MPEG4Writer.cpp revision 872a481558350634a3fd5cb67939de288af00ecb
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 <ctype.h> 24#include <pthread.h> 25 26#include <media/stagefright/MPEG4Writer.h> 27#include <media/stagefright/MediaBuffer.h> 28#include <media/stagefright/MetaData.h> 29#include <media/stagefright/MediaDebug.h> 30#include <media/stagefright/MediaDefs.h> 31#include <media/stagefright/MediaErrors.h> 32#include <media/stagefright/MediaSource.h> 33#include <media/stagefright/Utils.h> 34#include <media/mediarecorder.h> 35 36#include "include/ESDS.h" 37 38namespace android { 39 40static const int64_t kMax32BitFileSize = 0x007fffffffLL; 41static const uint8_t kNalUnitTypeSeqParamSet = 0x07; 42static const uint8_t kNalUnitTypePicParamSet = 0x08; 43static const int64_t kVideoMediaTimeAdjustPeriodTimeUs = 10000000LL; // 10s 44 45class MPEG4Writer::Track { 46public: 47 Track(MPEG4Writer *owner, const sp<MediaSource> &source); 48 49 ~Track(); 50 51 status_t start(MetaData *params); 52 status_t stop(); 53 status_t pause(); 54 bool reachedEOS(); 55 56 int64_t getDurationUs() const; 57 int64_t getEstimatedTrackSizeBytes() const; 58 void writeTrackHeader(int32_t trackID, bool use32BitOffset = true); 59 void bufferChunk(int64_t timestampUs); 60 bool isAvc() const { return mIsAvc; } 61 bool isAudio() const { return mIsAudio; } 62 bool isMPEG4() const { return mIsMPEG4; } 63 void addChunkOffset(off_t offset); 64 status_t dump(int fd, const Vector<String16>& args) const; 65 66private: 67 MPEG4Writer *mOwner; 68 sp<MetaData> mMeta; 69 sp<MediaSource> mSource; 70 volatile bool mDone; 71 volatile bool mPaused; 72 volatile bool mResumed; 73 bool mIsAvc; 74 bool mIsAudio; 75 bool mIsMPEG4; 76 int64_t mTrackDurationUs; 77 78 // For realtime applications, we need to adjust the media clock 79 // for video track based on the audio media clock 80 bool mIsRealTimeRecording; 81 int64_t mMaxTimeStampUs; 82 int64_t mEstimatedTrackSizeBytes; 83 int64_t mMdatSizeBytes; 84 int32_t mTimeScale; 85 86 pthread_t mThread; 87 88 // mNumSamples is used to track how many samples in mSampleSizes List. 89 // This is to reduce the cost associated with mSampleSizes.size() call, 90 // since it is O(n). Ideally, the fix should be in List class. 91 size_t mNumSamples; 92 List<size_t> mSampleSizes; 93 bool mSamplesHaveSameSize; 94 95 List<MediaBuffer *> mChunkSamples; 96 97 size_t mNumStcoTableEntries; 98 List<off_t> mChunkOffsets; 99 100 size_t mNumStscTableEntries; 101 struct StscTableEntry { 102 103 StscTableEntry(uint32_t chunk, uint32_t samples, uint32_t id) 104 : firstChunk(chunk), 105 samplesPerChunk(samples), 106 sampleDescriptionId(id) {} 107 108 uint32_t firstChunk; 109 uint32_t samplesPerChunk; 110 uint32_t sampleDescriptionId; 111 }; 112 List<StscTableEntry> mStscTableEntries; 113 114 size_t mNumStssTableEntries; 115 List<int32_t> mStssTableEntries; 116 117 size_t mNumSttsTableEntries; 118 struct SttsTableEntry { 119 120 SttsTableEntry(uint32_t count, uint32_t durationUs) 121 : sampleCount(count), sampleDurationUs(durationUs) {} 122 123 uint32_t sampleCount; 124 uint32_t sampleDurationUs; 125 }; 126 List<SttsTableEntry> mSttsTableEntries; 127 128 // Sequence parameter set or picture parameter set 129 struct AVCParamSet { 130 AVCParamSet(uint16_t length, const uint8_t *data) 131 : mLength(length), mData(data) {} 132 133 uint16_t mLength; 134 const uint8_t *mData; 135 }; 136 List<AVCParamSet> mSeqParamSets; 137 List<AVCParamSet> mPicParamSets; 138 uint8_t mProfileIdc; 139 uint8_t mProfileCompatible; 140 uint8_t mLevelIdc; 141 142 void *mCodecSpecificData; 143 size_t mCodecSpecificDataSize; 144 bool mGotAllCodecSpecificData; 145 bool mTrackingProgressStatus; 146 147 bool mReachedEOS; 148 int64_t mStartTimestampUs; 149 int64_t mPreviousTrackTimeUs; 150 int64_t mTrackEveryTimeDurationUs; 151 152 // Has the media time adjustment for video started? 153 bool mIsMediaTimeAdjustmentOn; 154 // The time stamp when previous media time adjustment period starts 155 int64_t mPrevMediaTimeAdjustTimestampUs; 156 // Number of vidoe frames whose time stamp may be adjusted 157 int64_t mMediaTimeAdjustNumFrames; 158 // The sample number when previous meida time adjustmnet period starts 159 int64_t mPrevMediaTimeAdjustSample; 160 // The total accumulated drift time within a period of 161 // kVideoMediaTimeAdjustPeriodTimeUs. 162 int64_t mTotalDriftTimeToAdjustUs; 163 // The total accumalated drift time since the start of the recording 164 // excluding the current time adjustment period 165 int64_t mPrevTotalAccumDriftTimeUs; 166 167 // Update the audio track's drift information. 168 void updateDriftTime(const sp<MetaData>& meta); 169 170 // Adjust the time stamp of the video track according to 171 // the drift time information from the audio track. 172 void adjustMediaTime(int64_t *timestampUs); 173 174 static void *ThreadWrapper(void *me); 175 status_t threadEntry(); 176 177 const uint8_t *parseParamSet( 178 const uint8_t *data, size_t length, int type, size_t *paramSetLen); 179 180 status_t makeAVCCodecSpecificData( 181 const uint8_t *data, size_t size); 182 status_t copyAVCCodecSpecificData( 183 const uint8_t *data, size_t size); 184 status_t parseAVCCodecSpecificData( 185 const uint8_t *data, size_t size); 186 187 // Track authoring progress status 188 void trackProgressStatus(int64_t timeUs, status_t err = OK); 189 void initTrackingProgressStatus(MetaData *params); 190 191 void getCodecSpecificDataFromInputFormatIfPossible(); 192 193 // Determine the track time scale 194 // If it is an audio track, try to use the sampling rate as 195 // the time scale; however, if user chooses the overwrite 196 // value, the user-supplied time scale will be used. 197 void setTimeScale(); 198 199 // Simple validation on the codec specific data 200 status_t checkCodecSpecificData() const; 201 202 void updateTrackSizeEstimate(); 203 void addOneStscTableEntry(size_t chunkId, size_t sampleId); 204 void addOneStssTableEntry(size_t sampleId); 205 void addOneSttsTableEntry(size_t sampleCount, int64_t durationUs); 206 207 Track(const Track &); 208 Track &operator=(const Track &); 209}; 210 211MPEG4Writer::MPEG4Writer(const char *filename) 212 : mFile(fopen(filename, "wb")), 213 mUse4ByteNalLength(true), 214 mUse32BitOffset(true), 215 mPaused(false), 216 mStarted(false), 217 mOffset(0), 218 mMdatOffset(0), 219 mEstimatedMoovBoxSize(0), 220 mInterleaveDurationUs(1000000) { 221 CHECK(mFile != NULL); 222} 223 224MPEG4Writer::MPEG4Writer(int fd) 225 : mFile(fdopen(fd, "wb")), 226 mUse4ByteNalLength(true), 227 mUse32BitOffset(true), 228 mPaused(false), 229 mStarted(false), 230 mOffset(0), 231 mMdatOffset(0), 232 mEstimatedMoovBoxSize(0), 233 mInterleaveDurationUs(1000000) { 234 CHECK(mFile != NULL); 235} 236 237MPEG4Writer::~MPEG4Writer() { 238 stop(); 239 240 while (!mTracks.empty()) { 241 List<Track *>::iterator it = mTracks.begin(); 242 delete *it; 243 (*it) = NULL; 244 mTracks.erase(it); 245 } 246 mTracks.clear(); 247} 248 249status_t MPEG4Writer::dump( 250 int fd, const Vector<String16>& args) { 251 const size_t SIZE = 256; 252 char buffer[SIZE]; 253 String8 result; 254 snprintf(buffer, SIZE, " MPEG4Writer %p\n", this); 255 result.append(buffer); 256 snprintf(buffer, SIZE, " mStarted: %s\n", mStarted? "true": "false"); 257 result.append(buffer); 258 ::write(fd, result.string(), result.size()); 259 for (List<Track *>::iterator it = mTracks.begin(); 260 it != mTracks.end(); ++it) { 261 (*it)->dump(fd, args); 262 } 263 return OK; 264} 265 266status_t MPEG4Writer::Track::dump( 267 int fd, const Vector<String16>& args) const { 268 const size_t SIZE = 256; 269 char buffer[SIZE]; 270 String8 result; 271 snprintf(buffer, SIZE, " %s track\n", mIsAudio? "Audio": "Video"); 272 result.append(buffer); 273 snprintf(buffer, SIZE, " reached EOS: %s\n", 274 mReachedEOS? "true": "false"); 275 result.append(buffer); 276 ::write(fd, result.string(), result.size()); 277 return OK; 278} 279 280status_t MPEG4Writer::addSource(const sp<MediaSource> &source) { 281 Track *track = new Track(this, source); 282 mTracks.push_back(track); 283 284 return OK; 285} 286 287status_t MPEG4Writer::startTracks(MetaData *params) { 288 for (List<Track *>::iterator it = mTracks.begin(); 289 it != mTracks.end(); ++it) { 290 status_t err = (*it)->start(params); 291 292 if (err != OK) { 293 for (List<Track *>::iterator it2 = mTracks.begin(); 294 it2 != it; ++it2) { 295 (*it2)->stop(); 296 } 297 298 return err; 299 } 300 } 301 return OK; 302} 303 304int64_t MPEG4Writer::estimateMoovBoxSize(int32_t bitRate) { 305 // This implementation is highly experimental/heurisitic. 306 // 307 // Statistical analysis shows that metadata usually accounts 308 // for a small portion of the total file size, usually < 0.6%. 309 // Currently, lets set to 0.4% for now. 310 311 // The default MIN_MOOV_BOX_SIZE is set to 0.4% x 1MB, 312 // where 1MB is the common file size limit for MMS application. 313 // The default MAX _MOOV_BOX_SIZE value is based on about 4 314 // minute video recording with a bit rate about 3 Mbps, because 315 // statistics also show that most of the video captured are going 316 // to be less than 3 minutes. 317 318 // If the estimation is wrong, we will pay the price of wasting 319 // some reserved space. This should not happen so often statistically. 320 static const int32_t factor = mUse32BitOffset? 1: 2; 321 static const int64_t MIN_MOOV_BOX_SIZE = 4 * 1024; // 4 KB 322 static const int64_t MAX_MOOV_BOX_SIZE = (180 * 3000000 * 6LL / 8000); 323 int64_t size = MIN_MOOV_BOX_SIZE; 324 325 if (mMaxFileSizeLimitBytes != 0) { 326 size = mMaxFileSizeLimitBytes * 4 / 1000; 327 } else if (mMaxFileDurationLimitUs != 0) { 328 if (bitRate <= 0) { 329 // We could not estimate the file size since bitRate is not set. 330 size = MIN_MOOV_BOX_SIZE; 331 } else { 332 size = ((mMaxFileDurationLimitUs * bitRate * 4) / 1000 / 8000000); 333 } 334 } 335 if (size < MIN_MOOV_BOX_SIZE) { 336 size = MIN_MOOV_BOX_SIZE; 337 } 338 339 // Any long duration recording will be probably end up with 340 // non-streamable mp4 file. 341 if (size > MAX_MOOV_BOX_SIZE) { 342 size = MAX_MOOV_BOX_SIZE; 343 } 344 345 LOGV("limits: %lld/%lld bytes/us, bit rate: %d bps and the estimated" 346 " moov size %lld bytes", 347 mMaxFileSizeLimitBytes, mMaxFileDurationLimitUs, bitRate, size); 348 return factor * size; 349} 350 351status_t MPEG4Writer::start(MetaData *param) { 352 if (mFile == NULL) { 353 return UNKNOWN_ERROR; 354 } 355 356 int32_t use64BitOffset; 357 if (param && 358 param->findInt32(kKey64BitFileOffset, &use64BitOffset) && 359 use64BitOffset) { 360 mUse32BitOffset = false; 361 } 362 363 if (mUse32BitOffset) { 364 // Implicit 32 bit file size limit 365 if (mMaxFileSizeLimitBytes == 0) { 366 mMaxFileSizeLimitBytes = kMax32BitFileSize; 367 } 368 369 // If file size is set to be larger than the 32 bit file 370 // size limit, treat it as an error. 371 if (mMaxFileSizeLimitBytes > kMax32BitFileSize) { 372 LOGW("32-bit file size limit (%lld bytes) too big. " 373 "It is changed to %lld bytes", 374 mMaxFileSizeLimitBytes, kMax32BitFileSize); 375 mMaxFileSizeLimitBytes = kMax32BitFileSize; 376 } 377 } 378 379 int32_t use2ByteNalLength; 380 if (param && 381 param->findInt32(kKey2ByteNalLength, &use2ByteNalLength) && 382 use2ByteNalLength) { 383 mUse4ByteNalLength = false; 384 } 385 386 mStartTimestampUs = -1; 387 388 if (mStarted) { 389 if (mPaused) { 390 mPaused = false; 391 return startTracks(param); 392 } 393 return OK; 394 } 395 396 if (!param || 397 !param->findInt32(kKeyTimeScale, &mTimeScale)) { 398 mTimeScale = 1000; 399 } 400 CHECK(mTimeScale > 0); 401 LOGV("movie time scale: %d", mTimeScale); 402 403 mStreamableFile = true; 404 mWriteMoovBoxToMemory = false; 405 mMoovBoxBuffer = NULL; 406 mMoovBoxBufferOffset = 0; 407 408 beginBox("ftyp"); 409 { 410 int32_t fileType; 411 if (param && param->findInt32(kKeyFileType, &fileType) && 412 fileType != OUTPUT_FORMAT_MPEG_4) { 413 writeFourcc("3gp4"); 414 } else { 415 writeFourcc("isom"); 416 } 417 } 418 writeInt32(0); 419 writeFourcc("isom"); 420 writeFourcc("3gp4"); 421 endBox(); 422 423 mFreeBoxOffset = mOffset; 424 425 if (mEstimatedMoovBoxSize == 0) { 426 int32_t bitRate = -1; 427 if (param) { 428 param->findInt32(kKeyBitRate, &bitRate); 429 } 430 mEstimatedMoovBoxSize = estimateMoovBoxSize(bitRate); 431 } 432 CHECK(mEstimatedMoovBoxSize >= 8); 433 fseeko(mFile, mFreeBoxOffset, SEEK_SET); 434 writeInt32(mEstimatedMoovBoxSize); 435 write("free", 4); 436 437 mMdatOffset = mFreeBoxOffset + mEstimatedMoovBoxSize; 438 mOffset = mMdatOffset; 439 fseeko(mFile, mMdatOffset, SEEK_SET); 440 if (mUse32BitOffset) { 441 write("????mdat", 8); 442 } else { 443 write("\x00\x00\x00\x01mdat????????", 16); 444 } 445 446 status_t err = startWriterThread(); 447 if (err != OK) { 448 return err; 449 } 450 451 err = startTracks(param); 452 if (err != OK) { 453 return err; 454 } 455 456 mStarted = true; 457 return OK; 458} 459 460bool MPEG4Writer::use32BitFileOffset() const { 461 return mUse32BitOffset; 462} 463 464status_t MPEG4Writer::pause() { 465 if (mFile == NULL) { 466 return OK; 467 } 468 mPaused = true; 469 status_t err = OK; 470 for (List<Track *>::iterator it = mTracks.begin(); 471 it != mTracks.end(); ++it) { 472 status_t status = (*it)->pause(); 473 if (status != OK) { 474 err = status; 475 } 476 } 477 return err; 478} 479 480void MPEG4Writer::stopWriterThread() { 481 LOGV("stopWriterThread"); 482 483 { 484 Mutex::Autolock autolock(mLock); 485 486 mDone = true; 487 mChunkReadyCondition.signal(); 488 } 489 490 void *dummy; 491 pthread_join(mThread, &dummy); 492} 493 494status_t MPEG4Writer::stop() { 495 if (mFile == NULL) { 496 return OK; 497 } 498 499 status_t err = OK; 500 int64_t maxDurationUs = 0; 501 for (List<Track *>::iterator it = mTracks.begin(); 502 it != mTracks.end(); ++it) { 503 status_t status = (*it)->stop(); 504 if (err == OK && status != OK) { 505 err = status; 506 } 507 508 int64_t durationUs = (*it)->getDurationUs(); 509 if (durationUs > maxDurationUs) { 510 maxDurationUs = durationUs; 511 } 512 } 513 514 stopWriterThread(); 515 516 // Do not write out movie header on error. 517 if (err != OK) { 518 fflush(mFile); 519 fclose(mFile); 520 mFile = NULL; 521 mStarted = false; 522 return err; 523 } 524 525 // Fix up the size of the 'mdat' chunk. 526 if (mUse32BitOffset) { 527 fseeko(mFile, mMdatOffset, SEEK_SET); 528 int32_t size = htonl(static_cast<int32_t>(mOffset - mMdatOffset)); 529 fwrite(&size, 1, 4, mFile); 530 } else { 531 fseeko(mFile, mMdatOffset + 8, SEEK_SET); 532 int64_t size = mOffset - mMdatOffset; 533 size = hton64(size); 534 fwrite(&size, 1, 8, mFile); 535 } 536 fseeko(mFile, mOffset, SEEK_SET); 537 538 time_t now = time(NULL); 539 const off_t moovOffset = mOffset; 540 mWriteMoovBoxToMemory = true; 541 mMoovBoxBuffer = (uint8_t *) malloc(mEstimatedMoovBoxSize); 542 mMoovBoxBufferOffset = 0; 543 CHECK(mMoovBoxBuffer != NULL); 544 int32_t duration = (maxDurationUs * mTimeScale + 5E5) / 1E6; 545 546 beginBox("moov"); 547 548 beginBox("mvhd"); 549 writeInt32(0); // version=0, flags=0 550 writeInt32(now); // creation time 551 writeInt32(now); // modification time 552 writeInt32(mTimeScale); // mvhd timescale 553 writeInt32(duration); 554 writeInt32(0x10000); // rate: 1.0 555 writeInt16(0x100); // volume 556 writeInt16(0); // reserved 557 writeInt32(0); // reserved 558 writeInt32(0); // reserved 559 writeInt32(0x10000); // matrix 560 writeInt32(0); 561 writeInt32(0); 562 writeInt32(0); 563 writeInt32(0x10000); 564 writeInt32(0); 565 writeInt32(0); 566 writeInt32(0); 567 writeInt32(0x40000000); 568 writeInt32(0); // predefined 569 writeInt32(0); // predefined 570 writeInt32(0); // predefined 571 writeInt32(0); // predefined 572 writeInt32(0); // predefined 573 writeInt32(0); // predefined 574 writeInt32(mTracks.size() + 1); // nextTrackID 575 endBox(); // mvhd 576 577 int32_t id = 1; 578 for (List<Track *>::iterator it = mTracks.begin(); 579 it != mTracks.end(); ++it, ++id) { 580 (*it)->writeTrackHeader(id, mUse32BitOffset); 581 } 582 endBox(); // moov 583 584 mWriteMoovBoxToMemory = false; 585 if (mStreamableFile) { 586 CHECK(mMoovBoxBufferOffset + 8 <= mEstimatedMoovBoxSize); 587 588 // Moov box 589 fseeko(mFile, mFreeBoxOffset, SEEK_SET); 590 mOffset = mFreeBoxOffset; 591 write(mMoovBoxBuffer, 1, mMoovBoxBufferOffset, mFile); 592 593 // Free box 594 fseeko(mFile, mOffset, SEEK_SET); 595 writeInt32(mEstimatedMoovBoxSize - mMoovBoxBufferOffset); 596 write("free", 4); 597 598 // Free temp memory 599 free(mMoovBoxBuffer); 600 mMoovBoxBuffer = NULL; 601 mMoovBoxBufferOffset = 0; 602 } else { 603 LOGI("The mp4 file will not be streamable."); 604 } 605 606 CHECK(mBoxes.empty()); 607 608 fflush(mFile); 609 fclose(mFile); 610 mFile = NULL; 611 mStarted = false; 612 return err; 613} 614 615status_t MPEG4Writer::setInterleaveDuration(uint32_t durationUs) { 616 mInterleaveDurationUs = durationUs; 617 return OK; 618} 619 620void MPEG4Writer::lock() { 621 mLock.lock(); 622} 623 624void MPEG4Writer::unlock() { 625 mLock.unlock(); 626} 627 628off_t MPEG4Writer::addSample_l(MediaBuffer *buffer) { 629 off_t old_offset = mOffset; 630 631 fwrite((const uint8_t *)buffer->data() + buffer->range_offset(), 632 1, buffer->range_length(), mFile); 633 634 mOffset += buffer->range_length(); 635 636 return old_offset; 637} 638 639static void StripStartcode(MediaBuffer *buffer) { 640 if (buffer->range_length() < 4) { 641 return; 642 } 643 644 const uint8_t *ptr = 645 (const uint8_t *)buffer->data() + buffer->range_offset(); 646 647 if (!memcmp(ptr, "\x00\x00\x00\x01", 4)) { 648 buffer->set_range( 649 buffer->range_offset() + 4, buffer->range_length() - 4); 650 } 651} 652 653off_t MPEG4Writer::addLengthPrefixedSample_l(MediaBuffer *buffer) { 654 off_t old_offset = mOffset; 655 656 size_t length = buffer->range_length(); 657 658 if (mUse4ByteNalLength) { 659 uint8_t x = length >> 24; 660 fwrite(&x, 1, 1, mFile); 661 x = (length >> 16) & 0xff; 662 fwrite(&x, 1, 1, mFile); 663 x = (length >> 8) & 0xff; 664 fwrite(&x, 1, 1, mFile); 665 x = length & 0xff; 666 fwrite(&x, 1, 1, mFile); 667 668 fwrite((const uint8_t *)buffer->data() + buffer->range_offset(), 669 1, length, mFile); 670 mOffset += length + 4; 671 } else { 672 CHECK(length < 65536); 673 674 uint8_t x = length >> 8; 675 fwrite(&x, 1, 1, mFile); 676 x = length & 0xff; 677 fwrite(&x, 1, 1, mFile); 678 fwrite((const uint8_t *)buffer->data() + buffer->range_offset(), 679 1, length, mFile); 680 mOffset += length + 2; 681 } 682 683 return old_offset; 684} 685 686size_t MPEG4Writer::write( 687 const void *ptr, size_t size, size_t nmemb, FILE *stream) { 688 689 const size_t bytes = size * nmemb; 690 if (mWriteMoovBoxToMemory) { 691 off_t moovBoxSize = 8 + mMoovBoxBufferOffset + bytes; 692 if (moovBoxSize > mEstimatedMoovBoxSize) { 693 for (List<off_t>::iterator it = mBoxes.begin(); 694 it != mBoxes.end(); ++it) { 695 (*it) += mOffset; 696 } 697 fseeko(mFile, mOffset, SEEK_SET); 698 fwrite(mMoovBoxBuffer, 1, mMoovBoxBufferOffset, stream); 699 fwrite(ptr, size, nmemb, stream); 700 mOffset += (bytes + mMoovBoxBufferOffset); 701 free(mMoovBoxBuffer); 702 mMoovBoxBuffer = NULL; 703 mMoovBoxBufferOffset = 0; 704 mWriteMoovBoxToMemory = false; 705 mStreamableFile = false; 706 } else { 707 memcpy(mMoovBoxBuffer + mMoovBoxBufferOffset, ptr, bytes); 708 mMoovBoxBufferOffset += bytes; 709 } 710 } else { 711 fwrite(ptr, size, nmemb, stream); 712 mOffset += bytes; 713 } 714 return bytes; 715} 716 717void MPEG4Writer::beginBox(const char *fourcc) { 718 CHECK_EQ(strlen(fourcc), 4); 719 720 mBoxes.push_back(mWriteMoovBoxToMemory? 721 mMoovBoxBufferOffset: mOffset); 722 723 writeInt32(0); 724 writeFourcc(fourcc); 725} 726 727void MPEG4Writer::endBox() { 728 CHECK(!mBoxes.empty()); 729 730 off_t offset = *--mBoxes.end(); 731 mBoxes.erase(--mBoxes.end()); 732 733 if (mWriteMoovBoxToMemory) { 734 int32_t x = htonl(mMoovBoxBufferOffset - offset); 735 memcpy(mMoovBoxBuffer + offset, &x, 4); 736 } else { 737 fseeko(mFile, offset, SEEK_SET); 738 writeInt32(mOffset - offset); 739 mOffset -= 4; 740 fseeko(mFile, mOffset, SEEK_SET); 741 } 742} 743 744void MPEG4Writer::writeInt8(int8_t x) { 745 write(&x, 1, 1, mFile); 746} 747 748void MPEG4Writer::writeInt16(int16_t x) { 749 x = htons(x); 750 write(&x, 1, 2, mFile); 751} 752 753void MPEG4Writer::writeInt32(int32_t x) { 754 x = htonl(x); 755 write(&x, 1, 4, mFile); 756} 757 758void MPEG4Writer::writeInt64(int64_t x) { 759 x = hton64(x); 760 write(&x, 1, 8, mFile); 761} 762 763void MPEG4Writer::writeCString(const char *s) { 764 size_t n = strlen(s); 765 write(s, 1, n + 1, mFile); 766} 767 768void MPEG4Writer::writeFourcc(const char *s) { 769 CHECK_EQ(strlen(s), 4); 770 write(s, 1, 4, mFile); 771} 772 773void MPEG4Writer::write(const void *data, size_t size) { 774 write(data, 1, size, mFile); 775} 776 777bool MPEG4Writer::exceedsFileSizeLimit() { 778 // No limit 779 if (mMaxFileSizeLimitBytes == 0) { 780 return false; 781 } 782 783 int64_t nTotalBytesEstimate = static_cast<int64_t>(mEstimatedMoovBoxSize); 784 for (List<Track *>::iterator it = mTracks.begin(); 785 it != mTracks.end(); ++it) { 786 nTotalBytesEstimate += (*it)->getEstimatedTrackSizeBytes(); 787 } 788 789 return (nTotalBytesEstimate + 1024 >= mMaxFileSizeLimitBytes); 790} 791 792bool MPEG4Writer::exceedsFileDurationLimit() { 793 // No limit 794 if (mMaxFileDurationLimitUs == 0) { 795 return false; 796 } 797 798 for (List<Track *>::iterator it = mTracks.begin(); 799 it != mTracks.end(); ++it) { 800 if ((*it)->getDurationUs() >= mMaxFileDurationLimitUs) { 801 return true; 802 } 803 } 804 return false; 805} 806 807bool MPEG4Writer::reachedEOS() { 808 bool allDone = true; 809 for (List<Track *>::iterator it = mTracks.begin(); 810 it != mTracks.end(); ++it) { 811 if (!(*it)->reachedEOS()) { 812 allDone = false; 813 break; 814 } 815 } 816 817 return allDone; 818} 819 820void MPEG4Writer::setStartTimestampUs(int64_t timeUs) { 821 LOGI("setStartTimestampUs: %lld", timeUs); 822 CHECK(timeUs >= 0); 823 Mutex::Autolock autoLock(mLock); 824 if (mStartTimestampUs < 0 || mStartTimestampUs > timeUs) { 825 mStartTimestampUs = timeUs; 826 LOGI("Earliest track starting time: %lld", mStartTimestampUs); 827 } 828} 829 830int64_t MPEG4Writer::getStartTimestampUs() { 831 Mutex::Autolock autoLock(mLock); 832 return mStartTimestampUs; 833} 834 835size_t MPEG4Writer::numTracks() { 836 Mutex::Autolock autolock(mLock); 837 return mTracks.size(); 838} 839 840//////////////////////////////////////////////////////////////////////////////// 841 842MPEG4Writer::Track::Track( 843 MPEG4Writer *owner, const sp<MediaSource> &source) 844 : mOwner(owner), 845 mMeta(source->getFormat()), 846 mSource(source), 847 mDone(false), 848 mPaused(false), 849 mResumed(false), 850 mTrackDurationUs(0), 851 mEstimatedTrackSizeBytes(0), 852 mSamplesHaveSameSize(true), 853 mCodecSpecificData(NULL), 854 mCodecSpecificDataSize(0), 855 mGotAllCodecSpecificData(false), 856 mReachedEOS(false) { 857 getCodecSpecificDataFromInputFormatIfPossible(); 858 859 const char *mime; 860 mMeta->findCString(kKeyMIMEType, &mime); 861 mIsAvc = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC); 862 mIsAudio = !strncasecmp(mime, "audio/", 6); 863 mIsMPEG4 = !strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) || 864 !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC); 865 866 setTimeScale(); 867} 868 869void MPEG4Writer::Track::updateTrackSizeEstimate() { 870 871 int64_t stcoBoxSizeBytes = mOwner->use32BitFileOffset() 872 ? mNumStcoTableEntries * 4 873 : mNumStcoTableEntries * 8; 874 875 int64_t stszBoxSizeBytes = mSamplesHaveSameSize? 4: (mNumSamples * 4); 876 877 mEstimatedTrackSizeBytes = mMdatSizeBytes + // media data size 878 mNumStscTableEntries * 12 + // stsc box size 879 mNumStssTableEntries * 4 + // stss box size 880 mNumSttsTableEntries * 8 + // stts box size 881 stcoBoxSizeBytes + // stco box size 882 stszBoxSizeBytes; // stsz box size 883} 884 885void MPEG4Writer::Track::addOneStscTableEntry( 886 size_t chunkId, size_t sampleId) { 887 888 StscTableEntry stscEntry(chunkId, sampleId, 1); 889 mStscTableEntries.push_back(stscEntry); 890 ++mNumStscTableEntries; 891} 892 893void MPEG4Writer::Track::addOneStssTableEntry(size_t sampleId) { 894 mStssTableEntries.push_back(sampleId); 895 ++mNumStssTableEntries; 896} 897 898void MPEG4Writer::Track::addOneSttsTableEntry( 899 size_t sampleCount, int64_t durationUs) { 900 901 SttsTableEntry sttsEntry(sampleCount, durationUs); 902 mSttsTableEntries.push_back(sttsEntry); 903 ++mNumSttsTableEntries; 904} 905 906void MPEG4Writer::Track::addChunkOffset(off_t offset) { 907 ++mNumStcoTableEntries; 908 mChunkOffsets.push_back(offset); 909} 910 911void MPEG4Writer::Track::setTimeScale() { 912 LOGV("setTimeScale"); 913 // Default time scale 914 mTimeScale = 90000; 915 916 if (mIsAudio) { 917 // Use the sampling rate as the default time scale for audio track. 918 int32_t sampleRate; 919 bool success = mMeta->findInt32(kKeySampleRate, &sampleRate); 920 CHECK(success); 921 mTimeScale = sampleRate; 922 } 923 924 // If someone would like to overwrite the timescale, use user-supplied value. 925 int32_t timeScale; 926 if (mMeta->findInt32(kKeyTimeScale, &timeScale)) { 927 mTimeScale = timeScale; 928 } 929 930 CHECK(mTimeScale > 0); 931} 932 933void MPEG4Writer::Track::getCodecSpecificDataFromInputFormatIfPossible() { 934 const char *mime; 935 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 936 937 if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_AVC)) { 938 uint32_t type; 939 const void *data; 940 size_t size; 941 if (mMeta->findData(kKeyAVCC, &type, &data, &size)) { 942 mCodecSpecificData = malloc(size); 943 mCodecSpecificDataSize = size; 944 memcpy(mCodecSpecificData, data, size); 945 mGotAllCodecSpecificData = true; 946 } 947 } else if (!strcasecmp(mime, MEDIA_MIMETYPE_VIDEO_MPEG4) 948 || !strcasecmp(mime, MEDIA_MIMETYPE_AUDIO_AAC)) { 949 uint32_t type; 950 const void *data; 951 size_t size; 952 if (mMeta->findData(kKeyESDS, &type, &data, &size)) { 953 ESDS esds(data, size); 954 if (esds.getCodecSpecificInfo(&data, &size) == OK) { 955 mCodecSpecificData = malloc(size); 956 mCodecSpecificDataSize = size; 957 memcpy(mCodecSpecificData, data, size); 958 mGotAllCodecSpecificData = true; 959 } 960 } 961 } 962} 963 964MPEG4Writer::Track::~Track() { 965 stop(); 966 967 if (mCodecSpecificData != NULL) { 968 free(mCodecSpecificData); 969 mCodecSpecificData = NULL; 970 } 971} 972 973void MPEG4Writer::Track::initTrackingProgressStatus(MetaData *params) { 974 LOGV("initTrackingProgressStatus"); 975 mPreviousTrackTimeUs = -1; 976 mTrackingProgressStatus = false; 977 mTrackEveryTimeDurationUs = 0; 978 { 979 int64_t timeUs; 980 if (params && params->findInt64(kKeyTrackTimeStatus, &timeUs)) { 981 LOGV("Receive request to track progress status for every %lld us", timeUs); 982 mTrackEveryTimeDurationUs = timeUs; 983 mTrackingProgressStatus = true; 984 } 985 } 986} 987 988// static 989void *MPEG4Writer::ThreadWrapper(void *me) { 990 LOGV("ThreadWrapper: %p", me); 991 MPEG4Writer *writer = static_cast<MPEG4Writer *>(me); 992 writer->threadFunc(); 993 return NULL; 994} 995 996void MPEG4Writer::bufferChunk(const Chunk& chunk) { 997 LOGV("bufferChunk: %p", chunk.mTrack); 998 Mutex::Autolock autolock(mLock); 999 CHECK_EQ(mDone, false); 1000 1001 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1002 it != mChunkInfos.end(); ++it) { 1003 1004 if (chunk.mTrack == it->mTrack) { // Found owner 1005 it->mChunks.push_back(chunk); 1006 mChunkReadyCondition.signal(); 1007 return; 1008 } 1009 } 1010 1011 CHECK("Received a chunk for a unknown track" == 0); 1012} 1013 1014void MPEG4Writer::writeFirstChunk(ChunkInfo* info) { 1015 LOGV("writeFirstChunk: %p", info->mTrack); 1016 1017 List<Chunk>::iterator chunkIt = info->mChunks.begin(); 1018 for (List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1019 it != chunkIt->mSamples.end(); ++it) { 1020 1021 off_t offset = info->mTrack->isAvc() 1022 ? addLengthPrefixedSample_l(*it) 1023 : addSample_l(*it); 1024 if (it == chunkIt->mSamples.begin()) { 1025 info->mTrack->addChunkOffset(offset); 1026 } 1027 } 1028 1029 // Done with the current chunk. 1030 // Release all the samples in this chunk. 1031 while (!chunkIt->mSamples.empty()) { 1032 List<MediaBuffer *>::iterator it = chunkIt->mSamples.begin(); 1033 (*it)->release(); 1034 (*it) = NULL; 1035 chunkIt->mSamples.erase(it); 1036 } 1037 chunkIt->mSamples.clear(); 1038 info->mChunks.erase(chunkIt); 1039} 1040 1041void MPEG4Writer::writeChunks() { 1042 LOGV("writeChunks"); 1043 size_t outstandingChunks = 0; 1044 while (!mChunkInfos.empty()) { 1045 List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1046 while (!it->mChunks.empty()) { 1047 CHECK_EQ(OK, writeOneChunk()); 1048 ++outstandingChunks; 1049 } 1050 it->mTrack = NULL; 1051 mChunkInfos.erase(it); 1052 } 1053 mChunkInfos.clear(); 1054 LOGD("%d chunks are written in the last batch", outstandingChunks); 1055} 1056 1057status_t MPEG4Writer::writeOneChunk() { 1058 LOGV("writeOneChunk"); 1059 1060 // Find the smallest timestamp, and write that chunk out 1061 // XXX: What if some track is just too slow? 1062 int64_t minTimestampUs = 0x7FFFFFFFFFFFFFFFLL; 1063 Track *track = NULL; 1064 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1065 it != mChunkInfos.end(); ++it) { 1066 if (!it->mChunks.empty()) { 1067 List<Chunk>::iterator chunkIt = it->mChunks.begin(); 1068 if (chunkIt->mTimeStampUs < minTimestampUs) { 1069 minTimestampUs = chunkIt->mTimeStampUs; 1070 track = it->mTrack; 1071 } 1072 } 1073 } 1074 1075 if (track == NULL) { 1076 LOGV("Nothing to be written after all"); 1077 return OK; 1078 } 1079 1080 if (mIsFirstChunk) { 1081 mIsFirstChunk = false; 1082 } 1083 for (List<ChunkInfo>::iterator it = mChunkInfos.begin(); 1084 it != mChunkInfos.end(); ++it) { 1085 if (it->mTrack == track) { 1086 writeFirstChunk(&(*it)); 1087 } 1088 } 1089 return OK; 1090} 1091 1092void MPEG4Writer::threadFunc() { 1093 LOGV("threadFunc"); 1094 1095 while (!mDone) { 1096 { 1097 Mutex::Autolock autolock(mLock); 1098 mChunkReadyCondition.wait(mLock); 1099 CHECK_EQ(writeOneChunk(), OK); 1100 } 1101 } 1102 1103 { 1104 // Write ALL samples 1105 Mutex::Autolock autolock(mLock); 1106 writeChunks(); 1107 } 1108} 1109 1110status_t MPEG4Writer::startWriterThread() { 1111 LOGV("startWriterThread"); 1112 1113 mDone = false; 1114 mIsFirstChunk = true; 1115 mDriftTimeUs = 0; 1116 for (List<Track *>::iterator it = mTracks.begin(); 1117 it != mTracks.end(); ++it) { 1118 ChunkInfo info; 1119 info.mTrack = *it; 1120 mChunkInfos.push_back(info); 1121 } 1122 1123 pthread_attr_t attr; 1124 pthread_attr_init(&attr); 1125 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1126 pthread_create(&mThread, &attr, ThreadWrapper, this); 1127 pthread_attr_destroy(&attr); 1128 return OK; 1129} 1130 1131 1132status_t MPEG4Writer::Track::start(MetaData *params) { 1133 if (!mDone && mPaused) { 1134 mPaused = false; 1135 mResumed = true; 1136 return OK; 1137 } 1138 1139 int64_t startTimeUs; 1140 if (params == NULL || !params->findInt64(kKeyTime, &startTimeUs)) { 1141 startTimeUs = 0; 1142 } 1143 1144 mIsRealTimeRecording = false; 1145 { 1146 int32_t isNotRealTime; 1147 if (params && params->findInt32(kKeyNotRealTime, &isNotRealTime)) { 1148 mIsRealTimeRecording = (isNotRealTime == 0); 1149 } 1150 } 1151 1152 initTrackingProgressStatus(params); 1153 1154 sp<MetaData> meta = new MetaData; 1155 meta->setInt64(kKeyTime, startTimeUs); 1156 status_t err = mSource->start(meta.get()); 1157 if (err != OK) { 1158 mDone = mReachedEOS = true; 1159 return err; 1160 } 1161 1162 pthread_attr_t attr; 1163 pthread_attr_init(&attr); 1164 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); 1165 1166 mDone = false; 1167 mTrackDurationUs = 0; 1168 mReachedEOS = false; 1169 mEstimatedTrackSizeBytes = 0; 1170 mNumStcoTableEntries = 0; 1171 mNumStssTableEntries = 0; 1172 mNumStscTableEntries = 0; 1173 mNumSttsTableEntries = 0; 1174 mMdatSizeBytes = 0; 1175 mIsMediaTimeAdjustmentOn = false; 1176 mPrevMediaTimeAdjustTimestampUs = 0; 1177 mMediaTimeAdjustNumFrames = 0; 1178 mPrevMediaTimeAdjustSample = 0; 1179 mTotalDriftTimeToAdjustUs = 0; 1180 mPrevTotalAccumDriftTimeUs = 0; 1181 1182 pthread_create(&mThread, &attr, ThreadWrapper, this); 1183 pthread_attr_destroy(&attr); 1184 1185 return OK; 1186} 1187 1188status_t MPEG4Writer::Track::pause() { 1189 mPaused = true; 1190 return OK; 1191} 1192 1193status_t MPEG4Writer::Track::stop() { 1194 if (mDone) { 1195 return OK; 1196 } 1197 1198 mDone = true; 1199 1200 void *dummy; 1201 pthread_join(mThread, &dummy); 1202 1203 status_t err = (status_t) dummy; 1204 1205 { 1206 status_t status = mSource->stop(); 1207 if (err == OK && status != OK && status != ERROR_END_OF_STREAM) { 1208 err = status; 1209 } 1210 } 1211 1212 return err; 1213} 1214 1215bool MPEG4Writer::Track::reachedEOS() { 1216 return mReachedEOS; 1217} 1218 1219// static 1220void *MPEG4Writer::Track::ThreadWrapper(void *me) { 1221 Track *track = static_cast<Track *>(me); 1222 1223 status_t err = track->threadEntry(); 1224 return (void *) err; 1225} 1226 1227static void getNalUnitType(uint8_t byte, uint8_t* type) { 1228 LOGV("getNalUnitType: %d", byte); 1229 1230 // nal_unit_type: 5-bit unsigned integer 1231 *type = (byte & 0x1F); 1232} 1233 1234static const uint8_t *findNextStartCode( 1235 const uint8_t *data, size_t length) { 1236 1237 LOGV("findNextStartCode: %p %d", data, length); 1238 1239 size_t bytesLeft = length; 1240 while (bytesLeft > 4 && 1241 memcmp("\x00\x00\x00\x01", &data[length - bytesLeft], 4)) { 1242 --bytesLeft; 1243 } 1244 if (bytesLeft <= 4) { 1245 bytesLeft = 0; // Last parameter set 1246 } 1247 return &data[length - bytesLeft]; 1248} 1249 1250const uint8_t *MPEG4Writer::Track::parseParamSet( 1251 const uint8_t *data, size_t length, int type, size_t *paramSetLen) { 1252 1253 LOGV("parseParamSet"); 1254 CHECK(type == kNalUnitTypeSeqParamSet || 1255 type == kNalUnitTypePicParamSet); 1256 1257 const uint8_t *nextStartCode = findNextStartCode(data, length); 1258 *paramSetLen = nextStartCode - data; 1259 if (*paramSetLen == 0) { 1260 LOGE("Param set is malformed, since its length is 0"); 1261 return NULL; 1262 } 1263 1264 AVCParamSet paramSet(*paramSetLen, data); 1265 if (type == kNalUnitTypeSeqParamSet) { 1266 if (*paramSetLen < 4) { 1267 LOGE("Seq parameter set malformed"); 1268 return NULL; 1269 } 1270 if (mSeqParamSets.empty()) { 1271 mProfileIdc = data[1]; 1272 mProfileCompatible = data[2]; 1273 mLevelIdc = data[3]; 1274 } else { 1275 if (mProfileIdc != data[1] || 1276 mProfileCompatible != data[2] || 1277 mLevelIdc != data[3]) { 1278 LOGE("Inconsistent profile/level found in seq parameter sets"); 1279 return NULL; 1280 } 1281 } 1282 mSeqParamSets.push_back(paramSet); 1283 } else { 1284 mPicParamSets.push_back(paramSet); 1285 } 1286 return nextStartCode; 1287} 1288 1289status_t MPEG4Writer::Track::copyAVCCodecSpecificData( 1290 const uint8_t *data, size_t size) { 1291 LOGV("copyAVCCodecSpecificData"); 1292 1293 // 2 bytes for each of the parameter set length field 1294 // plus the 7 bytes for the header 1295 if (size < 4 + 7) { 1296 LOGE("Codec specific data length too short: %d", size); 1297 return ERROR_MALFORMED; 1298 } 1299 1300 mCodecSpecificDataSize = size; 1301 mCodecSpecificData = malloc(size); 1302 memcpy(mCodecSpecificData, data, size); 1303 return OK; 1304} 1305 1306status_t MPEG4Writer::Track::parseAVCCodecSpecificData( 1307 const uint8_t *data, size_t size) { 1308 1309 LOGV("parseAVCCodecSpecificData"); 1310 // Data starts with a start code. 1311 // SPS and PPS are separated with start codes. 1312 // Also, SPS must come before PPS 1313 uint8_t type = kNalUnitTypeSeqParamSet; 1314 bool gotSps = false; 1315 bool gotPps = false; 1316 const uint8_t *tmp = data; 1317 const uint8_t *nextStartCode = data; 1318 size_t bytesLeft = size; 1319 size_t paramSetLen = 0; 1320 mCodecSpecificDataSize = 0; 1321 while (bytesLeft > 4 && !memcmp("\x00\x00\x00\x01", tmp, 4)) { 1322 getNalUnitType(*(tmp + 4), &type); 1323 if (type == kNalUnitTypeSeqParamSet) { 1324 if (gotPps) { 1325 LOGE("SPS must come before PPS"); 1326 return ERROR_MALFORMED; 1327 } 1328 if (!gotSps) { 1329 gotSps = true; 1330 } 1331 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1332 } else if (type == kNalUnitTypePicParamSet) { 1333 if (!gotSps) { 1334 LOGE("SPS must come before PPS"); 1335 return ERROR_MALFORMED; 1336 } 1337 if (!gotPps) { 1338 gotPps = true; 1339 } 1340 nextStartCode = parseParamSet(tmp + 4, bytesLeft - 4, type, ¶mSetLen); 1341 } else { 1342 LOGE("Only SPS and PPS Nal units are expected"); 1343 return ERROR_MALFORMED; 1344 } 1345 1346 if (nextStartCode == NULL) { 1347 return ERROR_MALFORMED; 1348 } 1349 1350 // Move on to find the next parameter set 1351 bytesLeft -= nextStartCode - tmp; 1352 tmp = nextStartCode; 1353 mCodecSpecificDataSize += (2 + paramSetLen); 1354 } 1355 1356 { 1357 // Check on the number of seq parameter sets 1358 size_t nSeqParamSets = mSeqParamSets.size(); 1359 if (nSeqParamSets == 0) { 1360 LOGE("Cound not find sequence parameter set"); 1361 return ERROR_MALFORMED; 1362 } 1363 1364 if (nSeqParamSets > 0x1F) { 1365 LOGE("Too many seq parameter sets (%d) found", nSeqParamSets); 1366 return ERROR_MALFORMED; 1367 } 1368 } 1369 1370 { 1371 // Check on the number of pic parameter sets 1372 size_t nPicParamSets = mPicParamSets.size(); 1373 if (nPicParamSets == 0) { 1374 LOGE("Cound not find picture parameter set"); 1375 return ERROR_MALFORMED; 1376 } 1377 if (nPicParamSets > 0xFF) { 1378 LOGE("Too many pic parameter sets (%d) found", nPicParamSets); 1379 return ERROR_MALFORMED; 1380 } 1381 } 1382 1383 { 1384 // Check on the profiles 1385 // These profiles requires additional parameter set extensions 1386 if (mProfileIdc == 100 || mProfileIdc == 110 || 1387 mProfileIdc == 122 || mProfileIdc == 144) { 1388 LOGE("Sorry, no support for profile_idc: %d!", mProfileIdc); 1389 return BAD_VALUE; 1390 } 1391 } 1392 1393 return OK; 1394} 1395 1396status_t MPEG4Writer::Track::makeAVCCodecSpecificData( 1397 const uint8_t *data, size_t size) { 1398 1399 if (mCodecSpecificData != NULL) { 1400 LOGE("Already have codec specific data"); 1401 return ERROR_MALFORMED; 1402 } 1403 1404 if (size < 4) { 1405 LOGE("Codec specific data length too short: %d", size); 1406 return ERROR_MALFORMED; 1407 } 1408 1409 // Data is in the form of AVCCodecSpecificData 1410 if (memcmp("\x00\x00\x00\x01", data, 4)) { 1411 return copyAVCCodecSpecificData(data, size); 1412 } 1413 1414 if (parseAVCCodecSpecificData(data, size) != OK) { 1415 return ERROR_MALFORMED; 1416 } 1417 1418 // ISO 14496-15: AVC file format 1419 mCodecSpecificDataSize += 7; // 7 more bytes in the header 1420 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1421 uint8_t *header = (uint8_t *)mCodecSpecificData; 1422 header[0] = 1; // version 1423 header[1] = mProfileIdc; // profile indication 1424 header[2] = mProfileCompatible; // profile compatibility 1425 header[3] = mLevelIdc; 1426 1427 // 6-bit '111111' followed by 2-bit to lengthSizeMinuusOne 1428 if (mOwner->useNalLengthFour()) { 1429 header[4] = 0xfc | 3; // length size == 4 bytes 1430 } else { 1431 header[4] = 0xfc | 1; // length size == 2 bytes 1432 } 1433 1434 // 3-bit '111' followed by 5-bit numSequenceParameterSets 1435 int nSequenceParamSets = mSeqParamSets.size(); 1436 header[5] = 0xe0 | nSequenceParamSets; 1437 header += 6; 1438 for (List<AVCParamSet>::iterator it = mSeqParamSets.begin(); 1439 it != mSeqParamSets.end(); ++it) { 1440 // 16-bit sequence parameter set length 1441 uint16_t seqParamSetLength = it->mLength; 1442 header[0] = seqParamSetLength >> 8; 1443 header[1] = seqParamSetLength & 0xff; 1444 1445 // SPS NAL unit (sequence parameter length bytes) 1446 memcpy(&header[2], it->mData, seqParamSetLength); 1447 header += (2 + seqParamSetLength); 1448 } 1449 1450 // 8-bit nPictureParameterSets 1451 int nPictureParamSets = mPicParamSets.size(); 1452 header[0] = nPictureParamSets; 1453 header += 1; 1454 for (List<AVCParamSet>::iterator it = mPicParamSets.begin(); 1455 it != mPicParamSets.end(); ++it) { 1456 // 16-bit picture parameter set length 1457 uint16_t picParamSetLength = it->mLength; 1458 header[0] = picParamSetLength >> 8; 1459 header[1] = picParamSetLength & 0xff; 1460 1461 // PPS Nal unit (picture parameter set length bytes) 1462 memcpy(&header[2], it->mData, picParamSetLength); 1463 header += (2 + picParamSetLength); 1464 } 1465 1466 return OK; 1467} 1468 1469/* 1470* The video track's media time adjustment for real-time applications 1471* is described as follows: 1472* 1473* First, the media time adjustment is done for every period of 1474* kVideoMediaTimeAdjustPeriodTimeUs. kVideoMediaTimeAdjustPeriodTimeUs 1475* is currently a fixed value chosen heuristically. The value of 1476* kVideoMediaTimeAdjustPeriodTimeUs should not be very large or very small 1477* for two considerations: on one hand, a relatively large value 1478* helps reduce large fluctuation of drift time in the audio encoding 1479* path; while on the other hand, a relatively small value helps keep 1480* restoring synchronization in audio/video more frequently. Note for the 1481* very first period of kVideoMediaTimeAdjustPeriodTimeUs, there is 1482* no media time adjustment for the video track. 1483* 1484* Second, the total accumulated audio track time drift found 1485* in a period of kVideoMediaTimeAdjustPeriodTimeUs is distributed 1486* over a stream of incoming video frames. The number of video frames 1487* affected is determined based on the number of recorded video frames 1488* within the past kVideoMediaTimeAdjustPeriodTimeUs period. 1489* We choose to distribute the drift time over only a portion 1490* (rather than all) of the total number of recorded video frames 1491* in order to make sure that the video track media time adjustment is 1492* completed for the current period before the next video track media 1493* time adjustment period starts. Currently, the portion chosen is a 1494* half (0.5). 1495* 1496* Last, various additional checks are performed to ensure that 1497* the actual audio encoding path does not have too much drift. 1498* In particular, 1) we want to limit the average incremental time 1499* adjustment for each video frame to be less than a threshold 1500* for a single period of kVideoMediaTimeAdjustPeriodTimeUs. 1501* Currently, the threshold is set to 5 ms. If the average incremental 1502* media time adjustment for a video frame is larger than the 1503* threshold, the audio encoding path has too much time drift. 1504* 2) We also want to limit the total time drift in the audio 1505* encoding path to be less than a threshold for a period of 1506* kVideoMediaTimeAdjustPeriodTimeUs. Currently, the threshold 1507* is 0.5% of kVideoMediaTimeAdjustPeriodTimeUs. If the time drift of 1508* the audio encoding path is larger than the threshold, the audio 1509* encoding path has too much time drift. We treat the large time 1510* drift of the audio encoding path as errors, since there is no 1511* way to keep audio/video in synchronization for real-time 1512* applications if the time drift is too large unless we drop some 1513* video frames, which has its own problems that we don't want 1514* to get into for the time being. 1515*/ 1516void MPEG4Writer::Track::adjustMediaTime(int64_t *timestampUs) { 1517 if (*timestampUs - mPrevMediaTimeAdjustTimestampUs >= 1518 kVideoMediaTimeAdjustPeriodTimeUs) { 1519 1520 LOGV("New media time adjustment period at %lld us", *timestampUs); 1521 mIsMediaTimeAdjustmentOn = true; 1522 mMediaTimeAdjustNumFrames = 1523 (mNumSamples - mPrevMediaTimeAdjustSample) >> 1; 1524 1525 mPrevMediaTimeAdjustTimestampUs = *timestampUs; 1526 mPrevMediaTimeAdjustSample = mNumSamples; 1527 int64_t totalAccumDriftTimeUs = mOwner->getDriftTimeUs(); 1528 mTotalDriftTimeToAdjustUs = 1529 totalAccumDriftTimeUs - mPrevTotalAccumDriftTimeUs; 1530 1531 mPrevTotalAccumDriftTimeUs = totalAccumDriftTimeUs; 1532 1533 // Check on incremental adjusted time per frame 1534 int64_t adjustTimePerFrameUs = 1535 mTotalDriftTimeToAdjustUs / mMediaTimeAdjustNumFrames; 1536 1537 if (adjustTimePerFrameUs < 0) { 1538 adjustTimePerFrameUs = -adjustTimePerFrameUs; 1539 } 1540 if (adjustTimePerFrameUs >= 5000) { 1541 LOGE("Adjusted time per video frame is %lld us", 1542 adjustTimePerFrameUs); 1543 CHECK(!"Video frame time adjustment is too large!"); 1544 } 1545 1546 // Check on total accumulated time drift within a period of 1547 // kVideoMediaTimeAdjustPeriodTimeUs. 1548 int64_t driftPercentage = (mTotalDriftTimeToAdjustUs * 1000) 1549 / kVideoMediaTimeAdjustPeriodTimeUs; 1550 1551 if (driftPercentage < 0) { 1552 driftPercentage = -driftPercentage; 1553 } 1554 if (driftPercentage > 5) { 1555 LOGE("Audio track has time drift %lld us over %lld us", 1556 mTotalDriftTimeToAdjustUs, 1557 kVideoMediaTimeAdjustPeriodTimeUs); 1558 1559 CHECK(!"The audio track media time drifts too much!"); 1560 } 1561 1562 } 1563 1564 if (mIsMediaTimeAdjustmentOn) { 1565 if (mNumSamples - mPrevMediaTimeAdjustSample <= 1566 mMediaTimeAdjustNumFrames) { 1567 1568 // Do media time incremental adjustment 1569 int64_t incrementalAdjustTimeUs = 1570 (mTotalDriftTimeToAdjustUs * 1571 (mNumSamples - mPrevMediaTimeAdjustSample)) 1572 / mMediaTimeAdjustNumFrames; 1573 1574 *timestampUs += 1575 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs); 1576 1577 LOGV("Incremental video frame media time adjustment: %lld us", 1578 (incrementalAdjustTimeUs + mPrevTotalAccumDriftTimeUs)); 1579 } else { 1580 // Within the remaining adjustment period, 1581 // no incremental adjustment is needed. 1582 *timestampUs += 1583 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs); 1584 1585 LOGV("Fixed video frame media time adjustment: %lld us", 1586 (mTotalDriftTimeToAdjustUs + mPrevTotalAccumDriftTimeUs)); 1587 } 1588 } 1589} 1590 1591/* 1592 * Updates the drift time from the audio track so that 1593 * the video track can get the updated drift time information 1594 * from the file writer. The fluctuation of the drift time of the audio 1595 * encoding path is smoothed out with a simple filter by giving a larger 1596 * weight to more recently drift time. The filter coefficients, 0.5 and 0.5, 1597 * are heuristically determined. 1598 */ 1599void MPEG4Writer::Track::updateDriftTime(const sp<MetaData>& meta) { 1600 int64_t driftTimeUs = 0; 1601 if (meta->findInt64(kKeyDriftTime, &driftTimeUs)) { 1602 int64_t prevDriftTimeUs = mOwner->getDriftTimeUs(); 1603 int64_t timeUs = (driftTimeUs + prevDriftTimeUs) >> 1; 1604 mOwner->setDriftTimeUs(timeUs); 1605 } 1606} 1607 1608status_t MPEG4Writer::Track::threadEntry() { 1609 int32_t count = 0; 1610 const int64_t interleaveDurationUs = mOwner->interleaveDuration(); 1611 int64_t chunkTimestampUs = 0; 1612 int32_t nChunks = 0; 1613 int32_t nZeroLengthFrames = 0; 1614 int64_t lastTimestampUs = 0; // Previous sample time stamp in ms 1615 int64_t lastDurationUs = 0; // Between the previous two samples in ms 1616 int64_t currDurationTicks = 0; // Timescale based ticks 1617 int64_t lastDurationTicks = 0; // Timescale based ticks 1618 int32_t sampleCount = 1; // Sample count in the current stts table entry 1619 uint32_t previousSampleSize = 0; // Size of the previous sample 1620 int64_t previousPausedDurationUs = 0; 1621 int64_t timestampUs; 1622 1623 sp<MetaData> meta_data; 1624 1625 mNumSamples = 0; 1626 status_t err = OK; 1627 MediaBuffer *buffer; 1628 while (!mDone && (err = mSource->read(&buffer)) == OK) { 1629 if (buffer->range_length() == 0) { 1630 buffer->release(); 1631 buffer = NULL; 1632 ++nZeroLengthFrames; 1633 continue; 1634 } 1635 1636 // If the codec specific data has not been received yet, delay pause. 1637 // After the codec specific data is received, discard what we received 1638 // when the track is to be paused. 1639 if (mPaused && !mResumed) { 1640 buffer->release(); 1641 buffer = NULL; 1642 continue; 1643 } 1644 1645 ++count; 1646 1647 int32_t isCodecConfig; 1648 if (buffer->meta_data()->findInt32(kKeyIsCodecConfig, &isCodecConfig) 1649 && isCodecConfig) { 1650 CHECK(!mGotAllCodecSpecificData); 1651 1652 if (mIsAvc) { 1653 status_t err = makeAVCCodecSpecificData( 1654 (const uint8_t *)buffer->data() 1655 + buffer->range_offset(), 1656 buffer->range_length()); 1657 CHECK_EQ(OK, err); 1658 } else if (mIsMPEG4) { 1659 mCodecSpecificDataSize = buffer->range_length(); 1660 mCodecSpecificData = malloc(mCodecSpecificDataSize); 1661 memcpy(mCodecSpecificData, 1662 (const uint8_t *)buffer->data() 1663 + buffer->range_offset(), 1664 buffer->range_length()); 1665 } 1666 1667 buffer->release(); 1668 buffer = NULL; 1669 1670 mGotAllCodecSpecificData = true; 1671 continue; 1672 } 1673 1674 // Make a deep copy of the MediaBuffer and Metadata and release 1675 // the original as soon as we can 1676 MediaBuffer *copy = new MediaBuffer(buffer->range_length()); 1677 memcpy(copy->data(), (uint8_t *)buffer->data() + buffer->range_offset(), 1678 buffer->range_length()); 1679 copy->set_range(0, buffer->range_length()); 1680 meta_data = new MetaData(*buffer->meta_data().get()); 1681 buffer->release(); 1682 buffer = NULL; 1683 1684 if (mIsAvc) StripStartcode(copy); 1685 1686 size_t sampleSize = copy->range_length(); 1687 if (mIsAvc) { 1688 if (mOwner->useNalLengthFour()) { 1689 sampleSize += 4; 1690 } else { 1691 sampleSize += 2; 1692 } 1693 } 1694 1695 // Max file size or duration handling 1696 mMdatSizeBytes += sampleSize; 1697 updateTrackSizeEstimate(); 1698 1699 if (mOwner->exceedsFileSizeLimit()) { 1700 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_FILESIZE_REACHED, 0); 1701 break; 1702 } 1703 if (mOwner->exceedsFileDurationLimit()) { 1704 mOwner->notify(MEDIA_RECORDER_EVENT_INFO, MEDIA_RECORDER_INFO_MAX_DURATION_REACHED, 0); 1705 break; 1706 } 1707 1708 1709 int32_t isSync = false; 1710 meta_data->findInt32(kKeyIsSyncFrame, &isSync); 1711 1712 /* 1713 * The original timestamp found in the data buffer will be modified as below: 1714 * 1715 * There is a playback offset into this track if the track's start time 1716 * is not the same as the movie start time, which will be recorded in edst 1717 * box of the output file. The playback offset is to make sure that the 1718 * starting time of the audio/video tracks are synchronized. Although the 1719 * track's media timestamp may be subject to various modifications 1720 * as outlined below, the track's playback offset time remains unchanged 1721 * once the first data buffer of the track is received. 1722 * 1723 * The media time stamp will be calculated by subtracting the playback offset 1724 * (and potential pause durations) from the original timestamp in the buffer. 1725 * 1726 * If this track is a video track for a real-time recording application with 1727 * both audio and video tracks, its media timestamp will subject to further 1728 * modification based on the media clock of the audio track. This modification 1729 * is needed for the purpose of maintaining good audio/video synchronization. 1730 * 1731 * If the recording session is paused and resumed multiple times, the track 1732 * media timestamp will be modified as if the recording session had never been 1733 * paused at all during playback of the recorded output file. In other words, 1734 * the output file will have no memory of pause/resume durations. 1735 * 1736 */ 1737 CHECK(meta_data->findInt64(kKeyTime, ×tampUs)); 1738 LOGV("%s timestampUs: %lld", mIsAudio? "Audio": "Video", timestampUs); 1739 1740//////////////////////////////////////////////////////////////////////////////// 1741 if (mSampleSizes.empty()) { 1742 mStartTimestampUs = timestampUs; 1743 mOwner->setStartTimestampUs(mStartTimestampUs); 1744 previousPausedDurationUs = mStartTimestampUs; 1745 } 1746 1747 if (mResumed) { 1748 int64_t durExcludingEarlierPausesUs = timestampUs - previousPausedDurationUs; 1749 CHECK(durExcludingEarlierPausesUs >= 0); 1750 int64_t pausedDurationUs = durExcludingEarlierPausesUs - mTrackDurationUs; 1751 CHECK(pausedDurationUs >= lastDurationUs); 1752 previousPausedDurationUs += pausedDurationUs - lastDurationUs; 1753 mResumed = false; 1754 } 1755 1756 timestampUs -= previousPausedDurationUs; 1757 CHECK(timestampUs >= 0); 1758 1759 // Media time adjustment for real-time applications 1760 if (mIsRealTimeRecording) { 1761 if (mIsAudio) { 1762 updateDriftTime(meta_data); 1763 } else { 1764 adjustMediaTime(×tampUs); 1765 } 1766 } 1767 1768 CHECK(timestampUs >= 0); 1769 if (mNumSamples > 1) { 1770 if (timestampUs <= lastTimestampUs) { 1771 LOGW("Frame arrives too late!"); 1772 // Don't drop the late frame, since dropping a frame may cause 1773 // problems later during playback 1774 1775 // The idea here is to avoid having two or more samples with the 1776 // same timestamp in the output file. 1777 if (mTimeScale >= 1000000LL) { 1778 timestampUs = lastTimestampUs + 1; 1779 } else { 1780 timestampUs = lastTimestampUs + (1000000LL + (mTimeScale >> 1)) / mTimeScale; 1781 } 1782 } 1783 } 1784 1785 LOGV("%s media time stamp: %lld and previous paused duration %lld", 1786 mIsAudio? "Audio": "Video", timestampUs, previousPausedDurationUs); 1787 if (timestampUs > mTrackDurationUs) { 1788 mTrackDurationUs = timestampUs; 1789 } 1790 1791 mSampleSizes.push_back(sampleSize); 1792 ++mNumSamples; 1793 if (mNumSamples > 2) { 1794 // We need to use the time scale based ticks, rather than the 1795 // timestamp itself to determine whether we have to use a new 1796 // stts entry, since we may have rounding errors. 1797 // The calculation is intended to reduce the accumulated 1798 // rounding errors. 1799 currDurationTicks = 1800 ((timestampUs * mTimeScale + 500000LL) / 1000000LL - 1801 (lastTimestampUs * mTimeScale + 500000LL) / 1000000LL); 1802 1803 if (currDurationTicks != lastDurationTicks) { 1804 addOneSttsTableEntry(sampleCount, lastDurationUs); 1805 sampleCount = 1; 1806 } else { 1807 ++sampleCount; 1808 } 1809 } 1810 if (mSamplesHaveSameSize) { 1811 if (mNumSamples >= 2 && previousSampleSize != sampleSize) { 1812 mSamplesHaveSameSize = false; 1813 } 1814 previousSampleSize = sampleSize; 1815 } 1816 lastDurationUs = timestampUs - lastTimestampUs; 1817 lastDurationTicks = currDurationTicks; 1818 lastTimestampUs = timestampUs; 1819 1820 if (isSync != 0) { 1821 addOneStssTableEntry(mNumSamples); 1822 } 1823 1824 if (mTrackingProgressStatus) { 1825 if (mPreviousTrackTimeUs <= 0) { 1826 mPreviousTrackTimeUs = mStartTimestampUs; 1827 } 1828 trackProgressStatus(timestampUs); 1829 } 1830 if (mOwner->numTracks() == 1) { 1831 off_t offset = mIsAvc? mOwner->addLengthPrefixedSample_l(copy) 1832 : mOwner->addSample_l(copy); 1833 if (mChunkOffsets.empty()) { 1834 addChunkOffset(offset); 1835 } 1836 copy->release(); 1837 copy = NULL; 1838 continue; 1839 } 1840 1841 mChunkSamples.push_back(copy); 1842 if (interleaveDurationUs == 0) { 1843 addOneStscTableEntry(++nChunks, 1); 1844 bufferChunk(timestampUs); 1845 } else { 1846 if (chunkTimestampUs == 0) { 1847 chunkTimestampUs = timestampUs; 1848 } else { 1849 if (timestampUs - chunkTimestampUs > interleaveDurationUs) { 1850 ++nChunks; 1851 if (nChunks == 1 || // First chunk 1852 (--(mStscTableEntries.end()))->samplesPerChunk != 1853 mChunkSamples.size()) { 1854 addOneStscTableEntry(nChunks, mChunkSamples.size()); 1855 } 1856 bufferChunk(timestampUs); 1857 chunkTimestampUs = timestampUs; 1858 } 1859 } 1860 } 1861 1862 } 1863 1864 if (mSampleSizes.empty() || // no samples written 1865 (!mIsAudio && mNumStssTableEntries == 0) || // no sync frames for video 1866 (OK != checkCodecSpecificData())) { // no codec specific data 1867 err = ERROR_MALFORMED; 1868 } 1869 mOwner->trackProgressStatus(this, -1, err); 1870 1871 // Last chunk 1872 if (mOwner->numTracks() == 1) { 1873 addOneStscTableEntry(1, mNumSamples); 1874 } else if (!mChunkSamples.empty()) { 1875 addOneStscTableEntry(++nChunks, mChunkSamples.size()); 1876 bufferChunk(timestampUs); 1877 } 1878 1879 // We don't really know how long the last frame lasts, since 1880 // there is no frame time after it, just repeat the previous 1881 // frame's duration. 1882 if (mNumSamples == 1) { 1883 lastDurationUs = 0; // A single sample's duration 1884 } else { 1885 ++sampleCount; // Count for the last sample 1886 } 1887 addOneSttsTableEntry(sampleCount, lastDurationUs); 1888 mTrackDurationUs += lastDurationUs; 1889 mReachedEOS = true; 1890 LOGI("Received total/0-length (%d/%d) buffers and encoded %d frames. - %s", 1891 count, nZeroLengthFrames, mNumSamples, mIsAudio? "audio": "video"); 1892 if (mIsAudio) { 1893 LOGI("Audio track drift time: %lld us", mOwner->getDriftTimeUs()); 1894 } 1895 1896 if (err == ERROR_END_OF_STREAM) { 1897 return OK; 1898 } 1899 return err; 1900} 1901 1902void MPEG4Writer::Track::trackProgressStatus(int64_t timeUs, status_t err) { 1903 LOGV("trackProgressStatus: %lld us", timeUs); 1904 if (mTrackEveryTimeDurationUs > 0 && 1905 timeUs - mPreviousTrackTimeUs >= mTrackEveryTimeDurationUs) { 1906 LOGV("Fire time tracking progress status at %lld us", timeUs); 1907 mOwner->trackProgressStatus(this, timeUs - mPreviousTrackTimeUs, err); 1908 mPreviousTrackTimeUs = timeUs; 1909 } 1910} 1911 1912void MPEG4Writer::trackProgressStatus( 1913 const MPEG4Writer::Track* track, int64_t timeUs, status_t err) { 1914 Mutex::Autolock lock(mLock); 1915 int32_t nTracks = mTracks.size(); 1916 CHECK(nTracks >= 1); 1917 CHECK(nTracks < 64); // Arbitrary number 1918 1919 int32_t trackNum = 0; 1920 CHECK(trackNum < nTracks); 1921 trackNum <<= 16; 1922 1923 // Error notification 1924 // Do not consider ERROR_END_OF_STREAM an error 1925 if (err != OK && err != ERROR_END_OF_STREAM) { 1926 notify(MEDIA_RECORDER_EVENT_ERROR, 1927 trackNum | MEDIA_RECORDER_ERROR_UNKNOWN, 1928 err); 1929 return; 1930 } 1931 1932 if (timeUs == -1) { 1933 // Send completion notification 1934 notify(MEDIA_RECORDER_EVENT_INFO, 1935 trackNum | MEDIA_RECORDER_INFO_COMPLETION_STATUS, 1936 err); 1937 return; 1938 } else { 1939 // Send progress status 1940 notify(MEDIA_RECORDER_EVENT_INFO, 1941 trackNum | MEDIA_RECORDER_INFO_PROGRESS_TIME_STATUS, 1942 timeUs / 1000); 1943 } 1944} 1945 1946void MPEG4Writer::setDriftTimeUs(int64_t driftTimeUs) { 1947 LOGV("setDriftTimeUs: %lld us", driftTimeUs); 1948 Mutex::Autolock autolock(mLock); 1949 mDriftTimeUs = driftTimeUs; 1950} 1951 1952int64_t MPEG4Writer::getDriftTimeUs() { 1953 LOGV("getDriftTimeUs: %lld us", mDriftTimeUs); 1954 Mutex::Autolock autolock(mLock); 1955 return mDriftTimeUs; 1956} 1957 1958bool MPEG4Writer::useNalLengthFour() { 1959 return mUse4ByteNalLength; 1960} 1961 1962void MPEG4Writer::Track::bufferChunk(int64_t timestampUs) { 1963 LOGV("bufferChunk"); 1964 1965 Chunk chunk(this, timestampUs, mChunkSamples); 1966 mOwner->bufferChunk(chunk); 1967 mChunkSamples.clear(); 1968} 1969 1970int64_t MPEG4Writer::Track::getDurationUs() const { 1971 return mTrackDurationUs; 1972} 1973 1974int64_t MPEG4Writer::Track::getEstimatedTrackSizeBytes() const { 1975 return mEstimatedTrackSizeBytes; 1976} 1977 1978status_t MPEG4Writer::Track::checkCodecSpecificData() const { 1979 const char *mime; 1980 CHECK(mMeta->findCString(kKeyMIMEType, &mime)); 1981 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime) || 1982 !strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime) || 1983 !strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 1984 if (!mCodecSpecificData || 1985 mCodecSpecificDataSize <= 0) { 1986 LOGE("Missing codec specific data"); 1987 return ERROR_MALFORMED; 1988 } 1989 } else { 1990 if (mCodecSpecificData || 1991 mCodecSpecificDataSize > 0) { 1992 LOGE("Unexepected codec specific data found"); 1993 return ERROR_MALFORMED; 1994 } 1995 } 1996 return OK; 1997} 1998 1999void MPEG4Writer::Track::writeTrackHeader( 2000 int32_t trackID, bool use32BitOffset) { 2001 const char *mime; 2002 bool success = mMeta->findCString(kKeyMIMEType, &mime); 2003 CHECK(success); 2004 2005 LOGV("%s track time scale: %d", 2006 mIsAudio? "Audio": "Video", mTimeScale); 2007 2008 time_t now = time(NULL); 2009 int32_t mvhdTimeScale = mOwner->getTimeScale(); 2010 int64_t trakDurationUs = getDurationUs(); 2011 2012 mOwner->beginBox("trak"); 2013 2014 mOwner->beginBox("tkhd"); 2015 // Flags = 7 to indicate that the track is enabled, and 2016 // part of the presentation 2017 mOwner->writeInt32(0x07); // version=0, flags=7 2018 mOwner->writeInt32(now); // creation time 2019 mOwner->writeInt32(now); // modification time 2020 mOwner->writeInt32(trackID); 2021 mOwner->writeInt32(0); // reserved 2022 int32_t tkhdDuration = 2023 (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2024 mOwner->writeInt32(tkhdDuration); // in mvhd timescale 2025 mOwner->writeInt32(0); // reserved 2026 mOwner->writeInt32(0); // reserved 2027 mOwner->writeInt16(0); // layer 2028 mOwner->writeInt16(0); // alternate group 2029 mOwner->writeInt16(mIsAudio ? 0x100 : 0); // volume 2030 mOwner->writeInt16(0); // reserved 2031 2032 mOwner->writeInt32(0x10000); // matrix 2033 mOwner->writeInt32(0); 2034 mOwner->writeInt32(0); 2035 mOwner->writeInt32(0); 2036 mOwner->writeInt32(0x10000); 2037 mOwner->writeInt32(0); 2038 mOwner->writeInt32(0); 2039 mOwner->writeInt32(0); 2040 mOwner->writeInt32(0x40000000); 2041 2042 if (mIsAudio) { 2043 mOwner->writeInt32(0); 2044 mOwner->writeInt32(0); 2045 } else { 2046 int32_t width, height; 2047 bool success = mMeta->findInt32(kKeyWidth, &width); 2048 success = success && mMeta->findInt32(kKeyHeight, &height); 2049 CHECK(success); 2050 2051 mOwner->writeInt32(width << 16); // 32-bit fixed-point value 2052 mOwner->writeInt32(height << 16); // 32-bit fixed-point value 2053 } 2054 mOwner->endBox(); // tkhd 2055 2056 int64_t moovStartTimeUs = mOwner->getStartTimestampUs(); 2057 if (mStartTimestampUs != moovStartTimeUs) { 2058 mOwner->beginBox("edts"); 2059 mOwner->beginBox("elst"); 2060 mOwner->writeInt32(0); // version=0, flags=0: 32-bit time 2061 mOwner->writeInt32(2); // never ends with an empty list 2062 2063 // First elst entry: specify the starting time offset 2064 int64_t offsetUs = mStartTimestampUs - moovStartTimeUs; 2065 LOGV("OffsetUs: %lld", offsetUs); 2066 int32_t seg = (offsetUs * mvhdTimeScale + 5E5) / 1E6; 2067 mOwner->writeInt32(seg); // in mvhd timecale 2068 mOwner->writeInt32(-1); // starting time offset 2069 mOwner->writeInt32(1 << 16); // rate = 1.0 2070 2071 // Second elst entry: specify the track duration 2072 seg = (trakDurationUs * mvhdTimeScale + 5E5) / 1E6; 2073 mOwner->writeInt32(seg); // in mvhd timescale 2074 mOwner->writeInt32(0); 2075 mOwner->writeInt32(1 << 16); 2076 mOwner->endBox(); 2077 mOwner->endBox(); 2078 } 2079 2080 mOwner->beginBox("mdia"); 2081 2082 mOwner->beginBox("mdhd"); 2083 mOwner->writeInt32(0); // version=0, flags=0 2084 mOwner->writeInt32(now); // creation time 2085 mOwner->writeInt32(now); // modification time 2086 mOwner->writeInt32(mTimeScale); // media timescale 2087 int32_t mdhdDuration = (trakDurationUs * mTimeScale + 5E5) / 1E6; 2088 mOwner->writeInt32(mdhdDuration); // use media timescale 2089 // Language follows the three letter standard ISO-639-2/T 2090 // 'e', 'n', 'g' for "English", for instance. 2091 // Each character is packed as the difference between its ASCII value and 0x60. 2092 // For "English", these are 00101, 01110, 00111. 2093 // XXX: Where is the padding bit located: 0x15C7? 2094 mOwner->writeInt16(0); // language code 2095 mOwner->writeInt16(0); // predefined 2096 mOwner->endBox(); 2097 2098 mOwner->beginBox("hdlr"); 2099 mOwner->writeInt32(0); // version=0, flags=0 2100 mOwner->writeInt32(0); // component type: should be mhlr 2101 mOwner->writeFourcc(mIsAudio ? "soun" : "vide"); // component subtype 2102 mOwner->writeInt32(0); // reserved 2103 mOwner->writeInt32(0); // reserved 2104 mOwner->writeInt32(0); // reserved 2105 // Removing "r" for the name string just makes the string 4 byte aligned 2106 mOwner->writeCString(mIsAudio ? "SoundHandle": "VideoHandle"); // name 2107 mOwner->endBox(); 2108 2109 mOwner->beginBox("minf"); 2110 if (mIsAudio) { 2111 mOwner->beginBox("smhd"); 2112 mOwner->writeInt32(0); // version=0, flags=0 2113 mOwner->writeInt16(0); // balance 2114 mOwner->writeInt16(0); // reserved 2115 mOwner->endBox(); 2116 } else { 2117 mOwner->beginBox("vmhd"); 2118 mOwner->writeInt32(0x01); // version=0, flags=1 2119 mOwner->writeInt16(0); // graphics mode 2120 mOwner->writeInt16(0); // opcolor 2121 mOwner->writeInt16(0); 2122 mOwner->writeInt16(0); 2123 mOwner->endBox(); 2124 } 2125 2126 mOwner->beginBox("dinf"); 2127 mOwner->beginBox("dref"); 2128 mOwner->writeInt32(0); // version=0, flags=0 2129 mOwner->writeInt32(1); // entry count (either url or urn) 2130 // The table index here refers to the sample description index 2131 // in the sample table entries. 2132 mOwner->beginBox("url "); 2133 mOwner->writeInt32(1); // version=0, flags=1 (self-contained) 2134 mOwner->endBox(); // url 2135 mOwner->endBox(); // dref 2136 mOwner->endBox(); // dinf 2137 2138 mOwner->beginBox("stbl"); 2139 2140 mOwner->beginBox("stsd"); 2141 mOwner->writeInt32(0); // version=0, flags=0 2142 mOwner->writeInt32(1); // entry count 2143 if (mIsAudio) { 2144 const char *fourcc = NULL; 2145 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime)) { 2146 fourcc = "samr"; 2147 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2148 fourcc = "sawb"; 2149 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2150 fourcc = "mp4a"; 2151 } else { 2152 LOGE("Unknown mime type '%s'.", mime); 2153 CHECK(!"should not be here, unknown mime type."); 2154 } 2155 2156 mOwner->beginBox(fourcc); // audio format 2157 mOwner->writeInt32(0); // reserved 2158 mOwner->writeInt16(0); // reserved 2159 mOwner->writeInt16(0x1); // data ref index 2160 mOwner->writeInt32(0); // reserved 2161 mOwner->writeInt32(0); // reserved 2162 int32_t nChannels; 2163 CHECK_EQ(true, mMeta->findInt32(kKeyChannelCount, &nChannels)); 2164 mOwner->writeInt16(nChannels); // channel count 2165 mOwner->writeInt16(16); // sample size 2166 mOwner->writeInt16(0); // predefined 2167 mOwner->writeInt16(0); // reserved 2168 2169 int32_t samplerate; 2170 bool success = mMeta->findInt32(kKeySampleRate, &samplerate); 2171 CHECK(success); 2172 mOwner->writeInt32(samplerate << 16); 2173 if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AAC, mime)) { 2174 mOwner->beginBox("esds"); 2175 CHECK(mCodecSpecificData); 2176 CHECK(mCodecSpecificDataSize > 0); 2177 2178 mOwner->writeInt32(0); // version=0, flags=0 2179 mOwner->writeInt8(0x03); // ES_DescrTag 2180 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2181 mOwner->writeInt16(0x0000);// ES_ID 2182 mOwner->writeInt8(0x00); 2183 2184 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2185 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2186 mOwner->writeInt8(0x40); // objectTypeIndication ISO/IEC 14492-2 2187 mOwner->writeInt8(0x15); // streamType AudioStream 2188 2189 mOwner->writeInt16(0x03); // XXX 2190 mOwner->writeInt8(0x00); // buffer size 24-bit 2191 mOwner->writeInt32(96000); // max bit rate 2192 mOwner->writeInt32(96000); // avg bit rate 2193 2194 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2195 mOwner->writeInt8(mCodecSpecificDataSize); 2196 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2197 2198 static const uint8_t kData2[] = { 2199 0x06, // SLConfigDescriptorTag 2200 0x01, 2201 0x02 2202 }; 2203 mOwner->write(kData2, sizeof(kData2)); 2204 2205 mOwner->endBox(); // esds 2206 } else if (!strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_NB, mime) || 2207 !strcasecmp(MEDIA_MIMETYPE_AUDIO_AMR_WB, mime)) { 2208 // 3gpp2 Spec AMRSampleEntry fields 2209 mOwner->beginBox("damr"); 2210 mOwner->writeCString(" "); // vendor: 4 bytes 2211 mOwner->writeInt8(0); // decoder version 2212 mOwner->writeInt16(0x83FF); // mode set: all enabled 2213 mOwner->writeInt8(0); // mode change period 2214 mOwner->writeInt8(1); // frames per sample 2215 mOwner->endBox(); 2216 } 2217 mOwner->endBox(); 2218 } else { 2219 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2220 mOwner->beginBox("mp4v"); 2221 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2222 mOwner->beginBox("s263"); 2223 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2224 mOwner->beginBox("avc1"); 2225 } else { 2226 LOGE("Unknown mime type '%s'.", mime); 2227 CHECK(!"should not be here, unknown mime type."); 2228 } 2229 2230 mOwner->writeInt32(0); // reserved 2231 mOwner->writeInt16(0); // reserved 2232 mOwner->writeInt16(1); // data ref index 2233 mOwner->writeInt16(0); // predefined 2234 mOwner->writeInt16(0); // reserved 2235 mOwner->writeInt32(0); // predefined 2236 mOwner->writeInt32(0); // predefined 2237 mOwner->writeInt32(0); // predefined 2238 2239 int32_t width, height; 2240 bool success = mMeta->findInt32(kKeyWidth, &width); 2241 success = success && mMeta->findInt32(kKeyHeight, &height); 2242 CHECK(success); 2243 2244 mOwner->writeInt16(width); 2245 mOwner->writeInt16(height); 2246 mOwner->writeInt32(0x480000); // horiz resolution 2247 mOwner->writeInt32(0x480000); // vert resolution 2248 mOwner->writeInt32(0); // reserved 2249 mOwner->writeInt16(1); // frame count 2250 mOwner->write(" ", 32); 2251 mOwner->writeInt16(0x18); // depth 2252 mOwner->writeInt16(-1); // predefined 2253 2254 CHECK(23 + mCodecSpecificDataSize < 128); 2255 2256 if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_MPEG4, mime)) { 2257 CHECK(mCodecSpecificData); 2258 CHECK(mCodecSpecificDataSize > 0); 2259 mOwner->beginBox("esds"); 2260 2261 mOwner->writeInt32(0); // version=0, flags=0 2262 2263 mOwner->writeInt8(0x03); // ES_DescrTag 2264 mOwner->writeInt8(23 + mCodecSpecificDataSize); 2265 mOwner->writeInt16(0x0000); // ES_ID 2266 mOwner->writeInt8(0x1f); 2267 2268 mOwner->writeInt8(0x04); // DecoderConfigDescrTag 2269 mOwner->writeInt8(15 + mCodecSpecificDataSize); 2270 mOwner->writeInt8(0x20); // objectTypeIndication ISO/IEC 14492-2 2271 mOwner->writeInt8(0x11); // streamType VisualStream 2272 2273 static const uint8_t kData[] = { 2274 0x01, 0x77, 0x00, 2275 0x00, 0x03, 0xe8, 0x00, 2276 0x00, 0x03, 0xe8, 0x00 2277 }; 2278 mOwner->write(kData, sizeof(kData)); 2279 2280 mOwner->writeInt8(0x05); // DecoderSpecificInfoTag 2281 2282 mOwner->writeInt8(mCodecSpecificDataSize); 2283 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2284 2285 static const uint8_t kData2[] = { 2286 0x06, // SLConfigDescriptorTag 2287 0x01, 2288 0x02 2289 }; 2290 mOwner->write(kData2, sizeof(kData2)); 2291 2292 mOwner->endBox(); // esds 2293 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_H263, mime)) { 2294 mOwner->beginBox("d263"); 2295 2296 mOwner->writeInt32(0); // vendor 2297 mOwner->writeInt8(0); // decoder version 2298 mOwner->writeInt8(10); // level: 10 2299 mOwner->writeInt8(0); // profile: 0 2300 2301 mOwner->endBox(); // d263 2302 } else if (!strcasecmp(MEDIA_MIMETYPE_VIDEO_AVC, mime)) { 2303 CHECK(mCodecSpecificData); 2304 CHECK(mCodecSpecificDataSize >= 5); 2305 2306 // Patch avcc's lengthSize field to match the number 2307 // of bytes we use to indicate the size of a nal unit. 2308 uint8_t *ptr = (uint8_t *)mCodecSpecificData; 2309 ptr[4] = 2310 (ptr[4] & 0xfc) 2311 | (mOwner->useNalLengthFour() ? 3 : 1); 2312 2313 mOwner->beginBox("avcC"); 2314 mOwner->write(mCodecSpecificData, mCodecSpecificDataSize); 2315 mOwner->endBox(); // avcC 2316 } 2317 2318 mOwner->beginBox("pasp"); 2319 // This is useful if the pixel is not square 2320 mOwner->writeInt32(1 << 16); // hspacing 2321 mOwner->writeInt32(1 << 16); // vspacing 2322 mOwner->endBox(); // pasp 2323 mOwner->endBox(); // mp4v, s263 or avc1 2324 } 2325 mOwner->endBox(); // stsd 2326 2327 mOwner->beginBox("stts"); 2328 mOwner->writeInt32(0); // version=0, flags=0 2329 mOwner->writeInt32(mNumSttsTableEntries); 2330 int64_t prevTimestampUs = 0; 2331 for (List<SttsTableEntry>::iterator it = mSttsTableEntries.begin(); 2332 it != mSttsTableEntries.end(); ++it) { 2333 mOwner->writeInt32(it->sampleCount); 2334 2335 // Make sure that we are calculating the sample duration the exactly 2336 // same way as we made decision on how to create stts entries. 2337 int64_t currTimestampUs = prevTimestampUs + it->sampleDurationUs; 2338 int32_t dur = ((currTimestampUs * mTimeScale + 500000LL) / 1000000LL - 2339 (prevTimestampUs * mTimeScale + 500000LL) / 1000000LL); 2340 prevTimestampUs += (it->sampleCount * it->sampleDurationUs); 2341 2342 mOwner->writeInt32(dur); 2343 } 2344 mOwner->endBox(); // stts 2345 2346 if (!mIsAudio) { 2347 mOwner->beginBox("stss"); 2348 mOwner->writeInt32(0); // version=0, flags=0 2349 mOwner->writeInt32(mNumStssTableEntries); // number of sync frames 2350 for (List<int32_t>::iterator it = mStssTableEntries.begin(); 2351 it != mStssTableEntries.end(); ++it) { 2352 mOwner->writeInt32(*it); 2353 } 2354 mOwner->endBox(); // stss 2355 } 2356 2357 mOwner->beginBox("stsz"); 2358 mOwner->writeInt32(0); // version=0, flags=0 2359 if (mSamplesHaveSameSize) { 2360 List<size_t>::iterator it = mSampleSizes.begin(); 2361 mOwner->writeInt32(*it); // default sample size 2362 } else { 2363 mOwner->writeInt32(0); 2364 } 2365 mOwner->writeInt32(mNumSamples); 2366 if (!mSamplesHaveSameSize) { 2367 for (List<size_t>::iterator it = mSampleSizes.begin(); 2368 it != mSampleSizes.end(); ++it) { 2369 mOwner->writeInt32(*it); 2370 } 2371 } 2372 mOwner->endBox(); // stsz 2373 2374 mOwner->beginBox("stsc"); 2375 mOwner->writeInt32(0); // version=0, flags=0 2376 mOwner->writeInt32(mNumStscTableEntries); 2377 for (List<StscTableEntry>::iterator it = mStscTableEntries.begin(); 2378 it != mStscTableEntries.end(); ++it) { 2379 mOwner->writeInt32(it->firstChunk); 2380 mOwner->writeInt32(it->samplesPerChunk); 2381 mOwner->writeInt32(it->sampleDescriptionId); 2382 } 2383 mOwner->endBox(); // stsc 2384 mOwner->beginBox(use32BitOffset? "stco": "co64"); 2385 mOwner->writeInt32(0); // version=0, flags=0 2386 mOwner->writeInt32(mNumStcoTableEntries); 2387 for (List<off_t>::iterator it = mChunkOffsets.begin(); 2388 it != mChunkOffsets.end(); ++it) { 2389 if (use32BitOffset) { 2390 mOwner->writeInt32(static_cast<int32_t>(*it)); 2391 } else { 2392 mOwner->writeInt64((*it)); 2393 } 2394 } 2395 mOwner->endBox(); // stco or co64 2396 2397 mOwner->endBox(); // stbl 2398 mOwner->endBox(); // minf 2399 mOwner->endBox(); // mdia 2400 mOwner->endBox(); // trak 2401} 2402 2403} // namespace android 2404