(false), 932 mPaused(false), 933 mResumed(false), 934 mTrackDurationUs(0), 935 mEstimatedTrackSizeBytes(0), 936 mSamplesHaveSameSize(true), 937 mCodecSpecificData(NULL), 938 mCodecSpecificDataSize(0), 939 mGotAllCodecSpecificData(false), 940 mReachedEOS(false), 941 mRotation(0) { 942 getCodecSpecificDataFromInputFormatIfPossible(); 943 944 const char *mime; 945 mMeta->findCString(kKeyMIMEType, &mime); 946 mIsAvc = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC); 947 mIsAudio = !strncasecmp(mime, "audio/", 6); 948 mIsMPEG4 = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) || 949 !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC); 950 951 setTimeScale(); 952} 953 954void MPEG4Writer::Track::updateTrackSizeEstimate() { 955 956 int64_t stcoBoxSizeBytes = mOwner->use32BitFileOffset() 957 ? mNumStcoTableEntries * 4 958 : mNumStcoTableEntries * 8; 959 960 int64_t stszBoxSizeBytes = mSamplesHaveSameSize? 4: (mNumSamples * 4); 961 962 mEstimatedTrackSizeBytes = mMdatSizeBytes; // media data size 963 if (!mOwner->isFileStreamable()) { 964 // Reserved free space is not large enough to hold 965 // all meta data and thus wasted. 966 mEstimatedTrackSizeBytes += mNumStscTableEntries * 12 + // stsc box size 967 mNumStssTableEntries * 4 + // stss box size 968 mNumSttsTableEntries * 8 + // stts box size 969 stcoBoxSizeBytes + // stco box size 970 stszBoxSizeBytes; // stsz box size 971 } 972} 973 974void MPEG4Writer::Track::addOneStscTableEntry( 975 size_t chunkId, size_t sampleId) { 976 977 StscTableEntry stscEntry(chunkId, sampleId, 1); 978 mStscTableEntries.push_back(stscEntry); 979 ++mNumStscTableEntries; 980} 981 982void MPEG4Writer::Track::addOneStssTableEntry(size_t sampleId) { 983 mStssTableEntries.push_back(sampleId); 984 ++mNumStssTableEntries; 985} 986 987void MPEG4Writer::Track::addOneSttsTableEntry( 988 size_t sampleCount, int64_t durationUs) { 989 990 SttsTableEntry sttsEntry(sampleCount, durationUs); 991 mSttsTableEntries.push_back(sttsEntry); 992 ++mNumSttsTableEntries; 993} 994 995void MPEG4Writer::Track::addChunkOffset(off64_t offset) { 996 ++mNumStcoTableEntries; 997 mChunkOffsets.push_back(offset); 998} 999 1000void MPEG4Writer::Track::setTimeScale() { 1001 LOGV("setTimeScale"); 1002 // Default time scale 1003 mTimeScale = 90000; 1004 1005 if (mIsAudio) { 1006 // Use the sampling rate as the default time scale for audio track. 1007 int32_t sampleRate; 1008 bool success = mMeta->findInt32(kKeySampleRate, &sampleRate); 1009 CHECK(success); 1010 mTimeScale = sampleRate; 1011 } 1012 1013 // If someone would like to overwrite the timescale, use user-supplied value. 1014 int32_t timeScale; 1015 if (mMeta->findInt32(kKeyTimeScale, &timeScale)) { 1016 mTimeScale = timeScale; 1017 } 1018 1019 CHECK(mTimeScale > 0); 1020} 1021 1022void MPEG4Writer::Track::getCodecSpecificDataFromInputFormatIfPossible() { 1023 const char *mime; 1024 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 1025 1026 if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC)) { 1027 uint32_t type; 1028 const void *data; 1029 size_t size; 1030 if (mMeta->findData(kKeyAVCC, &type, &data, &size)) { 1031 mCodecSpecificData = malloc(size); 1032 mCodecSpecificDataSize = size; 1033 memcpy(mCodecSpecificData, data, size); 1034 mGotAllCodecSpecificData = true; 1035 } 1036 } else if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) 1037 || !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC)) { 1038 uint32_t type; 1039 const void *data; 1040 size_t size; 1041 if (mMeta->findData(kKeyESDS, &type, &data, &size)) { 1042 ESDS esds(data, size); 1043 if (esds.getCodecSpecificInfo(&data, &size) == OK) { 1044 mCodecSpecificData = malloc(size); 1045 mCodecSpecificDataSize = size; 1046 memcpy(mCodecSpecificData, data, size); 1047 mGotAllCodecSpecificData = true; 1048 } 1049 } 1050 } 1051} 1052 1053MPEG4Writer::Track::~Track() { 1054 stop(); 1055 1056 if (mCodecSpecificData != NULL) { 1057 free(mCodecSpecificData); 1058 mCodecSpecificData = NULL; 1059 } 1060} 1061 1062void MPEG4Writer::Track::initTrackingProgressStatus(MetaData *params) { 1063 LOGV("initTrackingProgressStatus"); 1064 mPreviousTrackTimeUs = -1; 1065 mTrackingProgressStatus = false; 1066 mTrackEveryTimeDurationUs = 0; 1067 { 1068 int64_t timeUs; 1069 if (params && params->findInt64(kKeyTrackTimeStatus, &timeUs)) { 1070 LOGV("Receive request to track progress status for every %lld us", timeUs); 1071 mTrackEveryTimeDurationUs = timeUs; 1072 mTrackingProgressStatus = true; 1073 } 1074 } 1075} 1076 1077// static 1078void *MPEG4Writer::ThreadWrapper(void *me) { 1079 LOGV("ThreadWrapper: %p", me); 1080 MPEG4Writer *writer = static_cast<MPEG4Writer *>(me); 1081 writer->threadFunc(); 1082 return NULL; 1083} 1084 1085void MPEG4Writer::bufferChunk(const Chunk& chunk) { 1086 LOGV("bufferChunk: %p", chunk.mTrack); 1087 Mutex::Autolock autolock(mLock); 1088 CHECK_EQ(mDone, false); 1089 1090 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1091 it != mChunkInfos.end(); ++it) { 1092 1093 if (chunk.mTrack == it->mTrack) { // Found owner 1094 it->mChunks.push_back(chunk); 1095 mChunkReadyCondition.signal(); 1096 return; 1097 } 1098 } 1099 1100 CHECK("Received a chunk for a unknown track" == 0); 1101} 1102 1103void MPEG4Writer::writeFirstChunk(ChunkInfo* info) { 1104 LOGV("writeFirstChunk: %p", info->mTrack); 1105 1106 List<Chunk>::iterator chunkIt = info->mChunks.begin(); 1107 for (List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1108 it != chunkIt->mSamples.end(); ++it) { 1109 1110 off64_t offset = info->mTrack->isAvc() 1111 ? addLengthPrefixedSample_l(*it) 1112 : addSample_l(*it); 1113 if (it == chunkIt->mSamples.begin()) { 1114 info->mTrack->addChunkOffset(offset); 1115 } 1116 } 1117 1118 // Done with the current chunk. 1119 // Release all the samples in this chunk. 1120 while (!chunkIt->mSamples.empty()) { 1121 List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1122 (*it)->release(); 1123 (*it) = NULL; 1124 chunkIt->mSamples.erase(it); 1125 } 1126 chunkIt->mSamples.clear(); 1127 info->mChunks.erase(chunkIt); 1128} 1129 1130void MPEG4Writer::writeChunks() { 1131 LOGV("writeChunks"); 1132 size_t outstandingChunks = 0; 1133 while (!mChunkInfos.empty()) { 1134 List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1135 while (!it->mChunks.empty()) { 1136 CHECK_EQ(OK, writeOneChunk()); 1137 ++outstandingChunks; 1138 } 1139 it->mTrack = NULL; 1140 mChunkInfos.erase(it); 1141 } 1142 mChunkInfos.clear(); 1143 LOGD("%d chunks are written in the last batch", outstandingChunks); 1144} 1145 1146status_t MPEG4Writer::writeOneChunk() { 1147 LOGV("writeOneChunk"); 1148 1149 // Find the smallest timestamp, and write that chunk out 1150 // XXX: What if some track is just too slow? 1151 int64_t minTimestampUs = 0x7FFFFFFFFFFFFFFFLL; 1152 Track *track = NULL; 1153 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1154 it != mChunkInfos.end(); ++it) { 1155 if (!it->mChunks.empty()) { 1156 List<Chunk>::iterator chunkIt = it->mChunks.begin(); 1157 if (chunkIt->mTimeStampUs < minTimestampUs) { 1158 minTimestampUs = chunkIt->mTimeStampUs; 1159 track = it->mTrack; 1160 } 1161 } 1162 } 1163 1164 if (track == NULL) { 1165 LOGV("Nothing to be written after all"); 1166 return OK; 1167 } 1168 1169 if (mIsFirstChunk) { 1170 mIsFirstChunk = false; 1171 } 1172 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1173 it != mChunkInfos.end(); ++it) { 1174 if (it->mTrack == track) { 1175 writeFirstChunk(&(*it)); 1176 } 1177 } 1178 return OK; 1179} 1180 1181void MPEG4Writer::threadFunc() { 1182 LOGV("threadFunc"); 1183 1184 prctl(PR_SET_NAME, (unsigned long)"MPEG4Writer", 0, 0, 0); 1185 while (!mDone) { 1186 { 1187 Mutex::Autolock autolock(mLock); 1188 mChunkReadyCondition.wait(mLock); 1189 CHECK_EQ(writeOneChunk(), OK); 1190 } 1191 } 1192 1193 { 1194 // Write ALL samples 1195 Mutex::Autolock autolock(mLock); 1196 writeChunks(); 1197 } 1198} 1199 1200status_t MPEG4Writer::startWriterThread() { 1201 LOGV("startWriterThread"); 1202 1203 mDone = false; 1204 mIsFirstChunk = true; 1205 mDriftTimeUs = 0; 1206 for (List<Track *>::iterator it = mTracks.begin(); 1207 it != mTracks.end(); ++it) { 1208 ChunkInfo info; 1209 info.mTrack = *it; 1210 mChunkInfos.push_back(info); 1211 } 1212 1213 pthread_attr_t attr; 1214 pthread_attr_init(&attr); 1215 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1216 pthread_create(&mThread, &attr, ThreadWrapper, this); 1217 pthread_attr_destroy(&attr); 1218 return OK; 1219} 1220 1221 1222status_t MPEG4Writer::Track::start(MetaData *params) { 1223 if (!mDone && mPaused) { 1224 mPaused = false; 1225 mResumed = true; 1226 return OK; 1227 } 1228 1229 int64_t startTimeUs; 1230 if (params == NULL || !params->findInt64(kKeyTime, &startTimeUs)) { 1231 startTimeUs = 0; 1232 } 1233 1234 int32_t rotationDegrees; 1235 if (!mIsAudio && params && params->findInt32(kKeyRotation, &rotationDegrees)) { 1236 mRotation = rotationDegrees; 1237 } 1238 1239 mIsRealTimeRecording = true; 1240 { 1241 int32_t isNotRealTime; 1242 if (params && params->findInt32(kKeyNotRealTime, &isNotRealTime)) { 1243 mIsRealTimeRecording = (isNotRealTime == 0); 1244 } 1245 } 1246 1247 initTrackingProgressStatus(params); 1248 1249 sp<MetaData> meta = new MetaData; 1250 meta->setInt64(kKeyTime, startTimeUs); 1251 status_t err = mSource->start(meta.get()); 1252 if (err != OK) { 1253 mDone = mReachedEOS = true; 1254 return err; 1255 } 1256 1257 pthread_attr_t attr; 1258 pthread_attr_init(&attr); 1259 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1260 1261 mDone = false; 1262 mTrackDurationUs = 0; 1263 mReachedEOS = false; 1264 mEstimatedTrackSizeBytes = 0; 1265 mNumStcoTableEntries = 0; 1266 mNumStssTableEntries = 0; 1267 mNumStscTableEntries = 0; 1268 mNumSttsTableEntries = 0; 1269 mMdatSizeBytes = 0; 1270 mIsMediaTimeAdjustmentOn = false; 1271 mPrevMediaTimeAdjustTimestampUs = 0; 1272 mMediaTimeAdjustNumFrames = 0; 1273 mPrevMediaTimeAdjustSample = 0; 1274 mTotalDriftTimeToAdjustUs = 0; 1275 mPrevTotalAccumDriftTimeUs = 0; 1276 1277 pthread_create(&mThread, &attr, ThreadWrapper, this); 1278 pthread_attr_destroy(&attr); 1279 1280 return OK; 1281} 1282 1283status_t MPEG4Writer::Track::pause() { 1284 mPaused = true; 1285 return OK; 1286} 1287 1288status_t MPEG4Writer::Track::stop() { 1289 LOGD("Stopping %s track", mIsAudio? "Audio": "Video"); 1290 if (mDone) { 1291 return OK; 1292 } 1293 1294 mDone = true; 1295 1296 void *dummy; 1297 pthread_join(mThread, &dummy); 1298 1299 status_t err = (status_t) dummy; 1300 1301 LOGD("Stopping %s track source", mIsAudio? "Audio": "Video"); 1302 { 1303 status_t status = mSource->stop(); 1304 if (err == OK && status != OK && status != ERROR_END_OF_STREAM) { 1305 err = status; 1306 } 1307 } 1308 1309 LOGD("%s track stopped", mIsAudio? "Audio": "Video"); 1310 return err; 1311} 1312 1313bool MPEG4Writer::Track::reachedEOS() { 1314 return mReachedEOS; 1315} 1316 1317// static 1318void *MPEG4Writer::Track::ThreadWrapper(void *me) { 1319 Track *track = static_cast<Track *>(me); 1320 1321 status_t err = track->threadEntry(); 1322 return (void *) err; 1323} 1324 1325static void getNalUnitType(uint8_t byte, uint8_t* type) { 1326 LOGV("getNalUnitType: %d", byte); 1327 1328 // nal_unit_type: 5-bit unsigned integer 1329 *type = (byte & 0x1F); 1330} 1331 1332static const uint8_t *findNextStartCode( 1333 const uint8_t *data, size_t length) { 1334 1335 LOGV("findNextStartCode: %p %d", data, length); 1336 1337 size_t bytesLeft = length; 1338 while (bytesLeft > 4 && 1339 memcmp("\x00\x00\x00\x01", &data[length - bytesLeft], 4)) { 1340 --bytesLeft; 1341 } 1342 if (bytesLeft <= 4) { 1343 bytesLeft = 0; // Last parameter set 1344 } 1345 return &data[length - bytesLeft]; 1346} 1347 1348const uint8_t *MPEG4Writer::Track::parseParamSet( 1349 const uint8_t *data, size_t length, int type, size_t *paramSetLen) { 1350 1351 LOGV("parseParamSet"); 1352 CHECK(type == kNalUnitTypeSeqParamSet || 1353 type == kNalUnitTypePicParamSet); 1354 1355 const uint8_t *nextStartCode = findNextStartCode(data, length); 1356 *paramSetLen = nextStartCode - data; 1357 if (*paramSetLen == 0) { 1358 LOGE("Param set is malformed, since its length is 0"); 1359 return NULL; 1360 } 1361 1362 AVCParamSet paramSet(*paramSetLen, data); 1363 if (type == kNalUnitTypeSeqParamSet) { 1364 if (*paramSetLen < 4) { 1365 LOGE("Seq parameter set malformed"); 1366 return NULL; 1367 } 1368 if (mSeqParamSets.empty()) { 1369 mProfileIdc = data[1]; 1370 mProfileCompatible = data[2]; 1371 mLevelIdc = data[3]; 1372 } else { 1373 if (mProfileIdc != data[1] || 1374 mProfileCompatible != data[2] || 1375 mLevelIdc != data[3]) { 1376 LOGE("Inconsistent profile/level found in seq parameter sets"); 1377 return NULL; 1378 } 1379 } 1380 mSeqParamSets.push_back(paramSet); 1381 } else { 1382 mPicParamSets.push_back(paramSet); 1383 } 1384 return nextStartCode; 1385} 1386 1387status_t MPEG4Writer::Track::copyAVCCodecSpecificData( 1388 const uint8_t *data, size_t size) { 1389 LOGV("copyAVCCodecSpecificData"); 1390 1391 // 2 bytes for each of the parameter set length field 1392 // plus the 7 bytes for the header 1393 if (size < 4 + 7) { 1394 LOGE("Codec specific data length too short: %d", size); 1395 return ERROR_MALFORMED; 1396 } 1397 1398 mCodecSpecificDataSize = size; 1399 mCodecSpecificData = malloc(size); 1400 memcpy(mCodecSpecificData, data, size); 1401 return OK; 1402} 1403 1404status_t MPEG4Writer::Track::parseAVCCodecSpecificData( 1405 const uint8_t *data, size_t size) { 1406 1407 LOGV("parseAVCCodecSpecificData"); 1408 // Data starts with a start code. 1409 // SPS and PPS are separated with start codes. 1410 // Also, SPS must come before PPS 1411 uint8_t type = kNalUnitTypeSeqParamSet; 1412 bool gotSps = false; 1413 bool gotPps = false; 1414 const uint8_t *tmp = data; 1415 const uint8_t *nextStartCode = data; 1416 size_t bytesLeft = size; 1417 size_t paramSetLen = 0; 1418 mCodecSpecificDataSize = 0; 1419 while (bytesLeft > 4 && !memcmp("\x00\x00\x00\x01", tmp, 4)) { 1420 getNalUnitType(*(tmp + 4), &type); 1421 if (type == kNalUnitTypeSeqParamSet) { 1422 if (gotPps) { 1423 LOGE("SPS must come before PPS"); 1424 return ERROR_MALFORMED; 1425 } 1426 if (!gotSps) { 1427 gotSps = true; 1428 } 1429 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1430 } else if (type == kNalUnitTypePicParamSet) { 1431 if (!gotSps) { 1432 LOGE("SPS must come before PPS"); 1433 return ERROR_MALFORMED; 1434 } 1435 if (!gotPps) { 1436 gotPps = true; 1437 } 1438 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1439 } else { 1440 LOGE("Only SPS and PPS Nal units are expected"); 1441 return ERROR_MALFORMED; 1442 } 1443 1444 if (nextStartCode == NULL) { 1445 return ERROR_MALFORMED; 1446 } 1447 1448 // Move on to find the next parameter set 1449 bytesLeft -= nextStartCode - tmp; 1450 tmp = nextStartCode; 1451 mCodecSpecificDataSize += (2 + paramSetLen); 1452 } 1453 1454 { 1455 // Check on the number of seq parameter sets 1456 size_t nSeqParamSets = mSeqParamSets.size(); 1457 if (nSeqParamSets == 0) { 1458 LOGE("Cound not find sequence parameter set"); 1459 return ERROR_MALFORMED; 1460 } 1461 1462 if (nSeqParamSets > 0x1F) { 1463 LOGE("Too many seq parameter sets (%d) found", nSeqParamSets); 1464 return ERROR_MALFORMED; 1465 } 1466 } 1467 1468 { 1469 // Check on the number of pic parameter sets 1470 size_t nPicParamSets = mPicParamSets.size(); 1471 if (nPicParamSets == 0) { 1472 LOGE("Cound not find picture parameter set"); 1473 return ERROR_MALFORMED; 1474 } 1475 if (nPicParamSets > 0xFF) { 1476 LOGE("Too many pic parameter sets (%d) found", nPicParamSets); 1477 return ERROR_MALFORMED; 1478 } 1479 } 1480 1481 { 1482 // Check on the profiles 1483 // These profiles requires additional parameter set extensions 1484 if (mProfileIdc == 100 || mProfileIdc == 110 || 1485 mProfileIdc == 122 || mProfileIdc == 144) { 1486 LOGE("Sorry, no support for profile_idc: %d!", mProfileIdc); 1487 return BAD_VALUE; 1488 } 1489 } 1490 1491 return OK; 1492} 1493 1494status_t MPEG4Writer::Track::makeAVCCodecSpecificData( 1495 const uint8_t *data, size_t size) { 1496 1497 if (mCodecSpecificData != NULL) { 1498 LOGE("Already have codec specific data"); 1499 return ERROR_MALFORMED; 1500 } 1501 1502 if (size < 4) { 1503 LOGE("Codec specific data length too short: %d", size); 1504 return ERROR_MALFORMED; 1505 } 1506 1507 // Data is in the form of AVCCodecSpecificData 1508 if (memcmp("\x00\x00\x00\x01", data, 4)) { 1509 return copyAVCCodecSpecificData(data, size); 1510 } 1511 1512 if (parseAVCCodecSpecificData(data, size) != OK) { 1513 return ERROR_MALFORMED; 1514 } 1515 1516 // ISO 14496-15: AVC file format 1517 mCodecSpecificDataSize += 7; // 7 more bytes in the header 1518 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1519 uint8_t *header = (uint8_t *)mCodecSpecificData; 1520 header[0] = 1; // version 1521 header[1] = mProfileIdc; // profile indication 1522 header[2] = mProfileCompatible; // profile compatibility 1523 header[3] = mLevelIdc; 1524 1525 // 6-bit '111111' followed by 2-bit to lengthSizeMinuusOne 1526 if (mOwner->useNalLengthFour()) { 1527 header[4] = 0xfc | 3; // length size == 4 bytes 1528 } else { 1529 header[4] = 0xfc | 1; // length size == 2 bytes 1530 } 1531 1532 // 3-bit '111' followed by 5-bit numSequenceParameterSets 1533 int nSequenceParamSets = mSeqParamSets.size(); 1534 header[5] = 0xe0 | nSequenceParamSets; 1535 header += 6; 1536 for (List<AVCParamSet>::iterator it = mSeqParamSets.begin(); 1537 it != mSeqParamSets.end(); ++it) { 1538 // 16-bit sequence parameter set length 1539 uint16_t seqParamSetLength = it->mLength; 1540 header[0] = seqParamSetLength >> 8; 1541 header[1] = seqParamSetLength & 0xff; 1542 1543 // SPS NAL unit (sequence parameter length bytes) 1544 memcpy(&header[2], it->mData, seqParamSetLength); 1545 header += (2 + seqParamSetLength); 1546 } 1547 1548 // 8-bit nPictureParameterSets 1549 int nPictureParamSets = mPicParamSets.size(); 1550 header[0] = nPictureParamSets; 1551 header += 1; 1552 for (List<AVCParamSet>::iterator it = mPicParamSets.begin(); 1553 it != mPicParamSets.end(); ++it) { 1554 // 16-bit picture parameter set length 1555 uint16_t picParamSetLength = it->mLength; 1556 header[0] = picParamSetLength >> 8; 1557 header[1] = picParamSetLength & 0xff; 1558 1559 // PPS Nal unit (picture parameter set length bytes) 1560 memcpy(&header[2], it->mData, picParamSetLength); 1561 header += (2 + picParamSetLength); 1562 } 1563 1564 return OK; 1565} 1566 1567/* 1568* The video track's media time adjustment for real-time applications 1569* is described as follows: 1570* 1571* First, the media time adjustment is done for every period of 1572* kVideoMediaTimeAdjustPeriodTimeUs. kVideoMediaTimeAdjustPeriodTimeUs 1573* is currently a fixed value chosen heuristically. The value of 1574* kVideoMediaTimeAdjustPeriodTimeUs should not be very large or very small 1575* for two considerations: on one hand, a relatively large value 1576* helps reduce large fluctuation of drift time in the audio encoding 1577* path; while on the other hand, a relatively small value helps keep 1578* restoring synchronization in audio/video more frequently. Note for the 1579* very first period of kVideoMediaTimeAdjustPeriodTimeUs, there is 1580* no media time adjustment for the video track. 1581* 1582* Second, the total accumulated audio track time drift found 1583* in a period of kVideoMediaTimeAdjustPeriodTimeUs is distributed 1584* over a stream of incoming video frames. The number of video frames 1585* affected is determined based on the number of recorded video frames 1586* within the past kVideoMediaTimeAdjustPeriodTimeUs period. 1587* We choose to distribute the drift time over only a portion 1588* (rather than all) of the total number of recorded video frames 1589* in order to make sure that the video track media time adjustment is 1590* completed for the current period before the next video track media 1591* time adjustment period starts. Currently, the portion chosen is a 1592* half (0.5). 1593* 1594* Last, various additional checks are performed to ensure that 1595* the actual audio encoding path does not have too much drift. 1596* In particular, 1) we want to limit the average incremental time 1597* adjustment for each video frame to be less than a threshold 1598* for a single period of kVideoMediaTimeAdjustPeriodTimeUs. 1599* Currently, the threshold is set to 5 ms. If the average incremental 1600* media time adjustment for a video frame is larger than the 1601* threshold, the audio encoding path has too much time drift. 1602* 2) We also want to limit the total time drift in the audio 1603* encoding path to be less than a threshold for a period of 1604* kVideoMediaTimeAdjustPeriodTimeUs. Currently, the threshold 1605* is 0.5% of kVideoMediaTimeAdjustPeriodTimeUs. If the time drift of 1606* the audio encoding path is larger than the threshold, the audio 1607* encoding path has too much time drift. We treat the large time 1608* drift of the audio encoding path as errors, since there is no 1609* way to keep audio/video in synchronization for real-time 1610* applications if the time drift is too large unless we drop some 1611* video frames, which has its own problems that we don't want 1612* to get into for the time being. 1613*/ 1614void MPEG4Writer::Track::adjustMediaTime(int64_t *timestampUs) { 1615 if (*timestampUs - mPrevMediaTimeAdjustTimestampUs >= 1616 kVideoMediaTimeAdjustPeriodTimeUs) { 1617 1618 LOGV("New media time adjustment period at %lld us", *timestampUs); 1619 mIsMediaTimeAdjustmentOn = true; 1620 mMediaTimeAdjustNumFrames = 1621 (mNumSamples - mPrevMediaTimeAdjustSample) >> 1; 1622 1623 mPrevMediaTimeAdjustTimestampUs = *timestampUs; 1624 mPrevMediaTimeAdjustSample = mNumSamples; 1625 int64_t totalAccumDriftTimeUs = mOwner->getDriftTimeUs(); 1626 mTotalDriftTimeToAdjustUs = 1627 totalAccumDriftTimeUs - mPrevTotalAccumDriftTimeUs; 1628 1629 mPrevTotalAccumDriftTimeUs = totalAccumDriftTimeUs; 1630 1631 // Check on incremental adjusted time per frame 1632 int64_t adjustTimePerFrameUs = 1633 mTotalDriftTimeToAdjustUs / mMediaTimeAdjustNumFrames; 1634 1635 if (adjustTimePerFrameUs < 0) { 1636 adjustTimePerFrameUs = -adjustTimePerFrameUs; 1637 } 1638 if (adjustTimePerFrameUs >= 5000) { 1639 LOGE("Adjusted time per video frame is %lld us", 1640 adjustTimePerFrameUs); 1641 CHECK(!"Video frame time adjustment is too large!"); 1642 } 1643 1644 // Check on total accumulated time drift within a period of 1645 // kVideoMediaTimeAdjustPeriodTimeUs. 1646 int64_t driftPercentage = (mTotalDriftTimeToAdjustUs * 1000) 1647 / kVideoMediaTimeAdjustPeriodTimeUs; 1648 1649 if (driftPercentage < 0) { 1650 driftPercentage = -driftPercentage; 1651 } 1652 if (driftPercentage > 5) { 1653 LOGE("Audio track has time drift %lld us over %lld us", 1654 mTotalDriftTimeToAdjustUs, 1655 kVideoMediaTimeAdjustPeriodTimeUs); 1656 1657 CHECK(!"The audio track media time drifts too much!"); 1658 } 1659 1660 } 1661 1662 if (mIsMediaTimeAdjustmentOn) { 1663 if (mNumSamples - mPrevMediaTimeAdjustSample <= 1664 mMediaTimeAdjustNumFrames) { 1665 1666 // Do media time incremental adjustment 1667 int64_t incrementalAdjustTimeUs = 1668 (mTotalDriftTimeToAdjustUs * 1669 (mNumSamples - mPrevMediaTimeAdjustSample)) 1670 / mMediaTimeAdjustNumFrames; 1671 1672 *timestampUs += 1673 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs); 1674 1675 LOGV("Incremental video frame media time adjustment: %lld us", 1676 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs)); 1677 } else { 1678 // Within the remaining adjustment period, 1679 // no incremental adjustment is needed. 1680 *timestampUs += 1681 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs); 1682 1683 LOGV("Fixed video frame media time adjustment: %lld us", 1684 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs)); 1685 } 1686 } 1687} 1688 1689/* 1690 * Updates the drift time from the audio track so that 1691 * the video track can get the updated drift time information 1692 * from the file writer. The fluctuation of the drift time of the audio 1693 * encoding path is smoothed out with a simple filter by giving a larger 1694 * weight to more recently drift time. The filter coefficients, 0.5 and 0.5, 1695 * are heuristically determined. 1696 */ 1697void MPEG4Writer::Track::updateDriftTime(const sp<MetaData>& meta) { 1698 int64_t driftTimeUs = 0; 1699 if (meta->findInt64(kKeyDriftTime, &driftTimeUs)) { 1700 int64_t prevDriftTimeUs = mOwner->getDriftTimeUs(); 1701 int64_t timeUs = (driftTimeUs + prevDriftTimeUs) >> 1; 1702 mOwner->setDriftTimeUs(timeUs); 1703 } 1704} 1705 1706status_t MPEG4Writer::Track::threadEntry() { 1707 int32_t count = 0; 1708 const int64_t interleaveDurationUs = mOwner->interleaveDuration(); 1709 int64_t chunkTimestampUs = 0; 1710 int32_t nChunks = 0; 1711 int32_t nZeroLengthFrames = 0; 1712 int64_t lastTimestampUs = 0; // Previous sample time stamp in ms 1713 int64_t lastDurationUs = 0; // Between the previous two samples in ms 1714 int64_t currDurationTicks = 0; // Timescale based ticks 1715 int64_t lastDurationTicks = 0; // Timescale based ticks 1716 int32_t sampleCount = 1; // Sample count in the current stts table entry 1717 uint32_t previousSampleSize = 0; // Size of the previous sample 1718 int64_t previousPausedDurationUs = 0; 1719 int64_t timestampUs; 1720 1721 if (mIsAudio) { 1722 prctl(PR_SET_NAME, (unsigned long)"AudioTrackEncoding", 0, 0, 0); 1723 } else { 1724 prctl(PR_SET_NAME, (unsigned long)"VideoTrackEncoding", 0, 0, 0); 1725 } 1726 sp<MetaData> meta_data; 1727 1728 mNumSamples = 0; 1729 status_t err = OK; 1730 MediaBuffer *buffer; 1731 while (!mDone && (err = mSource->read(&buffer)) == OK) { 1732 if (buffer->range_length() == 0) { 1733 buffer->release(); 1734 buffer = NULL; 1735 ++nZeroLengthFrames; 1736 continue; 1737 } 1738 1739 // If the codec specific data has not been received yet, delay pause. 1740 // After the codec specific data is received, discard what we received 1741 // when the track is to be paused. 1742 if (mPaused && !mResumed) { 1743 buffer->release(); 1744 buffer = NULL; 1745 continue; 1746 } 1747 1748 ++count; 1749 1750 int32_t isCodecConfig; 1751 if (buffer->meta_data()->findInt32(kKeyIsCodecConfig, &isCodecConfig) 1752 && isCodecConfig) { 1753 CHECK(!mGotAllCodecSpecificData); 1754 1755 if (mIsAvc) { 1756 status_t err = makeAVCCodecSpecificData( 1757 (const uint8_t *)buffer->data() 1758 + buffer->range_offset(), 1759 buffer->range_length()); 1760 CHECK_EQ(OK, err); 1761 } else if (mIsMPEG4) { 1762 mCodecSpecificDataSize = buffer->range_length(); 1763 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1764 memcpy(mCodecSpecificData, 1765 (const uint8_t *)buffer->data() 1766 + buffer->range_offset(), 1767 buffer->range_length()); 1768 } 1769 1770 buffer->release(); 1771 buffer = NULL; 1772 1773 mGotAllCodecSpecificData = true; 1774 continue; 1775 } 1776 1777 // Make a deep copy of the MediaBuffer and Metadata and release 1778 // the original as soon as we can 1779 MediaBuffer *copy = new MediaBuffer(buffer->range_length()); 1780 memcpy(copy->data(), (uint8_t *)buffer->data() + buffer->range_offset(), 1781 buffer->range_length()); 1782 copy->set_range(0, buffer->range_length()); 1783 meta_data = new MetaData(*buffer->meta_data().get()); 1784 buffer->release(); 1785 buffer = NULL; 1786 1787 if (mIsAvc) StripStartcode(copy); 1788 1789 size_t sampleSize = copy->range_length(); 1790 if (mIsAvc) { 1791 if (mOwner->useNalLengthFour()) { 1792 sampleSize += 4; 1793 } else { 1794 sampleSize += 2; 1795 } 1796 } 1797 1798 // Max file size or duration handling 1799 mMdatSizeBytes += sampleSize; 1800 updateTrackSizeEstimate(); 1801 1802 if (mOwner->exceedsFileSizeLimit()) { 1803 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_FILESIZE_REACHED, 0); 1804 break; 1805 } 1806 if (mOwner->exceedsFileDurationLimit()) { 1807 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_DURATION_REACHED, 0); 1808 break; 1809 } 1810 1811 1812 int32_t isSync = false; 1813 meta_data->findInt32(kKeyIsSyncFrame, &isSync); 1814 1815 /* 1816 * The original timestamp found in the data buffer will be modified as below: 1817 * 1818 * There is a playback offset into this track if the track's start time 1819 * is not the same as the movie start time, which will be recorded in edst 1820 * box of the output file. The playback offset is to make sure that the 1821 * starting time of the audio/video tracks are synchronized. Although the 1822 * track's media timestamp may be subject to various modifications 1823 * as outlined below, the track's playback offset time remains unchanged 1824 * once the first data buffer of the track is received. 1825 * 1826 * The media time stamp will be calculated by subtracting the playback offset 1827 * (and potential pause durations) from the original timestamp in the buffer. 1828 * 1829 * If this track is a video track for a real-time recording application with 1830 * both audio and video tracks, its media timestamp will subject to further 1831 * modification based on the media clock of the audio track. This modification 1832 * is needed for the purpose of maintaining good audio/video synchronization. 1833 * 1834 * If the recording session is paused and resumed multiple times, the track 1835 * media timestamp will be modified as if the recording session had never been 1836 * paused at all during playback of the recorded output file. In other words, 1837 * the output file will have no memory of pause/resume durations. 1838 * 1839 */ 1840 CHECK(meta_data->findInt64(kKeyTime, ×tampUs)); 1841 LOGV("%s timestampUs: %lld", mIsAudio? "Audio": "Video", timestampUs); 1842 1843//////////////////////////////////////////////////////////////////////////////// 1844 if (mSampleSizes.empty()) { 1845 mStartTimestampUs = timestampUs; 1846 mOwner->setStartTimestampUs(mStartTimestampUs); 1847 previousPausedDurationUs = mStartTimestampUs; 1848 } 1849 1850 if (mResumed) { 1851 int64_t durExcludingEarlierPausesUs = timestampUs - previousPausedDurationUs; 1852 CHECK(durExcludingEarlierPausesUs >= 0); 1853 int64_t pausedDurationUs = durExcludingEarlierPausesUs - mTrackDurationUs; 1854 CHECK(pausedDurationUs >= lastDurationUs); 1855 previousPausedDurationUs += pausedDurationUs - lastDurationUs; 1856 mResumed = false; 1857 } 1858 1859 timestampUs -= previousPausedDurationUs; 1860 CHECK(timestampUs >= 0); 1861 1862 // Media time adjustment for real-time applications 1863 if (mIsRealTimeRecording) { 1864 if (mIsAudio) { 1865 updateDriftTime(meta_data); 1866 } else { 1867 adjustMediaTime(×tampUs); 1868 } 1869 } 1870 1871 CHECK(timestampUs >= 0); 1872 if (mNumSamples > 1) { 1873 if (timestampUs <= lastTimestampUs) { 1874 LOGW("Frame arrives too late!"); 1875 // Don't drop the late frame, since dropping a frame may cause 1876 // problems later during playback 1877 1878 // The idea here is to avoid having two or more samples with the 1879 // same timestamp in the output file. 1880 if (mTimeScale >= 1000000LL) { 1881 timestampUs = lastTimestampUs + 1; 1882 } else { 1883 timestampUs = lastTimestampUs + (1000000LL + (mTimeScale >> 1)) / mTimeScale; 1884 } 1885 } 1886 } 1887 1888 LOGV("%s media time stamp: %lld and previous paused duration %lld", 1889 mIsAudio? "Audio": "Video", timestampUs, previousPausedDurationUs); 1890 if (timestampUs > mTrackDurationUs) { 1891 mTrackDurationUs = timestampUs; 1892 } 1893 1894 mSampleSizes.push_back(sampleSize); 1895 ++mNumSamples; 1896 if (mNumSamples > 2) { 1897 // We need to use the time scale based ticks, rather than the 1898 // timestamp itself to determine whether we have to use a new 1899 // stts entry, since we may have rounding errors. 1900 // The calculation is intended to reduce the accumulated 1901 // rounding errors. 1902 currDurationTicks = 1903 ((timestampUs * mTimeScale + 500000LL) / 1000000LL - 1904 (lastTimestampUs * mTimeScale + 500000LL) / 1000000LL); 1905 1906 if (currDurationTicks != lastDurationTicks) { 1907 addOneSttsTableEntry(sampleCount, lastDurationUs); 1908 sampleCount = 1; 1909 } else { 1910 ++sampleCount; 1911 } 1912 } 1913 if (mSamplesHaveSameSize) { 1914 if (mNumSamples >= 2 && previousSampleSize != sampleSize) { 1915 mSamplesHaveSameSize = false; 1916 } 1917 previousSampleSize = sampleSize; 1918 } 1919 lastDurationUs = timestampUs - lastTimestampUs; 1920 lastDurationTicks = currDurationTicks; 1921 lastTimestampUs = timestampUs; 1922 1923 if (isSync != 0) { 1924 addOneStssTableEntry(mNumSamples); 1925 } 1926 1927 if (mTrackingProgressStatus) { 1928 if (mPreviousTrackTimeUs <= 0) { 1929 mPreviousTrackTimeUs = mStartTimestampUs; 1930 } 1931 trackProgressStatus(timestampUs); 1932 } 1933 if (mOwner->numTracks() == 1) { 1934 off64_t offset = mIsAvc? mOwner->addLengthPrefixedSample_l(copy) 1935 : mOwner->addSample_l(copy); 1936 if (mChunkOffsets.empty()) { 1937 addChunkOffset(offset); 1938 } 1939 copy->release(); 1940 copy = NULL; 1941 continue; 1942 } 1943 1944 mChunkSamples.push_back(copy); 1945 if (interleaveDurationUs == 0) { 1946 addOneStscTableEntry(++nChunks, 1); 1947 bufferChunk(timestampUs); 1948 } else { 1949 if (chunkTimestampUs == 0) { 1950 chunkTimestampUs = timestampUs; 1951 } else { 1952 if (timestampUs - chunkTimestampUs > interleaveDurationUs) { 1953 ++nChunks; 1954 if (nChunks == 1 || // First chunk 1955 (--(mStscTableEntries.end()))->samplesPerChunk != 1956 mChunkSamples.size()) { 1957 addOneStscTableEntry(nChunks, mChunkSamples.size()); 1958 } 1959 bufferChunk(timestampUs); 1960 chunkTimestampUs = timestampUs; 1961 } 1962 } 1963 } 1964 1965 } 1966 1967 if (mSampleSizes.empty() || // no samples written 1968 (!mIsAudio && mNumStssTableEntries == 0) || // no sync frames for video 1969 (OK != checkCodecSpecificData())) { // no codec specific data 1970 err = ERROR_MALFORMED; 1971 } 1972 mOwner->trackProgressStatus(this, -1, err); 1973 1974 // Last chunk 1975 if (mOwner->numTracks() == 1) { 1976 addOneStscTableEntry(1, mNumSamples); 1977 } else if (!mChunkSamples.empty()) { 1978 addOneStscTableEntry(++nChunks, mChunkSamples.size()); 1979 bufferChunk(timestampUs); 1980 } 1981 1982 // We don't really know how long the last frame lasts, since 1983 // there is no frame time after it, just repeat the previous 1984 // frame's duration. 1985 if (mNumSamples == 1) { 1986 lastDurationUs = 0; // A single sample's duration 1987 } else { 1988 ++sampleCount; // Count for the last sample 1989 } 1990 addOneSttsTableEntry(sampleCount, lastDurationUs); 1991 mTrackDurationUs += lastDurationUs; 1992 mReachedEOS = true; 1993 LOGI("Received total/0-length (%d/%d) buffers and encoded %d frames. - %s", 1994 count, nZeroLengthFrames, mNumSamples, mIsAudio? "audio": "video"); 1995 if (mIsAudio) { 1996 LOGI("Audio track drift time: %lld us", mOwner->getDriftTimeUs()); 1997 } 1998 1999 if (err == ERROR_END_OF_STREAM) { 2000 return OK; 2001 } 2002 return err; 2003} 2004 2005void MPEG4Writer::Track::trackProgressStatus(int64_t timeUs, status_t err) { 2006 LOGV("trackProgressStatus: %lld us", timeUs); 2007 if (mTrackEveryTimeDurationUs > 0 && 2008 timeUs - mPreviousTrackTimeUs >= mTrackEveryTimeDurationUs) { 2009 LOGV("Fire time tracking progress status at %lld us", timeUs); 2010 mOwner->trackProgressStatus(this, timeUs - mPreviousTrackTimeUs, err); 2011 mPreviousTrackTimeUs = timeUs; 2012 } 2013} 2014 2015void MPEG4Writer::trackProgressStatus( 2016 const MPEG4Writer::Track* track, int64_t timeUs, status_t err) { 2017 Mutex::Autolock lock(mLock); 2018 int32_t nTracks = mTracks.size(); 2019 CHECK(nTracks >= 1); 2020 CHECK(nTracks < 64); // Arbitrary number 2021 2022 int32_t trackNum = 0; 2023 CHECK(trackNum < nTracks); 2024 trackNum <<= 16; 2025 2026 // Error notification 2027 // Do not consider ERROR_END_OF_STREAM an error 2028 if (err != OK && err != ERROR_END_OF_STREAM) { 2029 notify(MEDIA_RECORDER_EVENT_ERROR, 2030 trackNum | MEDIA_RECORDER_ERROR_UNKNOWN, 2031 err); 2032 return; 2033 } 2034 2035 if (timeUs == -1) { 2036 // Send completion notification 2037 notify(MEDIA_RECORDER_EVENT_INFO, 2038 trackNum | MEDIA_RECORDER_INFO_COMPLETION_STATUS, 2039 err); 2040 return; 2041 } else { 2042 // Send progress status 2043 notify(MEDIA_RECORDER_EVENT_INFO, 2044 trackNum | MEDIA_RECORDER_INFO_PROGRESS_TIME_STATUS, 2045 timeUs / 1000); 2046 } 2047} 2048 2049void MPEG4Writer::setDriftTimeUs(int64_t driftTimeUs) { 2050 LOGV("setDriftTimeUs: %lld us", driftTimeUs); 2051 Mutex::Autolock autolock(mLock); 2052 mDriftTimeUs = driftTimeUs; 2053} 2054 2055int64_t MPEG4Writer::getDriftTimeUs() { 2056 LOGV("getDriftTimeUs: %lld us", mDriftTimeUs); 2057 Mutex::Autolock autolock(mLock); 2058 return mDriftTimeUs; 2059} 2060 2061bool MPEG4Writer::useNalLengthFour() { 2062 return mUse4ByteNalLength; 2063} 2064 2065void MPEG4Writer::Track::bufferChunk(int64_t timestampUs) { 2066 LOGV("bufferChunk"); 2067 2068 Chunk chunk(this, timestampUs, mChunkSamples); 2069 mOwner->bufferChunk(chunk); 2070 mChunkSamples.clear(); 2071} 2072 2073int64_t MPEG4Writer::Track::getDurationUs() const { 2074 return mTrackDurationUs; 2075} 2076 2077int64_t MPEG4Writer::Track::getEstimatedTrackSizeBytes() const { 2078 return mEstimatedTrackSizeBytes; 2079} 2080 2081status_t MPEG4Writer::Track::checkCodecSpecificData() const { 2082 const char *mime; 2083 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 2084 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime) || 2085 !strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime) || 2086 !strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2087 if (!mCodecSpecificData || 2088 mCodecSpecificDataSize <= 0) { 2089 LOGE("Missing codec specific data"); 2090 return ERROR_MALFORMED; 2091 } 2092 } else { 2093 if (mCodecSpecificData || 2094 mCodecSpecificDataSize > 0) { 2095 LOGE("Unexepected codec specific data found"); 2096 return ERROR_MALFORMED; 2097 } 2098 } 2099 return OK; 2100} 2101 2102void MPEG4Writer::Track::writeTrackHeader( 2103 int32_t trackID, bool use32BitOffset) { 2104 const char *mime; 2105 bool success = mMeta->findCString(kKeyMIMEType, &mime); 2106 CHECK(success); 2107 2108 LOGV("%s track time scale: %d", 2109 mIsAudio? "Audio": "Video", mTimeScale); 2110 2111 time_t now = time(NULL); 2112 int32_t mvhdTimeScale = mOwner->getTimeScale(); 2113 int64_t trakDurationUs = getDurationUs(); 2114 2115 mOwner->beginBox("trak"); 2116 2117 mOwner->beginBox("tkhd"); 2118 // Flags = 7 to indicate that the track is enabled, and 2119 // part of the presentation 2120 mOwner->writeInt32(0x07); // version=0, flags=7 2121 mOwner->writeInt32(now); // creation time 2122 mOwner->writeInt32(now); // modification time 2123 mOwner->writeInt32(trackID); 2124 mOwner->writeInt32(0); // reserved 2125 int32_t tkhdDuration = 2126 (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2127 mOwner->writeInt32(tkhdDuration); // in mvhd timescale 2128 mOwner->writeInt32(0); // reserved 2129 mOwner->writeInt32(0); // reserved 2130 mOwner->writeInt16(0); // layer 2131 mOwner->writeInt16(0); // alternate group 2132 mOwner->writeInt16(mIsAudio ? 0x100 : 0); // volume 2133 mOwner->writeInt16(0); // reserved 2134 2135 mOwner->writeCompositionMatrix(mRotation); // matrix 2136 2137 if (mIsAudio) { 2138 mOwner->writeInt32(0); 2139 mOwner->writeInt32(0); 2140 } else { 2141 int32_t width, height; 2142 bool success = mMeta->findInt32(kKeyWidth, &width); 2143 success = success && mMeta->findInt32(kKeyHeight, &height); 2144 CHECK(success); 2145 2146 mOwner->writeInt32(width << 16); // 32-bit fixed-point value 2147 mOwner->writeInt32(height << 16); // 32-bit fixed-point value 2148 } 2149 mOwner->endBox(); // tkhd 2150 2151 int64_t moovStartTimeUs = mOwner->getStartTimestampUs(); 2152 if (mStartTimestampUs != moovStartTimeUs) { 2153 mOwner->beginBox("edts"); 2154 mOwner->beginBox("elst"); 2155 mOwner->writeInt32(0); // version=0, flags=0: 32-bit time 2156 mOwner->writeInt32(2); // never ends with an empty list 2157 2158 // First elst entry: specify the starting time offset 2159 int64_t offsetUs = mStartTimestampUs - moovStartTimeUs; 2160 LOGV("OffsetUs: %lld", offsetUs); 2161 int32_t seg = (offsetUs * mvhdTimeScale + 5E5) / 1E6; 2162 mOwner->writeInt32(seg); // in mvhd timecale 2163 mOwner->writeInt32(-1); // starting time offset 2164 mOwner->writeInt32(1 << 16); // rate = 1.0 2165 2166 // Second elst entry: specify the track duration 2167 seg = (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2168 mOwner->writeInt32(seg); // in mvhd timescale 2169 mOwner->writeInt32(0); 2170 mOwner->writeInt32(1 << 16); 2171 mOwner->endBox(); 2172 mOwner->endBox(); 2173 } 2174 2175 mOwner->beginBox("mdia"); 2176 2177 mOwner->beginBox("mdhd"); 2178 mOwner->writeInt32(0); // version=0, flags=0 2179 mOwner->writeInt32(now); // creation time 2180 mOwner->writeInt32(now); // modification time 2181 mOwner->writeInt32(mTimeScale); // media timescale 2182 int32_t mdhdDuration = (trakDurationUs * mTimeScale + 5E5) / 1E6; 2183 mOwner->writeInt32(mdhdDuration); // use media timescale 2184 // Language follows the three letter standard ISO-639-2/T 2185 // 'e', 'n', 'g' for "English", for instance. 2186 // Each character is packed as the difference between its ASCII value and 0x60. 2187 // For "English", these are 00101, 01110, 00111. 2188 // XXX: Where is the padding bit located: 0x15C7? 2189 mOwner->writeInt16(0); // language code 2190 mOwner->writeInt16(0); // predefined 2191 mOwner->endBox(); 2192 2193 mOwner->beginBox("hdlr"); 2194 mOwner->writeInt32(0); // version=0, flags=0 2195 mOwner->writeInt32(0); // component type: should be mhlr 2196 mOwner->writeFourcc(mIsAudio ? "soun" : "vide"); // component subtype 2197 mOwner->writeInt32(0); // reserved 2198 mOwner->writeInt32(0); // reserved 2199 mOwner->writeInt32(0); // reserved 2200 // Removing "r" for the name string just makes the string 4 byte aligned 2201 mOwner->writeCString(mIsAudio ? "SoundHandle": "VideoHandle"); // name 2202 mOwner->endBox(); 2203 2204 mOwner->beginBox("minf"); 2205 if (mIsAudio) { 2206 mOwner->beginBox("smhd"); 2207 mOwner->writeInt32(0); // version=0, flags=0 2208 mOwner->writeInt16(0); // balance 2209 mOwner->writeInt16(0); // reserved 2210 mOwner->endBox(); 2211 } else { 2212 mOwner->beginBox("vmhd"); 2213 mOwner->writeInt32(0x01); // version=0, flags=1 2214 mOwner->writeInt16(0); // graphics mode 2215 mOwner->writeInt16(0); // opcolor 2216 mOwner->writeInt16(0); 2217 mOwner->writeInt16(0); 2218 mOwner->endBox(); 2219 } 2220 2221 mOwner->beginBox("dinf"); 2222 mOwner->beginBox("dref"); 2223 mOwner->writeInt32(0); // version=0, flags=0 2224 mOwner->writeInt32(1); // entry count (either url or urn) 2225 // The table index here refers to the sample description index 2226 // in the sample table entries. 2227 mOwner->beginBox("url "); 2228 mOwner->writeInt32(1); // version=0, flags=1 (self-contained) 2229 mOwner->endBox(); // url 2230 mOwner->endBox(); // dref 2231 mOwner->endBox(); // dinf 2232 2233 mOwner->beginBox("stbl"); 2234 2235 mOwner->beginBox("stsd"); 2236 mOwner->writeInt32(0); // version=0, flags=0 2237 mOwner->writeInt32(1); // entry count 2238 if (mIsAudio) { 2239 const char *fourcc = NULL; 2240 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime)) { 2241 fourcc = "samr"; 2242 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2243 fourcc = "sawb"; 2244 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2245 fourcc = "mp4a"; 2246 } else { 2247 LOGE("Unknown mime type '%s'.", mime); 2248 CHECK(!"should not be here, unknown mime type."); 2249 } 2250 2251 mOwner->beginBox(fourcc); // audio format 2252 mOwner->writeInt32(0); // reserved 2253 mOwner->writeInt16(0); // reserved 2254 mOwner->writeInt16(0x1); // data ref index 2255 mOwner->writeInt32(0); // reserved 2256 mOwner->writeInt32(0); // reserved 2257 int32_t nChannels; 2258 CHECK_EQ(true, mMeta->findInt32(kKeyChannelCount, &nChannels)); 2259 mOwner->writeInt16(nChannels); // channel count 2260 mOwner->writeInt16(16); // sample size 2261 mOwner->writeInt16(0); // predefined 2262 mOwner->writeInt16(0); // reserved 2263 2264 int32_t samplerate; 2265 bool success = mMeta->findInt32(kKeySampleRate, &samplerate); 2266 CHECK(success); 2267 mOwner->writeInt32(samplerate << 16); 2268 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2269 mOwner->beginBox("esds"); 2270 CHECK(mCodecSpecificData); 2271 CHECK(mCodecSpecificDataSize > 0); 2272 2273 // Make sure all sizes encode to a single byte. 2274 CHECK(mCodecSpecificDataSize + 23 < 128); 2275 2276 mOwner->writeInt32(0); // version=0, flags=0 2277 mOwner->writeInt8(0x03); // ES_DescrTag 2278 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2279 mOwner->writeInt16(0x0000);// ES_ID 2280 mOwner->writeInt8(0x00); 2281 2282 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2283 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2284 mOwner->writeInt8(0x40); // objectTypeIndication ISO/IEC 14492-2 2285 mOwner->writeInt8(0x15); // streamType AudioStream 2286 2287 mOwner->writeInt16(0x03); // XXX 2288 mOwner->writeInt8(0x00); // buffer size 24-bit 2289 mOwner->writeInt32(96000); // max bit rate 2290 mOwner->writeInt32(96000); // avg bit rate 2291 2292 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2293 mOwner->writeInt8(mCodecSpecificDataSize); 2294 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2295 2296 static const uint8_t kData2[] = { 2297 0x06, // SLConfigDescriptorTag 2298 0x01, 2299 0x02 2300 }; 2301 mOwner->write(kData2, sizeof(kData2)); 2302 2303 mOwner->endBox(); // esds 2304 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime) || 2305 !strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2306 // 3gpp2 Spec AMRSampleEntry fields 2307 mOwner->beginBox("damr"); 2308 mOwner->writeCString(" "); // vendor: 4 bytes 2309 mOwner->writeInt8(0); // decoder version 2310 mOwner->writeInt16(0x83FF); // mode set: all enabled 2311 mOwner->writeInt8(0); // mode change period 2312 mOwner->writeInt8(1); // frames per sample 2313 mOwner->endBox(); 2314 } 2315 mOwner->endBox(); 2316 } else { 2317 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2318 mOwner->beginBox("mp4v"); 2319 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2320 mOwner->beginBox("s263"); 2321 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2322 mOwner->beginBox("avc1"); 2323 } else { 2324 LOGE("Unknown mime type '%s'.", mime); 2325 CHECK(!"should not be here, unknown mime type."); 2326 } 2327 2328 mOwner->writeInt32(0); // reserved 2329 mOwner->writeInt16(0); // reserved 2330 mOwner->writeInt16(1); // data ref index 2331 mOwner->writeInt16(0); // predefined 2332 mOwner->writeInt16(0); // reserved 2333 mOwner->writeInt32(0); // predefined 2334 mOwner->writeInt32(0); // predefined 2335 mOwner->writeInt32(0); // predefined 2336 2337 int32_t width, height; 2338 bool success = mMeta->findInt32(kKeyWidth, &width); 2339 success = success && mMeta->findInt32(kKeyHeight, &height); 2340 CHECK(success); 2341 2342 mOwner->writeInt16(width); 2343 mOwner->writeInt16(height); 2344 mOwner->writeInt32(0x480000); // horiz resolution 2345 mOwner->writeInt32(0x480000); // vert resolution 2346 mOwner->writeInt32(0); // reserved 2347 mOwner->writeInt16(1); // frame count 2348 mOwner->write(" ", 32); 2349 mOwner->writeInt16(0x18); // depth 2350 mOwner->writeInt16(-1); // predefined 2351 2352 CHECK(23 + mCodecSpecificDataSize < 128); 2353 2354 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2355 CHECK(mCodecSpecificData); 2356 CHECK(mCodecSpecificDataSize > 0); 2357 mOwner->beginBox("esds"); 2358 2359 mOwner->writeInt32(0); // version=0, flags=0 2360 2361 mOwner->writeInt8(0x03); // ES_DescrTag 2362 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2363 mOwner->writeInt16(0x0000); // ES_ID 2364 mOwner->writeInt8(0x1f); 2365 2366 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2367 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2368 mOwner->writeInt8(0x20); // objectTypeIndication ISO/IEC 14492-2 2369 mOwner->writeInt8(0x11); // streamType VisualStream 2370 2371 static const uint8_t kData[] = { 2372 0x01, 0x77, 0x00, 2373 0x00, 0x03, 0xe8, 0x00, 2374 0x00, 0x03, 0xe8, 0x00 2375 }; 2376 mOwner->write(kData, sizeof(kData)); 2377 2378 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2379 2380 mOwner->writeInt8(mCodecSpecificDataSize); 2381 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2382 2383 static const uint8_t kData2[] = { 2384 0x06, // SLConfigDescriptorTag 2385 0x01, 2386 0x02 2387 }; 2388 mOwner->write(kData2, sizeof(kData2)); 2389 2390 mOwner->endBox(); // esds 2391 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2392 mOwner->beginBox("d263"); 2393 2394 mOwner->writeInt32(0); // vendor 2395 mOwner->writeInt8(0); // decoder version 2396 mOwner->writeInt8(10); // level: 10 2397 mOwner->writeInt8(0); // profile: 0 2398 2399 mOwner->endBox(); // d263 2400 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2401 CHECK(mCodecSpecificData); 2402 CHECK(mCodecSpecificDataSize >= 5); 2403 2404 // Patch avcc's lengthSize field to match the number 2405 // of bytes we use to indicate the size of a nal unit. 2406 uint8_t *ptr = (uint8_t *)mCodecSpecificData; 2407 ptr[4] = 2408 (ptr[4] & 0xfc) 2409 | (mOwner->useNalLengthFour() ? 3 : 1); 2410 2411 mOwner->beginBox("avcC"); 2412 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2413 mOwner->endBox(); // avcC 2414 } 2415 2416 mOwner->beginBox("pasp"); 2417 // This is useful if the pixel is not square 2418 mOwner->writeInt32(1 << 16); // hspacing 2419 mOwner->writeInt32(1 << 16); // vspacing 2420 mOwner->endBox(); // pasp 2421 mOwner->endBox(); // mp4v, s263 or avc1 2422 } 2423 mOwner->endBox(); // stsd 2424 2425 mOwner->beginBox("stts"); 2426 mOwner->writeInt32(0); // version=0, flags=0 2427 mOwner->writeInt32(mNumSttsTableEntries); 2428 int64_t prevTimestampUs = 0; 2429 for (List<SttsTableEntry>::iterator it = mSttsTableEntries.begin(); 2430 it != mSttsTableEntries.end(); ++it) { 2431 mOwner->writeInt32(it->sampleCount); 2432 2433 // Make sure that we are calculating the sample duration the exactly 2434 // same way as we made decision on how to create stts entries. 2435 int64_t currTimestampUs = prevTimestampUs + it->sampleDurationUs; 2436 int32_t dur = ((currTimestampUs * mTimeScale + 500000LL) / 1000000LL - 2437 (prevTimestampUs * mTimeScale + 500000LL) / 1000000LL); 2438 prevTimestampUs += (it->sampleCount * it->sampleDurationUs); 2439 2440 mOwner->writeInt32(dur); 2441 } 2442 mOwner->endBox(); // stts 2443 2444 if (!mIsAudio) { 2445 mOwner->beginBox("stss"); 2446 mOwner->writeInt32(0); // version=0, flags=0 2447 mOwner->writeInt32(mNumStssTableEntries); // number of sync frames 2448 for (List<int32_t>::iterator it = mStssTableEntries.begin(); 2449 it != mStssTableEntries.end(); ++it) { 2450 mOwner->writeInt32(*it); 2451 } 2452 mOwner->endBox(); // stss 2453 } 2454 2455 mOwner->beginBox("stsz"); 2456 mOwner->writeInt32(0); // version=0, flags=0 2457 if (mSamplesHaveSameSize) { 2458 List<size_t>::iterator it = mSampleSizes.begin(); 2459 mOwner->writeInt32(*it); // default sample size 2460 } else { 2461 mOwner->writeInt32(0); 2462 } 2463 mOwner->writeInt32(mNumSamples); 2464 if (!mSamplesHaveSameSize) { 2465 for (List<size_t>::iterator it = mSampleSizes.begin(); 2466 it != mSampleSizes.end(); ++it) { 2467 mOwner->writeInt32(*it); 2468 } 2469 } 2470 mOwner->endBox(); // stsz 2471 2472 mOwner->beginBox("stsc"); 2473 mOwner->writeInt32(0); // version=0, flags=0 2474 mOwner->writeInt32(mNumStscTableEntries); 2475 for (List<StscTableEntry>::iterator it = mStscTableEntries.begin(); 2476 it != mStscTableEntries.end(); ++it) { 2477 mOwner->writeInt32(it->firstChunk); 2478 mOwner->writeInt32(it->samplesPerChunk); 2479 mOwner->writeInt32(it->sampleDescriptionId); 2480 } 2481 mOwner->endBox(); // stsc 2482 mOwner->beginBox(use32BitOffset? "stco": "co64"); 2483 mOwner->writeInt32(0); // version=0, flags=0 2484 mOwner->writeInt32(mNumStcoTableEntries); 2485 for (List<off64_t>::iterator it = mChunkOffsets.begin(); 2486 it != mChunkOffsets.end(); ++it) { 2487 if (use32BitOffset) { 2488 mOwner->writeInt32(static_cast<int32_t>(*it)); 2489 } else { 2490 mOwner->writeInt64((*it)); 2491 } 2492 } 2493 mOwner->endBox(); // stco or co64 2494 2495 mOwner->endBox(); // stbl 2496 mOwner->endBox(); // minf 2497 mOwner->endBox(); // mdia 2498 mOwner->endBox(); // trak 2499} 2500 2501} // namespace android 2502