Android多媒体框架（3）—— libstagefright中MediaCodec源码分析
libstagefright中MediaCodec源码分析
和前两篇一样，我们按照MediaCodec的各个状态来分析libstagefright中MediaCodec的源代码。
- configure
首先我们看一下configure在libstagefright中MediaCodec中的定义：
438status_t MediaCodec::configure(
439 const sp &format,
440 const sp &surface,
441 const sp &crypto,
442 uint32_t flags) {
443 sp msg = new AMessage(kWhatConfigure, this);
444
445 if (mIsVideo) {
446 format->findInt32("width", &mVideoWidth);
447 format->findInt32("height", &mVideoHeight);
448 if (!format->findInt32("rotation-degrees", &mRotationDegrees)) {
449 mRotationDegrees = 0;
450 }
451
452 // Prevent possible integer overflow in downstream code.
453 if (mInitIsEncoder
454 && (uint64_t)mVideoWidth * mVideoHeight > (uint64_t)INT32_MAX / 4) {
455 ALOGE("buffer size is too big, width=%d, height=%d", mVideoWidth, mVideoHeight);
456 return BAD_VALUE;
457 }
458 }
459
460 msg->setMessage("format", format);
461 msg->setInt32("flags", flags);
462 msg->setObject("surface", surface);
463
464 if (crypto != NULL) {
465 msg->setPointer("crypto", crypto.get());
466 }
467
468 // save msg for reset
469 mConfigureMsg = msg;
470
471 status_t err;
472 Vector resources;
473 MediaResource::Type type = (mFlags & kFlagIsSecure) ?
474 MediaResource::kSecureCodec : MediaResource::kNonSecureCodec;
475 MediaResource::SubType subtype =
476 mIsVideo ? MediaResource::kVideoCodec : MediaResource::kAudioCodec;
477 resources.push_back(MediaResource(type, subtype, 1));
478 // Don't know the buffer size at this point, but it's fine to use 1 because
479 // the reclaimResource call doesn't consider the requester's buffer size for now.
480 resources.push_back(MediaResource(MediaResource::kGraphicMemory, 1));
481 for (int i = 0; i <= kMaxRetry; ++i) {
482 if (i > 0) {
483 // Don't try to reclaim resource for the first time.
484 if (!mResourceManagerService->reclaimResource(resources)) {
485 break;
486 }
487 }
488
489 sp response;
490 err = PostAndAwaitResponse(msg, &response);
491 if (err != OK && err != INVALID_OPERATION) {
492 // MediaCodec now set state to UNINITIALIZED upon any fatal error.
493 // To maintain backward-compatibility, do a reset() to put codec
494 // back into INITIALIZED state.
495 // But don't reset if the err is INVALID_OPERATION, which means
496 // the configure failure is due to wrong state.
497
498 ALOGE("configure failed with err 0x%08x, resetting...", err);
499 reset();
500 }
501 if (!isResourceError(err)) {
502 break;
503 }
504 }
505 return err;
506}
首先，生成一个AMessage的心的强引用计数msg，AMessage的构造函数的两个参数，一个是枚举值kWhatConfigure（“init”）,在MediaCodec.h中定义，另一个参数应该是AHandler类型，而MediaCodec就是AHandler的子类。现在来看一下是否是video（mIsVideo)，如果是，要从参数format中取出视频的width和height，分别把值赋给类属性mVideoWidth和mVideoHeight；获取旋转角度，赋值给mRotationsDegrees，如果不存在该属性，就把mRotationsDegrees设成0度。如果encoder，那么宽高相乘要小于INT32_MAX的四分之一。一次设置msg的format，flags和surface。如果参数crypto不为NULL，设置msg的crypto。把msg赋值给类属性mConfigrueMsg。判断是否安全编码，把结果保存在type中；判断是video还是audio，结果保存在subtype中。生成两种MediaResource，根据type，subtype，另一种根据MediaResource::kGraphicMemory，把他们都放入之前生命的MediaResource类型的Vector resources中。进行一个循环，除了第一个循环，每次循环开始，都需要reclaim resource。调用PostAndAwaitResponse(msg, &response)（这会调用AMessage类中的同名方法，最终通过ALooper发送消息），发送msg，并且把响应存储到response，把返回值存储到err，如果err不等于NO_MEMORY，跳出循环。最后，返回err的值。
- start
下面，我们来看看start状态。
status_t MediaCodec::start() {
568 sp msg = new AMessage(kWhatStart, this);
569
570 status_t err;
571 Vector resources;
572 MediaResource::Type type = (mFlags & kFlagIsSecure) ?
573 MediaResource::kSecureCodec : MediaResource::kNonSecureCodec;
574 MediaResource::SubType subtype =
575 mIsVideo ? MediaResource::kVideoCodec : MediaResource::kAudioCodec;
576 resources.push_back(MediaResource(type, subtype, 1));
577 // Don't know the buffer size at this point, but it's fine to use 1 because
578 // the reclaimResource call doesn't consider the requester's buffer size for now.
579 resources.push_back(MediaResource(MediaResource::kGraphicMemory, 1));
580 for (int i = 0; i <= kMaxRetry; ++i) {
581 if (i > 0) {
582 // Don't try to reclaim resource for the first time.
583 if (!mResourceManagerService->reclaimResource(resources)) {
584 break;
585 }
586 // Recover codec from previous error before retry start.
587 err = reset();
588 if (err != OK) {
589 ALOGE("retrying start: failed to reset codec");
590 break;
591 }
592 sp response;
593 err = PostAndAwaitResponse(mConfigureMsg, &response);
594 if (err != OK) {
595 ALOGE("retrying start: failed to configure codec");
596 break;
597 }
598 }
599
600 sp response;
601 err = PostAndAwaitResponse(msg, &response);
602 if (!isResourceError(err)) {
603 break;
604 }
605 }
606 return err;
607}
首先以kWhatStart状态创建一个AMessage的强引用计数msg。同configure一样，type和subtype一样，分别表示是否安全编码和音频或视频编码。分别以type和subtype创建MediaResource，存入向量resources中。在循环之中，如果不是首次循环，需要对资源进行回收再利用。每次循环都要进行reset。如果reset失败，终止循环。调用PostAndAwaitResponse(mConfigureMsg, &response)发送消息并等待返回，mConfigureMsg在configure阶段已经赋值。如果返回值不等于OK，终止循环。调用PostAndAwaitResponse(msg, &response)发送消息并等待返回。如果返回值不是NO_MEMORY，跳出循环。最后，函数返回最后设置的err的值。
- dequeueInputBuffe
现在来看看输入缓冲区的出列处理。
status_t MediaCodec::dequeueInputBuffer(size_t *index, int64_t timeoutUs) {
748 sp msg = new AMessage(kWhatDequeueInputBuffer, this);
749 msg->setInt64("timeoutUs", timeoutUs);
750
751 sp response;
752 status_t err;
753 if ((err = PostAndAwaitResponse(msg, &response)) != OK) {
754 return err;
755 }
756
757 CHECK(response->findSize("index", index));
758
759 return OK;
760}
同样是，生成AMessage的强引用计数msg和response。调用PostAndAwaitResponse(msg, &response))发送消息并等待返回。如果返回不是OK，就把返回的error code作为函数的返回值。否则，返回OK。
- queueInputBuffer
将数据压入输入缓冲区队列。
status_t MediaCodec::queueInputBuffer(
689 size_t index,
690 size_t offset,
691 size_t size,
692 int64_t presentationTimeUs,
693 uint32_t flags,
694 AString *errorDetailMsg) {
695 if (errorDetailMsg != NULL) {
696 errorDetailMsg->clear();
697 }
698
699 sp msg = new AMessage(kWhatQueueInputBuffer, this);
700 msg->setSize("index", index);
701 msg->setSize("offset", offset);
702 msg->setSize("size", size);
703 msg->setInt64("timeUs", presentationTimeUs);
704 msg->setInt32("flags", flags);
705 msg->setPointer("errorDetailMsg", errorDetailMsg);
706
707 sp response;
708 return PostAndAwaitResponse(msg, &response);
709}
首先，创建AMessage的强引用计数msg，用参数index，offset，size，presentationTimeUs，flags，errorDetailMsg来初始化msg的index，offset，size，timeUs，flags，errorDetailMsg。PostAndAwaitResponse(msg, &response)发送并等待消息返回，并且，用这个调用的返回值作为函数的返回值。
- stop
stop的代码比较简单。
status_t MediaCodec::stop() {
610 sp msg = new AMessage(kWhatStop, this);
611
612 sp response;
613 return PostAndAwaitResponse(msg, &response);
614}
同样是，生成AMessage的强引用计数msg和response。调用PostAndAwaitResponse(msg, &response))发送消息并等待返回。
- reset
reset的代码如下：
status_t MediaCodec::reset() {
654 /* When external-facing MediaCodec object is created,
655 it is already initialized. Thus, reset is essentially
656 release() followed by init(), plus clearing the state */
657
658 status_t err = release();
659
660 // unregister handlers
661 if (mCodec != NULL) {
662 if (mCodecLooper != NULL) {
663 mCodecLooper->unregisterHandler(mCodec->id());
664 } else {
665 mLooper->unregisterHandler(mCodec->id());
666 }
667 mCodec = NULL;
668 }
669 mLooper->unregisterHandler(id());
670
671 mFlags = 0; // clear all flags
672 mStickyError = OK;
673
674 // reset state not reset by setState(UNINITIALIZED)
675 mReplyID = 0;
676 mDequeueInputReplyID = 0;
677 mDequeueOutputReplyID = 0;
678 mDequeueInputTimeoutGeneration = 0;
679 mDequeueOutputTimeoutGeneration = 0;
680 mHaveInputSurface = false;
681
682 if (err == OK) {
683 err = init(mInitName, mInitNameIsType, mInitIsEncoder);
684 }
685 return err;
686}
reset首先调用release释放codec。如果mCodec（CodecBase类型）不为NULL，如果mCodecLooper也不为NULL，首先对mCodecLooper反注册Handler，否则，对mLooper反注册Handler，并且把mCodec置为NULL。接下来进行一系列重新赋值操作，都赋值为初始化值。如果err等于OK，调用init进行初始化。
- release
我们看看reset中调用，并且可以做为一个独立状态的release。
status_t MediaCodec::release() {
647 sp msg = new AMessage(kWhatRelease, this);
648
649 sp response;
650 return PostAndAwaitResponse(msg, &response);
651}
同样是，生成AMessage的强引用计数msg和response。调用PostAndAwaitResponse(msg, &response))发送消息并等待返回。