MPEG4Writer.cpp revision cb09c03450bb09304ad5afd2d223254bf2e6e38c
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(off_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<off_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; 205 206 void updateTrackSizeEstimate(); 207 void addOneStscTableEntry(size_t chunkId, size_t sampleId); 208 void addOneStssTableEntry(size_t sampleId); 209 void addOneSttsTableEntry(size_t sampleCount, int64_t durationUs); 210 211 Track(const Track &); 212 Track &operator=(const Track &); 213}; 214 215MPEG4Writer::MPEG4Writer(const char *filename) 216 : mFile(fopen(filename, "wb")), 217 mUse4ByteNalLength(true), 218 mUse32BitOffset(true), 219 mIsFileSizeLimitExplicitlyRequested(false), 220 mPaused(false), 221 mStarted(false), 222 mOffset(0), 223 mMdatOffset(0), 224 mEstimatedMoovBoxSize(0), 225 mInterleaveDurationUs(1000000) { 226 CHECK(mFile != NULL); 227} 228 229MPEG4Writer::MPEG4Writer(int fd) 230 : mFile(fdopen(fd, "wb")), 231 mUse4ByteNalLength(true), 232 mUse32BitOffset(true), 233 mIsFileSizeLimitExplicitlyRequested(false), 234 mPaused(false), 235 mStarted(false), 236 mOffset(0), 237 mMdatOffset(0), 238 mEstimatedMoovBoxSize(0), 239 mInterleaveDurationUs(1000000) { 240 CHECK(mFile != NULL); 241} 242 243MPEG4Writer::~MPEG4Writer() { 244 stop(); 245 246 while (!mTracks.empty()) { 247 List<Track *>::iterator it = mTracks.begin(); 248 delete *it; 249 (*it) = NULL; 250 mTracks.erase(it); 251 } 252 mTracks.clear(); 253} 254 255status_t MPEG4Writer::dump( 256 int fd, const Vector<String16>& args) { 257 const size_t SIZE = 256; 258 char buffer[SIZE]; 259 String8 result; 260 snprintf(buffer, SIZE, " MPEG4Writer %p\n", this); 261 result.append(buffer); 262 snprintf(buffer, SIZE, " mStarted: %s\n", mStarted? "true": "false"); 263 result.append(buffer); 264 ::write(fd, result.string(), result.size()); 265 for (List<Track *>::iterator it = mTracks.begin(); 266 it != mTracks.end(); ++it) { 267 (*it)->dump(fd, args); 268 } 269 return OK; 270} 271 272status_t MPEG4Writer::Track::dump( 273 int fd, const Vector<String16>& args) const { 274 const size_t SIZE = 256; 275 char buffer[SIZE]; 276 String8 result; 277 snprintf(buffer, SIZE, " %s track\n", mIsAudio? "Audio": "Video"); 278 result.append(buffer); 279 snprintf(buffer, SIZE, " reached EOS: %s\n", 280 mReachedEOS? "true": "false"); 281 result.append(buffer); 282 ::write(fd, result.string(), result.size()); 283 return OK; 284} 285 286status_t MPEG4Writer::addSource(const sp<MediaSource> &source) { 287 Track *track = new Track(this, source); 288 mTracks.push_back(track); 289 290 return OK; 291} 292 293status_t MPEG4Writer::startTracks(MetaData *params) { 294 for (List<Track *>::iterator it = mTracks.begin(); 295 it != mTracks.end(); ++it) { 296 status_t err = (*it)->start(params); 297 298 if (err != OK) { 299 for (List<Track *>::iterator it2 = mTracks.begin(); 300 it2 != it; ++it2) { 301 (*it2)->stop(); 302 } 303 304 return err; 305 } 306 } 307 return OK; 308} 309 310int64_t MPEG4Writer::estimateMoovBoxSize(int32_t bitRate) { 311 // This implementation is highly experimental/heurisitic. 312 // 313 // Statistical analysis shows that metadata usually accounts 314 // for a small portion of the total file size, usually < 0.6%. 315 316 // The default MIN_MOOV_BOX_SIZE is set to 0.6% x 1MB / 2, 317 // where 1MB is the common file size limit for MMS application. 318 // The default MAX _MOOV_BOX_SIZE value is based on about 3 319 // minute video recording with a bit rate about 3 Mbps, because 320 // statistics also show that most of the video captured are going 321 // to be less than 3 minutes. 322 323 // If the estimation is wrong, we will pay the price of wasting 324 // some reserved space. This should not happen so often statistically. 325 static const int32_t factor = mUse32BitOffset? 1: 2; 326 static const int64_t MIN_MOOV_BOX_SIZE = 3 * 1024; // 3 KB 327 static const int64_t MAX_MOOV_BOX_SIZE = (180 * 3000000 * 6LL / 8000); 328 int64_t size = MIN_MOOV_BOX_SIZE; 329 330 // Max file size limit is set 331 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 332 size = mMaxFileSizeLimitBytes * 6 / 1000; 333 } 334 335 // Max file duration limit is set 336 if (mMaxFileDurationLimitUs != 0) { 337 if (bitRate > 0) { 338 int64_t size2 = 339 ((mMaxFileDurationLimitUs * bitRate * 6) / 1000 / 8000000); 340 if (mMaxFileSizeLimitBytes != 0 && mIsFileSizeLimitExplicitlyRequested) { 341 // When both file size and duration limits are set, 342 // we use the smaller limit of the two. 343 if (size > size2) { 344 size = size2; 345 } 346 } else { 347 // Only max file duration limit is set 348 size = size2; 349 } 350 } 351 } 352 353 if (size < MIN_MOOV_BOX_SIZE) { 354 size = MIN_MOOV_BOX_SIZE; 355 } 356 357 // Any long duration recording will be probably end up with 358 // non-streamable mp4 file. 359 if (size > MAX_MOOV_BOX_SIZE) { 360 size = MAX_MOOV_BOX_SIZE; 361 } 362 363 LOGI("limits: %lld/%lld bytes/us, bit rate: %d bps and the estimated" 364 " moov size %lld bytes", 365 mMaxFileSizeLimitBytes, mMaxFileDurationLimitUs, bitRate, size); 366 return factor * size; 367} 368 369status_t MPEG4Writer::start(MetaData *param) { 370 if (mFile == NULL) { 371 return UNKNOWN_ERROR; 372 } 373 374 /* 375 * Check mMaxFileSizeLimitBytes at the beginning 376 * since mMaxFileSizeLimitBytes may be implicitly 377 * changed later for 32-bit file offset even if 378 * user does not ask to set it explicitly. 379 */ 380 if (mMaxFileSizeLimitBytes != 0) { 381 mIsFileSizeLimitExplicitlyRequested = true; 382 } 383 384 int32_t use64BitOffset; 385 if (param && 386 param->findInt32(kKey64BitFileOffset, &use64BitOffset) && 387 use64BitOffset) { 388 mUse32BitOffset = false; 389 } 390 391 if (mUse32BitOffset) { 392 // Implicit 32 bit file size limit 393 if (mMaxFileSizeLimitBytes == 0) { 394 mMaxFileSizeLimitBytes = kMax32BitFileSize; 395 } 396 397 // If file size is set to be larger than the 32 bit file 398 // size limit, treat it as an error. 399 if (mMaxFileSizeLimitBytes > kMax32BitFileSize) { 400 LOGW("32-bit file size limit (%lld bytes) too big. " 401 "It is changed to %lld bytes", 402 mMaxFileSizeLimitBytes, kMax32BitFileSize); 403 mMaxFileSizeLimitBytes = kMax32BitFileSize; 404 } 405 } 406 407 int32_t use2ByteNalLength; 408 if (param && 409 param->findInt32(kKey2ByteNalLength, &use2ByteNalLength) && 410 use2ByteNalLength) { 411 mUse4ByteNalLength = false; 412 } 413 414 mStartTimestampUs = -1; 415 416 if (mStarted) { 417 if (mPaused) { 418 mPaused = false; 419 return startTracks(param); 420 } 421 return OK; 422 } 423 424 if (!param || 425 !param->findInt32(kKeyTimeScale, &mTimeScale)) { 426 mTimeScale = 1000; 427 } 428 CHECK(mTimeScale > 0); 429 LOGV("movie time scale: %d", mTimeScale); 430 431 mStreamableFile = true; 432 mWriteMoovBoxToMemory = false; 433 mMoovBoxBuffer = NULL; 434 mMoovBoxBufferOffset = 0; 435 436 beginBox("ftyp"); 437 { 438 int32_t fileType; 439 if (param && param->findInt32(kKeyFileType, &fileType) && 440 fileType != OUTPUT_FORMAT_MPEG_4) { 441 writeFourcc("3gp4"); 442 } else { 443 writeFourcc("isom"); 444 } 445 } 446 writeInt32(0); 447 writeFourcc("isom"); 448 writeFourcc("3gp4"); 449 endBox(); 450 451 mFreeBoxOffset = mOffset; 452 453 if (mEstimatedMoovBoxSize == 0) { 454 int32_t bitRate = -1; 455 if (param) { 456 param->findInt32(kKeyBitRate, &bitRate); 457 } 458 mEstimatedMoovBoxSize = estimateMoovBoxSize(bitRate); 459 } 460 CHECK(mEstimatedMoovBoxSize >= 8); 461 fseeko(mFile, mFreeBoxOffset, SEEK_SET); 462 writeInt32(mEstimatedMoovBoxSize); 463 write("free", 4); 464 465 mMdatOffset = mFreeBoxOffset + mEstimatedMoovBoxSize; 466 mOffset = mMdatOffset; 467 fseeko(mFile, mMdatOffset, SEEK_SET); 468 if (mUse32BitOffset) { 469 write("????mdat", 8); 470 } else { 471 write("\x00\x00\x00\x01mdat????????", 16); 472 } 473 474 status_t err = startWriterThread(); 475 if (err != OK) { 476 return err; 477 } 478 479 err = startTracks(param); 480 if (err != OK) { 481 return err; 482 } 483 484 mStarted = true; 485 return OK; 486} 487 488bool MPEG4Writer::use32BitFileOffset() const { 489 return mUse32BitOffset; 490} 491 492status_t MPEG4Writer::pause() { 493 if (mFile == NULL) { 494 return OK; 495 } 496 mPaused = true; 497 status_t err = OK; 498 for (List<Track *>::iterator it = mTracks.begin(); 499 it != mTracks.end(); ++it) { 500 status_t status = (*it)->pause(); 501 if (status != OK) { 502 err = status; 503 } 504 } 505 return err; 506} 507 508void MPEG4Writer::stopWriterThread() { 509 LOGD("Stopping writer thread"); 510 511 { 512 Mutex::Autolock autolock(mLock); 513 514 mDone = true; 515 mChunkReadyCondition.signal(); 516 } 517 518 void *dummy; 519 pthread_join(mThread, &dummy); 520 LOGD("Writer thread stopped"); 521} 522 523status_t MPEG4Writer::stop() { 524 if (mFile == NULL) { 525 return OK; 526 } 527 528 status_t err = OK; 529 int64_t maxDurationUs = 0; 530 for (List<Track *>::iterator it = mTracks.begin(); 531 it != mTracks.end(); ++it) { 532 status_t status = (*it)->stop(); 533 if (err == OK && status != OK) { 534 err = status; 535 } 536 537 int64_t durationUs = (*it)->getDurationUs(); 538 if (durationUs > maxDurationUs) { 539 maxDurationUs = durationUs; 540 } 541 } 542 543 stopWriterThread(); 544 545 // Do not write out movie header on error. 546 if (err != OK) { 547 fflush(mFile); 548 fclose(mFile); 549 mFile = NULL; 550 mStarted = false; 551 return err; 552 } 553 554 // Fix up the size of the 'mdat' chunk. 555 if (mUse32BitOffset) { 556 fseeko(mFile, mMdatOffset, SEEK_SET); 557 int32_t size = htonl(static_cast<int32_t>(mOffset - mMdatOffset)); 558 fwrite(&size, 1, 4, mFile); 559 } else { 560 fseeko(mFile, mMdatOffset + 8, SEEK_SET); 561 int64_t size = mOffset - mMdatOffset; 562 size = hton64(size); 563 fwrite(&size, 1, 8, mFile); 564 } 565 fseeko(mFile, mOffset, SEEK_SET); 566 567 time_t now = time(NULL); 568 const off_t moovOffset = mOffset; 569 mWriteMoovBoxToMemory = true; 570 mMoovBoxBuffer = (uint8_t *) malloc(mEstimatedMoovBoxSize); 571 mMoovBoxBufferOffset = 0; 572 CHECK(mMoovBoxBuffer != NULL); 573 int32_t duration = (maxDurationUs * mTimeScale + 5E5) / 1E6; 574 575 beginBox("moov"); 576 577 beginBox("mvhd"); 578 writeInt32(0); // version=0, flags=0 579 writeInt32(now); // creation time 580 writeInt32(now); // modification time 581 writeInt32(mTimeScale); // mvhd timescale 582 writeInt32(duration); 583 writeInt32(0x10000); // rate: 1.0 584 writeInt16(0x100); // volume 585 writeInt16(0); // reserved 586 writeInt32(0); // reserved 587 writeInt32(0); // reserved 588 writeInt32(0x10000); // matrix 589 writeInt32(0); 590 writeInt32(0); 591 writeInt32(0); 592 writeInt32(0x10000); 593 writeInt32(0); 594 writeInt32(0); 595 writeInt32(0); 596 writeInt32(0x40000000); 597 writeInt32(0); // predefined 598 writeInt32(0); // predefined 599 writeInt32(0); // predefined 600 writeInt32(0); // predefined 601 writeInt32(0); // predefined 602 writeInt32(0); // predefined 603 writeInt32(mTracks.size() + 1); // nextTrackID 604 endBox(); // mvhd 605 606 int32_t id = 1; 607 for (List<Track *>::iterator it = mTracks.begin(); 608 it != mTracks.end(); ++it, ++id) { 609 (*it)->writeTrackHeader(id, mUse32BitOffset); 610 } 611 endBox(); // moov 612 613 mWriteMoovBoxToMemory = false; 614 if (mStreamableFile) { 615 CHECK(mMoovBoxBufferOffset + 8 <= mEstimatedMoovBoxSize); 616 617 // Moov box 618 fseeko(mFile, mFreeBoxOffset, SEEK_SET); 619 mOffset = mFreeBoxOffset; 620 write(mMoovBoxBuffer, 1, mMoovBoxBufferOffset, mFile); 621 622 // Free box 623 fseeko(mFile, mOffset, SEEK_SET); 624 writeInt32(mEstimatedMoovBoxSize - mMoovBoxBufferOffset); 625 write("free", 4); 626 627 // Free temp memory 628 free(mMoovBoxBuffer); 629 mMoovBoxBuffer = NULL; 630 mMoovBoxBufferOffset = 0; 631 } else { 632 LOGI("The mp4 file will not be streamable."); 633 } 634 635 CHECK(mBoxes.empty()); 636 637 fflush(mFile); 638 fclose(mFile); 639 mFile = NULL; 640 mStarted = false; 641 return err; 642} 643 644status_t MPEG4Writer::setInterleaveDuration(uint32_t durationUs) { 645 mInterleaveDurationUs = durationUs; 646 return OK; 647} 648 649void MPEG4Writer::lock() { 650 mLock.lock(); 651} 652 653void MPEG4Writer::unlock() { 654 mLock.unlock(); 655} 656 657off_t MPEG4Writer::addSample_l(MediaBuffer *buffer) { 658 off_t old_offset = mOffset; 659 660 fwrite((const uint8_t *)buffer->data() + buffer->range_offset(), 661 1, buffer->range_length(), mFile); 662 663 mOffset += buffer->range_length(); 664 665 return old_offset; 666} 667 668static void StripStartcode(MediaBuffer *buffer) { 669 if (buffer->range_length() < 4) { 670 return; 671 } 672 673 const uint8_t *ptr = 674 (const uint8_t *)buffer->data() + buffer->range_offset(); 675 676 if (!memcmp(ptr, "\x00\x00\x00\x01", 4)) { 677 buffer->set_range( 678 buffer->range_offset() + 4, buffer->range_length() - 4); 679 } 680} 681 682off_t MPEG4Writer::addLengthPrefixedSample_l(MediaBuffer *buffer) { 683 off_t old_offset = mOffset; 684 685 size_t length = buffer->range_length(); 686 687 if (mUse4ByteNalLength) { 688 uint8_t x = length >> 24; 689 fwrite(&x, 1, 1, mFile); 690 x = (length >> 16) & 0xff; 691 fwrite(&x, 1, 1, mFile); 692 x = (length >> 8) & 0xff; 693 fwrite(&x, 1, 1, mFile); 694 x = length & 0xff; 695 fwrite(&x, 1, 1, mFile); 696 697 fwrite((const uint8_t *)buffer->data() + buffer->range_offset(), 698 1, length, mFile); 699 mOffset += length + 4; 700 } else { 701 CHECK(length < 65536); 702 703 uint8_t x = length >> 8; 704 fwrite(&x, 1, 1, mFile); 705 x = length & 0xff; 706 fwrite(&x, 1, 1, mFile); 707 fwrite((const uint8_t *)buffer->data() + buffer->range_offset(), 708 1, length, mFile); 709 mOffset += length + 2; 710 } 711 712 return old_offset; 713} 714 715size_t MPEG4Writer::write( 716 const void *ptr, size_t size, size_t nmemb, FILE *stream) { 717 718 const size_t bytes = size * nmemb; 719 if (mWriteMoovBoxToMemory) { 720 off_t moovBoxSize = 8 + mMoovBoxBufferOffset + bytes; 721 if (moovBoxSize > mEstimatedMoovBoxSize) { 722 for (List<off_t>::iterator it = mBoxes.begin(); 723 it != mBoxes.end(); ++it) { 724 (*it) += mOffset; 725 } 726 fseeko(mFile, mOffset, SEEK_SET); 727 fwrite(mMoovBoxBuffer, 1, mMoovBoxBufferOffset, stream); 728 fwrite(ptr, size, nmemb, stream); 729 mOffset += (bytes + mMoovBoxBufferOffset); 730 free(mMoovBoxBuffer); 731 mMoovBoxBuffer = NULL; 732 mMoovBoxBufferOffset = 0; 733 mWriteMoovBoxToMemory = false; 734 mStreamableFile = false; 735 } else { 736 memcpy(mMoovBoxBuffer + mMoovBoxBufferOffset, ptr, bytes); 737 mMoovBoxBufferOffset += bytes; 738 } 739 } else { 740 fwrite(ptr, size, nmemb, stream); 741 mOffset += bytes; 742 } 743 return bytes; 744} 745 746void MPEG4Writer::beginBox(const char *fourcc) { 747 CHECK_EQ(strlen(fourcc), 4); 748 749 mBoxes.push_back(mWriteMoovBoxToMemory? 750 mMoovBoxBufferOffset: mOffset); 751 752 writeInt32(0); 753 writeFourcc(fourcc); 754} 755 756void MPEG4Writer::endBox() { 757 CHECK(!mBoxes.empty()); 758 759 off_t offset = *--mBoxes.end(); 760 mBoxes.erase(--mBoxes.end()); 761 762 if (mWriteMoovBoxToMemory) { 763 int32_t x = htonl(mMoovBoxBufferOffset - offset); 764 memcpy(mMoovBoxBuffer + offset, &x, 4); 765 } else { 766 fseeko(mFile, offset, SEEK_SET); 767 writeInt32(mOffset - offset); 768 mOffset -= 4; 769 fseeko(mFile, mOffset, SEEK_SET); 770 } 771} 772 773void MPEG4Writer::writeInt8(int8_t x) { 774 write(&x, 1, 1, mFile); 775} 776 777void MPEG4Writer::writeInt16(int16_t x) { 778 x = htons(x); 779 write(&x, 1, 2, mFile); 780} 781 782void MPEG4Writer::writeInt32(int32_t x) { 783 x = htonl(x); 784 write(&x, 1, 4, mFile); 785} 786 787void MPEG4Writer::writeInt64(int64_t x) { 788 x = hton64(x); 789 write(&x, 1, 8, mFile); 790} 791 792void MPEG4Writer::writeCString(const char *s) { 793 size_t n = strlen(s); 794 write(s, 1, n + 1, mFile); 795} 796 797void MPEG4Writer::writeFourcc(const char *s) { 798 CHECK_EQ(strlen(s), 4); 799 write(s, 1, 4, mFile); 800} 801 802void MPEG4Writer::write(const void *data, size_t size) { 803 write(data, 1, size, mFile); 804} 805 806bool MPEG4Writer::isFileStreamable() const { 807 return mStreamableFile; 808} 809 810bool MPEG4Writer::exceedsFileSizeLimit() { 811 // No limit 812 if (mMaxFileSizeLimitBytes == 0) { 813 return false; 814 } 815 816 int64_t nTotalBytesEstimate = static_cast<int64_t>(mEstimatedMoovBoxSize); 817 for (List<Track *>::iterator it = mTracks.begin(); 818 it != mTracks.end(); ++it) { 819 nTotalBytesEstimate += (*it)->getEstimatedTrackSizeBytes(); 820 } 821 822 return (nTotalBytesEstimate >= mMaxFileSizeLimitBytes); 823} 824 825bool MPEG4Writer::exceedsFileDurationLimit() { 826 // No limit 827 if (mMaxFileDurationLimitUs == 0) { 828 return false; 829 } 830 831 for (List<Track *>::iterator it = mTracks.begin(); 832 it != mTracks.end(); ++it) { 833 if ((*it)->getDurationUs() >= mMaxFileDurationLimitUs) { 834 return true; 835 } 836 } 837 return false; 838} 839 840bool MPEG4Writer::reachedEOS() { 841 bool allDone = true; 842 for (List<Track *>::iterator it = mTracks.begin(); 843 it != mTracks.end(); ++it) { 844 if (!(*it)->reachedEOS()) { 845 allDone = false; 846 break; 847 } 848 } 849 850 return allDone; 851} 852 853void MPEG4Writer::setStartTimestampUs(int64_t timeUs) { 854 LOGI("setStartTimestampUs: %lld", timeUs); 855 CHECK(timeUs >= 0); 856 Mutex::Autolock autoLock(mLock); 857 if (mStartTimestampUs < 0 || mStartTimestampUs > timeUs) { 858 mStartTimestampUs = timeUs; 859 LOGI("Earliest track starting time: %lld", mStartTimestampUs); 860 } 861} 862 863int64_t MPEG4Writer::getStartTimestampUs() { 864 Mutex::Autolock autoLock(mLock); 865 return mStartTimestampUs; 866} 867 868size_t MPEG4Writer::numTracks() { 869 Mutex::Autolock autolock(mLock); 870 return mTracks.size(); 871} 872 873//////////////////////////////////////////////////////////////////////////////// 874 875MPEG4Writer::Track::Track( 876 MPEG4Writer *owner, const sp<MediaSource> &source) 877 : mOwner(owner), 878 mMeta(source->getFormat()), 879 mSource(source), 880 mDone(false), 881 mPaused(false), 882 mResumed(false), 883 mTrackDurationUs(0), 884 mEstimatedTrackSizeBytes(0), 885 mSamplesHaveSameSize(true), 886 mCodecSpecificData(NULL), 887 mCodecSpecificDataSize(0), 888 mGotAllCodecSpecificData(false), 889 mReachedEOS(false) { 890 getCodecSpecificDataFromInputFormatIfPossible(); 891 892 const char *mime; 893 mMeta->findCString(kKeyMIMEType, &mime); 894 mIsAvc = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC); 895 mIsAudio = !strncasecmp(mime, "audio/", 6); 896 mIsMPEG4 = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) || 897 !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC); 898 899 setTimeScale(); 900} 901 902void MPEG4Writer::Track::updateTrackSizeEstimate() { 903 904 int64_t stcoBoxSizeBytes = mOwner->use32BitFileOffset() 905 ? mNumStcoTableEntries * 4 906 : mNumStcoTableEntries * 8; 907 908 int64_t stszBoxSizeBytes = mSamplesHaveSameSize? 4: (mNumSamples * 4); 909 910 mEstimatedTrackSizeBytes = mMdatSizeBytes; // media data size 911 if (!mOwner->isFileStreamable()) { 912 // Reserved free space is not large enough to hold 913 // all meta data and thus wasted. 914 mEstimatedTrackSizeBytes += mNumStscTableEntries * 12 + // stsc box size 915 mNumStssTableEntries * 4 + // stss box size 916 mNumSttsTableEntries * 8 + // stts box size 917 stcoBoxSizeBytes + // stco box size 918 stszBoxSizeBytes; // stsz box size 919 } 920} 921 922void MPEG4Writer::Track::addOneStscTableEntry( 923 size_t chunkId, size_t sampleId) { 924 925 StscTableEntry stscEntry(chunkId, sampleId, 1); 926 mStscTableEntries.push_back(stscEntry); 927 ++mNumStscTableEntries; 928} 929 930void MPEG4Writer::Track::addOneStssTableEntry(size_t sampleId) { 931 mStssTableEntries.push_back(sampleId); 932 ++mNumStssTableEntries; 933} 934 935void MPEG4Writer::Track::addOneSttsTableEntry( 936 size_t sampleCount, int64_t durationUs) { 937 938 SttsTableEntry sttsEntry(sampleCount, durationUs); 939 mSttsTableEntries.push_back(sttsEntry); 940 ++mNumSttsTableEntries; 941} 942 943void MPEG4Writer::Track::addChunkOffset(off_t offset) { 944 ++mNumStcoTableEntries; 945 mChunkOffsets.push_back(offset); 946} 947 948void MPEG4Writer::Track::setTimeScale() { 949 LOGV("setTimeScale"); 950 // Default time scale 951 mTimeScale = 90000; 952 953 if (mIsAudio) { 954 // Use the sampling rate as the default time scale for audio track. 955 int32_t sampleRate; 956 bool success = mMeta->findInt32(kKeySampleRate, &sampleRate); 957 CHECK(success); 958 mTimeScale = sampleRate; 959 } 960 961 // If someone would like to overwrite the timescale, use user-supplied value. 962 int32_t timeScale; 963 if (mMeta->findInt32(kKeyTimeScale, &timeScale)) { 964 mTimeScale = timeScale; 965 } 966 967 CHECK(mTimeScale > 0); 968} 969 970void MPEG4Writer::Track::getCodecSpecificDataFromInputFormatIfPossible() { 971 const char *mime; 972 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 973 974 if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC)) { 975 uint32_t type; 976 const void *data; 977 size_t size; 978 if (mMeta->findData(kKeyAVCC, &type, &data, &size)) { 979 mCodecSpecificData = malloc(size); 980 mCodecSpecificDataSize = size; 981 memcpy(mCodecSpecificData, data, size); 982 mGotAllCodecSpecificData = true; 983 } 984 } else if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) 985 || !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC)) { 986 uint32_t type; 987 const void *data; 988 size_t size; 989 if (mMeta->findData(kKeyESDS, &type, &data, &size)) { 990 ESDS esds(data, size); 991 if (esds.getCodecSpecificInfo(&data, &size) == OK) { 992 mCodecSpecificData = malloc(size); 993 mCodecSpecificDataSize = size; 994 memcpy(mCodecSpecificData, data, size); 995 mGotAllCodecSpecificData = true; 996 } 997 } 998 } 999} 1000 1001MPEG4Writer::Track::~Track() { 1002 stop(); 1003 1004 if (mCodecSpecificData != NULL) { 1005 free(mCodecSpecificData); 1006 mCodecSpecificData = NULL; 1007 } 1008} 1009 1010void MPEG4Writer::Track::initTrackingProgressStatus(MetaData *params) { 1011 LOGV("initTrackingProgressStatus"); 1012 mPreviousTrackTimeUs = -1; 1013 mTrackingProgressStatus = false; 1014 mTrackEveryTimeDurationUs = 0; 1015 { 1016 int64_t timeUs; 1017 if (params && params->findInt64(kKeyTrackTimeStatus, &timeUs)) { 1018 LOGV("Receive request to track progress status for every %lld us", timeUs); 1019 mTrackEveryTimeDurationUs = timeUs; 1020 mTrackingProgressStatus = true; 1021 } 1022 } 1023} 1024 1025// static 1026void *MPEG4Writer::ThreadWrapper(void *me) { 1027 LOGV("ThreadWrapper: %p", me); 1028 MPEG4Writer *writer = static_cast<MPEG4Writer *>(me); 1029 writer->threadFunc(); 1030 return NULL; 1031} 1032 1033void MPEG4Writer::bufferChunk(const Chunk& chunk) { 1034 LOGV("bufferChunk: %p", chunk.mTrack); 1035 Mutex::Autolock autolock(mLock); 1036 CHECK_EQ(mDone, false); 1037 1038 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1039 it != mChunkInfos.end(); ++it) { 1040 1041 if (chunk.mTrack == it->mTrack) { // Found owner 1042 it->mChunks.push_back(chunk); 1043 mChunkReadyCondition.signal(); 1044 return; 1045 } 1046 } 1047 1048 CHECK("Received a chunk for a unknown track" == 0); 1049} 1050 1051void MPEG4Writer::writeFirstChunk(ChunkInfo* info) { 1052 LOGV("writeFirstChunk: %p", info->mTrack); 1053 1054 List<Chunk>::iterator chunkIt = info->mChunks.begin(); 1055 for (List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1056 it != chunkIt->mSamples.end(); ++it) { 1057 1058 off_t offset = info->mTrack->isAvc() 1059 ? addLengthPrefixedSample_l(*it) 1060 : addSample_l(*it); 1061 if (it == chunkIt->mSamples.begin()) { 1062 info->mTrack->addChunkOffset(offset); 1063 } 1064 } 1065 1066 // Done with the current chunk. 1067 // Release all the samples in this chunk. 1068 while (!chunkIt->mSamples.empty()) { 1069 List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1070 (*it)->release(); 1071 (*it) = NULL; 1072 chunkIt->mSamples.erase(it); 1073 } 1074 chunkIt->mSamples.clear(); 1075 info->mChunks.erase(chunkIt); 1076} 1077 1078void MPEG4Writer::writeChunks() { 1079 LOGV("writeChunks"); 1080 size_t outstandingChunks = 0; 1081 while (!mChunkInfos.empty()) { 1082 List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1083 while (!it->mChunks.empty()) { 1084 CHECK_EQ(OK, writeOneChunk()); 1085 ++outstandingChunks; 1086 } 1087 it->mTrack = NULL; 1088 mChunkInfos.erase(it); 1089 } 1090 mChunkInfos.clear(); 1091 LOGD("%d chunks are written in the last batch", outstandingChunks); 1092} 1093 1094status_t MPEG4Writer::writeOneChunk() { 1095 LOGV("writeOneChunk"); 1096 1097 // Find the smallest timestamp, and write that chunk out 1098 // XXX: What if some track is just too slow? 1099 int64_t minTimestampUs = 0x7FFFFFFFFFFFFFFFLL; 1100 Track *track = NULL; 1101 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1102 it != mChunkInfos.end(); ++it) { 1103 if (!it->mChunks.empty()) { 1104 List<Chunk>::iterator chunkIt = it->mChunks.begin(); 1105 if (chunkIt->mTimeStampUs < minTimestampUs) { 1106 minTimestampUs = chunkIt->mTimeStampUs; 1107 track = it->mTrack; 1108 } 1109 } 1110 } 1111 1112 if (track == NULL) { 1113 LOGV("Nothing to be written after all"); 1114 return OK; 1115 } 1116 1117 if (mIsFirstChunk) { 1118 mIsFirstChunk = false; 1119 } 1120 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1121 it != mChunkInfos.end(); ++it) { 1122 if (it->mTrack == track) { 1123 writeFirstChunk(&(*it)); 1124 } 1125 } 1126 return OK; 1127} 1128 1129void MPEG4Writer::threadFunc() { 1130 LOGV("threadFunc"); 1131 1132 prctl(PR_SET_NAME, (unsigned long)"MPEG4Writer", 0, 0, 0); 1133 while (!mDone) { 1134 { 1135 Mutex::Autolock autolock(mLock); 1136 mChunkReadyCondition.wait(mLock); 1137 CHECK_EQ(writeOneChunk(), OK); 1138 } 1139 } 1140 1141 { 1142 // Write ALL samples 1143 Mutex::Autolock autolock(mLock); 1144 writeChunks(); 1145 } 1146} 1147 1148status_t MPEG4Writer::startWriterThread() { 1149 LOGV("startWriterThread"); 1150 1151 mDone = false; 1152 mIsFirstChunk = true; 1153 mDriftTimeUs = 0; 1154 for (List<Track *>::iterator it = mTracks.begin(); 1155 it != mTracks.end(); ++it) { 1156 ChunkInfo info; 1157 info.mTrack = *it; 1158 mChunkInfos.push_back(info); 1159 } 1160 1161 pthread_attr_t attr; 1162 pthread_attr_init(&attr); 1163 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1164 pthread_create(&mThread, &attr, ThreadWrapper, this); 1165 pthread_attr_destroy(&attr); 1166 return OK; 1167} 1168 1169 1170status_t MPEG4Writer::Track::start(MetaData *params) { 1171 if (!mDone && mPaused) { 1172 mPaused = false; 1173 mResumed = true; 1174 return OK; 1175 } 1176 1177 int64_t startTimeUs; 1178 if (params == NULL || !params->findInt64(kKeyTime, &startTimeUs)) { 1179 startTimeUs = 0; 1180 } 1181 1182 mIsRealTimeRecording = true; 1183 { 1184 int32_t isNotRealTime; 1185 if (params && params->findInt32(kKeyNotRealTime, &isNotRealTime)) { 1186 mIsRealTimeRecording = (isNotRealTime == 0); 1187 } 1188 } 1189 1190 initTrackingProgressStatus(params); 1191 1192 sp<MetaData> meta = new MetaData; 1193 meta->setInt64(kKeyTime, startTimeUs); 1194 status_t err = mSource->start(meta.get()); 1195 if (err != OK) { 1196 mDone = mReachedEOS = true; 1197 return err; 1198 } 1199 1200 pthread_attr_t attr; 1201 pthread_attr_init(&attr); 1202 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1203 1204 mDone = false; 1205 mTrackDurationUs = 0; 1206 mReachedEOS = false; 1207 mEstimatedTrackSizeBytes = 0; 1208 mNumStcoTableEntries = 0; 1209 mNumStssTableEntries = 0; 1210 mNumStscTableEntries = 0; 1211 mNumSttsTableEntries = 0; 1212 mMdatSizeBytes = 0; 1213 mIsMediaTimeAdjustmentOn = false; 1214 mPrevMediaTimeAdjustTimestampUs = 0; 1215 mMediaTimeAdjustNumFrames = 0; 1216 mPrevMediaTimeAdjustSample = 0; 1217 mTotalDriftTimeToAdjustUs = 0; 1218 mPrevTotalAccumDriftTimeUs = 0; 1219 1220 pthread_create(&mThread, &attr, ThreadWrapper, this); 1221 pthread_attr_destroy(&attr); 1222 1223 return OK; 1224} 1225 1226status_t MPEG4Writer::Track::pause() { 1227 mPaused = true; 1228 return OK; 1229} 1230 1231status_t MPEG4Writer::Track::stop() { 1232 LOGD("Stopping %s track", mIsAudio? "Audio": "Video"); 1233 if (mDone) { 1234 return OK; 1235 } 1236 1237 mDone = true; 1238 1239 void *dummy; 1240 pthread_join(mThread, &dummy); 1241 1242 status_t err = (status_t) dummy; 1243 1244 LOGD("Stopping %s track source", mIsAudio? "Audio": "Video"); 1245 { 1246 status_t status = mSource->stop(); 1247 if (err == OK && status != OK && status != ERROR_END_OF_STREAM) { 1248 err = status; 1249 } 1250 } 1251 1252 LOGD("%s track stopped", mIsAudio? "Audio": "Video"); 1253 return err; 1254} 1255 1256bool MPEG4Writer::Track::reachedEOS() { 1257 return mReachedEOS; 1258} 1259 1260// static 1261void *MPEG4Writer::Track::ThreadWrapper(void *me) { 1262 Track *track = static_cast<Track *>(me); 1263 1264 status_t err = track->threadEntry(); 1265 return (void *) err; 1266} 1267 1268static void getNalUnitType(uint8_t byte, uint8_t* type) { 1269 LOGV("getNalUnitType: %d", byte); 1270 1271 // nal_unit_type: 5-bit unsigned integer 1272 *type = (byte & 0x1F); 1273} 1274 1275static const uint8_t *findNextStartCode( 1276 const uint8_t *data, size_t length) { 1277 1278 LOGV("findNextStartCode: %p %d", data, length); 1279 1280 size_t bytesLeft = length; 1281 while (bytesLeft > 4 && 1282 memcmp("\x00\x00\x00\x01", &data[length - bytesLeft], 4)) { 1283 --bytesLeft; 1284 } 1285 if (bytesLeft <= 4) { 1286 bytesLeft = 0; // Last parameter set 1287 } 1288 return &data[length - bytesLeft]; 1289} 1290 1291const uint8_t *MPEG4Writer::Track::parseParamSet( 1292 const uint8_t *data, size_t length, int type, size_t *paramSetLen) { 1293 1294 LOGV("parseParamSet"); 1295 CHECK(type == kNalUnitTypeSeqParamSet || 1296 type == kNalUnitTypePicParamSet); 1297 1298 const uint8_t *nextStartCode = findNextStartCode(data, length); 1299 *paramSetLen = nextStartCode - data; 1300 if (*paramSetLen == 0) { 1301 LOGE("Param set is malformed, since its length is 0"); 1302 return NULL; 1303 } 1304 1305 AVCParamSet paramSet(*paramSetLen, data); 1306 if (type == kNalUnitTypeSeqParamSet) { 1307 if (*paramSetLen < 4) { 1308 LOGE("Seq parameter set malformed"); 1309 return NULL; 1310 } 1311 if (mSeqParamSets.empty()) { 1312 mProfileIdc = data[1]; 1313 mProfileCompatible = data[2]; 1314 mLevelIdc = data[3]; 1315 } else { 1316 if (mProfileIdc != data[1] || 1317 mProfileCompatible != data[2] || 1318 mLevelIdc != data[3]) { 1319 LOGE("Inconsistent profile/level found in seq parameter sets"); 1320 return NULL; 1321 } 1322 } 1323 mSeqParamSets.push_back(paramSet); 1324 } else { 1325 mPicParamSets.push_back(paramSet); 1326 } 1327 return nextStartCode; 1328} 1329 1330status_t MPEG4Writer::Track::copyAVCCodecSpecificData( 1331 const uint8_t *data, size_t size) { 1332 LOGV("copyAVCCodecSpecificData"); 1333 1334 // 2 bytes for each of the parameter set length field 1335 // plus the 7 bytes for the header 1336 if (size < 4 + 7) { 1337 LOGE("Codec specific data length too short: %d", size); 1338 return ERROR_MALFORMED; 1339 } 1340 1341 mCodecSpecificDataSize = size; 1342 mCodecSpecificData = malloc(size); 1343 memcpy(mCodecSpecificData, data, size); 1344 return OK; 1345} 1346 1347status_t MPEG4Writer::Track::parseAVCCodecSpecificData( 1348 const uint8_t *data, size_t size) { 1349 1350 LOGV("parseAVCCodecSpecificData"); 1351 // Data starts with a start code. 1352 // SPS and PPS are separated with start codes. 1353 // Also, SPS must come before PPS 1354 uint8_t type = kNalUnitTypeSeqParamSet; 1355 bool gotSps = false; 1356 bool gotPps = false; 1357 const uint8_t *tmp = data; 1358 const uint8_t *nextStartCode = data; 1359 size_t bytesLeft = size; 1360 size_t paramSetLen = 0; 1361 mCodecSpecificDataSize = 0; 1362 while (bytesLeft > 4 && !memcmp("\x00\x00\x00\x01", tmp, 4)) { 1363 getNalUnitType(*(tmp + 4), &type); 1364 if (type == kNalUnitTypeSeqParamSet) { 1365 if (gotPps) { 1366 LOGE("SPS must come before PPS"); 1367 return ERROR_MALFORMED; 1368 } 1369 if (!gotSps) { 1370 gotSps = true; 1371 } 1372 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1373 } else if (type == kNalUnitTypePicParamSet) { 1374 if (!gotSps) { 1375 LOGE("SPS must come before PPS"); 1376 return ERROR_MALFORMED; 1377 } 1378 if (!gotPps) { 1379 gotPps = true; 1380 } 1381 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1382 } else { 1383 LOGE("Only SPS and PPS Nal units are expected"); 1384 return ERROR_MALFORMED; 1385 } 1386 1387 if (nextStartCode == NULL) { 1388 return ERROR_MALFORMED; 1389 } 1390 1391 // Move on to find the next parameter set 1392 bytesLeft -= nextStartCode - tmp; 1393 tmp = nextStartCode; 1394 mCodecSpecificDataSize += (2 + paramSetLen); 1395 } 1396 1397 { 1398 // Check on the number of seq parameter sets 1399 size_t nSeqParamSets = mSeqParamSets.size(); 1400 if (nSeqParamSets == 0) { 1401 LOGE("Cound not find sequence parameter set"); 1402 return ERROR_MALFORMED; 1403 } 1404 1405 if (nSeqParamSets > 0x1F) { 1406 LOGE("Too many seq parameter sets (%d) found", nSeqParamSets); 1407 return ERROR_MALFORMED; 1408 } 1409 } 1410 1411 { 1412 // Check on the number of pic parameter sets 1413 size_t nPicParamSets = mPicParamSets.size(); 1414 if (nPicParamSets == 0) { 1415 LOGE("Cound not find picture parameter set"); 1416 return ERROR_MALFORMED; 1417 } 1418 if (nPicParamSets > 0xFF) { 1419 LOGE("Too many pic parameter sets (%d) found", nPicParamSets); 1420 return ERROR_MALFORMED; 1421 } 1422 } 1423 1424 { 1425 // Check on the profiles 1426 // These profiles requires additional parameter set extensions 1427 if (mProfileIdc == 100 || mProfileIdc == 110 || 1428 mProfileIdc == 122 || mProfileIdc == 144) { 1429 LOGE("Sorry, no support for profile_idc: %d!", mProfileIdc); 1430 return BAD_VALUE; 1431 } 1432 } 1433 1434 return OK; 1435} 1436 1437status_t MPEG4Writer::Track::makeAVCCodecSpecificData( 1438 const uint8_t *data, size_t size) { 1439 1440 if (mCodecSpecificData != NULL) { 1441 LOGE("Already have codec specific data"); 1442 return ERROR_MALFORMED; 1443 } 1444 1445 if (size < 4) { 1446 LOGE("Codec specific data length too short: %d", size); 1447 return ERROR_MALFORMED; 1448 } 1449 1450 // Data is in the form of AVCCodecSpecificData 1451 if (memcmp("\x00\x00\x00\x01", data, 4)) { 1452 return copyAVCCodecSpecificData(data, size); 1453 } 1454 1455 if (parseAVCCodecSpecificData(data, size) != OK) { 1456 return ERROR_MALFORMED; 1457 } 1458 1459 // ISO 14496-15: AVC file format 1460 mCodecSpecificDataSize += 7; // 7 more bytes in the header 1461 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1462 uint8_t *header = (uint8_t *)mCodecSpecificData; 1463 header[0] = 1; // version 1464 header[1] = mProfileIdc; // profile indication 1465 header[2] = mProfileCompatible; // profile compatibility 1466 header[3] = mLevelIdc; 1467 1468 // 6-bit '111111' followed by 2-bit to lengthSizeMinuusOne 1469 if (mOwner->useNalLengthFour()) { 1470 header[4] = 0xfc | 3; // length size == 4 bytes 1471 } else { 1472 header[4] = 0xfc | 1; // length size == 2 bytes 1473 } 1474 1475 // 3-bit '111' followed by 5-bit numSequenceParameterSets 1476 int nSequenceParamSets = mSeqParamSets.size(); 1477 header[5] = 0xe0 | nSequenceParamSets; 1478 header += 6; 1479 for (List<AVCParamSet>::iterator it = mSeqParamSets.begin(); 1480 it != mSeqParamSets.end(); ++it) { 1481 // 16-bit sequence parameter set length 1482 uint16_t seqParamSetLength = it->mLength; 1483 header[0] = seqParamSetLength >> 8; 1484 header[1] = seqParamSetLength & 0xff; 1485 1486 // SPS NAL unit (sequence parameter length bytes) 1487 memcpy(&header[2], it->mData, seqParamSetLength); 1488 header += (2 + seqParamSetLength); 1489 } 1490 1491 // 8-bit nPictureParameterSets 1492 int nPictureParamSets = mPicParamSets.size(); 1493 header[0] = nPictureParamSets; 1494 header += 1; 1495 for (List<AVCParamSet>::iterator it = mPicParamSets.begin(); 1496 it != mPicParamSets.end(); ++it) { 1497 // 16-bit picture parameter set length 1498 uint16_t picParamSetLength = it->mLength; 1499 header[0] = picParamSetLength >> 8; 1500 header[1] = picParamSetLength & 0xff; 1501 1502 // PPS Nal unit (picture parameter set length bytes) 1503 memcpy(&header[2], it->mData, picParamSetLength); 1504 header += (2 + picParamSetLength); 1505 } 1506 1507 return OK; 1508} 1509 1510/* 1511* The video track's media time adjustment for real-time applications 1512* is described as follows: 1513* 1514* First, the media time adjustment is done for every period of 1515* kVideoMediaTimeAdjustPeriodTimeUs. kVideoMediaTimeAdjustPeriodTimeUs 1516* is currently a fixed value chosen heuristically. The value of 1517* kVideoMediaTimeAdjustPeriodTimeUs should not be very large or very small 1518* for two considerations: on one hand, a relatively large value 1519* helps reduce large fluctuation of drift time in the audio encoding 1520* path; while on the other hand, a relatively small value helps keep 1521* restoring synchronization in audio/video more frequently. Note for the 1522* very first period of kVideoMediaTimeAdjustPeriodTimeUs, there is 1523* no media time adjustment for the video track. 1524* 1525* Second, the total accumulated audio track time drift found 1526* in a period of kVideoMediaTimeAdjustPeriodTimeUs is distributed 1527* over a stream of incoming video frames. The number of video frames 1528* affected is determined based on the number of recorded video frames 1529* within the past kVideoMediaTimeAdjustPeriodTimeUs period. 1530* We choose to distribute the drift time over only a portion 1531* (rather than all) of the total number of recorded video frames 1532* in order to make sure that the video track media time adjustment is 1533* completed for the current period before the next video track media 1534* time adjustment period starts. Currently, the portion chosen is a 1535* half (0.5). 1536* 1537* Last, various additional checks are performed to ensure that 1538* the actual audio encoding path does not have too much drift. 1539* In particular, 1) we want to limit the average incremental time 1540* adjustment for each video frame to be less than a threshold 1541* for a single period of kVideoMediaTimeAdjustPeriodTimeUs. 1542* Currently, the threshold is set to 5 ms. If the average incremental 1543* media time adjustment for a video frame is larger than the 1544* threshold, the audio encoding path has too much time drift. 1545* 2) We also want to limit the total time drift in the audio 1546* encoding path to be less than a threshold for a period of 1547* kVideoMediaTimeAdjustPeriodTimeUs. Currently, the threshold 1548* is 0.5% of kVideoMediaTimeAdjustPeriodTimeUs. If the time drift of 1549* the audio encoding path is larger than the threshold, the audio 1550* encoding path has too much time drift. We treat the large time 1551* drift of the audio encoding path as errors, since there is no 1552* way to keep audio/video in synchronization for real-time 1553* applications if the time drift is too large unless we drop some 1554* video frames, which has its own problems that we don't want 1555* to get into for the time being. 1556*/ 1557void MPEG4Writer::Track::adjustMediaTime(int64_t *timestampUs) { 1558 if (*timestampUs - mPrevMediaTimeAdjustTimestampUs >= 1559 kVideoMediaTimeAdjustPeriodTimeUs) { 1560 1561 LOGV("New media time adjustment period at %lld us", *timestampUs); 1562 mIsMediaTimeAdjustmentOn = true; 1563 mMediaTimeAdjustNumFrames = 1564 (mNumSamples - mPrevMediaTimeAdjustSample) >> 1; 1565 1566 mPrevMediaTimeAdjustTimestampUs = *timestampUs; 1567 mPrevMediaTimeAdjustSample = mNumSamples; 1568 int64_t totalAccumDriftTimeUs = mOwner->getDriftTimeUs(); 1569 mTotalDriftTimeToAdjustUs = 1570 totalAccumDriftTimeUs - mPrevTotalAccumDriftTimeUs; 1571 1572 mPrevTotalAccumDriftTimeUs = totalAccumDriftTimeUs; 1573 1574 // Check on incremental adjusted time per frame 1575 int64_t adjustTimePerFrameUs = 1576 mTotalDriftTimeToAdjustUs / mMediaTimeAdjustNumFrames; 1577 1578 if (adjustTimePerFrameUs < 0) { 1579 adjustTimePerFrameUs = -adjustTimePerFrameUs; 1580 } 1581 if (adjustTimePerFrameUs >= 5000) { 1582 LOGE("Adjusted time per video frame is %lld us", 1583 adjustTimePerFrameUs); 1584 CHECK(!"Video frame time adjustment is too large!"); 1585 } 1586 1587 // Check on total accumulated time drift within a period of 1588 // kVideoMediaTimeAdjustPeriodTimeUs. 1589 int64_t driftPercentage = (mTotalDriftTimeToAdjustUs * 1000) 1590 / kVideoMediaTimeAdjustPeriodTimeUs; 1591 1592 if (driftPercentage < 0) { 1593 driftPercentage = -driftPercentage; 1594 } 1595 if (driftPercentage > 5) { 1596 LOGE("Audio track has time drift %lld us over %lld us", 1597 mTotalDriftTimeToAdjustUs, 1598 kVideoMediaTimeAdjustPeriodTimeUs); 1599 1600 CHECK(!"The audio track media time drifts too much!"); 1601 } 1602 1603 } 1604 1605 if (mIsMediaTimeAdjustmentOn) { 1606 if (mNumSamples - mPrevMediaTimeAdjustSample <= 1607 mMediaTimeAdjustNumFrames) { 1608 1609 // Do media time incremental adjustment 1610 int64_t incrementalAdjustTimeUs = 1611 (mTotalDriftTimeToAdjustUs * 1612 (mNumSamples - mPrevMediaTimeAdjustSample)) 1613 / mMediaTimeAdjustNumFrames; 1614 1615 *timestampUs += 1616 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs); 1617 1618 LOGV("Incremental video frame media time adjustment: %lld us", 1619 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs)); 1620 } else { 1621 // Within the remaining adjustment period, 1622 // no incremental adjustment is needed. 1623 *timestampUs += 1624 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs); 1625 1626 LOGV("Fixed video frame media time adjustment: %lld us", 1627 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs)); 1628 } 1629 } 1630} 1631 1632/* 1633 * Updates the drift time from the audio track so that 1634 * the video track can get the updated drift time information 1635 * from the file writer. The fluctuation of the drift time of the audio 1636 * encoding path is smoothed out with a simple filter by giving a larger 1637 * weight to more recently drift time. The filter coefficients, 0.5 and 0.5, 1638 * are heuristically determined. 1639 */ 1640void MPEG4Writer::Track::updateDriftTime(const sp<MetaData>& meta) { 1641 int64_t driftTimeUs = 0; 1642 if (meta->findInt64(kKeyDriftTime, &driftTimeUs)) { 1643 int64_t prevDriftTimeUs = mOwner->getDriftTimeUs(); 1644 int64_t timeUs = (driftTimeUs + prevDriftTimeUs) >> 1; 1645 mOwner->setDriftTimeUs(timeUs); 1646 } 1647} 1648 1649status_t MPEG4Writer::Track::threadEntry() { 1650 int32_t count = 0; 1651 const int64_t interleaveDurationUs = mOwner->interleaveDuration(); 1652 int64_t chunkTimestampUs = 0; 1653 int32_t nChunks = 0; 1654 int32_t nZeroLengthFrames = 0; 1655 int64_t lastTimestampUs = 0; // Previous sample time stamp in ms 1656 int64_t lastDurationUs = 0; // Between the previous two samples in ms 1657 int64_t currDurationTicks = 0; // Timescale based ticks 1658 int64_t lastDurationTicks = 0; // Timescale based ticks 1659 int32_t sampleCount = 1; // Sample count in the current stts table entry 1660 uint32_t previousSampleSize = 0; // Size of the previous sample 1661 int64_t previousPausedDurationUs = 0; 1662 int64_t timestampUs; 1663 1664 if (mIsAudio) { 1665 prctl(PR_SET_NAME, (unsigned long)"AudioTrackEncoding", 0, 0, 0); 1666 } else { 1667 prctl(PR_SET_NAME, (unsigned long)"VideoTrackEncoding", 0, 0, 0); 1668 } 1669 sp<MetaData> meta_data; 1670 1671 mNumSamples = 0; 1672 status_t err = OK; 1673 MediaBuffer *buffer; 1674 while (!mDone && (err = mSource->read(&buffer)) == OK) { 1675 if (buffer->range_length() == 0) { 1676 buffer->release(); 1677 buffer = NULL; 1678 ++nZeroLengthFrames; 1679 continue; 1680 } 1681 1682 // If the codec specific data has not been received yet, delay pause. 1683 // After the codec specific data is received, discard what we received 1684 // when the track is to be paused. 1685 if (mPaused && !mResumed) { 1686 buffer->release(); 1687 buffer = NULL; 1688 continue; 1689 } 1690 1691 ++count; 1692 1693 int32_t isCodecConfig; 1694 if (buffer->meta_data()->findInt32(kKeyIsCodecConfig, &isCodecConfig) 1695 && isCodecConfig) { 1696 CHECK(!mGotAllCodecSpecificData); 1697 1698 if (mIsAvc) { 1699 status_t err = makeAVCCodecSpecificData( 1700 (const uint8_t *)buffer->data() 1701 + buffer->range_offset(), 1702 buffer->range_length()); 1703 CHECK_EQ(OK, err); 1704 } else if (mIsMPEG4) { 1705 mCodecSpecificDataSize = buffer->range_length(); 1706 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1707 memcpy(mCodecSpecificData, 1708 (const uint8_t *)buffer->data() 1709 + buffer->range_offset(), 1710 buffer->range_length()); 1711 } 1712 1713 buffer->release(); 1714 buffer = NULL; 1715 1716 mGotAllCodecSpecificData = true; 1717 continue; 1718 } 1719 1720 // Make a deep copy of the MediaBuffer and Metadata and release 1721 // the original as soon as we can 1722 MediaBuffer *copy = new MediaBuffer(buffer->range_length()); 1723 memcpy(copy->data(), (uint8_t *)buffer->data() + buffer->range_offset(), 1724 buffer->range_length()); 1725 copy->set_range(0, buffer->range_length()); 1726 meta_data = new MetaData(*buffer->meta_data().get()); 1727 buffer->release(); 1728 buffer = NULL; 1729 1730 if (mIsAvc) StripStartcode(copy); 1731 1732 size_t sampleSize = copy->range_length(); 1733 if (mIsAvc) { 1734 if (mOwner->useNalLengthFour()) { 1735 sampleSize += 4; 1736 } else { 1737 sampleSize += 2; 1738 } 1739 } 1740 1741 // Max file size or duration handling 1742 mMdatSizeBytes += sampleSize; 1743 updateTrackSizeEstimate(); 1744 1745 if (mOwner->exceedsFileSizeLimit()) { 1746 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_FILESIZE_REACHED, 0); 1747 break; 1748 } 1749 if (mOwner->exceedsFileDurationLimit()) { 1750 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_DURATION_REACHED, 0); 1751 break; 1752 } 1753 1754 1755 int32_t isSync = false; 1756 meta_data->findInt32(kKeyIsSyncFrame, &isSync); 1757 1758 /* 1759 * The original timestamp found in the data buffer will be modified as below: 1760 * 1761 * There is a playback offset into this track if the track's start time 1762 * is not the same as the movie start time, which will be recorded in edst 1763 * box of the output file. The playback offset is to make sure that the 1764 * starting time of the audio/video tracks are synchronized. Although the 1765 * track's media timestamp may be subject to various modifications 1766 * as outlined below, the track's playback offset time remains unchanged 1767 * once the first data buffer of the track is received. 1768 * 1769 * The media time stamp will be calculated by subtracting the playback offset 1770 * (and potential pause durations) from the original timestamp in the buffer. 1771 * 1772 * If this track is a video track for a real-time recording application with 1773 * both audio and video tracks, its media timestamp will subject to further 1774 * modification based on the media clock of the audio track. This modification 1775 * is needed for the purpose of maintaining good audio/video synchronization. 1776 * 1777 * If the recording session is paused and resumed multiple times, the track 1778 * media timestamp will be modified as if the recording session had never been 1779 * paused at all during playback of the recorded output file. In other words, 1780 * the output file will have no memory of pause/resume durations. 1781 * 1782 */ 1783 CHECK(meta_data->findInt64(kKeyTime, ×tampUs)); 1784 LOGV("%s timestampUs: %lld", mIsAudio? "Audio": "Video", timestampUs); 1785 1786//////////////////////////////////////////////////////////////////////////////// 1787 if (mSampleSizes.empty()) { 1788 mStartTimestampUs = timestampUs; 1789 mOwner->setStartTimestampUs(mStartTimestampUs); 1790 previousPausedDurationUs = mStartTimestampUs; 1791 } 1792 1793 if (mResumed) { 1794 int64_t durExcludingEarlierPausesUs = timestampUs - previousPausedDurationUs; 1795 CHECK(durExcludingEarlierPausesUs >= 0); 1796 int64_t pausedDurationUs = durExcludingEarlierPausesUs - mTrackDurationUs; 1797 CHECK(pausedDurationUs >= lastDurationUs); 1798 previousPausedDurationUs += pausedDurationUs - lastDurationUs; 1799 mResumed = false; 1800 } 1801 1802 timestampUs -= previousPausedDurationUs; 1803 CHECK(timestampUs >= 0); 1804 1805 // Media time adjustment for real-time applications 1806 if (mIsRealTimeRecording) { 1807 if (mIsAudio) { 1808 updateDriftTime(meta_data); 1809 } else { 1810 adjustMediaTime(×tampUs); 1811 } 1812 } 1813 1814 CHECK(timestampUs >= 0); 1815 if (mNumSamples > 1) { 1816 if (timestampUs <= lastTimestampUs) { 1817 LOGW("Frame arrives too late!"); 1818 // Don't drop the late frame, since dropping a frame may cause 1819 // problems later during playback 1820 1821 // The idea here is to avoid having two or more samples with the 1822 // same timestamp in the output file. 1823 if (mTimeScale >= 1000000LL) { 1824 timestampUs = lastTimestampUs + 1; 1825 } else { 1826 timestampUs = lastTimestampUs + (1000000LL + (mTimeScale >> 1)) / mTimeScale; 1827 } 1828 } 1829 } 1830 1831 LOGV("%s media time stamp: %lld and previous paused duration %lld", 1832 mIsAudio? "Audio": "Video", timestampUs, previousPausedDurationUs); 1833 if (timestampUs > mTrackDurationUs) { 1834 mTrackDurationUs = timestampUs; 1835 } 1836 1837 mSampleSizes.push_back(sampleSize); 1838 ++mNumSamples; 1839 if (mNumSamples > 2) { 1840 // We need to use the time scale based ticks, rather than the 1841 // timestamp itself to determine whether we have to use a new 1842 // stts entry, since we may have rounding errors. 1843 // The calculation is intended to reduce the accumulated 1844 // rounding errors. 1845 currDurationTicks = 1846 ((timestampUs * mTimeScale + 500000LL) / 1000000LL - 1847 (lastTimestampUs * mTimeScale + 500000LL) / 1000000LL); 1848 1849 if (currDurationTicks != lastDurationTicks) { 1850 addOneSttsTableEntry(sampleCount, lastDurationUs); 1851 sampleCount = 1; 1852 } else { 1853 ++sampleCount; 1854 } 1855 } 1856 if (mSamplesHaveSameSize) { 1857 if (mNumSamples >= 2 && previousSampleSize != sampleSize) { 1858 mSamplesHaveSameSize = false; 1859 } 1860 previousSampleSize = sampleSize; 1861 } 1862 lastDurationUs = timestampUs - lastTimestampUs; 1863 lastDurationTicks = currDurationTicks; 1864 lastTimestampUs = timestampUs; 1865 1866 if (isSync != 0) { 1867 addOneStssTableEntry(mNumSamples); 1868 } 1869 1870 if (mTrackingProgressStatus) { 1871 if (mPreviousTrackTimeUs <= 0) { 1872 mPreviousTrackTimeUs = mStartTimestampUs; 1873 } 1874 trackProgressStatus(timestampUs); 1875 } 1876 if (mOwner->numTracks() == 1) { 1877 off_t offset = mIsAvc? mOwner->addLengthPrefixedSample_l(copy) 1878 : mOwner->addSample_l(copy); 1879 if (mChunkOffsets.empty()) { 1880 addChunkOffset(offset); 1881 } 1882 copy->release(); 1883 copy = NULL; 1884 continue; 1885 } 1886 1887 mChunkSamples.push_back(copy); 1888 if (interleaveDurationUs == 0) { 1889 addOneStscTableEntry(++nChunks, 1); 1890 bufferChunk(timestampUs); 1891 } else { 1892 if (chunkTimestampUs == 0) { 1893 chunkTimestampUs = timestampUs; 1894 } else { 1895 if (timestampUs - chunkTimestampUs > interleaveDurationUs) { 1896 ++nChunks; 1897 if (nChunks == 1 || // First chunk 1898 (--(mStscTableEntries.end()))->samplesPerChunk != 1899 mChunkSamples.size()) { 1900 addOneStscTableEntry(nChunks, mChunkSamples.size()); 1901 } 1902 bufferChunk(timestampUs); 1903 chunkTimestampUs = timestampUs; 1904 } 1905 } 1906 } 1907 1908 } 1909 1910 if (mSampleSizes.empty() || // no samples written 1911 (!mIsAudio && mNumStssTableEntries == 0) || // no sync frames for video 1912 (OK != checkCodecSpecificData())) { // no codec specific data 1913 err = ERROR_MALFORMED; 1914 } 1915 mOwner->trackProgressStatus(this, -1, err); 1916 1917 // Last chunk 1918 if (mOwner->numTracks() == 1) { 1919 addOneStscTableEntry(1, mNumSamples); 1920 } else if (!mChunkSamples.empty()) { 1921 addOneStscTableEntry(++nChunks, mChunkSamples.size()); 1922 bufferChunk(timestampUs); 1923 } 1924 1925 // We don't really know how long the last frame lasts, since 1926 // there is no frame time after it, just repeat the previous 1927 // frame's duration. 1928 if (mNumSamples == 1) { 1929 lastDurationUs = 0; // A single sample's duration 1930 } else { 1931 ++sampleCount; // Count for the last sample 1932 } 1933 addOneSttsTableEntry(sampleCount, lastDurationUs); 1934 mTrackDurationUs += lastDurationUs; 1935 mReachedEOS = true; 1936 LOGI("Received total/0-length (%d/%d) buffers and encoded %d frames. - %s", 1937 count, nZeroLengthFrames, mNumSamples, mIsAudio? "audio": "video"); 1938 if (mIsAudio) { 1939 LOGI("Audio track drift time: %lld us", mOwner->getDriftTimeUs()); 1940 } 1941 1942 if (err == ERROR_END_OF_STREAM) { 1943 return OK; 1944 } 1945 return err; 1946} 1947 1948void MPEG4Writer::Track::trackProgressStatus(int64_t timeUs, status_t err) { 1949 LOGV("trackProgressStatus: %lld us", timeUs); 1950 if (mTrackEveryTimeDurationUs > 0 && 1951 timeUs - mPreviousTrackTimeUs >= mTrackEveryTimeDurationUs) { 1952 LOGV("Fire time tracking progress status at %lld us", timeUs); 1953 mOwner->trackProgressStatus(this, timeUs - mPreviousTrackTimeUs, err); 1954 mPreviousTrackTimeUs = timeUs; 1955 } 1956} 1957 1958void MPEG4Writer::trackProgressStatus( 1959 const MPEG4Writer::Track* track, int64_t timeUs, status_t err) { 1960 Mutex::Autolock lock(mLock); 1961 int32_t nTracks = mTracks.size(); 1962 CHECK(nTracks >= 1); 1963 CHECK(nTracks < 64); // Arbitrary number 1964 1965 int32_t trackNum = 0; 1966 CHECK(trackNum < nTracks); 1967 trackNum <<= 16; 1968 1969 // Error notification 1970 // Do not consider ERROR_END_OF_STREAM an error 1971 if (err != OK && err != ERROR_END_OF_STREAM) { 1972 notify(MEDIA_RECORDER_EVENT_ERROR, 1973 trackNum | MEDIA_RECORDER_ERROR_UNKNOWN, 1974 err); 1975 return; 1976 } 1977 1978 if (timeUs == -1) { 1979 // Send completion notification 1980 notify(MEDIA_RECORDER_EVENT_INFO, 1981 trackNum | MEDIA_RECORDER_INFO_COMPLETION_STATUS, 1982 err); 1983 return; 1984 } else { 1985 // Send progress status 1986 notify(MEDIA_RECORDER_EVENT_INFO, 1987 trackNum | MEDIA_RECORDER_INFO_PROGRESS_TIME_STATUS, 1988 timeUs / 1000); 1989 } 1990} 1991 1992void MPEG4Writer::setDriftTimeUs(int64_t driftTimeUs) { 1993 LOGV("setDriftTimeUs: %lld us", driftTimeUs); 1994 Mutex::Autolock autolock(mLock); 1995 mDriftTimeUs = driftTimeUs; 1996} 1997 1998int64_t MPEG4Writer::getDriftTimeUs() { 1999 LOGV("getDriftTimeUs: %lld us", mDriftTimeUs); 2000 Mutex::Autolock autolock(mLock); 2001 return mDriftTimeUs; 2002} 2003 2004bool MPEG4Writer::useNalLengthFour() { 2005 return mUse4ByteNalLength; 2006} 2007 2008void MPEG4Writer::Track::bufferChunk(int64_t timestampUs) { 2009 LOGV("bufferChunk"); 2010 2011 Chunk chunk(this, timestampUs, mChunkSamples); 2012 mOwner->bufferChunk(chunk); 2013 mChunkSamples.clear(); 2014} 2015 2016int64_t MPEG4Writer::Track::getDurationUs() const { 2017 return mTrackDurationUs; 2018} 2019 2020int64_t MPEG4Writer::Track::getEstimatedTrackSizeBytes() const { 2021 return mEstimatedTrackSizeBytes; 2022} 2023 2024status_t MPEG4Writer::Track::checkCodecSpecificData() const { 2025 const char *mime; 2026 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 2027 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime) || 2028 !strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime) || 2029 !strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2030 if (!mCodecSpecificData || 2031 mCodecSpecificDataSize <= 0) { 2032 LOGE("Missing codec specific data"); 2033 return ERROR_MALFORMED; 2034 } 2035 } else { 2036 if (mCodecSpecificData || 2037 mCodecSpecificDataSize > 0) { 2038 LOGE("Unexepected codec specific data found"); 2039 return ERROR_MALFORMED; 2040 } 2041 } 2042 return OK; 2043} 2044 2045void MPEG4Writer::Track::writeTrackHeader( 2046 int32_t trackID, bool use32BitOffset) { 2047 const char *mime; 2048 bool success = mMeta->findCString(kKeyMIMEType, &mime); 2049 CHECK(success); 2050 2051 LOGV("%s track time scale: %d", 2052 mIsAudio? "Audio": "Video", mTimeScale); 2053 2054 time_t now = time(NULL); 2055 int32_t mvhdTimeScale = mOwner->getTimeScale(); 2056 int64_t trakDurationUs = getDurationUs(); 2057 2058 mOwner->beginBox("trak"); 2059 2060 mOwner->beginBox("tkhd"); 2061 // Flags = 7 to indicate that the track is enabled, and 2062 // part of the presentation 2063 mOwner->writeInt32(0x07); // version=0, flags=7 2064 mOwner->writeInt32(now); // creation time 2065 mOwner->writeInt32(now); // modification time 2066 mOwner->writeInt32(trackID); 2067 mOwner->writeInt32(0); // reserved 2068 int32_t tkhdDuration = 2069 (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2070 mOwner->writeInt32(tkhdDuration); // in mvhd timescale 2071 mOwner->writeInt32(0); // reserved 2072 mOwner->writeInt32(0); // reserved 2073 mOwner->writeInt16(0); // layer 2074 mOwner->writeInt16(0); // alternate group 2075 mOwner->writeInt16(mIsAudio ? 0x100 : 0); // volume 2076 mOwner->writeInt16(0); // reserved 2077 2078 mOwner->writeInt32(0x10000); // matrix 2079 mOwner->writeInt32(0); 2080 mOwner->writeInt32(0); 2081 mOwner->writeInt32(0); 2082 mOwner->writeInt32(0x10000); 2083 mOwner->writeInt32(0); 2084 mOwner->writeInt32(0); 2085 mOwner->writeInt32(0); 2086 mOwner->writeInt32(0x40000000); 2087 2088 if (mIsAudio) { 2089 mOwner->writeInt32(0); 2090 mOwner->writeInt32(0); 2091 } else { 2092 int32_t width, height; 2093 bool success = mMeta->findInt32(kKeyWidth, &width); 2094 success = success && mMeta->findInt32(kKeyHeight, &height); 2095 CHECK(success); 2096 2097 mOwner->writeInt32(width << 16); // 32-bit fixed-point value 2098 mOwner->writeInt32(height << 16); // 32-bit fixed-point value 2099 } 2100 mOwner->endBox(); // tkhd 2101 2102 int64_t moovStartTimeUs = mOwner->getStartTimestampUs(); 2103 if (mStartTimestampUs != moovStartTimeUs) { 2104 mOwner->beginBox("edts"); 2105 mOwner->beginBox("elst"); 2106 mOwner->writeInt32(0); // version=0, flags=0: 32-bit time 2107 mOwner->writeInt32(2); // never ends with an empty list 2108 2109 // First elst entry: specify the starting time offset 2110 int64_t offsetUs = mStartTimestampUs - moovStartTimeUs; 2111 LOGV("OffsetUs: %lld", offsetUs); 2112 int32_t seg = (offsetUs * mvhdTimeScale + 5E5) / 1E6; 2113 mOwner->writeInt32(seg); // in mvhd timecale 2114 mOwner->writeInt32(-1); // starting time offset 2115 mOwner->writeInt32(1 << 16); // rate = 1.0 2116 2117 // Second elst entry: specify the track duration 2118 seg = (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2119 mOwner->writeInt32(seg); // in mvhd timescale 2120 mOwner->writeInt32(0); 2121 mOwner->writeInt32(1 << 16); 2122 mOwner->endBox(); 2123 mOwner->endBox(); 2124 } 2125 2126 mOwner->beginBox("mdia"); 2127 2128 mOwner->beginBox("mdhd"); 2129 mOwner->writeInt32(0); // version=0, flags=0 2130 mOwner->writeInt32(now); // creation time 2131 mOwner->writeInt32(now); // modification time 2132 mOwner->writeInt32(mTimeScale); // media timescale 2133 int32_t mdhdDuration = (trakDurationUs * mTimeScale + 5E5) / 1E6; 2134 mOwner->writeInt32(mdhdDuration); // use media timescale 2135 // Language follows the three letter standard ISO-639-2/T 2136 // 'e', 'n', 'g' for "English", for instance. 2137 // Each character is packed as the difference between its ASCII value and 0x60. 2138 // For "English", these are 00101, 01110, 00111. 2139 // XXX: Where is the padding bit located: 0x15C7? 2140 mOwner->writeInt16(0); // language code 2141 mOwner->writeInt16(0); // predefined 2142 mOwner->endBox(); 2143 2144 mOwner->beginBox("hdlr"); 2145 mOwner->writeInt32(0); // version=0, flags=0 2146 mOwner->writeInt32(0); // component type: should be mhlr 2147 mOwner->writeFourcc(mIsAudio ? "soun" : "vide"); // component subtype 2148 mOwner->writeInt32(0); // reserved 2149 mOwner->writeInt32(0); // reserved 2150 mOwner->writeInt32(0); // reserved 2151 // Removing "r" for the name string just makes the string 4 byte aligned 2152 mOwner->writeCString(mIsAudio ? "SoundHandle": "VideoHandle"); // name 2153 mOwner->endBox(); 2154 2155 mOwner->beginBox("minf"); 2156 if (mIsAudio) { 2157 mOwner->beginBox("smhd"); 2158 mOwner->writeInt32(0); // version=0, flags=0 2159 mOwner->writeInt16(0); // balance 2160 mOwner->writeInt16(0); // reserved 2161 mOwner->endBox(); 2162 } else { 2163 mOwner->beginBox("vmhd"); 2164 mOwner->writeInt32(0x01); // version=0, flags=1 2165 mOwner->writeInt16(0); // graphics mode 2166 mOwner->writeInt16(0); // opcolor 2167 mOwner->writeInt16(0); 2168 mOwner->writeInt16(0); 2169 mOwner->endBox(); 2170 } 2171 2172 mOwner->beginBox("dinf"); 2173 mOwner->beginBox("dref"); 2174 mOwner->writeInt32(0); // version=0, flags=0 2175 mOwner->writeInt32(1); // entry count (either url or urn) 2176 // The table index here refers to the sample description index 2177 // in the sample table entries. 2178 mOwner->beginBox("url "); 2179 mOwner->writeInt32(1); // version=0, flags=1 (self-contained) 2180 mOwner->endBox(); // url 2181 mOwner->endBox(); // dref 2182 mOwner->endBox(); // dinf 2183 2184 mOwner->beginBox("stbl"); 2185 2186 mOwner->beginBox("stsd"); 2187 mOwner->writeInt32(0); // version=0, flags=0 2188 mOwner->writeInt32(1); // entry count 2189 if (mIsAudio) { 2190 const char *fourcc = NULL; 2191 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime)) { 2192 fourcc = "samr"; 2193 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2194 fourcc = "sawb"; 2195 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2196 fourcc = "mp4a"; 2197 } else { 2198 LOGE("Unknown mime type '%s'.", mime); 2199 CHECK(!"should not be here, unknown mime type."); 2200 } 2201 2202 mOwner->beginBox(fourcc); // audio format 2203 mOwner->writeInt32(0); // reserved 2204 mOwner->writeInt16(0); // reserved 2205 mOwner->writeInt16(0x1); // data ref index 2206 mOwner->writeInt32(0); // reserved 2207 mOwner->writeInt32(0); // reserved 2208 int32_t nChannels; 2209 CHECK_EQ(true, mMeta->findInt32(kKeyChannelCount, &nChannels)); 2210 mOwner->writeInt16(nChannels); // channel count 2211 mOwner->writeInt16(16); // sample size 2212 mOwner->writeInt16(0); // predefined 2213 mOwner->writeInt16(0); // reserved 2214 2215 int32_t samplerate; 2216 bool success = mMeta->findInt32(kKeySampleRate, &samplerate); 2217 CHECK(success); 2218 mOwner->writeInt32(samplerate << 16); 2219 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2220 mOwner->beginBox("esds"); 2221 CHECK(mCodecSpecificData); 2222 CHECK(mCodecSpecificDataSize > 0); 2223 2224 // Make sure all sizes encode to a single byte. 2225 CHECK(mCodecSpecificDataSize + 23 < 128); 2226 2227 mOwner->writeInt32(0); // version=0, flags=0 2228 mOwner->writeInt8(0x03); // ES_DescrTag 2229 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2230 mOwner->writeInt16(0x0000);// ES_ID 2231 mOwner->writeInt8(0x00); 2232 2233 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2234 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2235 mOwner->writeInt8(0x40); // objectTypeIndication ISO/IEC 14492-2 2236 mOwner->writeInt8(0x15); // streamType AudioStream 2237 2238 mOwner->writeInt16(0x03); // XXX 2239 mOwner->writeInt8(0x00); // buffer size 24-bit 2240 mOwner->writeInt32(96000); // max bit rate 2241 mOwner->writeInt32(96000); // avg bit rate 2242 2243 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2244 mOwner->writeInt8(mCodecSpecificDataSize); 2245 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2246 2247 static const uint8_t kData2[] = { 2248 0x06, // SLConfigDescriptorTag 2249 0x01, 2250 0x02 2251 }; 2252 mOwner->write(kData2, sizeof(kData2)); 2253 2254 mOwner->endBox(); // esds 2255 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime) || 2256 !strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2257 // 3gpp2 Spec AMRSampleEntry fields 2258 mOwner->beginBox("damr"); 2259 mOwner->writeCString(" "); // vendor: 4 bytes 2260 mOwner->writeInt8(0); // decoder version 2261 mOwner->writeInt16(0x83FF); // mode set: all enabled 2262 mOwner->writeInt8(0); // mode change period 2263 mOwner->writeInt8(1); // frames per sample 2264 mOwner->endBox(); 2265 } 2266 mOwner->endBox(); 2267 } else { 2268 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2269 mOwner->beginBox("mp4v"); 2270 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2271 mOwner->beginBox("s263"); 2272 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2273 mOwner->beginBox("avc1"); 2274 } else { 2275 LOGE("Unknown mime type '%s'.", mime); 2276 CHECK(!"should not be here, unknown mime type."); 2277 } 2278 2279 mOwner->writeInt32(0); // reserved 2280 mOwner->writeInt16(0); // reserved 2281 mOwner->writeInt16(1); // data ref index 2282 mOwner->writeInt16(0); // predefined 2283 mOwner->writeInt16(0); // reserved 2284 mOwner->writeInt32(0); // predefined 2285 mOwner->writeInt32(0); // predefined 2286 mOwner->writeInt32(0); // predefined 2287 2288 int32_t width, height; 2289 bool success = mMeta->findInt32(kKeyWidth, &width); 2290 success = success && mMeta->findInt32(kKeyHeight, &height); 2291 CHECK(success); 2292 2293 mOwner->writeInt16(width); 2294 mOwner->writeInt16(height); 2295 mOwner->writeInt32(0x480000); // horiz resolution 2296 mOwner->writeInt32(0x480000); // vert resolution 2297 mOwner->writeInt32(0); // reserved 2298 mOwner->writeInt16(1); // frame count 2299 mOwner->write(" ", 32); 2300 mOwner->writeInt16(0x18); // depth 2301 mOwner->writeInt16(-1); // predefined 2302 2303 CHECK(23 + mCodecSpecificDataSize < 128); 2304 2305 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2306 CHECK(mCodecSpecificData); 2307 CHECK(mCodecSpecificDataSize > 0); 2308 mOwner->beginBox("esds"); 2309 2310 mOwner->writeInt32(0); // version=0, flags=0 2311 2312 mOwner->writeInt8(0x03); // ES_DescrTag 2313 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2314 mOwner->writeInt16(0x0000); // ES_ID 2315 mOwner->writeInt8(0x1f); 2316 2317 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2318 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2319 mOwner->writeInt8(0x20); // objectTypeIndication ISO/IEC 14492-2 2320 mOwner->writeInt8(0x11); // streamType VisualStream 2321 2322 static const uint8_t kData[] = { 2323 0x01, 0x77, 0x00, 2324 0x00, 0x03, 0xe8, 0x00, 2325 0x00, 0x03, 0xe8, 0x00 2326 }; 2327 mOwner->write(kData, sizeof(kData)); 2328 2329 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2330 2331 mOwner->writeInt8(mCodecSpecificDataSize); 2332 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2333 2334 static const uint8_t kData2[] = { 2335 0x06, // SLConfigDescriptorTag 2336 0x01, 2337 0x02 2338 }; 2339 mOwner->write(kData2, sizeof(kData2)); 2340 2341 mOwner->endBox(); // esds 2342 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2343 mOwner->beginBox("d263"); 2344 2345 mOwner->writeInt32(0); // vendor 2346 mOwner->writeInt8(0); // decoder version 2347 mOwner->writeInt8(10); // level: 10 2348 mOwner->writeInt8(0); // profile: 0 2349 2350 mOwner->endBox(); // d263 2351 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2352 CHECK(mCodecSpecificData); 2353 CHECK(mCodecSpecificDataSize >= 5); 2354 2355 // Patch avcc's lengthSize field to match the number 2356 // of bytes we use to indicate the size of a nal unit. 2357 uint8_t *ptr = (uint8_t *)mCodecSpecificData; 2358 ptr[4] = 2359 (ptr[4] & 0xfc) 2360 | (mOwner->useNalLengthFour() ? 3 : 1); 2361 2362 mOwner->beginBox("avcC"); 2363 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2364 mOwner->endBox(); // avcC 2365 } 2366 2367 mOwner->beginBox("pasp"); 2368 // This is useful if the pixel is not square 2369 mOwner->writeInt32(1 << 16); // hspacing 2370 mOwner->writeInt32(1 << 16); // vspacing 2371 mOwner->endBox(); // pasp 2372 mOwner->endBox(); // mp4v, s263 or avc1 2373 } 2374 mOwner->endBox(); // stsd 2375 2376 mOwner->beginBox("stts"); 2377 mOwner->writeInt32(0); // version=0, flags=0 2378 mOwner->writeInt32(mNumSttsTableEntries); 2379 int64_t prevTimestampUs = 0; 2380 for (List<SttsTableEntry>::iterator it = mSttsTableEntries.begin(); 2381 it != mSttsTableEntries.end(); ++it) { 2382 mOwner->writeInt32(it->sampleCount); 2383 2384 // Make sure that we are calculating the sample duration the exactly 2385 // same way as we made decision on how to create stts entries. 2386 int64_t currTimestampUs = prevTimestampUs + it->sampleDurationUs; 2387 int32_t dur = ((currTimestampUs * mTimeScale + 500000LL) / 1000000LL - 2388 (prevTimestampUs * mTimeScale + 500000LL) / 1000000LL); 2389 prevTimestampUs += (it->sampleCount * it->sampleDurationUs); 2390 2391 mOwner->writeInt32(dur); 2392 } 2393 mOwner->endBox(); // stts 2394 2395 if (!mIsAudio) { 2396 mOwner->beginBox("stss"); 2397 mOwner->writeInt32(0); // version=0, flags=0 2398 mOwner->writeInt32(mNumStssTableEntries); // number of sync frames 2399 for (List<int32_t>::iterator it = mStssTableEntries.begin(); 2400 it != mStssTableEntries.end(); ++it) { 2401 mOwner->writeInt32(*it); 2402 } 2403 mOwner->endBox(); // stss 2404 } 2405 2406 mOwner->beginBox("stsz"); 2407 mOwner->writeInt32(0); // version=0, flags=0 2408 if (mSamplesHaveSameSize) { 2409 List<size_t>::iterator it = mSampleSizes.begin(); 2410 mOwner->writeInt32(*it); // default sample size 2411 } else { 2412 mOwner->writeInt32(0); 2413 } 2414 mOwner->writeInt32(mNumSamples); 2415 if (!mSamplesHaveSameSize) { 2416 for (List<size_t>::iterator it = mSampleSizes.begin(); 2417 it != mSampleSizes.end(); ++it) { 2418 mOwner->writeInt32(*it); 2419 } 2420 } 2421 mOwner->endBox(); // stsz 2422 2423 mOwner->beginBox("stsc"); 2424 mOwner->writeInt32(0); // version=0, flags=0 2425 mOwner->writeInt32(mNumStscTableEntries); 2426 for (List<StscTableEntry>::iterator it = mStscTableEntries.begin(); 2427 it != mStscTableEntries.end(); ++it) { 2428 mOwner->writeInt32(it->firstChunk); 2429 mOwner->writeInt32(it->samplesPerChunk); 2430 mOwner->writeInt32(it->sampleDescriptionId); 2431 } 2432 mOwner->endBox(); // stsc 2433 mOwner->beginBox(use32BitOffset? "stco": "co64"); 2434 mOwner->writeInt32(0); // version=0, flags=0 2435 mOwner->writeInt32(mNumStcoTableEntries); 2436 for (List<off_t>::iterator it = mChunkOffsets.begin(); 2437 it != mChunkOffsets.end(); ++it) { 2438 if (use32BitOffset) { 2439 mOwner->writeInt32(static_cast<int32_t>(*it)); 2440 } else { 2441 mOwner->writeInt64((*it)); 2442 } 2443 } 2444 mOwner->endBox(); // stco or co64 2445 2446 mOwner->endBox(); // stbl 2447 mOwner->endBox(); // minf 2448 mOwner->endBox(); // mdia 2449 mOwner->endBox(); // trak 2450} 2451 2452} // namespace android 2453