Android8.0 Audio系统之AudioTrack

继上一篇Android硬件抽象层和HAL层Binder框架，我们这次选择Audio系统来研究，之所以选择Audio系统，并非Audio系统简单，恰恰是因为Audio系统复杂。Audio系统与Media系统，Surface系统，通话系统，蓝牙系统…都有交互，通过Audio系统的复杂度，让我们去领略它的设计思想。我们将从四个层面去剖析Audio系统：1. AudioTrack , 2. AudioFlinger , 3. AudioPolicy, 4.Audio硬件抽象层。

从整个Audio系统来看，要完成音频的路由，合成，解码，AP和BP处理器等功能避免不了复杂度的上升。为此需要寻找一个切入点去分析，一来不脱离主线，二来简化分析流程；本篇我们主要分析AudioTrack。

1. Java层的AudioTrack

以API的使用为切入点分析AudioTrack, 逐步转移到AudioPolicy, AudioFlinger的分析

//1.查询硬件所能分配的最小缓冲区
int bufsize = AudioTrack.getMinBufferSize(16000,
						 AudioFormat.CHANNEL_CONFIGURATION_STEREO,
						 AudioFormat.ENCODING_PCM_16BIT);
//2.创建音轨
AudioTrack trackPlayer = new AudioTrack(AudioManager.STREAM_MUSIC, 16000,
						 AudioFormat.CHANNEL_CONFIGURATION_STEREO,
						 AudioFormat.ENCODING_PCM_16BIT,buffersize,
						 AudioFormat.MODE_STREAM);
//3.播放PCM流
trackPlayer.play();
//4.写入PCM数据
trackplayer.write(bytes, bytres.length);
//5.停止和释放资源
trackplayer.stop();
reackerplayer.release();

1.1 getMinBufferSize()

sampleRateInHz： 采样率； channelConfig : 声道数； audioFormat : 采样精度

static public int getMinBufferSize(int sampleRateInHz, int channelConfig, int audioFormat) {
        int channelCount = 0;
        switch(channelConfig) {
        case AudioFormat.CHANNEL_OUT_MONO:
        case AudioFormat.CHANNEL_CONFIGURATION_MONO:
            channelCount = 1;
            break;
        case AudioFormat.CHANNEL_OUT_STEREO:
        case AudioFormat.CHANNEL_CONFIGURATION_STEREO:
            channelCount = 2;
            break;
        default:
            if (!isMultichannelConfigSupported(channelConfig)) {
                loge("getMinBufferSize(): Invalid channel configuration.");
                return ERROR_BAD_VALUE;
            } else {
                channelCount = AudioFormat.channelCountFromOutChannelMask(channelConfig);
            }
        }

        if (!AudioFormat.isPublicEncoding(audioFormat)) { //采样精度参数校验
            loge("getMinBufferSize(): Invalid audio format.");
            return ERROR_BAD_VALUE;
        }

        // sample rate, note these values are subject to change
        // Note: AudioFormat.SAMPLE_RATE_UNSPECIFIED is not allowed
        if ( (sampleRateInHz < AudioFormat.SAMPLE_RATE_HZ_MIN) || //采样率参数校验
                (sampleRateInHz > AudioFormat.SAMPLE_RATE_HZ_MAX) ) {
            loge("getMinBufferSize(): " + sampleRateInHz + " Hz is not a supported sample rate.");
            return ERROR_BAD_VALUE;
        }
        //调用JNI
        int size = native_get_min_buff_size(sampleRateInHz, channelCount, audioFormat);
        
        if (size <= 0) {
            loge("getMinBufferSize(): error querying hardware");
            return ERROR;
        }
        else {
            return size;
        }
    }

frameworks\base\core\jni\android_media_AudioTrack.cpp

static jint android_media_AudioTrack_get_min_buff_size(JNIEnv *env,  jobject thiz,
    jint sampleRateInHertz, jint channelCount, jint audioFormat) {

    size_t frameCount;
    const status_t status = AudioTrack::getMinFrameCount(&frameCount, AUDIO_STREAM_DEFAULT,
            sampleRateInHertz); //获取音频帧
    const audio_format_t format = audioFormatToNative(audioFormat);
    if (audio_has_proportional_frames(format)) {
        const size_t bytesPerSample = audio_bytes_per_sample(format);
        return frameCount * channelCount * bytesPerSample; //计算最小缓冲大小
    } else {
        return frameCount;
    }
}

frameworks\av\media\libaudioclient\AudioTrack.cpp

AudioSystem::getOutputSamplingRate 将调用AudioFlinger查询参数

status_t AudioTrack::getMinFrameCount(
        size_t* frameCount,
        audio_stream_type_t streamType,
        uint32_t sampleRate)
{
    if (frameCount == NULL) {
        return BAD_VALUE;
    }
    ......
    uint32_t afSampleRate;
    status_t status;
    status = AudioSystem::getOutputSamplingRate(&afSampleRate, streamType); //采样率
    size_t afFrameCount;
    status = AudioSystem::getOutputFrameCount(&afFrameCount, streamType); //音频帧
    uint32_t afLatency;
    status = AudioSystem::getOutputLatency(&afLatency, streamType);//音频延迟

    // When called from createTrack, speed is 1.0f (normal speed).
    // This is rechecked again on setting playback rate (TODO: on setting sample rate, too).
    *frameCount = calculateMinFrameCount(afLatency, afFrameCount, afSampleRate, sampleRate, 1.0f
            /*, 0 notificationsPerBufferReq*/);
    ......
    return NO_ERROR;
}

计算采样缓冲大小

static size_t calculateMinFrameCount(
        uint32_t afLatencyMs, uint32_t afFrameCount, uint32_t afSampleRate,
        uint32_t sampleRate, float speed /*, uint32_t notificationsPerBufferReq*/)
{
    // Ensure that buffer depth covers at least audio hardware latency
    uint32_t minBufCount = afLatencyMs / ((1000 * afFrameCount) / afSampleRate);
    if (minBufCount < 2) {
        minBufCount = 2; //至少需要两个缓冲
    }
#if 0
    if (minBufCount < notificationsPerBufferReq) {
        minBufCount = notificationsPerBufferReq;
    }
#endif
    return minBufCount * sourceFramesNeededWithTimestretch(
            sampleRate, afFrameCount, afSampleRate, speed);
}

1.2 创建AudioTrack

public AudioTrack(AudioAttributes attributes, AudioFormat format, int bufferSizeInBytes,
            int mode, int sessionId)
                    throws IllegalArgumentException {
        super(attributes, AudioPlaybackConfiguration.PLAYER_TYPE_JAM_AUDIOTRACK);
        ......
        // native initialization
        int initResult = native_setup(new WeakReference<AudioTrack>(this), mAttributes,
                sampleRate, mChannelMask, mChannelIndexMask, mAudioFormat,
                mNativeBufferSizeInBytes, mDataLoadMode, session, 0 /*nativeTrackInJavaObj*/);
        ......
    }

使用内存共享和生产者消费者模式传递数据

static jint android_media_AudioTrack_setup(JNIEnv *env, jobject thiz, jobject weak_this, jobject jaa,
        jintArray jSampleRate, jint channelPositionMask, jint channelIndexMask,
        jint audioFormat, jint buffSizeInBytes, jint memoryMode, jintArray jSession,
        jlong nativeAudioTrack) {
    sp<AudioTrack> lpTrack = 0;

    jint* nSession = (jint *) env->GetPrimitiveArrayCritical(jSession, NULL);
    audio_session_t sessionId = (audio_session_t) nSession[0];
    env->ReleasePrimitiveArrayCritical(jSession, nSession, 0);
    nSession = NULL;
    // 环形内存共享，消息队列和管道基本上都是4次拷贝，而共享内存（mmap, shmget）只有两次
    AudioTrackJniStorage* lpJniStorage = NULL; 
    audio_attributes_t *paa = NULL;
    jclass clazz = env->GetObjectClass(thiz);
    // if we pass in an existing *Native* AudioTrack, we don't need to create/initialize one.
    if (nativeAudioTrack == 0) {
        ......
        // create the native AudioTrack object
        lpTrack = new AudioTrack(); //创建AudioTrack对象
        // this data will be passed with every AudioTrack callback
        lpJniStorage = new AudioTrackJniStorage(); 
        lpJniStorage->mCallbackData.audioTrack_class = (jclass)env->NewGlobalRef(clazz);
        // we use a weak reference so the AudioTrack object can be garbage collected.
        lpJniStorage->mCallbackData.audioTrack_ref = env->NewGlobalRef(weak_this);
        lpJniStorage->mCallbackData.busy = false;

        // initialize the native AudioTrack object
        status_t status = NO_ERROR;
        switch (memoryMode) {
        case MODE_STREAM:
            status = lpTrack->set( //创建Track
                    AUDIO_STREAM_DEFAULT,
                    sampleRateInHertz,
                    format,// word length, PCM
                    nativeChannelMask,
                    frameCount,
                    AUDIO_OUTPUT_FLAG_NONE,
                    audioCallback, &(lpJniStorage->mCallbackData),//callback, callback data (user)
                    0,
                    0,// shared mem
                    true,// thread can call Java
                    sessionId,// audio session ID
                    AudioTrack::TRANSFER_SYNC,
                    NULL,                         // default offloadInfo
                    -1, -1,                       // default uid, pid values
                    paa);
            break;
            ......
        }
    } else {  // end if (nativeAudioTrack == 0)
        lpTrack = (AudioTrack*)nativeAudioTrack; //取本地
        // initialize the callback information:
        // this data will be passed with every AudioTrack callback
        lpJniStorage = new AudioTrackJniStorage();
        lpJniStorage->mCallbackData.audioTrack_class = (jclass)env->NewGlobalRef(clazz);
        // we use a weak reference so the AudioTrack object can be garbage collected.
        lpJniStorage->mCallbackData.audioTrack_ref = env->NewGlobalRef(weak_this);
        lpJniStorage->mCallbackData.busy = false;
    }
    
    ......
    
    return (jint) AUDIO_JAVA_SUCCESS;
    
}

1.3 play()

static void
android_media_AudioTrack_start(JNIEnv *env, jobject thiz)
{
    sp<AudioTrack> lpTrack = getAudioTrack(env, thiz);
    if (lpTrack == NULL) {
        jniThrowException(env, "java/lang/IllegalStateException",
            "Unable to retrieve AudioTrack pointer for start()");
        return;
    }

    lpTrack->start(); //调用native的Track
}

取已保存的AudioTrack对象

static sp<AudioTrack> getAudioTrack(JNIEnv* env, jobject thiz)
{
    Mutex::Autolock l(sLock);
    AudioTrack* const at =
            (AudioTrack*)env->GetLongField(thiz, javaAudioTrackFields.nativeTrackInJavaObj);
    return sp<AudioTrack>(at);
}

1.4 write()

template <typename T>
static jint android_media_AudioTrack_writeArray(JNIEnv *env, jobject thiz,
                                                T javaAudioData,
                                                jint offsetInSamples, jint sizeInSamples,
                                                jint javaAudioFormat,
                                                jboolean isWriteBlocking) {
    //ALOGV("android_media_AudioTrack_writeArray(offset=%d, sizeInSamples=%d) called",
    //        offsetInSamples, sizeInSamples);
    sp<AudioTrack> lpTrack = getAudioTrack(env, thiz);
    ......
    jint samplesWritten = writeToTrack(lpTrack, javaAudioFormat, cAudioData,
            offsetInSamples, sizeInSamples, isWriteBlocking == JNI_TRUE /* blocking */);
    return samplesWritten;
}

最终会调用到这里

template <typename T>
static jint writeToTrack(const sp<AudioTrack>& track, jint audioFormat, const T *data,
                         jint offsetInSamples, jint sizeInSamples, bool blocking) {
    // give the data to the native AudioTrack object (the data starts at the offset)
    ssize_t written = 0;
    // regular write() or copy the data to the AudioTrack's shared memory?
    size_t sizeInBytes = sizeInSamples * sizeof(T);
    if (track->sharedBuffer() == 0) { //STREAM模式
        written = track->write(data + offsetInSamples, sizeInBytes, blocking); //写数据
        // for compatibility with earlier behavior of write(), return 0 in this case
        if (written == (ssize_t) WOULD_BLOCK) {
            written = 0;
        }
    } else { //STATIC模式
        // writing to shared memory, check for capacity
        if ((size_t)sizeInBytes > track->sharedBuffer()->size()) {
            sizeInBytes = track->sharedBuffer()->size();
        }
        //静态模式下直接拷贝到内存
        memcpy(track->sharedBuffer()->pointer(), data + offsetInSamples, sizeInBytes);
        written = sizeInBytes;
    }
    if (written >= 0) {
        return written / sizeof(T);
    }
    return interpretWriteSizeError(written);
}

到此AudioTrack的Java层分析结束，四个流程中还留涉及到AudioFlinger，AudioPolicyService相关的部分我们先放一放。

2. Native层的AudioTrack

继上面AudioTrack对象创建的set函数，创建 AudioTrackThread线程

 if (cbf != NULL) {
        mAudioTrackThread = new AudioTrackThread(*this, threadCanCallJava);
        mAudioTrackThread->run("AudioTrack", ANDROID_PRIORITY_AUDIO, 0 /*stack*/);
        // thread begins in paused state, and will not reference us until start()
    }

    // create the IAudioTrack
    status_t status = createTrack_l(); //创建Track()

    if (status != NO_ERROR) {
        if (mAudioTrackThread != 0) {
            mAudioTrackThread->requestExit();   // see comment in AudioTrack.h
            mAudioTrackThread->requestExitAndWait();
            mAudioTrackThread.clear();
        }
        return status;
    }

createTrack_l()函数将调用AudioFlinger创建Track()

const sp<IAudioFlinger>& audioFlinger = AudioSystem::get_audio_flinger();
sp<IAudioTrack> track = audioFlinger->createTrack(......); //后续分析
//线性环形缓冲
audio_track_cblk_t* cblk = static_cast<audio_track_cblk_t*>(iMemPointer);
  

AudioTrackThread启动后就开始处理数据

bool AudioTrack::AudioTrackThread::threadLoop()
{
   nsecs_t ns = mReceiver.processAudioBuffer(); 
}

处理线性环形缓冲数据

nsecs_t AudioTrack::processAudioBuffer()
{

    // perform callbacks while unlocked
    if (newUnderrun) {
        mCbf(EVENT_UNDERRUN, mUserData, NULL);
    }
    while (loopCountNotifications > 0) {
        mCbf(EVENT_LOOP_END, mUserData, NULL);
        --loopCountNotifications;
    }
    if (flags & CBLK_BUFFER_END) {
        mCbf(EVENT_BUFFER_END, mUserData, NULL);
    }
    if (markerReached) {
        mCbf(EVENT_MARKER, mUserData, &markerPosition);
    }
    //循环播放通知
    while (newPosCount > 0) {
        size_t temp = newPosition.value(); // FIXME size_t != uint32_t
        mCbf(EVENT_NEW_POS, mUserData, &temp);
        newPosition += updatePeriod;
        newPosCount--;
    }

    if (mObservedSequence != sequence) {
        mObservedSequence = sequence;
        mCbf(EVENT_NEW_IAUDIOTRACK, mUserData, NULL);
        // for offloaded tracks, just wait for the upper layers to recreate the track
        if (isOffloadedOrDirect()) {
            return NS_INACTIVE;
        }
    }

    // if inactive, then don't run me again until re-started
    if (!active) {
        return NS_INACTIVE;
    }

    // Compute the estimated time until the next timed event (position, markers, loops)
    // FIXME only for non-compressed audio
    //数据使用超过警戒线
    if (!markerReached && position < markerPosition) {
        minFrames = (markerPosition - position).value();
    }
    if (loopPeriod > 0 && loopPeriod < minFrames) {
        // loopPeriod is already adjusted for actual position.
        minFrames = loopPeriod;
    }
    if (updatePeriod > 0) { //进度通知
        minFrames = min(minFrames, (newPosition - position).value());
    }

    // If > 0, poll periodically to recover from a stuck server.  A good value is 2.
    static const uint32_t kPoll = 0;
    if (kPoll > 0 && mTransfer == TRANSFER_CALLBACK && kPoll * notificationFrames < minFrames) {
        minFrames = kPoll * notificationFrames;
    }

    ......
    
    size_t writtenFrames = 0;
    while (mRemainingFrames > 0) {

        Buffer audioBuffer;
        audioBuffer.frameCount = mRemainingFrames;
        size_t nonContig;
        //得到一块可写的缓冲
        status_t err = obtainBuffer(&audioBuffer, requested, NULL, &nonContig);
        LOG_ALWAYS_FATAL_IF((err != NO_ERROR) != (audioBuffer.frameCount == 0),
                "obtainBuffer() err=%d frameCount=%zu", err, audioBuffer.frameCount);
        ......
        size_t reqSize = audioBuffer.size;
        mCbf(EVENT_MORE_DATA, mUserData, &audioBuffer); //从用户那里PUll数据
        size_t writtenSize = audioBuffer.size;
        ......
        releaseBuffer(&audioBuffer); //写完毕，释放缓冲
        writtenFrames += releasedFrames;
        ......
    return 0;
}

写PCM数据，取一块空闲内存，将数据拷贝到内存

ssize_t AudioTrack::write(const void* buffer, size_t userSize, bool blocking)
{
    ......
    size_t written = 0;
    Buffer audioBuffer;

    while (userSize >= mFrameSize) {
        audioBuffer.frameCount = userSize / mFrameSize;
        //获得一块空闲内存
        status_t err = obtainBuffer(&audioBuffer,
                blocking ? &ClientProxy::kForever : &ClientProxy::kNonBlocking);
        if (err < 0) {
            if (written > 0) {
                break;
            }
            if (err == TIMED_OUT || err == -EINTR) {
                err = WOULD_BLOCK;
            }
            return ssize_t(err);
        }

        size_t toWrite = audioBuffer.size;
        memcpy(audioBuffer.i8, buffer, toWrite); //传递数据
        buffer = ((const char *) buffer) + toWrite; 
        userSize -= toWrite;
        written += toWrite;
        //释放
        releaseBuffer(&audioBuffer);
    }

    if (written > 0) {
        mFramesWritten += written / mFrameSize;
    }
    return written;
}

最终释放缓冲

void AudioTrack::stop()
{
    AutoMutex lock(mLock);
    if (mState != STATE_ACTIVE && mState != STATE_PAUSED) {
        return;
    }

    if (isOffloaded_l()) {
        mState = STATE_STOPPING;
    } else {
        mState = STATE_STOPPED;
        ALOGD_IF(mSharedBuffer == nullptr,
                "stop() called with %u frames delivered", mReleased.value());
        mReleased = 0;
    }

    mProxy->interrupt();
    mAudioTrack->stop();

    // Note: legacy handling - stop does not clear playback marker
    // and periodic update counter, but flush does for streaming tracks.

    if (mSharedBuffer != 0) {
        // clear buffer position and loop count.
        //清空循环播放设置
        mStaticProxy->setBufferPositionAndLoop(0 /* position */,
                0 /* loopStart */, 0 /* loopEnd */, 0 /* loopCount */);
    }

    sp<AudioTrackThread> t = mAudioTrackThread;
    if (t != 0) {
        if (!isOffloaded_l()) {
            t->pause(); //线程暂停
        }
    } else {
        setpriority(PRIO_PROCESS, 0, mPreviousPriority);
        set_sched_policy(0, mPreviousSchedulingGroup);
    }
}

OK，到此AudioTrack的分析到此打住，我们仅仅分析了AudioTrack的创建，数据处理。由于涉及到与AudioFlinger，AudioPolicy的交互，等我们分析完这两个以后再回过头来看AudioTrack。到此结束。