diff --git a/camera/libcameraservice/CameraService.cpp b/camera/libcameraservice/CameraService.cpp
index df59dcf..6419a5c 100644
--- a/camera/libcameraservice/CameraService.cpp
+++ b/camera/libcameraservice/CameraService.cpp
@@ -669,8 +669,14 @@
     LOGD("startRecording (pid %d)", getCallingPid());
 
     if (mMediaPlayerBeep.get() != NULL) {
-        mMediaPlayerBeep->seekTo(0);
-        mMediaPlayerBeep->start();
+        // do not play record jingle if stream volume is 0
+        // (typically because ringer mode is silent).
+        int index;
+        AudioSystem::getStreamVolumeIndex(AudioSystem::ENFORCED_AUDIBLE, &index);
+        if (index != 0) {
+            mMediaPlayerBeep->seekTo(0);
+            mMediaPlayerBeep->start();
+        }
     }
 
     mHardware->enableMsgType(CAMERA_MSG_VIDEO_FRAME);
@@ -888,8 +894,14 @@
 {
     // Play shutter sound.
     if (mMediaPlayerClick.get() != NULL) {
-        mMediaPlayerClick->seekTo(0);
-        mMediaPlayerClick->start();
+        // do not play shutter sound if stream volume is 0
+        // (typically because ringer mode is silent).
+        int index;
+        AudioSystem::getStreamVolumeIndex(AudioSystem::ENFORCED_AUDIBLE, &index);
+        if (index != 0) {
+            mMediaPlayerClick->seekTo(0);
+            mMediaPlayerClick->start();
+        }
     }
 
     // Screen goes black after the buffer is unregistered.
diff --git a/include/ui/Surface.h b/include/ui/Surface.h
index 70303cd..008c297 100644
--- a/include/ui/Surface.h
+++ b/include/ui/Surface.h
@@ -109,7 +109,7 @@
 
     ~SurfaceControl();
 
-    status_t validate(SharedClient const* cblk) const;
+    status_t validate() const;
     void destroy();
     
     sp<SurfaceComposerClient>   mClient;
@@ -190,7 +190,7 @@
 
     status_t getBufferLocked(int index, int usage);
    
-           status_t validate(SharedClient const* cblk) const;
+           status_t validate() const;
 
     inline const GraphicBufferMapper& getBufferMapper() const { return mBufferMapper; }
     inline GraphicBufferMapper& getBufferMapper() { return mBufferMapper; }
diff --git a/libs/audioflinger/AudioFlinger.cpp b/libs/audioflinger/AudioFlinger.cpp
index ebd470f..ecfe1e0 100644
--- a/libs/audioflinger/AudioFlinger.cpp
+++ b/libs/audioflinger/AudioFlinger.cpp
@@ -674,32 +674,13 @@
 }
 
 // audioConfigChanged_l() must be called with AudioFlinger::mLock held
-void AudioFlinger::audioConfigChanged_l(int event, const sp<ThreadBase>& thread, void *param2) {
-    int ioHandle = 0;
-
-    for (size_t i = 0; i < mPlaybackThreads.size(); i++) {
-        if (mPlaybackThreads.valueAt(i) == thread) {
-            ioHandle = mPlaybackThreads.keyAt(i);
-            break;
-        }
-    }
-    if (ioHandle == 0) {
-        for (size_t i = 0; i < mRecordThreads.size(); i++) {
-            if (mRecordThreads.valueAt(i) == thread) {
-                ioHandle = mRecordThreads.keyAt(i);
-                break;
-            }
-        }
-    }
-
-    if (ioHandle != 0) {
-        size_t size = mNotificationClients.size();
-        for (size_t i = 0; i < size; i++) {
-            sp<IBinder> binder = mNotificationClients.itemAt(i);
-            LOGV("audioConfigChanged_l() Notifying change to client %p", binder.get());
-            sp<IAudioFlingerClient> client = interface_cast<IAudioFlingerClient> (binder);
-            client->ioConfigChanged(event, ioHandle, param2);
-        }
+void AudioFlinger::audioConfigChanged_l(int event, int ioHandle, void *param2) {
+    size_t size = mNotificationClients.size();
+    for (size_t i = 0; i < size; i++) {
+        sp<IBinder> binder = mNotificationClients.itemAt(i);
+        LOGV("audioConfigChanged_l() Notifying change to client %p", binder.get());
+        sp<IAudioFlingerClient> client = interface_cast<IAudioFlingerClient> (binder);
+        client->ioConfigChanged(event, ioHandle, param2);
     }
 }
 
@@ -712,10 +693,10 @@
 
 // ----------------------------------------------------------------------------
 
-AudioFlinger::ThreadBase::ThreadBase(const sp<AudioFlinger>& audioFlinger)
+AudioFlinger::ThreadBase::ThreadBase(const sp<AudioFlinger>& audioFlinger, int id)
     :   Thread(false),
         mAudioFlinger(audioFlinger), mSampleRate(0), mFrameCount(0), mChannelCount(0),
-        mFormat(0), mFrameSize(1), mStandby(false)
+        mFormat(0), mFrameSize(1), mStandby(false), mId(id), mExiting(false)
 {
 }
 
@@ -734,6 +715,7 @@
     LOGV("ThreadBase::exit");
     {
         AutoMutex lock(&mLock);
+        mExiting = true;
         requestExit();
         mWaitWorkCV.signal();
     }
@@ -870,8 +852,8 @@
 
 // ----------------------------------------------------------------------------
 
-AudioFlinger::PlaybackThread::PlaybackThread(const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output)
-    :   ThreadBase(audioFlinger),
+AudioFlinger::PlaybackThread::PlaybackThread(const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output, int id)
+    :   ThreadBase(audioFlinger, id),
         mMixBuffer(0), mSuspended(0), mBytesWritten(0), mOutput(output),
         mLastWriteTime(0), mNumWrites(0), mNumDelayedWrites(0), mInWrite(false)
 {
@@ -950,6 +932,8 @@
     result.append(buffer);
     snprintf(buffer, SIZE, "blocked in write: %d\n", mInWrite);
     result.append(buffer);
+    snprintf(buffer, SIZE, "suspend count: %d\n", mSuspended);
+    result.append(buffer);
     write(fd, result.string(), result.size());
 
     dumpBase(fd, args);
@@ -1106,15 +1090,6 @@
 {
     status_t status = ALREADY_EXISTS;
 
-    // here the track could be either new, or restarted
-    // in both cases "unstop" the track
-    if (track->isPaused()) {
-        track->mState = TrackBase::RESUMING;
-        LOGV("PAUSED => RESUMING (%d) on thread %p", track->name(), this);
-    } else {
-        track->mState = TrackBase::ACTIVE;
-        LOGV("? => ACTIVE (%d) on thread %p", track->name(), this);
-    }
     // set retry count for buffer fill
     track->mRetryCount = kMaxTrackStartupRetries;
     if (mActiveTracks.indexOf(track) < 0) {
@@ -1138,7 +1113,6 @@
 {
     track->mState = TrackBase::TERMINATED;
     if (mActiveTracks.indexOf(track) < 0) {
-        LOGV("remove track (%d) and delete from mixer", track->name());
         mTracks.remove(track);
         deleteTrackName_l(track->name());
     }
@@ -1173,7 +1147,7 @@
         break;
     }
     Mutex::Autolock _l(mAudioFlinger->mLock);
-    mAudioFlinger->audioConfigChanged_l(event, this, param2);
+    mAudioFlinger->audioConfigChanged_l(event, mId, param2);
 }
 
 void AudioFlinger::PlaybackThread::readOutputParameters()
@@ -1185,7 +1159,6 @@
     mFrameSize = mOutput->frameSize();
     mFrameCount = mOutput->bufferSize() / mFrameSize;
 
-    mMinBytesToWrite = (mOutput->latency() * mSampleRate * mFrameSize) / 1000;
     // FIXME - Current mixer implementation only supports stereo output: Always
     // Allocate a stereo buffer even if HW output is mono.
     if (mMixBuffer != NULL) delete mMixBuffer;
@@ -1195,8 +1168,8 @@
 
 // ----------------------------------------------------------------------------
 
-AudioFlinger::MixerThread::MixerThread(const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output)
-    :   PlaybackThread(audioFlinger, output),
+AudioFlinger::MixerThread::MixerThread(const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output, int id)
+    :   PlaybackThread(audioFlinger, output, id),
         mAudioMixer(0)
 {
     mType = PlaybackThread::MIXER;
@@ -1215,23 +1188,25 @@
 
 bool AudioFlinger::MixerThread::threadLoop()
 {
-    uint32_t sleepTime = 1000;
-    uint32_t maxBufferRecoveryInUsecs = getMaxBufferRecoveryInUsecs();
     int16_t* curBuf = mMixBuffer;
     Vector< sp<Track> > tracksToRemove;
-    size_t enabledTracks = 0;
+    uint32_t mixerStatus = MIXER_IDLE;
     nsecs_t standbyTime = systemTime();
     size_t mixBufferSize = mFrameCount * mFrameSize;
     // FIXME: Relaxed timing because of a certain device that can't meet latency
     // Should be reduced to 2x after the vendor fixes the driver issue
     nsecs_t maxPeriod = seconds(mFrameCount) / mSampleRate * 3;
     nsecs_t lastWarning = 0;
+    bool longStandbyExit = false;
+    uint32_t activeSleepTime = activeSleepTimeUs();
+    uint32_t idleSleepTime = idleSleepTimeUs();
+    uint32_t sleepTime = idleSleepTime;
 
     while (!exitPending())
     {
         processConfigEvents();
 
-        enabledTracks = 0;
+        mixerStatus = MIXER_IDLE;
         { // scope for mLock
 
             Mutex::Autolock _l(mLock);
@@ -1241,7 +1216,8 @@
                 // FIXME: Relaxed timing because of a certain device that can't meet latency
                 // Should be reduced to 2x after the vendor fixes the driver issue
                 maxPeriod = seconds(mFrameCount) / mSampleRate * 3;
-                maxBufferRecoveryInUsecs = getMaxBufferRecoveryInUsecs();
+                activeSleepTime = activeSleepTimeUs();
+                idleSleepTime = idleSleepTimeUs();
             }
 
             const SortedVector< wp<Track> >& activeTracks = mActiveTracks;
@@ -1277,15 +1253,15 @@
                     }
 
                     standbyTime = systemTime() + kStandbyTimeInNsecs;
-                    sleepTime = 1000;
+                    sleepTime = idleSleepTime;
                     continue;
                 }
             }
 
-            enabledTracks = prepareTracks_l(activeTracks, &tracksToRemove);
+            mixerStatus = prepareTracks_l(activeTracks, &tracksToRemove);
        }
 
-        if (LIKELY(enabledTracks)) {
+        if (LIKELY(mixerStatus == MIXER_TRACKS_READY)) {
             // mix buffers...
             mAudioMixer->process(curBuf);
             sleepTime = 0;
@@ -1294,15 +1270,21 @@
             // If no tracks are ready, sleep once for the duration of an output
             // buffer size, then write 0s to the output
             if (sleepTime == 0) {
-                sleepTime = maxBufferRecoveryInUsecs;
-            } else if (mBytesWritten != 0) {
+                if (mixerStatus == MIXER_TRACKS_ENABLED) {
+                    sleepTime = activeSleepTime;
+                } else {
+                    sleepTime = idleSleepTime;
+                }
+            } else if (mBytesWritten != 0 ||
+                       (mixerStatus == MIXER_TRACKS_ENABLED && longStandbyExit)) {
                 memset (curBuf, 0, mixBufferSize);
                 sleepTime = 0;
+                LOGV_IF((mBytesWritten == 0 && (mixerStatus == MIXER_TRACKS_ENABLED && longStandbyExit)), "anticipated start");
             }
         }
 
         if (mSuspended) {
-            sleepTime = maxBufferRecoveryInUsecs;
+            sleepTime = idleSleepTime;
         }
         // sleepTime == 0 means we must write to audio hardware
         if (sleepTime == 0) {
@@ -1312,7 +1294,6 @@
             if (bytesWritten > 0) mBytesWritten += bytesWritten;
             mNumWrites++;
             mInWrite = false;
-            mStandby = false;
             nsecs_t now = systemTime();
             nsecs_t delta = now - mLastWriteTime;
             if (delta > maxPeriod) {
@@ -1322,7 +1303,11 @@
                             ns2ms(delta), mNumDelayedWrites, this);
                     lastWarning = now;
                 }
+                if (mStandby) {
+                    longStandbyExit = true;
+                }
             }
+            mStandby = false;
         } else {
             usleep(sleepTime);
         }
@@ -1342,10 +1327,10 @@
 }
 
 // prepareTracks_l() must be called with ThreadBase::mLock held
-size_t AudioFlinger::MixerThread::prepareTracks_l(const SortedVector< wp<Track> >& activeTracks, Vector< sp<Track> > *tracksToRemove)
+uint32_t AudioFlinger::MixerThread::prepareTracks_l(const SortedVector< wp<Track> >& activeTracks, Vector< sp<Track> > *tracksToRemove)
 {
 
-    size_t enabledTracks = 0;
+    uint32_t mixerStatus = MIXER_IDLE;
     // find out which tracks need to be processed
     size_t count = activeTracks.size();
     for (size_t i=0 ; i<count ; i++) {
@@ -1361,7 +1346,7 @@
         if (cblk->framesReady() && (track->isReady() || track->isStopped()) &&
                 !track->isPaused())
         {
-            //LOGV("track %d u=%08x, s=%08x [OK]", track->name(), cblk->user, cblk->server);
+            //LOGV("track %d u=%08x, s=%08x [OK] on thread %p", track->name(), cblk->user, cblk->server, this);
 
             // compute volume for this track
             int16_t left, right;
@@ -1415,9 +1400,9 @@
 
             // reset retry count
             track->mRetryCount = kMaxTrackRetries;
-            enabledTracks++;
+            mixerStatus = MIXER_TRACKS_READY;
         } else {
-            //LOGV("track %d u=%08x, s=%08x [NOT READY]", track->name(), cblk->user, cblk->server);
+            //LOGV("track %d u=%08x, s=%08x [NOT READY] on thread %p", track->name(), cblk->user, cblk->server, this);
             if (track->isStopped()) {
                 track->reset();
             }
@@ -1432,16 +1417,11 @@
                 if (--(track->mRetryCount) <= 0) {
                     LOGV("BUFFER TIMEOUT: remove(%d) from active list on thread %p", track->name(), this);
                     tracksToRemove->add(track);
+                } else if (mixerStatus != MIXER_TRACKS_READY) {
+                    mixerStatus = MIXER_TRACKS_ENABLED;
                 }
-                // For tracks using static shared memory buffer, make sure that we have
-                // written enough data to audio hardware before disabling the track
-                // NOTE: this condition with arrive before track->mRetryCount <= 0 so we
-                // don't care about code removing track from active list above.
-                if ((track->mSharedBuffer == 0) || (mBytesWritten >= mMinBytesToWrite)) {
-                    mAudioMixer->disable(AudioMixer::MIXING);
-                } else {
-                    enabledTracks++;
-                }
+
+                mAudioMixer->disable(AudioMixer::MIXING);
             }
         }
     }
@@ -1459,7 +1439,7 @@
         }
     }
 
-    return enabledTracks;
+    return mixerStatus;
 }
 
 void AudioFlinger::MixerThread::getTracks(
@@ -1532,6 +1512,7 @@
 // deleteTrackName_l() must be called with ThreadBase::mLock held
 void AudioFlinger::MixerThread::deleteTrackName_l(int name)
 {
+    LOGV("remove track (%d) and delete from mixer", name);
     mAudioMixer->deleteTrackName(name);
 }
 
@@ -1621,19 +1602,19 @@
     return NO_ERROR;
 }
 
-uint32_t AudioFlinger::MixerThread::getMaxBufferRecoveryInUsecs()
+uint32_t AudioFlinger::MixerThread::activeSleepTimeUs()
 {
-    uint32_t time = ((mFrameCount * 1000) / mSampleRate) * 1000;
-    // Add some margin with regard to scheduling precision
-    if (time > 10000) {
-        time -= 10000;
-    }
-    return time;
+    return (uint32_t)(mOutput->latency() * 1000) / 2;
+}
+
+uint32_t AudioFlinger::MixerThread::idleSleepTimeUs()
+{
+    return (uint32_t)((mFrameCount * 1000) / mSampleRate) * 1000;
 }
 
 // ----------------------------------------------------------------------------
-AudioFlinger::DirectOutputThread::DirectOutputThread(const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output)
-    :   PlaybackThread(audioFlinger, output),
+AudioFlinger::DirectOutputThread::DirectOutputThread(const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output, int id)
+    :   PlaybackThread(audioFlinger, output, id),
     mLeftVolume (1.0), mRightVolume(1.0)
 {
     mType = PlaybackThread::DIRECT;
@@ -1646,25 +1627,31 @@
 
 bool AudioFlinger::DirectOutputThread::threadLoop()
 {
-    uint32_t sleepTime = 1000;
-    uint32_t maxBufferRecoveryInUsecs = getMaxBufferRecoveryInUsecs();
+    uint32_t mixerStatus = MIXER_IDLE;
     sp<Track> trackToRemove;
     sp<Track> activeTrack;
     nsecs_t standbyTime = systemTime();
     int8_t *curBuf;
     size_t mixBufferSize = mFrameCount*mFrameSize;
+    uint32_t activeSleepTime = activeSleepTimeUs();
+    uint32_t idleSleepTime = idleSleepTimeUs();
+    uint32_t sleepTime = idleSleepTime;
+
 
     while (!exitPending())
     {
         processConfigEvents();
 
+        mixerStatus = MIXER_IDLE;
+
         { // scope for the mLock
 
             Mutex::Autolock _l(mLock);
 
             if (checkForNewParameters_l()) {
                 mixBufferSize = mFrameCount*mFrameSize;
-                maxBufferRecoveryInUsecs = getMaxBufferRecoveryInUsecs();
+                activeSleepTime = activeSleepTimeUs();
+                idleSleepTime = idleSleepTimeUs();
             }
 
             // put audio hardware into standby after short delay
@@ -1698,7 +1685,7 @@
                     }
 
                     standbyTime = systemTime() + kStandbyTimeInNsecs;
-                    sleepTime = 1000;
+                    sleepTime = idleSleepTime;
                     continue;
                 }
             }
@@ -1753,6 +1740,7 @@
                     // reset retry count
                     track->mRetryCount = kMaxTrackRetries;
                     activeTrack = t;
+                    mixerStatus = MIXER_TRACKS_READY;
                 } else {
                     //LOGV("track %d u=%08x, s=%08x [NOT READY]", track->name(), cblk->user, cblk->server);
                     if (track->isStopped()) {
@@ -1768,16 +1756,10 @@
                         if (--(track->mRetryCount) <= 0) {
                             LOGV("BUFFER TIMEOUT: remove(%d) from active list", track->name());
                             trackToRemove = track;
+                        } else {
+                            mixerStatus = MIXER_TRACKS_ENABLED;
                         }
-
-                        // For tracks using static shared memry buffer, make sure that we have
-                        // written enough data to audio hardware before disabling the track
-                        // NOTE: this condition with arrive before track->mRetryCount <= 0 so we
-                        // don't care about code removing track from active list above.
-                        if ((track->mSharedBuffer != 0) && (mBytesWritten < mMinBytesToWrite)) {
-                            activeTrack = t;
-                        }
-                     }
+                    }
                 }
             }
 
@@ -1791,7 +1773,7 @@
             }
        }
 
-        if (activeTrack != 0) {
+        if (LIKELY(mixerStatus == MIXER_TRACKS_READY)) {
             AudioBufferProvider::Buffer buffer;
             size_t frameCount = mFrameCount;
             curBuf = (int8_t *)mMixBuffer;
@@ -1812,7 +1794,11 @@
             standbyTime = systemTime() + kStandbyTimeInNsecs;
         } else {
             if (sleepTime == 0) {
-                sleepTime = maxBufferRecoveryInUsecs;
+                if (mixerStatus == MIXER_TRACKS_ENABLED) {
+                    sleepTime = activeSleepTime;
+                } else {
+                    sleepTime = idleSleepTime;
+                }
             } else if (mBytesWritten != 0 && AudioSystem::isLinearPCM(mFormat)) {
                 memset (mMixBuffer, 0, mFrameCount * mFrameSize);
                 sleepTime = 0;
@@ -1820,7 +1806,7 @@
         }
 
         if (mSuspended) {
-            sleepTime = maxBufferRecoveryInUsecs;
+            sleepTime = idleSleepTime;
         }
         // sleepTime == 0 means we must write to audio hardware
         if (sleepTime == 0) {
@@ -1905,15 +1891,22 @@
     return reconfig;
 }
 
-uint32_t AudioFlinger::DirectOutputThread::getMaxBufferRecoveryInUsecs()
+uint32_t AudioFlinger::DirectOutputThread::activeSleepTimeUs()
 {
     uint32_t time;
     if (AudioSystem::isLinearPCM(mFormat)) {
-        time = ((mFrameCount * 1000) / mSampleRate) * 1000;
-        // Add some margin with regard to scheduling precision
-        if (time > 10000) {
-            time -= 10000;
-        }
+        time = (uint32_t)(mOutput->latency() * 1000) / 2;
+    } else {
+        time = 10000;
+    }
+    return time;
+}
+
+uint32_t AudioFlinger::DirectOutputThread::idleSleepTimeUs()
+{
+    uint32_t time;
+    if (AudioSystem::isLinearPCM(mFormat)) {
+        time = (uint32_t)((mFrameCount * 1000) / mSampleRate) * 1000;
     } else {
         time = 10000;
     }
@@ -1922,8 +1915,8 @@
 
 // ----------------------------------------------------------------------------
 
-AudioFlinger::DuplicatingThread::DuplicatingThread(const sp<AudioFlinger>& audioFlinger, AudioFlinger::MixerThread* mainThread)
-    :   MixerThread(audioFlinger, mainThread->getOutput())
+AudioFlinger::DuplicatingThread::DuplicatingThread(const sp<AudioFlinger>& audioFlinger, AudioFlinger::MixerThread* mainThread, int id)
+    :   MixerThread(audioFlinger, mainThread->getOutput(), id), mWaitTimeMs(UINT_MAX)
 {
     mType = PlaybackThread::DUPLICATING;
     addOutputTrack(mainThread);
@@ -1931,33 +1924,39 @@
 
 AudioFlinger::DuplicatingThread::~DuplicatingThread()
 {
+    for (size_t i = 0; i < mOutputTracks.size(); i++) {
+        mOutputTracks[i]->destroy();
+    }
     mOutputTracks.clear();
 }
 
 bool AudioFlinger::DuplicatingThread::threadLoop()
 {
-    uint32_t sleepTime = 1000;
-    uint32_t maxBufferRecoveryInUsecs = getMaxBufferRecoveryInUsecs();
     int16_t* curBuf = mMixBuffer;
     Vector< sp<Track> > tracksToRemove;
-    size_t enabledTracks = 0;
+    uint32_t mixerStatus = MIXER_IDLE;
     nsecs_t standbyTime = systemTime();
     size_t mixBufferSize = mFrameCount*mFrameSize;
     SortedVector< sp<OutputTrack> > outputTracks;
     uint32_t writeFrames = 0;
+    uint32_t activeSleepTime = activeSleepTimeUs();
+    uint32_t idleSleepTime = idleSleepTimeUs();
+    uint32_t sleepTime = idleSleepTime;
 
     while (!exitPending())
     {
         processConfigEvents();
 
-        enabledTracks = 0;
+        mixerStatus = MIXER_IDLE;
         { // scope for the mLock
 
             Mutex::Autolock _l(mLock);
 
             if (checkForNewParameters_l()) {
                 mixBufferSize = mFrameCount*mFrameSize;
-                maxBufferRecoveryInUsecs = getMaxBufferRecoveryInUsecs();
+                updateWaitTime();
+                activeSleepTime = activeSleepTimeUs();
+                idleSleepTime = idleSleepTimeUs();
             }
 
             const SortedVector< wp<Track> >& activeTracks = mActiveTracks;
@@ -1997,22 +1996,30 @@
                     }
 
                     standbyTime = systemTime() + kStandbyTimeInNsecs;
-                    sleepTime = 1000;
+                    sleepTime = idleSleepTime;
                     continue;
                 }
             }
 
-            enabledTracks = prepareTracks_l(activeTracks, &tracksToRemove);
+            mixerStatus = prepareTracks_l(activeTracks, &tracksToRemove);
         }
 
-        if (LIKELY(enabledTracks)) {
+        if (LIKELY(mixerStatus == MIXER_TRACKS_READY)) {
             // mix buffers...
-            mAudioMixer->process(curBuf);
+            if (outputsReady(outputTracks)) {
+                mAudioMixer->process(curBuf);
+            } else {
+                memset(curBuf, 0, mixBufferSize);
+            }
             sleepTime = 0;
             writeFrames = mFrameCount;
         } else {
             if (sleepTime == 0) {
-                sleepTime = maxBufferRecoveryInUsecs;
+                if (mixerStatus == MIXER_TRACKS_ENABLED) {
+                    sleepTime = activeSleepTime;
+                } else {
+                    sleepTime = idleSleepTime;
+                }
             } else if (mBytesWritten != 0) {
                 // flush remaining overflow buffers in output tracks
                 for (size_t i = 0; i < outputTracks.size(); i++) {
@@ -2026,7 +2033,7 @@
         }
 
         if (mSuspended) {
-            sleepTime = maxBufferRecoveryInUsecs;
+            sleepTime = idleSleepTime;
         }
         // sleepTime == 0 means we must write to audio hardware
         if (sleepTime == 0) {
@@ -2047,17 +2054,6 @@
         outputTracks.clear();
     }
 
-    { // scope for the mLock
-
-        Mutex::Autolock _l(mLock);
-        if (!mStandby) {
-            LOGV("DuplicatingThread() exiting out of standby");
-            for (size_t i = 0; i < mOutputTracks.size(); i++) {
-                mOutputTracks[i]->destroy();
-            }
-        }
-    }
-
     return false;
 }
 
@@ -2065,6 +2061,7 @@
 {
     int frameCount = (3 * mFrameCount * mSampleRate) / thread->sampleRate();
     OutputTrack *outputTrack = new OutputTrack((ThreadBase *)thread,
+                                            this,
                                             mSampleRate,
                                             mFormat,
                                             mChannelCount,
@@ -2073,6 +2070,7 @@
         thread->setStreamVolume(AudioSystem::NUM_STREAM_TYPES, 1.0f);
         mOutputTracks.add(outputTrack);
         LOGV("addOutputTrack() track %p, on thread %p", outputTrack, thread);
+        updateWaitTime();
     }
 }
 
@@ -2083,12 +2081,50 @@
         if (mOutputTracks[i]->thread() == (ThreadBase *)thread) {
             mOutputTracks[i]->destroy();
             mOutputTracks.removeAt(i);
+            updateWaitTime();
             return;
         }
     }
     LOGV("removeOutputTrack(): unkonwn thread: %p", thread);
 }
 
+void AudioFlinger::DuplicatingThread::updateWaitTime()
+{
+    mWaitTimeMs = UINT_MAX;
+    for (size_t i = 0; i < mOutputTracks.size(); i++) {
+        sp<ThreadBase> strong = mOutputTracks[i]->thread().promote();
+        if (strong != NULL) {
+            uint32_t waitTimeMs = (strong->frameCount() * 2 * 1000) / strong->sampleRate();
+            if (waitTimeMs < mWaitTimeMs) {
+                mWaitTimeMs = waitTimeMs;
+            }
+        }
+    }
+}
+
+
+bool AudioFlinger::DuplicatingThread::outputsReady(SortedVector< sp<OutputTrack> > &outputTracks)
+{
+    for (size_t i = 0; i < outputTracks.size(); i++) {
+        sp <ThreadBase> thread = outputTracks[i]->thread().promote();
+        if (thread == 0) {
+            LOGW("DuplicatingThread::outputsReady() could not promote thread on output track %p", outputTracks[i].get());
+            return false;
+        }
+        PlaybackThread *playbackThread = (PlaybackThread *)thread.get();
+        if (playbackThread->standby() && !playbackThread->isSuspended()) {
+            LOGV("DuplicatingThread output track %p on thread %p Not Ready", outputTracks[i].get(), thread.get());
+            return false;
+        }
+    }
+    return true;
+}
+
+uint32_t AudioFlinger::DuplicatingThread::activeSleepTimeUs()
+{
+    return (mWaitTimeMs * 1000) / 2;
+}
+
 // ----------------------------------------------------------------------------
 
 // TrackBase constructor must be called with AudioFlinger::mLock held
@@ -2299,6 +2335,12 @@
     { // scope for mLock
         sp<ThreadBase> thread = mThread.promote();
         if (thread != 0) {
+            if (!isOutputTrack()) {
+                if (mState == ACTIVE || mState == RESUMING) {
+                    AudioSystem::stopOutput(thread->id(), (AudioSystem::stream_type)mStreamType);
+                }
+                AudioSystem::releaseOutput(thread->id());
+            }
             Mutex::Autolock _l(thread->mLock);
             PlaybackThread *playbackThread = (PlaybackThread *)thread.get();
             playbackThread->destroyTrack_l(this);
@@ -2380,14 +2422,37 @@
 
 status_t AudioFlinger::PlaybackThread::Track::start()
 {
+    status_t status = NO_ERROR;
     LOGV("start(%d), calling thread %d", mName, IPCThreadState::self()->getCallingPid());
     sp<ThreadBase> thread = mThread.promote();
     if (thread != 0) {
         Mutex::Autolock _l(thread->mLock);
-        PlaybackThread *playbackThread = (PlaybackThread *)thread.get();
-        playbackThread->addTrack_l(this);
+        int state = mState;
+        // here the track could be either new, or restarted
+        // in both cases "unstop" the track
+        if (mState == PAUSED) {
+            mState = TrackBase::RESUMING;
+            LOGV("PAUSED => RESUMING (%d) on thread %p", mName, this);
+        } else {
+            mState = TrackBase::ACTIVE;
+            LOGV("? => ACTIVE (%d) on thread %p", mName, this);
+        }
+
+        if (!isOutputTrack() && state != ACTIVE && state != RESUMING) {
+            thread->mLock.unlock();
+            status = AudioSystem::startOutput(thread->id(), (AudioSystem::stream_type)mStreamType);
+            thread->mLock.lock();
+        }
+        if (status == NO_ERROR) {
+            PlaybackThread *playbackThread = (PlaybackThread *)thread.get();
+            playbackThread->addTrack_l(this);
+        } else {
+            mState = state;
+        }
+    } else {
+        status = BAD_VALUE;
     }
-    return NO_ERROR;
+    return status;
 }
 
 void AudioFlinger::PlaybackThread::Track::stop()
@@ -2396,6 +2461,7 @@
     sp<ThreadBase> thread = mThread.promote();
     if (thread != 0) {
         Mutex::Autolock _l(thread->mLock);
+        int state = mState;
         if (mState > STOPPED) {
             mState = STOPPED;
             // If the track is not active (PAUSED and buffers full), flush buffers
@@ -2405,6 +2471,11 @@
             }
             LOGV("(> STOPPED) => STOPPED (%d) on thread %p", mName, playbackThread);
         }
+        if (!isOutputTrack() && (state == ACTIVE || state == RESUMING)) {
+            thread->mLock.unlock();
+            AudioSystem::stopOutput(thread->id(), (AudioSystem::stream_type)mStreamType);
+            thread->mLock.lock();
+        }
     }
 }
 
@@ -2417,6 +2488,11 @@
         if (mState == ACTIVE || mState == RESUMING) {
             mState = PAUSING;
             LOGV("ACTIVE/RESUMING => PAUSING (%d) on thread %p", mName, thread.get());
+            if (!isOutputTrack()) {
+                thread->mLock.unlock();
+                AudioSystem::stopOutput(thread->id(), (AudioSystem::stream_type)mStreamType);
+                thread->mLock.lock();
+            }
         }
     }
 }
@@ -2500,6 +2576,10 @@
 
 AudioFlinger::RecordThread::RecordTrack::~RecordTrack()
 {
+    sp<ThreadBase> thread = mThread.promote();
+    if (thread != 0) {
+        AudioSystem::releaseInput(thread->id());
+    }
 }
 
 status_t AudioFlinger::RecordThread::RecordTrack::getNextBuffer(AudioBufferProvider::Buffer* buffer)
@@ -2547,8 +2627,9 @@
     if (thread != 0) {
         RecordThread *recordThread = (RecordThread *)thread.get();
         return recordThread->start(this);
+    } else {
+        return BAD_VALUE;
     }
-    return NO_INIT;
 }
 
 void AudioFlinger::RecordThread::RecordTrack::stop()
@@ -2582,12 +2663,13 @@
 
 AudioFlinger::PlaybackThread::OutputTrack::OutputTrack(
             const wp<ThreadBase>& thread,
+            DuplicatingThread *sourceThread,
             uint32_t sampleRate,
             int format,
             int channelCount,
             int frameCount)
     :   Track(thread, NULL, AudioSystem::NUM_STREAM_TYPES, sampleRate, format, channelCount, frameCount, NULL),
-    mActive(false)
+    mActive(false), mSourceThread(sourceThread)
 {
 
     PlaybackThread *playbackThread = (PlaybackThread *)thread.unsafe_get();
@@ -2596,10 +2678,9 @@
         mCblk->buffers = (char*)mCblk + sizeof(audio_track_cblk_t);
         mCblk->volume[0] = mCblk->volume[1] = 0x1000;
         mOutBuffer.frameCount = 0;
-        mWaitTimeMs = (playbackThread->frameCount() * 2 * 1000) / playbackThread->sampleRate();
         playbackThread->mTracks.add(this);
-        LOGV("OutputTrack constructor mCblk %p, mBuffer %p, mCblk->buffers %p, mCblk->frameCount %d, mCblk->sampleRate %d, mCblk->channels %d mBufferEnd %p mWaitTimeMs %d",
-                mCblk, mBuffer, mCblk->buffers, mCblk->frameCount, mCblk->sampleRate, mCblk->channels, mBufferEnd, mWaitTimeMs);
+        LOGV("OutputTrack constructor mCblk %p, mBuffer %p, mCblk->buffers %p, mCblk->frameCount %d, mCblk->sampleRate %d, mCblk->channels %d mBufferEnd %p",
+                mCblk, mBuffer, mCblk->buffers, mCblk->frameCount, mCblk->sampleRate, mCblk->channels, mBufferEnd);
     } else {
         LOGW("Error creating output track on thread %p", playbackThread);
     }
@@ -2639,7 +2720,7 @@
     inBuffer.frameCount = frames;
     inBuffer.i16 = data;
 
-    uint32_t waitTimeLeftMs = mWaitTimeMs;
+    uint32_t waitTimeLeftMs = mSourceThread->waitTimeMs();
 
     if (!mActive && frames != 0) {
         start();
@@ -2678,12 +2759,11 @@
             mOutBuffer.frameCount = pInBuffer->frameCount;
             nsecs_t startTime = systemTime();
             if (obtainBuffer(&mOutBuffer, waitTimeLeftMs) == (status_t)AudioTrack::NO_MORE_BUFFERS) {
-                LOGV ("OutputTrack::write() %p no more output buffers", this);
+                LOGV ("OutputTrack::write() %p thread %p no more output buffers", this, mThread.unsafe_get());
                 outputBufferFull = true;
                 break;
             }
             uint32_t waitTimeMs = (uint32_t)ns2ms(systemTime() - startTime);
-            LOGV("OutputTrack::write() to thread %p waitTimeMs %d waitTimeLeftMs %d", mThread.unsafe_get(), waitTimeMs, waitTimeLeftMs);
             if (waitTimeLeftMs >= waitTimeMs) {
                 waitTimeLeftMs -= waitTimeMs;
             } else {
@@ -2704,7 +2784,7 @@
                 mBufferQueue.removeAt(0);
                 delete [] pInBuffer->mBuffer;
                 delete pInBuffer;
-                LOGV("OutputTrack::write() %p released overflow buffer %d", this, mBufferQueue.size());
+                LOGV("OutputTrack::write() %p thread %p released overflow buffer %d", this, mThread.unsafe_get(), mBufferQueue.size());
             } else {
                 break;
             }
@@ -2713,16 +2793,19 @@
 
     // If we could not write all frames, allocate a buffer and queue it for next time.
     if (inBuffer.frameCount) {
-        if (mBufferQueue.size() < kMaxOverFlowBuffers) {
-            pInBuffer = new Buffer;
-            pInBuffer->mBuffer = new int16_t[inBuffer.frameCount * channels];
-            pInBuffer->frameCount = inBuffer.frameCount;
-            pInBuffer->i16 = pInBuffer->mBuffer;
-            memcpy(pInBuffer->raw, inBuffer.raw, inBuffer.frameCount * channels * sizeof(int16_t));
-            mBufferQueue.add(pInBuffer);
-            LOGV("OutputTrack::write() %p adding overflow buffer %d", this, mBufferQueue.size());
-        } else {
-            LOGW("OutputTrack::write() %p no more overflow buffers", this);
+        sp<ThreadBase> thread = mThread.promote();
+        if (thread != 0 && !thread->standby()) {
+            if (mBufferQueue.size() < kMaxOverFlowBuffers) {
+                pInBuffer = new Buffer;
+                pInBuffer->mBuffer = new int16_t[inBuffer.frameCount * channels];
+                pInBuffer->frameCount = inBuffer.frameCount;
+                pInBuffer->i16 = pInBuffer->mBuffer;
+                memcpy(pInBuffer->raw, inBuffer.raw, inBuffer.frameCount * channels * sizeof(int16_t));
+                mBufferQueue.add(pInBuffer);
+                LOGV("OutputTrack::write() %p thread %p adding overflow buffer %d", this, mThread.unsafe_get(), mBufferQueue.size());
+            } else {
+                LOGW("OutputTrack::write() %p thread %p no more overflow buffers", mThread.unsafe_get(), this);
+            }
         }
     }
 
@@ -2985,8 +3068,8 @@
 
 // ----------------------------------------------------------------------------
 
-AudioFlinger::RecordThread::RecordThread(const sp<AudioFlinger>& audioFlinger, AudioStreamIn *input, uint32_t sampleRate, uint32_t channels) :
-    ThreadBase(audioFlinger),
+AudioFlinger::RecordThread::RecordThread(const sp<AudioFlinger>& audioFlinger, AudioStreamIn *input, uint32_t sampleRate, uint32_t channels, int id) :
+    ThreadBase(audioFlinger, id),
     mInput(input), mResampler(0), mRsmpOutBuffer(0), mRsmpInBuffer(0)
 {
     mReqChannelCount = AudioSystem::popCount(channels);
@@ -3014,6 +3097,7 @@
 
     run(buffer, PRIORITY_URGENT_AUDIO);
 }
+
 bool AudioFlinger::RecordThread::threadLoop()
 {
     AudioBufferProvider::Buffer buffer;
@@ -3043,22 +3127,37 @@
             }
             if (mActiveTrack != 0) {
                 if (mActiveTrack->mState == TrackBase::PAUSING) {
+                    if (!mStandby) {
+                        mInput->standby();
+                        mStandby = true;
+                    }
                     mActiveTrack.clear();
                     mStartStopCond.broadcast();
                 } else if (mActiveTrack->mState == TrackBase::RESUMING) {
-                    mRsmpInIndex = mFrameCount;
                     if (mReqChannelCount != mActiveTrack->channelCount()) {
                         mActiveTrack.clear();
-                    } else {
-                        mActiveTrack->mState = TrackBase::ACTIVE;
+                        mStartStopCond.broadcast();
+                    } else if (mBytesRead != 0) {
+                        // record start succeeds only if first read from audio input
+                        // succeeds
+                        if (mBytesRead > 0) {
+                            mActiveTrack->mState = TrackBase::ACTIVE;
+                        } else {
+                            mActiveTrack.clear();
+                        }
+                        mStartStopCond.broadcast();
                     }
-                    mStartStopCond.broadcast();
+                    mStandby = false;
                 }
-                mStandby = false;
             }
         }
 
         if (mActiveTrack != 0) {
+            if (mActiveTrack->mState != TrackBase::ACTIVE &&
+                mActiveTrack->mState != TrackBase::RESUMING) {
+                usleep(5000);
+                continue;
+            }
             buffer.frameCount = mFrameCount;
             if (LIKELY(mActiveTrack->getNextBuffer(&buffer) == NO_ERROR)) {
                 size_t framesOut = buffer.frameCount;
@@ -3093,18 +3192,19 @@
                             }
                         }
                         if (framesOut && mFrameCount == mRsmpInIndex) {
-                            ssize_t bytesRead;
                             if (framesOut == mFrameCount &&
                                 (mChannelCount == mReqChannelCount || mFormat != AudioSystem::PCM_16_BIT)) {
-                                bytesRead = mInput->read(buffer.raw, mInputBytes);
+                                mBytesRead = mInput->read(buffer.raw, mInputBytes);
                                 framesOut = 0;
                             } else {
-                                bytesRead = mInput->read(mRsmpInBuffer, mInputBytes);
+                                mBytesRead = mInput->read(mRsmpInBuffer, mInputBytes);
                                 mRsmpInIndex = 0;
                             }
-                            if (bytesRead < 0) {
+                            if (mBytesRead < 0) {
                                 LOGE("Error reading audio input");
-                                sleep(1);
+                                if (mActiveTrack->mState == TrackBase::ACTIVE) {
+                                    sleep(1);
+                                }
                                 mRsmpInIndex = mFrameCount;
                                 framesOut = 0;
                                 buffer.frameCount = 0;
@@ -3156,6 +3256,8 @@
     }
     mActiveTrack.clear();
 
+    mStartStopCond.broadcast();
+
     LOGV("RecordThread %p exiting", this);
     return false;
 }
@@ -3163,37 +3265,74 @@
 status_t AudioFlinger::RecordThread::start(RecordThread::RecordTrack* recordTrack)
 {
     LOGV("RecordThread::start");
-    AutoMutex lock(&mLock);
+    sp <ThreadBase> strongMe = this;
+    status_t status = NO_ERROR;
+    {
+        AutoMutex lock(&mLock);
+        if (mActiveTrack != 0) {
+            if (recordTrack != mActiveTrack.get()) {
+                status = -EBUSY;
+            } else if (mActiveTrack->mState == TrackBase::PAUSING) {
+                mActiveTrack->mState = TrackBase::ACTIVE;
+            }
+            return status;
+        }
 
-    if (mActiveTrack != 0) {
-        if (recordTrack != mActiveTrack.get()) return -EBUSY;
-
-        if (mActiveTrack->mState == TrackBase::PAUSING) mActiveTrack->mState = TrackBase::RESUMING;
-
-        return NO_ERROR;
-    }
-
-    mActiveTrack = recordTrack;
-    mActiveTrack->mState = TrackBase::RESUMING;
-    // signal thread to start
-    LOGV("Signal record thread");
-    mWaitWorkCV.signal();
-    mStartStopCond.wait(mLock);
-    if (mActiveTrack != 0) {
+        recordTrack->mState = TrackBase::IDLE;
+        mActiveTrack = recordTrack;
+        mLock.unlock();
+        status_t status = AudioSystem::startInput(mId);
+        mLock.lock();
+        if (status != NO_ERROR) {
+            mActiveTrack.clear();
+            return status;
+        }
+        mActiveTrack->mState = TrackBase::RESUMING;
+        mRsmpInIndex = mFrameCount;
+        mBytesRead = 0;
+        // signal thread to start
+        LOGV("Signal record thread");
+        mWaitWorkCV.signal();
+        // do not wait for mStartStopCond if exiting
+        if (mExiting) {
+            mActiveTrack.clear();
+            status = INVALID_OPERATION;
+            goto startError;
+        }
+        mStartStopCond.wait(mLock);
+        if (mActiveTrack == 0) {
+            LOGV("Record failed to start");
+            status = BAD_VALUE;
+            goto startError;
+        }
         LOGV("Record started OK");
-        return NO_ERROR;
-    } else {
-        LOGV("Record failed to start");
-        return BAD_VALUE;
+        return status;
     }
+startError:
+    AudioSystem::stopInput(mId);
+    return status;
 }
 
 void AudioFlinger::RecordThread::stop(RecordThread::RecordTrack* recordTrack) {
     LOGV("RecordThread::stop");
-    AutoMutex lock(&mLock);
-    if (mActiveTrack != 0 && recordTrack == mActiveTrack.get()) {
-        mActiveTrack->mState = TrackBase::PAUSING;
-        mStartStopCond.wait(mLock);
+    sp <ThreadBase> strongMe = this;
+    {
+        AutoMutex lock(&mLock);
+        if (mActiveTrack != 0 && recordTrack == mActiveTrack.get()) {
+            mActiveTrack->mState = TrackBase::PAUSING;
+            // do not wait for mStartStopCond if exiting
+            if (mExiting) {
+                return;
+            }
+            mStartStopCond.wait(mLock);
+            // if we have been restarted, recordTrack == mActiveTrack.get() here
+            if (mActiveTrack == 0 || recordTrack != mActiveTrack.get()) {
+                mLock.unlock();
+                AudioSystem::stopInput(mId);
+                mLock.lock();
+                LOGV("Record stopped OK");
+            }
+        }
     }
 }
 
@@ -3242,10 +3381,12 @@
     int channelCount;
 
     if (framesReady == 0) {
-        ssize_t bytesRead = mInput->read(mRsmpInBuffer, mInputBytes);
-        if (bytesRead < 0) {
+        mBytesRead = mInput->read(mRsmpInBuffer, mInputBytes);
+        if (mBytesRead < 0) {
             LOGE("RecordThread::getNextBuffer() Error reading audio input");
-            sleep(1);
+            if (mActiveTrack->mState == TrackBase::ACTIVE) {
+                sleep(1);
+            }
             buffer->raw = 0;
             buffer->frameCount = 0;
             return NOT_ENOUGH_DATA;
@@ -3363,7 +3504,7 @@
         break;
     }
     Mutex::Autolock _l(mAudioFlinger->mLock);
-    mAudioFlinger->audioConfigChanged_l(event, this, param2);
+    mAudioFlinger->audioConfigChanged_l(event, mId, param2);
 }
 
 void AudioFlinger::RecordThread::readInputParameters()
@@ -3451,21 +3592,23 @@
         if ((flags & AudioSystem::OUTPUT_FLAG_DIRECT) ||
             (format != AudioSystem::PCM_16_BIT) ||
             (channels != AudioSystem::CHANNEL_OUT_STEREO)) {
-            thread = new DirectOutputThread(this, output);
-            LOGV("openOutput() created direct output: ID %d thread %p", (mNextThreadId + 1), thread);
+            thread = new DirectOutputThread(this, output, ++mNextThreadId);
+            LOGV("openOutput() created direct output: ID %d thread %p", mNextThreadId, thread);
         } else {
-            thread = new MixerThread(this, output);
-            LOGV("openOutput() created mixer output: ID %d thread %p", (mNextThreadId + 1), thread);
+            thread = new MixerThread(this, output, ++mNextThreadId);
+            LOGV("openOutput() created mixer output: ID %d thread %p", mNextThreadId, thread);
         }
-        mPlaybackThreads.add(++mNextThreadId, thread);
+        mPlaybackThreads.add(mNextThreadId, thread);
 
         if (pSamplingRate) *pSamplingRate = samplingRate;
         if (pFormat) *pFormat = format;
         if (pChannels) *pChannels = channels;
         if (pLatencyMs) *pLatencyMs = thread->latency();
+
+        return mNextThreadId;
     }
 
-    return mNextThreadId;
+    return 0;
 }
 
 int AudioFlinger::openDuplicateOutput(int output1, int output2)
@@ -3480,9 +3623,9 @@
     }
 
 
-    DuplicatingThread *thread = new DuplicatingThread(this, thread1);
+    DuplicatingThread *thread = new DuplicatingThread(this, thread1, ++mNextThreadId);
     thread->addOutputTrack(thread2);
-    mPlaybackThreads.add(++mNextThreadId, thread);
+    mPlaybackThreads.add(mNextThreadId, thread);
     return mNextThreadId;
 }
 
@@ -3509,7 +3652,7 @@
             }
         }
         void *param2 = 0;
-        audioConfigChanged_l(AudioSystem::OUTPUT_CLOSED, thread, param2);
+        audioConfigChanged_l(AudioSystem::OUTPUT_CLOSED, output, param2);
         mPlaybackThreads.removeItem(output);
     }
     thread->exit();
@@ -3603,17 +3746,19 @@
 
     if (input != 0) {
          // Start record thread
-        thread = new RecordThread(this, input, reqSamplingRate, reqChannels);
-        mRecordThreads.add(++mNextThreadId, thread);
+        thread = new RecordThread(this, input, reqSamplingRate, reqChannels, ++mNextThreadId);
+        mRecordThreads.add(mNextThreadId, thread);
         LOGV("openInput() created record thread: ID %d thread %p", mNextThreadId, thread);
         if (pSamplingRate) *pSamplingRate = reqSamplingRate;
         if (pFormat) *pFormat = format;
         if (pChannels) *pChannels = reqChannels;
 
         input->standby();
+
+        return mNextThreadId;
     }
 
-    return mNextThreadId;
+    return 0;
 }
 
 status_t AudioFlinger::closeInput(int input)
@@ -3630,7 +3775,7 @@
 
         LOGV("closeInput() %d", input);
         void *param2 = 0;
-        audioConfigChanged_l(AudioSystem::INPUT_CLOSED, thread, param2);
+        audioConfigChanged_l(AudioSystem::INPUT_CLOSED, input, param2);
         mRecordThreads.removeItem(input);
     }
     thread->exit();
diff --git a/libs/audioflinger/AudioFlinger.h b/libs/audioflinger/AudioFlinger.h
index 22d15c9..12c90eb 100644
--- a/libs/audioflinger/AudioFlinger.h
+++ b/libs/audioflinger/AudioFlinger.h
@@ -20,6 +20,7 @@
 
 #include <stdint.h>
 #include <sys/types.h>
+#include <limits.h>
 
 #include <media/IAudioFlinger.h>
 #include <media/IAudioFlingerClient.h>
@@ -208,12 +209,13 @@
     class PlaybackThread;
     class MixerThread;
     class DirectOutputThread;
+    class DuplicatingThread;
     class Track;
     class RecordTrack;
 
     class ThreadBase : public Thread {
     public:
-        ThreadBase (const sp<AudioFlinger>& audioFlinger);
+        ThreadBase (const sp<AudioFlinger>& audioFlinger, int id);
         virtual             ~ThreadBase();
 
         status_t dumpBase(int fd, const Vector<String16>& args);
@@ -323,6 +325,8 @@
                     void        sendConfigEvent(int event, int param = 0);
                     void        sendConfigEvent_l(int event, int param = 0);
                     void        processConfigEvents();
+                    int         id() const { return mId;}
+                    bool        standby() { return mStandby; }
 
         mutable     Mutex                   mLock;
 
@@ -349,6 +353,8 @@
                     status_t                mParamStatus;
                     Vector<ConfigEvent *>   mConfigEvents;
                     bool                    mStandby;
+                    int                     mId;
+                    bool                    mExiting;
     };
 
     // --- PlaybackThread ---
@@ -361,6 +367,12 @@
             DUPLICATING
         };
 
+        enum mixer_state {
+            MIXER_IDLE,
+            MIXER_TRACKS_ENABLED,
+            MIXER_TRACKS_READY
+        };
+
         // playback track
         class Track : public TrackBase {
         public:
@@ -415,6 +427,10 @@
             void setPaused() { mState = PAUSED; }
             void reset();
 
+            bool isOutputTrack() const {
+                return (mStreamType == AudioSystem::NUM_STREAM_TYPES);
+            }
+
             // we don't really need a lock for these
             float               mVolume[2];
             volatile bool       mMute;
@@ -439,6 +455,7 @@
             };
 
                                 OutputTrack(  const wp<ThreadBase>& thread,
+                                        DuplicatingThread *sourceThread,
                                         uint32_t sampleRate,
                                         int format,
                                         int channelCount,
@@ -458,16 +475,15 @@
             void                clearBufferQueue();
 
             // Maximum number of pending buffers allocated by OutputTrack::write()
-            static const uint8_t kMaxOverFlowBuffers = 3;
+            static const uint8_t kMaxOverFlowBuffers = 10;
 
             Vector < Buffer* >          mBufferQueue;
             AudioBufferProvider::Buffer mOutBuffer;
-            uint32_t                    mWaitTimeMs;
             bool                        mActive;
-
+            DuplicatingThread*          mSourceThread;
         };  // end of OutputTrack
 
-        PlaybackThread (const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output);
+        PlaybackThread (const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output, int id);
         virtual             ~PlaybackThread();
 
         virtual     status_t    dump(int fd, const Vector<String16>& args);
@@ -507,6 +523,7 @@
         virtual     int         type() const { return mType; }
                     void        suspend() { mSuspended++; }
                     void        restore() { if (mSuspended) mSuspended--; }
+                    bool        isSuspended() { return (mSuspended != 0); }
         virtual     String8     getParameters(const String8& keys);
         virtual     void        audioConfigChanged(int event, int param = 0);
 
@@ -530,7 +547,8 @@
 
         virtual int             getTrackName_l() = 0;
         virtual void            deleteTrackName_l(int name) = 0;
-        virtual uint32_t        getMaxBufferRecoveryInUsecs() = 0;
+        virtual uint32_t        activeSleepTimeUs() = 0;
+        virtual uint32_t        idleSleepTimeUs() = 0;
 
     private:
 
@@ -562,12 +580,11 @@
         int                             mNumWrites;
         int                             mNumDelayedWrites;
         bool                            mInWrite;
-        int                             mMinBytesToWrite;
     };
 
     class MixerThread : public PlaybackThread {
     public:
-        MixerThread (const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output);
+        MixerThread (const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output, int id);
         virtual             ~MixerThread();
 
         // Thread virtuals
@@ -582,10 +599,11 @@
         virtual     status_t    dumpInternals(int fd, const Vector<String16>& args);
 
     protected:
-        size_t prepareTracks_l(const SortedVector< wp<Track> >& activeTracks, Vector< sp<Track> > *tracksToRemove);
+                    uint32_t    prepareTracks_l(const SortedVector< wp<Track> >& activeTracks, Vector< sp<Track> > *tracksToRemove);
         virtual     int         getTrackName_l();
         virtual     void        deleteTrackName_l(int name);
-        virtual     uint32_t    getMaxBufferRecoveryInUsecs();
+        virtual     uint32_t    activeSleepTimeUs();
+        virtual     uint32_t    idleSleepTimeUs();
 
         AudioMixer*                     mAudioMixer;
     };
@@ -593,7 +611,7 @@
     class DirectOutputThread : public PlaybackThread {
     public:
 
-        DirectOutputThread (const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output);
+        DirectOutputThread (const sp<AudioFlinger>& audioFlinger, AudioStreamOut* output, int id);
         ~DirectOutputThread();
 
         // Thread virtuals
@@ -604,7 +622,8 @@
     protected:
         virtual     int         getTrackName_l();
         virtual     void        deleteTrackName_l(int name);
-        virtual     uint32_t    getMaxBufferRecoveryInUsecs();
+        virtual     uint32_t    activeSleepTimeUs();
+        virtual     uint32_t    idleSleepTimeUs();
 
     private:
         float mLeftVolume;
@@ -613,23 +632,30 @@
 
     class DuplicatingThread : public MixerThread {
     public:
-        DuplicatingThread (const sp<AudioFlinger>& audioFlinger, MixerThread* mainThread);
+        DuplicatingThread (const sp<AudioFlinger>& audioFlinger, MixerThread* mainThread, int id);
         ~DuplicatingThread();
 
         // Thread virtuals
         virtual     bool        threadLoop();
                     void        addOutputTrack(MixerThread* thread);
                     void        removeOutputTrack(MixerThread* thread);
+                    uint32_t    waitTimeMs() { return mWaitTimeMs; }
+    protected:
+        virtual     uint32_t    activeSleepTimeUs();
 
     private:
+                    bool        outputsReady(SortedVector< sp<OutputTrack> > &outputTracks);
+                    void        updateWaitTime();
+
         SortedVector < sp<OutputTrack> >  mOutputTracks;
+                    uint32_t    mWaitTimeMs;
     };
 
               PlaybackThread *checkPlaybackThread_l(int output) const;
               MixerThread *checkMixerThread_l(int output) const;
               RecordThread *checkRecordThread_l(int input) const;
               float streamVolumeInternal(int stream) const { return mStreamTypes[stream].volume; }
-              void audioConfigChanged_l(int event, const sp<ThreadBase>& thread, void *param2);
+              void audioConfigChanged_l(int event, int ioHandle, void *param2);
 
     friend class AudioBuffer;
 
@@ -697,7 +723,8 @@
                 RecordThread(const sp<AudioFlinger>& audioFlinger,
                         AudioStreamIn *input,
                         uint32_t sampleRate,
-                        uint32_t channels);
+                        uint32_t channels,
+                        int id);
                 ~RecordThread();
 
         virtual bool        threadLoop();
@@ -728,6 +755,7 @@
                 size_t                              mInputBytes;
                 int                                 mReqChannelCount;
                 uint32_t                            mReqSampleRate;
+                ssize_t                             mBytesRead;
     };
 
     class RecordHandle : public android::BnAudioRecord {
diff --git a/libs/binder/ProcessState.cpp b/libs/binder/ProcessState.cpp
index d7daf73..2d4e10d 100644
--- a/libs/binder/ProcessState.cpp
+++ b/libs/binder/ProcessState.cpp
@@ -41,7 +41,7 @@
 #include <sys/mman.h>
 #include <sys/stat.h>
 
-#define BINDER_VM_SIZE (1*1024*1024)
+#define BINDER_VM_SIZE ((1*1024*1024) - (4096 *2))
 
 static bool gSingleProcess = false;
 
diff --git a/libs/surfaceflinger/Android.mk b/libs/surfaceflinger/Android.mk
index eb51c22..b3fed58 100644
--- a/libs/surfaceflinger/Android.mk
+++ b/libs/surfaceflinger/Android.mk
@@ -22,9 +22,6 @@
 ifeq ($(TARGET_BOARD_PLATFORM), msm7k)
 	LOCAL_CFLAGS += -DDIM_WITH_TEXTURE
 endif
-ifeq ($(TARGET_BOARD_PLATFORM), qsd8k)
-	LOCAL_CFLAGS += -DDIM_WITH_TEXTURE
-endif
 
 # need "-lrt" on Linux simulator to pick up clock_gettime
 ifeq ($(TARGET_SIMULATOR),true)
diff --git a/libs/surfaceflinger/DisplayHardware/DisplayHardware.cpp b/libs/surfaceflinger/DisplayHardware/DisplayHardware.cpp
index 1abfd68..d9d9bfe 100644
--- a/libs/surfaceflinger/DisplayHardware/DisplayHardware.cpp
+++ b/libs/surfaceflinger/DisplayHardware/DisplayHardware.cpp
@@ -246,10 +246,11 @@
     LOGI("version   : %s", glGetString(GL_VERSION));
     LOGI("extensions: %s", gl_extensions);
 
-    if (strstr(gl_renderer, "PowerVR SGX 530")) {
+    if (strstr(gl_renderer, "Adreno")) {
         LOGD("Assuming uncached graphics buffers.");
         mFlags &= ~CACHED_BUFFERS;
     }
+
     if (strstr(gl_extensions, "GL_ARB_texture_non_power_of_two")) {
         mFlags |= NPOT_EXTENSION;
     }
diff --git a/libs/surfaceflinger/Layer.cpp b/libs/surfaceflinger/Layer.cpp
index f5a5a0b..1870d3a 100644
--- a/libs/surfaceflinger/Layer.cpp
+++ b/libs/surfaceflinger/Layer.cpp
@@ -93,6 +93,7 @@
 status_t Layer::ditch()
 {
     // the layer is not on screen anymore. free as much resources as possible
+    mFreezeLock.clear();
     destroy();
     return NO_ERROR;
 }
@@ -134,7 +135,14 @@
 {
     Mutex::Autolock _l(mLock);
     sp<GraphicBuffer> buffer(getFrontBufferLocked());
-    int index = mFrontBufferIndex;
+    if (buffer == NULL) {
+        // this situation can happen if we ran out of memory for instance.
+        // not much we can do. continue to use whatever texture was bound
+        // to this context.
+        return;
+    }
+
+    const int index = mFrontBufferIndex;
 
     // create the new texture name if needed
     if (UNLIKELY(mTextures[index].name == -1U)) {
@@ -515,6 +523,11 @@
         dirtyRegion.andSelf(visibleRegionScreen);
         outDirtyRegion.orSelf(dirtyRegion);
     }
+    if (visibleRegionScreen.isEmpty()) {
+        // an invisible layer should not hold a freeze-lock
+        // (because it may never be updated and thereore never release it)
+        mFreezeLock.clear();
+    }
 }
 
 void Layer::finishPageFlip()
diff --git a/libs/surfaceflinger/LayerBuffer.cpp b/libs/surfaceflinger/LayerBuffer.cpp
index a36304c..2ff6167 100644
--- a/libs/surfaceflinger/LayerBuffer.cpp
+++ b/libs/surfaceflinger/LayerBuffer.cpp
@@ -26,6 +26,8 @@
 #include <ui/GraphicBuffer.h>
 #include <ui/PixelFormat.h>
 #include <ui/FramebufferNativeWindow.h>
+#include <ui/Rect.h>
+#include <ui/Region.h>
 
 #include <hardware/copybit.h>
 
@@ -46,12 +48,15 @@
 LayerBuffer::LayerBuffer(SurfaceFlinger* flinger, DisplayID display,
         const sp<Client>& client, int32_t i)
     : LayerBaseClient(flinger, display, client, i),
-      mNeedsBlending(false)
+      mNeedsBlending(false), mBlitEngine(0)
 {
 }
 
 LayerBuffer::~LayerBuffer()
 {
+    if (mBlitEngine) {
+        copybit_close(mBlitEngine);
+    }
 }
 
 void LayerBuffer::onFirstRef()
@@ -69,6 +74,10 @@
             sGrallocModule = (gralloc_module_t const *)module;
         }
     }
+
+    if (hw_get_module(COPYBIT_HARDWARE_MODULE_ID, &module) == 0) {
+        copybit_open(module, &mBlitEngine);
+    }
 }
 
 sp<LayerBaseClient::Surface> LayerBuffer::createSurface() const
@@ -109,7 +118,10 @@
     sp<Source> source(getSource());
     if (source != 0)
         source->onTransaction(flags);
-    return LayerBase::doTransaction(flags);    
+    uint32_t res = LayerBase::doTransaction(flags);
+    // we always want filtering for these surfaces
+    mUseLinearFiltering = !(mFlags & DisplayHardware::SLOW_CONFIG);
+    return res;
 }
 
 void LayerBuffer::unlockPageFlip(const Transform& planeTransform,
@@ -252,7 +264,16 @@
     : mBufferHeap(buffers)
 {
     NativeBuffer& src(mNativeBuffer);
-    src.img.handle = 0;
+    src.crop.l = 0;
+    src.crop.t = 0;
+    src.crop.r = buffers.w;
+    src.crop.b = buffers.h;
+
+    src.img.w       = buffers.hor_stride ?: buffers.w;
+    src.img.h       = buffers.ver_stride ?: buffers.h;
+    src.img.format  = buffers.format;
+    src.img.base    = (void*)(intptr_t(buffers.heap->base()) + offset);
+    src.img.handle  = 0;
 
     gralloc_module_t const * module = LayerBuffer::getGrallocModule();
     if (module && module->perform) {
@@ -262,19 +283,12 @@
                 offset, buffers.heap->base(),
                 &src.img.handle);
 
-        if (err == NO_ERROR) {
-            src.crop.l = 0;
-            src.crop.t = 0;
-            src.crop.r = buffers.w;
-            src.crop.b = buffers.h;
-
-            src.img.w       = buffers.hor_stride ?: buffers.w;
-            src.img.h       = buffers.ver_stride ?: buffers.h;
-            src.img.format  = buffers.format;
-            src.img.base    = (void*)(intptr_t(buffers.heap->base()) + offset);
-        }
+        LOGE_IF(err, "CREATE_HANDLE_FROM_BUFFER (heapId=%d, size=%d, "
+             "offset=%ld, base=%p) failed (%s)",
+                buffers.heap->heapID(), buffers.heap->getSize(),
+                offset, buffers.heap->base(), strerror(-err));
     }
-}
+ }
 
 LayerBuffer::Buffer::~Buffer()
 {
@@ -438,15 +452,24 @@
 
 #if defined(EGL_ANDROID_image_native_buffer)
     if (mLayer.mFlags & DisplayHardware::DIRECT_TEXTURE) {
-         // NOTE: Assume the buffer is  allocated with the proper USAGE flags
-        sp<GraphicBuffer> graphicBuffer = new GraphicBuffer(
-                src.crop.r, src.crop.b, src.img.format, 
-                GraphicBuffer::USAGE_HW_TEXTURE,
-                src.img.w, src.img.handle, false);
+        copybit_device_t* copybit = mLayer.mBlitEngine;
+        if (copybit) {
+            // create our EGLImageKHR the first time
+            err = initTempBuffer();
+            if (err == NO_ERROR) {
+                // NOTE: Assume the buffer is allocated with the proper USAGE flags
+                const NativeBuffer& dst(mTempBuffer);
+                region_iterator clip(Region(Rect(dst.crop.r, dst.crop.b)));
+                copybit->set_parameter(copybit, COPYBIT_TRANSFORM, 0);
+                copybit->set_parameter(copybit, COPYBIT_PLANE_ALPHA, 0xFF);
+                copybit->set_parameter(copybit, COPYBIT_DITHER, COPYBIT_ENABLE);
+                err = copybit->stretch(copybit, &dst.img, &src.img,
+                        &dst.crop, &src.crop, &clip);
 
-        graphicBuffer->setVerticalStride(src.img.h);
-
-        err = mLayer.initializeEglImage(graphicBuffer, &mTexture);
+            }
+        } else {
+            err = INVALID_OPERATION;
+        }
     }
 #endif
     else {
@@ -471,6 +494,72 @@
     mLayer.drawWithOpenGL(clip, mTexture);
 }
 
+status_t LayerBuffer::BufferSource::initTempBuffer() const
+{
+    // figure out the size we need now
+    const ISurface::BufferHeap& buffers(mBufferHeap);
+    uint32_t w = mLayer.mTransformedBounds.width();
+    uint32_t h = mLayer.mTransformedBounds.height();
+    if (buffers.w * h != buffers.h * w) {
+        int t = w; w = h; h = t;
+    }
+
+    if (mTexture.image != EGL_NO_IMAGE_KHR) {
+        // we have an EGLImage, make sure the needed size didn't change
+        if (w!=mTexture.width || h!= mTexture.height) {
+            // delete the EGLImage and texture
+            EGLDisplay dpy(mLayer.mFlinger->graphicPlane(0).getEGLDisplay());
+            glDeleteTextures(1, &mTexture.name);
+            eglDestroyImageKHR(dpy, mTexture.image);
+            Texture defaultTexture;
+            mTexture = defaultTexture;
+            mTempGraphicBuffer.clear();
+        } else {
+            // we're good, we have an EGLImageKHR and it's (still) the
+            // right size
+            return NO_ERROR;
+        }
+    }
+
+    // figure out if we need linear filtering
+    if (buffers.w * h == buffers.h * w) {
+        // same pixel area, don't use filtering
+        mLayer.mUseLinearFiltering = false;
+    }
+
+    // Allocate a temporary buffer and create the corresponding EGLImageKHR
+
+    status_t err;
+    mTempGraphicBuffer.clear();
+    mTempGraphicBuffer = new GraphicBuffer(
+            w, h, HAL_PIXEL_FORMAT_RGB_565,
+            GraphicBuffer::USAGE_HW_TEXTURE |
+            GraphicBuffer::USAGE_HW_2D);
+
+    err = mTempGraphicBuffer->initCheck();
+    if (err == NO_ERROR) {
+        NativeBuffer& dst(mTempBuffer);
+        dst.img.w = mTempGraphicBuffer->getStride();
+        dst.img.h = h;
+        dst.img.format = mTempGraphicBuffer->getPixelFormat();
+        dst.img.handle = (native_handle_t *)mTempGraphicBuffer->handle;
+        dst.img.base = 0;
+        dst.crop.l = 0;
+        dst.crop.t = 0;
+        dst.crop.r = w;
+        dst.crop.b = h;
+
+        err = mLayer.initializeEglImage(
+                mTempGraphicBuffer, &mTexture);
+        // once the EGLImage has been created (whether it fails
+        // or not) we don't need the graphic buffer reference
+        // anymore.
+        mTempGraphicBuffer.clear();
+    }
+
+    return err;
+}
+
 // ---------------------------------------------------------------------------
 
 LayerBuffer::OverlaySource::OverlaySource(LayerBuffer& layer,
diff --git a/libs/surfaceflinger/LayerBuffer.h b/libs/surfaceflinger/LayerBuffer.h
index 47482f4..2ca63ac 100644
--- a/libs/surfaceflinger/LayerBuffer.h
+++ b/libs/surfaceflinger/LayerBuffer.h
@@ -130,13 +130,15 @@
         virtual bool transformed() const;
         virtual void destroy() { }
     private:
+        status_t initTempBuffer() const;
         mutable Mutex                   mBufferSourceLock;
         sp<Buffer>                      mBuffer;
         status_t                        mStatus;
         ISurface::BufferHeap            mBufferHeap;
         size_t                          mBufferSize;
-        mutable sp<GraphicBuffer>       mTempBitmap;
         mutable LayerBase::Texture      mTexture;
+        mutable NativeBuffer            mTempBuffer;
+        mutable sp<GraphicBuffer>       mTempGraphicBuffer;
     };
     
     class OverlaySource : public Source {
@@ -205,6 +207,7 @@
     sp<Surface>     mSurface;
     bool            mInvalidate;
     bool            mNeedsBlending;
+    copybit_device_t* mBlitEngine;
 };
 
 // ---------------------------------------------------------------------------
diff --git a/libs/surfaceflinger/SurfaceFlinger.cpp b/libs/surfaceflinger/SurfaceFlinger.cpp
index 9694cf1..965b7dd 100644
--- a/libs/surfaceflinger/SurfaceFlinger.cpp
+++ b/libs/surfaceflinger/SurfaceFlinger.cpp
@@ -417,9 +417,9 @@
 {
     while (true) {
         nsecs_t timeout = -1;
+        const nsecs_t freezeDisplayTimeout = ms2ns(5000);
         if (UNLIKELY(isFrozen())) {
             // wait 5 seconds
-            const nsecs_t freezeDisplayTimeout = ms2ns(5000);
             const nsecs_t now = systemTime();
             if (mFreezeDisplayTime == 0) {
                 mFreezeDisplayTime = now;
@@ -429,23 +429,27 @@
         }
 
         MessageList::value_type msg = mEventQueue.waitMessage(timeout);
+
+        // see if we timed out
+        if (isFrozen()) {
+            const nsecs_t now = systemTime();
+            nsecs_t frozenTime = (now - mFreezeDisplayTime);
+            if (frozenTime >= freezeDisplayTimeout) {
+                // we timed out and are still frozen
+                LOGW("timeout expired mFreezeDisplay=%d, mFreezeCount=%d",
+                        mFreezeDisplay, mFreezeCount);
+                mFreezeDisplayTime = 0;
+                mFreezeCount = 0;
+                mFreezeDisplay = false;
+            }
+        }
+
         if (msg != 0) {
-            mFreezeDisplayTime = 0;
             switch (msg->what) {
                 case MessageQueue::INVALIDATE:
                     // invalidate message, just return to the main loop
                     return;
             }
-        } else {
-            // we timed out
-            if (isFrozen()) {
-                // we timed out and are still frozen
-                LOGW("timeout expired mFreezeDisplay=%d, mFreezeCount=%d",
-                        mFreezeDisplay, mFreezeCount);
-                mFreezeCount = 0;
-                mFreezeDisplay = false;
-                return;
-            }
         }
     }
 }
@@ -1646,6 +1650,7 @@
             }
             case 1007: // set mFreezeCount
                 mFreezeCount = data.readInt32();
+                mFreezeDisplayTime = 0;
                 return NO_ERROR;
             case 1010:  // interrogate.
                 reply->writeInt32(0);
diff --git a/libs/surfaceflinger/SurfaceFlinger.h b/libs/surfaceflinger/SurfaceFlinger.h
index f9bfe6c..c0ab73d 100644
--- a/libs/surfaceflinger/SurfaceFlinger.h
+++ b/libs/surfaceflinger/SurfaceFlinger.h
@@ -291,7 +291,11 @@
 
             friend class FreezeLock;
             sp<FreezeLock> getFreezeLock() const;
-            inline void incFreezeCount() { mFreezeCount++; }
+            inline void incFreezeCount() {
+                if (mFreezeCount == 0)
+                    mFreezeDisplayTime = 0;
+                mFreezeCount++;
+            }
             inline void decFreezeCount() { if (mFreezeCount > 0) mFreezeCount--; }
             inline bool hasFreezeRequest() const { return mFreezeDisplay; }
             inline bool isFrozen() const { 
diff --git a/libs/ui/EventHub.cpp b/libs/ui/EventHub.cpp
index e39a357..4aac455 100644
--- a/libs/ui/EventHub.cpp
+++ b/libs/ui/EventHub.cpp
@@ -489,6 +489,7 @@
 {
     int version;
     int fd;
+    int attempt;
     struct pollfd *new_mFDs;
     device_t **new_devices;
     char **new_device_names;
@@ -500,12 +501,17 @@
     LOGV("Opening device: %s", deviceName);
 
     AutoMutex _l(mLock);
-    
-    fd = open(deviceName, O_RDWR);
+
+    for (attempt = 0; attempt < 10; attempt++) {
+        fd = open(deviceName, O_RDWR);
+        if (fd >= 0) break;
+        usleep(100);
+    }
     if(fd < 0) {
         LOGE("could not open %s, %s\n", deviceName, strerror(errno));
         return -1;
     }
+    LOGV("Opened device: %s (%d failures)", deviceName, attempt);
 
     if(ioctl(fd, EVIOCGVERSION, &version)) {
         LOGE("could not get driver version for %s, %s\n", deviceName, strerror(errno));
diff --git a/libs/ui/FramebufferNativeWindow.cpp b/libs/ui/FramebufferNativeWindow.cpp
index 0efba9c..c5e22e5 100644
--- a/libs/ui/FramebufferNativeWindow.cpp
+++ b/libs/ui/FramebufferNativeWindow.cpp
@@ -118,8 +118,6 @@
         LOGE_IF(err, "fb buffer 1 allocation failed w=%d, h=%d, err=%s",
                 fbDev->width, fbDev->height, strerror(-err));
 
-        LOGE("xDpi %d", fbDev->xdpi);
-        LOGE("yDpi %d", fbDev->ydpi);
         const_cast<uint32_t&>(android_native_window_t::flags) = fbDev->flags; 
         const_cast<float&>(android_native_window_t::xdpi) = fbDev->xdpi;
         const_cast<float&>(android_native_window_t::ydpi) = fbDev->ydpi;
diff --git a/libs/ui/Surface.cpp b/libs/ui/Surface.cpp
index f51ca7a..24ae27f 100644
--- a/libs/ui/Surface.cpp
+++ b/libs/ui/Surface.cpp
@@ -152,96 +152,85 @@
 
 status_t SurfaceControl::setLayer(int32_t layer) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setLayer(mToken, layer);
 }
 status_t SurfaceControl::setPosition(int32_t x, int32_t y) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setPosition(mToken, x, y);
 }
 status_t SurfaceControl::setSize(uint32_t w, uint32_t h) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setSize(mToken, w, h);
 }
 status_t SurfaceControl::hide() {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->hide(mToken);
 }
 status_t SurfaceControl::show(int32_t layer) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->show(mToken, layer);
 }
 status_t SurfaceControl::freeze() {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->freeze(mToken);
 }
 status_t SurfaceControl::unfreeze() {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->unfreeze(mToken);
 }
 status_t SurfaceControl::setFlags(uint32_t flags, uint32_t mask) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setFlags(mToken, flags, mask);
 }
 status_t SurfaceControl::setTransparentRegionHint(const Region& transparent) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setTransparentRegionHint(mToken, transparent);
 }
 status_t SurfaceControl::setAlpha(float alpha) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setAlpha(mToken, alpha);
 }
 status_t SurfaceControl::setMatrix(float dsdx, float dtdx, float dsdy, float dtdy) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setMatrix(mToken, dsdx, dtdx, dsdy, dtdy);
 }
 status_t SurfaceControl::setFreezeTint(uint32_t tint) {
     const sp<SurfaceComposerClient>& client(mClient);
-    if (client == 0) return NO_INIT;
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err < 0) return err;
     return client->setFreezeTint(mToken, tint);
 }
 
-status_t SurfaceControl::validate(SharedClient const* cblk) const
+status_t SurfaceControl::validate() const
 {
     if (mToken<0 || mClient==0) {
         LOGE("invalid token (%d, identity=%u) or client (%p)", 
                 mToken, mIdentity, mClient.get());
         return NO_INIT;
     }
+    SharedClient const* cblk = mClient->mControl;
     if (cblk == 0) {
         LOGE("cblk is null (surface id=%d, identity=%u)", mToken, mIdentity);
         return NO_INIT;
@@ -394,7 +383,7 @@
     return mToken>=0 && mClient!=0;
 }
 
-status_t Surface::validate(SharedClient const* cblk) const
+status_t Surface::validate() const
 {
     sp<SurfaceComposerClient> client(getClient());
     if (mToken<0 || mClient==0) {
@@ -402,6 +391,7 @@
                 mToken, mIdentity, client.get());
         return NO_INIT;
     }
+    SharedClient const* cblk = mClient->mControl;
     if (cblk == 0) {
         LOGE("cblk is null (surface id=%d, identity=%u)", mToken, mIdentity);
         return NO_INIT;
@@ -488,7 +478,7 @@
 int Surface::dequeueBuffer(android_native_buffer_t** buffer)
 {
     sp<SurfaceComposerClient> client(getClient());
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err != NO_ERROR)
         return err;
 
@@ -533,7 +523,7 @@
 int Surface::lockBuffer(android_native_buffer_t* buffer)
 {
     sp<SurfaceComposerClient> client(getClient());
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err != NO_ERROR)
         return err;
 
@@ -546,7 +536,7 @@
 int Surface::queueBuffer(android_native_buffer_t* buffer)
 {   
     sp<SurfaceComposerClient> client(getClient());
-    status_t err = validate(client->mControl);
+    status_t err = validate();
     if (err != NO_ERROR)
         return err;
 
diff --git a/libs/utils/ResourceTypes.cpp b/libs/utils/ResourceTypes.cpp
index 872b2bc..450af8d 100644
--- a/libs/utils/ResourceTypes.cpp
+++ b/libs/utils/ResourceTypes.cpp
@@ -42,6 +42,7 @@
 #define TABLE_GETENTRY(x) //x
 #define TABLE_SUPER_NOISY(x) //x
 #define LOAD_TABLE_NOISY(x) //x
+#define TABLE_THEME(x) //x
 
 namespace android {
 
@@ -1447,18 +1448,23 @@
         const uint32_t t = Res_GETTYPE(resID);
         const uint32_t e = Res_GETENTRY(resID);
 
-        TABLE_NOISY(LOGV("Looking up attr 0x%08x in theme %p", resID, this));
+        TABLE_THEME(LOGI("Looking up attr 0x%08x in theme %p", resID, this));
 
         if (p >= 0) {
             const package_info* const pi = mPackages[p];
+            TABLE_THEME(LOGI("Found package: %p", pi));
             if (pi != NULL) {
+                TABLE_THEME(LOGI("Desired type index is %ld in avail %d", t, pi->numTypes));
                 if (t < pi->numTypes) {
                     const type_info& ti = pi->types[t];
+                    TABLE_THEME(LOGI("Desired entry index is %ld in avail %d", e, ti.numEntries));
                     if (e < ti.numEntries) {
                         const theme_entry& te = ti.entries[e];
-                            if (outTypeSpecFlags != NULL) {
-                                *outTypeSpecFlags |= te.typeSpecFlags;
-                            }
+                        if (outTypeSpecFlags != NULL) {
+                            *outTypeSpecFlags |= te.typeSpecFlags;
+                        }
+                        TABLE_THEME(LOGI("Theme value: type=0x%x, data=0x%08x",
+                                te.value.dataType, te.value.data));
                         const uint8_t type = te.value.dataType;
                         if (type == Res_value::TYPE_ATTRIBUTE) {
                             if (cnt > 0) {
@@ -1492,6 +1498,8 @@
     if (inOutValue->dataType == Res_value::TYPE_ATTRIBUTE) {
         uint32_t newTypeSpecFlags;
         blockIndex = getAttribute(inOutValue->data, inOutValue, &newTypeSpecFlags);
+        TABLE_THEME(LOGI("Resolving attr reference: blockIndex=%d, type=0x%x, data=%p\n",
+             (int)blockIndex, (int)inOutValue->dataType, (void*)inOutValue->data));
         if (inoutTypeSpecFlags != NULL) *inoutTypeSpecFlags |= newTypeSpecFlags;
         //printf("Retrieved attribute new type=0x%x\n", inOutValue->dataType);
         if (blockIndex < 0) {
@@ -1911,8 +1919,8 @@
         uint32_t newFlags = 0;
         const ssize_t newIndex = getResource(value->data, value, true, &newFlags,
                 outConfig);
-        //LOGI("Resolving reference d=%p: newIndex=%d, t=0x%02x, d=%p\n",
-        //     (void*)lastRef, (int)newIndex, (int)value->dataType, (void*)value->data);
+        TABLE_THEME(LOGI("Resolving reference %p: newIndex=%d, type=0x%x, data=%p\n",
+             (void*)lastRef, (int)newIndex, (int)value->dataType, (void*)value->data));
         //printf("Getting reference 0x%08x: newIndex=%d\n", value->data, newIndex);
         if (inoutTypeSpecFlags != NULL) *inoutTypeSpecFlags |= newFlags;
         if (newIndex < 0) {
diff --git a/opengl/include/ETC1/etc1.h b/opengl/include/ETC1/etc1.h
new file mode 100644
index 0000000..0d38905
--- /dev/null
+++ b/opengl/include/ETC1/etc1.h
@@ -0,0 +1,106 @@
+// Copyright 2009 Google Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef __etc1_h__
+#define __etc1_h__
+
+#define ETC1_ENCODED_BLOCK_SIZE 8
+#define ETC1_DECODED_BLOCK_SIZE 48
+
+#ifndef ETC1_RGB8_OES
+#define ETC1_RGB8_OES 0x8D64
+#endif
+
+typedef unsigned char etc1_byte;
+typedef int etc1_bool;
+typedef unsigned int etc1_uint32;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Encode a block of pixels.
+//
+// pIn is a pointer to a ETC_DECODED_BLOCK_SIZE array of bytes that represent a
+// 4 x 4 square of 3-byte pixels in form R, G, B. Byte (3 * (x + 4 * y) is the R
+// value of pixel (x, y).
+//
+// validPixelMask is a 16-bit mask where bit (1 << (x + y * 4)) indicates whether
+// the corresponding (x,y) pixel is valid. Invalid pixel color values are ignored when compressing.
+//
+// pOut is an ETC1 compressed version of the data.
+
+void etc1_encode_block(const etc1_byte* pIn, etc1_uint32 validPixelMask, etc1_byte* pOut);
+
+// Decode a block of pixels.
+//
+// pIn is an ETC1 compressed version of the data.
+//
+// pOut is a pointer to a ETC_DECODED_BLOCK_SIZE array of bytes that represent a
+// 4 x 4 square of 3-byte pixels in form R, G, B. Byte (3 * (x + 4 * y) is the R
+// value of pixel (x, y).
+
+void etc1_decode_block(const etc1_byte* pIn, etc1_byte* pOut);
+
+// Return the size of the encoded image data (does not include size of PKM header).
+
+etc1_uint32 etc1_get_encoded_data_size(etc1_uint32 width, etc1_uint32 height);
+
+// Encode an entire image.
+// pIn - pointer to the image data. Formatted such that
+//       pixel (x,y) is at pIn + pixelSize * x + stride * y;
+// pOut - pointer to encoded data. Must be large enough to store entire encoded image.
+// pixelSize can be 2 or 3. 2 is an GL_UNSIGNED_SHORT_5_6_5 image, 3 is a GL_BYTE RGB image.
+// returns non-zero if there is an error.
+
+int etc1_encode_image(const etc1_byte* pIn, etc1_uint32 width, etc1_uint32 height,
+        etc1_uint32 pixelSize, etc1_uint32 stride, etc1_byte* pOut);
+
+// Decode an entire image.
+// pIn - pointer to encoded data.
+// pOut - pointer to the image data. Will be written such that
+//        pixel (x,y) is at pIn + pixelSize * x + stride * y. Must be
+//        large enough to store entire image.
+// pixelSize can be 2 or 3. 2 is an GL_UNSIGNED_SHORT_5_6_5 image, 3 is a GL_BYTE RGB image.
+// returns non-zero if there is an error.
+
+int etc1_decode_image(const etc1_byte* pIn, etc1_byte* pOut,
+        etc1_uint32 width, etc1_uint32 height,
+        etc1_uint32 pixelSize, etc1_uint32 stride);
+
+// Size of a PKM header, in bytes.
+
+#define ETC_PKM_HEADER_SIZE 16
+
+// Format a PKM header
+
+void etc1_pkm_format_header(etc1_byte* pHeader, etc1_uint32 width, etc1_uint32 height);
+
+// Check if a PKM header is correctly formatted.
+
+etc1_bool etc1_pkm_is_valid(const etc1_byte* pHeader);
+
+// Read the image width from a PKM header
+
+etc1_uint32 etc1_pkm_get_width(const etc1_byte* pHeader);
+
+// Read the image height from a PKM header
+
+etc1_uint32 etc1_pkm_get_height(const etc1_byte* pHeader);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/opengl/libagl/array.cpp b/opengl/libagl/array.cpp
index 4878722..71825c5 100644
--- a/opengl/libagl/array.cpp
+++ b/opengl/libagl/array.cpp
@@ -1548,24 +1548,36 @@
         GLuint name = buffers[i];
         if (name) {
             // unbind bound deleted buffers...
-            if (c->arrays.element_array_buffer->name == name) {
-                c->arrays.element_array_buffer = 0;
+            if (c->arrays.element_array_buffer) {
+                if (c->arrays.element_array_buffer->name == name) {
+                    c->arrays.element_array_buffer = 0;
+                }
             }
-            if (c->arrays.array_buffer->name == name) {
-                c->arrays.array_buffer = 0;
+            if (c->arrays.array_buffer) {
+                if (c->arrays.array_buffer->name == name) {
+                    c->arrays.array_buffer = 0;
+                }
             }
-            if (c->arrays.vertex.bo->name == name) {
-                c->arrays.vertex.bo = 0;
+            if (c->arrays.vertex.bo) {
+                if (c->arrays.vertex.bo->name == name) {
+                    c->arrays.vertex.bo = 0;
+                }
             }
-            if (c->arrays.normal.bo->name == name) {
-                c->arrays.normal.bo = 0;
+            if (c->arrays.normal.bo) {
+                if (c->arrays.normal.bo->name == name) {
+                    c->arrays.normal.bo = 0;
+                }
             }
-            if (c->arrays.color.bo->name == name) {
-                c->arrays.color.bo = 0;
+            if (c->arrays.color.bo) {
+                if (c->arrays.color.bo->name == name) {
+                    c->arrays.color.bo = 0;
+                }
             }
             for (int t=0 ; t<GGL_TEXTURE_UNIT_COUNT ; t++) {
-                if (c->arrays.texture[t].bo->name == name) {
-                    c->arrays.texture[t].bo = 0;
+                if (c->arrays.texture[t].bo) {
+                    if (c->arrays.texture[t].bo->name == name) {
+                        c->arrays.texture[t].bo = 0;
+                    }
                 }
             }
         }
diff --git a/opengl/libagl/copybit.cpp b/opengl/libagl/copybit.cpp
index a68750e..4cacc11 100644
--- a/opengl/libagl/copybit.cpp
+++ b/opengl/libagl/copybit.cpp
@@ -38,7 +38,7 @@
 #include <ui/Rect.h>
 
 
-#define DEBUG_COPYBIT true
+#define DEBUG_COPYBIT false
 
 // ----------------------------------------------------------------------------
 
diff --git a/opengl/libagl/egl.cpp b/opengl/libagl/egl.cpp
index 8cfa084..81864bd 100644
--- a/opengl/libagl/egl.cpp
+++ b/opengl/libagl/egl.cpp
@@ -876,7 +876,7 @@
 #define VERSION_MAJOR 1
 #define VERSION_MINOR 2
 static char const * const gVendorString     = "Google Inc.";
-static char const * const gVersionString    = "1.2 Android Driver";
+static char const * const gVersionString    = "1.2 Android Driver 1.1.0";
 static char const * const gClientApiString  = "OpenGL ES";
 static char const * const gExtensionsString =
         "EGL_KHR_image_base "
diff --git a/opengl/libagl/state.cpp b/opengl/libagl/state.cpp
index a59b3b0..0f1f27d 100644
--- a/opengl/libagl/state.cpp
+++ b/opengl/libagl/state.cpp
@@ -37,7 +37,7 @@
 // ----------------------------------------------------------------------------
 
 static char const * const gVendorString     = "Android";
-static char const * const gRendererString   = "Android PixelFlinger 1.1";
+static char const * const gRendererString   = "Android PixelFlinger 1.2";
 static char const * const gVersionString    = "OpenGL ES-CM 1.0";
 static char const * const gExtensionsString =
     "GL_OES_byte_coordinates "              // OK
diff --git a/opengl/libs/Android.mk b/opengl/libs/Android.mk
index 6d20e80..37b9a43 100644
--- a/opengl/libs/Android.mk
+++ b/opengl/libs/Android.mk
@@ -120,3 +120,19 @@
 endif
 
 include $(BUILD_SHARED_LIBRARY)
+
+###############################################################################
+# Build the ETC1 host static library
+#
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES:= 		\
+	ETC1/etc1.cpp 	\
+#
+
+LOCAL_LDLIBS := -lpthread -ldl
+LOCAL_MODULE:= libETC1
+
+include $(BUILD_HOST_STATIC_LIBRARY)
+
diff --git a/opengl/libs/ETC1/etc1.cpp b/opengl/libs/ETC1/etc1.cpp
new file mode 100644
index 0000000..5ed2c3c
--- /dev/null
+++ b/opengl/libs/ETC1/etc1.cpp
@@ -0,0 +1,670 @@
+// Copyright 2009 Google Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <ETC1/etc1.h>
+
+#include <string.h>
+
+/* From http://www.khronos.org/registry/gles/extensions/OES/OES_compressed_ETC1_RGB8_texture.txt
+
+ The number of bits that represent a 4x4 texel block is 64 bits if
+ <internalformat> is given by ETC1_RGB8_OES.
+
+ The data for a block is a number of bytes,
+
+ {q0, q1, q2, q3, q4, q5, q6, q7}
+
+ where byte q0 is located at the lowest memory address and q7 at
+ the highest. The 64 bits specifying the block is then represented
+ by the following 64 bit integer:
+
+ int64bit = 256*(256*(256*(256*(256*(256*(256*q0+q1)+q2)+q3)+q4)+q5)+q6)+q7;
+
+ ETC1_RGB8_OES:
+
+ a) bit layout in bits 63 through 32 if diffbit = 0
+
+ 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48
+ -----------------------------------------------
+ | base col1 | base col2 | base col1 | base col2 |
+ | R1 (4bits)| R2 (4bits)| G1 (4bits)| G2 (4bits)|
+ -----------------------------------------------
+
+ 47 46 45 44 43 42 41 40 39 38 37 36 35 34  33  32
+ ---------------------------------------------------
+ | base col1 | base col2 | table  | table  |diff|flip|
+ | B1 (4bits)| B2 (4bits)| cw 1   | cw 2   |bit |bit |
+ ---------------------------------------------------
+
+
+ b) bit layout in bits 63 through 32 if diffbit = 1
+
+ 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48
+ -----------------------------------------------
+ | base col1    | dcol 2 | base col1    | dcol 2 |
+ | R1' (5 bits) | dR2    | G1' (5 bits) | dG2    |
+ -----------------------------------------------
+
+ 47 46 45 44 43 42 41 40 39 38 37 36 35 34  33  32
+ ---------------------------------------------------
+ | base col 1   | dcol 2 | table  | table  |diff|flip|
+ | B1' (5 bits) | dB2    | cw 1   | cw 2   |bit |bit |
+ ---------------------------------------------------
+
+
+ c) bit layout in bits 31 through 0 (in both cases)
+
+ 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16
+ -----------------------------------------------
+ |       most significant pixel index bits       |
+ | p| o| n| m| l| k| j| i| h| g| f| e| d| c| b| a|
+ -----------------------------------------------
+
+ 15 14 13 12 11 10  9  8  7  6  5  4  3   2   1  0
+ --------------------------------------------------
+ |         least significant pixel index bits       |
+ | p| o| n| m| l| k| j| i| h| g| f| e| d| c | b | a |
+ --------------------------------------------------
+
+
+ Add table 3.17.2: Intensity modifier sets for ETC1 compressed textures:
+
+ table codeword                modifier table
+ ------------------        ----------------------
+ 0                     -8  -2  2   8
+ 1                    -17  -5  5  17
+ 2                    -29  -9  9  29
+ 3                    -42 -13 13  42
+ 4                    -60 -18 18  60
+ 5                    -80 -24 24  80
+ 6                   -106 -33 33 106
+ 7                   -183 -47 47 183
+
+
+ Add table 3.17.3 Mapping from pixel index values to modifier values for
+ ETC1 compressed textures:
+
+ pixel index value
+ ---------------
+ msb     lsb           resulting modifier value
+ -----   -----          -------------------------
+ 1       1            -b (large negative value)
+ 1       0            -a (small negative value)
+ 0       0             a (small positive value)
+ 0       1             b (large positive value)
+
+
+ */
+
+static const int kModifierTable[] = {
+/* 0 */2, 8, -2, -8,
+/* 1 */5, 17, -5, -17,
+/* 2 */9, 29, -9, -29,
+/* 3 */13, 42, -13, -42,
+/* 4 */18, 60, -18, -60,
+/* 5 */24, 80, -24, -80,
+/* 6 */33, 106, -33, -106,
+/* 7 */47, 183, -47, -183 };
+
+static const int kLookup[8] = { 0, 1, 2, 3, -4, -3, -2, -1 };
+
+static inline etc1_byte clamp(int x) {
+    return (etc1_byte) (x >= 0 ? (x < 255 ? x : 255) : 0);
+}
+
+static
+inline int convert4To8(int b) {
+    int c = b & 0xf;
+    return (c << 4) | c;
+}
+
+static
+inline int convert5To8(int b) {
+    int c = b & 0x1f;
+    return (c << 3) | (c >> 2);
+}
+
+static
+inline int convert6To8(int b) {
+    int c = b & 0x3f;
+    return (c << 2) | (c >> 4);
+}
+
+static
+inline int divideBy255(int d) {
+    return (d + 128 + (d >> 8)) >> 8;
+}
+
+static
+inline int convert8To4(int b) {
+    int c = b & 0xff;
+    return divideBy255(b * 15);
+}
+
+static
+inline int convert8To5(int b) {
+    int c = b & 0xff;
+    return divideBy255(b * 31);
+}
+
+static
+inline int convertDiff(int base, int diff) {
+    return convert5To8((0x1f & base) + kLookup[0x7 & diff]);
+}
+
+static
+void decode_subblock(etc1_byte* pOut, int r, int g, int b, const int* table,
+        etc1_uint32 low, bool second, bool flipped) {
+    int baseX = 0;
+    int baseY = 0;
+    if (second) {
+        if (flipped) {
+            baseY = 2;
+        } else {
+            baseX = 2;
+        }
+    }
+    for (int i = 0; i < 8; i++) {
+        int x, y;
+        if (flipped) {
+            x = baseX + (i >> 1);
+            y = baseY + (i & 1);
+        } else {
+            x = baseX + (i >> 2);
+            y = baseY + (i & 3);
+        }
+        int k = y + (x * 4);
+        int offset = ((low >> k) & 1) | ((low >> (k + 15)) & 2);
+        int delta = table[offset];
+        etc1_byte* q = pOut + 3 * (x + 4 * y);
+        *q++ = clamp(r + delta);
+        *q++ = clamp(g + delta);
+        *q++ = clamp(b + delta);
+    }
+}
+
+// Input is an ETC1 compressed version of the data.
+// Output is a 4 x 4 square of 3-byte pixels in form R, G, B
+
+void etc1_decode_block(const etc1_byte* pIn, etc1_byte* pOut) {
+    etc1_uint32 high = (pIn[0] << 24) | (pIn[1] << 16) | (pIn[2] << 8) | pIn[3];
+    etc1_uint32 low = (pIn[4] << 24) | (pIn[5] << 16) | (pIn[6] << 8) | pIn[7];
+    int r1, r2, g1, g2, b1, b2;
+    if (high & 2) {
+        // differential
+        int rBase = high >> 27;
+        int gBase = high >> 19;
+        int bBase = high >> 11;
+        r1 = convert5To8(rBase);
+        r2 = convertDiff(rBase, high >> 24);
+        g1 = convert5To8(gBase);
+        g2 = convertDiff(gBase, high >> 16);
+        b1 = convert5To8(bBase);
+        b2 = convertDiff(bBase, high >> 8);
+    } else {
+        // not differential
+        r1 = convert4To8(high >> 28);
+        r2 = convert4To8(high >> 24);
+        g1 = convert4To8(high >> 20);
+        g2 = convert4To8(high >> 16);
+        b1 = convert4To8(high >> 12);
+        b2 = convert4To8(high >> 8);
+    }
+    int tableIndexA = 7 & (high >> 5);
+    int tableIndexB = 7 & (high >> 2);
+    const int* tableA = kModifierTable + tableIndexA * 4;
+    const int* tableB = kModifierTable + tableIndexB * 4;
+    bool flipped = (high & 1) != 0;
+    decode_subblock(pOut, r1, g1, b1, tableA, low, false, flipped);
+    decode_subblock(pOut, r2, g2, b2, tableB, low, true, flipped);
+}
+
+typedef struct {
+    etc1_uint32 high;
+    etc1_uint32 low;
+    etc1_uint32 score; // Lower is more accurate
+} etc_compressed;
+
+static
+inline void take_best(etc_compressed* a, const etc_compressed* b) {
+    if (a->score > b->score) {
+        *a = *b;
+    }
+}
+
+static
+void etc_average_colors_subblock(const etc1_byte* pIn, etc1_uint32 inMask,
+        etc1_byte* pColors, bool flipped, bool second) {
+    int r = 0;
+    int g = 0;
+    int b = 0;
+
+    if (flipped) {
+        int by = 0;
+        if (second) {
+            by = 2;
+        }
+        for (int y = 0; y < 2; y++) {
+            int yy = by + y;
+            for (int x = 0; x < 4; x++) {
+                int i = x + 4 * yy;
+                if (inMask & (1 << i)) {
+                    const etc1_byte* p = pIn + i * 3;
+                    r += *(p++);
+                    g += *(p++);
+                    b += *(p++);
+                }
+            }
+        }
+    } else {
+        int bx = 0;
+        if (second) {
+            bx = 2;
+        }
+        for (int y = 0; y < 4; y++) {
+            for (int x = 0; x < 2; x++) {
+                int xx = bx + x;
+                int i = xx + 4 * y;
+                if (inMask & (1 << i)) {
+                    const etc1_byte* p = pIn + i * 3;
+                    r += *(p++);
+                    g += *(p++);
+                    b += *(p++);
+                }
+            }
+        }
+    }
+    pColors[0] = (etc1_byte)((r + 4) >> 3);
+    pColors[1] = (etc1_byte)((g + 4) >> 3);
+    pColors[2] = (etc1_byte)((b + 4) >> 3);
+}
+
+static
+inline int square(int x) {
+    return x * x;
+}
+
+static etc1_uint32 chooseModifier(const etc1_byte* pBaseColors,
+        const etc1_byte* pIn, etc1_uint32 *pLow, int bitIndex,
+        const int* pModifierTable) {
+    etc1_uint32 bestScore = ~0;
+    int bestIndex = 0;
+    int pixelR = pIn[0];
+    int pixelG = pIn[1];
+    int pixelB = pIn[2];
+    int r = pBaseColors[0];
+    int g = pBaseColors[1];
+    int b = pBaseColors[2];
+    for (int i = 0; i < 4; i++) {
+        int modifier = pModifierTable[i];
+        int decodedG = clamp(g + modifier);
+        etc1_uint32 score = (etc1_uint32) (6 * square(decodedG - pixelG));
+        if (score >= bestScore) {
+            continue;
+        }
+        int decodedR = clamp(r + modifier);
+        score += (etc1_uint32) (3 * square(decodedR - pixelR));
+        if (score >= bestScore) {
+            continue;
+        }
+        int decodedB = clamp(b + modifier);
+        score += (etc1_uint32) square(decodedB - pixelB);
+        if (score < bestScore) {
+            bestScore = score;
+            bestIndex = i;
+        }
+    }
+    etc1_uint32 lowMask = (((bestIndex >> 1) << 16) | (bestIndex & 1))
+            << bitIndex;
+    *pLow |= lowMask;
+    return bestScore;
+}
+
+static
+void etc_encode_subblock_helper(const etc1_byte* pIn, etc1_uint32 inMask,
+        etc_compressed* pCompressed, bool flipped, bool second,
+        const etc1_byte* pBaseColors, const int* pModifierTable) {
+    int score = pCompressed->score;
+    if (flipped) {
+        int by = 0;
+        if (second) {
+            by = 2;
+        }
+        for (int y = 0; y < 2; y++) {
+            int yy = by + y;
+            for (int x = 0; x < 4; x++) {
+                int i = x + 4 * yy;
+                if (inMask & (1 << i)) {
+                    score += chooseModifier(pBaseColors, pIn + i * 3,
+                            &pCompressed->low, yy + x * 4, pModifierTable);
+                }
+            }
+        }
+    } else {
+        int bx = 0;
+        if (second) {
+            bx = 2;
+        }
+        for (int y = 0; y < 4; y++) {
+            for (int x = 0; x < 2; x++) {
+                int xx = bx + x;
+                int i = xx + 4 * y;
+                if (inMask & (1 << i)) {
+                    score += chooseModifier(pBaseColors, pIn + i * 3,
+                            &pCompressed->low, y + xx * 4, pModifierTable);
+                }
+            }
+        }
+    }
+    pCompressed->score = score;
+}
+
+static bool inRange4bitSigned(int color) {
+    return color >= -4 && color <= 3;
+}
+
+static void etc_encodeBaseColors(etc1_byte* pBaseColors,
+        const etc1_byte* pColors, etc_compressed* pCompressed) {
+    int r1, g1, b1, r2, g2, b2; // 8 bit base colors for sub-blocks
+    bool differential;
+    {
+        int r51 = convert8To5(pColors[0]);
+        int g51 = convert8To5(pColors[1]);
+        int b51 = convert8To5(pColors[2]);
+        int r52 = convert8To5(pColors[3]);
+        int g52 = convert8To5(pColors[4]);
+        int b52 = convert8To5(pColors[5]);
+
+        r1 = convert5To8(r51);
+        g1 = convert5To8(g51);
+        b1 = convert5To8(b51);
+
+        int dr = r52 - r51;
+        int dg = g52 - g51;
+        int db = b52 - b51;
+
+        differential = inRange4bitSigned(dr) && inRange4bitSigned(dg)
+                && inRange4bitSigned(db);
+        if (differential) {
+            r2 = convert5To8(r51 + dr);
+            g2 = convert5To8(g51 + dg);
+            b2 = convert5To8(b51 + db);
+            pCompressed->high |= (r51 << 27) | ((7 & dr) << 24) | (g51 << 19)
+                    | ((7 & dg) << 16) | (b51 << 11) | ((7 & db) << 8) | 2;
+        }
+    }
+
+    if (!differential) {
+        int r41 = convert8To4(pColors[0]);
+        int g41 = convert8To4(pColors[1]);
+        int b41 = convert8To4(pColors[2]);
+        int r42 = convert8To4(pColors[3]);
+        int g42 = convert8To4(pColors[4]);
+        int b42 = convert8To4(pColors[5]);
+        r1 = convert4To8(r41);
+        g1 = convert4To8(g41);
+        b1 = convert4To8(b41);
+        r2 = convert4To8(r42);
+        g2 = convert4To8(g42);
+        b2 = convert4To8(b42);
+        pCompressed->high |= (r41 << 28) | (r42 << 24) | (g41 << 20) | (g42
+                << 16) | (b41 << 12) | (b42 << 8);
+    }
+    pBaseColors[0] = r1;
+    pBaseColors[1] = g1;
+    pBaseColors[2] = b1;
+    pBaseColors[3] = r2;
+    pBaseColors[4] = g2;
+    pBaseColors[5] = b2;
+}
+
+static
+void etc_encode_block_helper(const etc1_byte* pIn, etc1_uint32 inMask,
+        const etc1_byte* pColors, etc_compressed* pCompressed, bool flipped) {
+    pCompressed->score = ~0;
+    pCompressed->high = (flipped ? 1 : 0);
+    pCompressed->low = 0;
+
+    etc1_byte pBaseColors[6];
+
+    etc_encodeBaseColors(pBaseColors, pColors, pCompressed);
+
+    int originalHigh = pCompressed->high;
+
+    const int* pModifierTable = kModifierTable;
+    for (int i = 0; i < 8; i++, pModifierTable += 4) {
+        etc_compressed temp;
+        temp.score = 0;
+        temp.high = originalHigh | (i << 5);
+        temp.low = 0;
+        etc_encode_subblock_helper(pIn, inMask, &temp, flipped, false,
+                pBaseColors, pModifierTable);
+        take_best(pCompressed, &temp);
+    }
+    pModifierTable = kModifierTable;
+    etc_compressed firstHalf = *pCompressed;
+    for (int i = 0; i < 8; i++, pModifierTable += 4) {
+        etc_compressed temp;
+        temp.score = firstHalf.score;
+        temp.high = firstHalf.high | (i << 2);
+        temp.low = firstHalf.low;
+        etc_encode_subblock_helper(pIn, inMask, &temp, flipped, true,
+                pBaseColors + 3, pModifierTable);
+        if (i == 0) {
+            *pCompressed = temp;
+        } else {
+            take_best(pCompressed, &temp);
+        }
+    }
+}
+
+static void writeBigEndian(etc1_byte* pOut, etc1_uint32 d) {
+    pOut[0] = (etc1_byte)(d >> 24);
+    pOut[1] = (etc1_byte)(d >> 16);
+    pOut[2] = (etc1_byte)(d >> 8);
+    pOut[3] = (etc1_byte) d;
+}
+
+// Input is a 4 x 4 square of 3-byte pixels in form R, G, B
+// inmask is a 16-bit mask where bit (1 << (x + y * 4)) tells whether the corresponding (x,y)
+// pixel is valid or not. Invalid pixel color values are ignored when compressing.
+// Output is an ETC1 compressed version of the data.
+
+void etc1_encode_block(const etc1_byte* pIn, etc1_uint32 inMask,
+        etc1_byte* pOut) {
+    etc1_byte colors[6];
+    etc1_byte flippedColors[6];
+    etc_average_colors_subblock(pIn, inMask, colors, false, false);
+    etc_average_colors_subblock(pIn, inMask, colors + 3, false, true);
+    etc_average_colors_subblock(pIn, inMask, flippedColors, true, false);
+    etc_average_colors_subblock(pIn, inMask, flippedColors + 3, true, true);
+
+    etc_compressed a, b;
+    etc_encode_block_helper(pIn, inMask, colors, &a, false);
+    etc_encode_block_helper(pIn, inMask, flippedColors, &b, true);
+    take_best(&a, &b);
+    writeBigEndian(pOut, a.high);
+    writeBigEndian(pOut + 4, a.low);
+}
+
+// Return the size of the encoded image data (does not include size of PKM header).
+
+etc1_uint32 etc1_get_encoded_data_size(etc1_uint32 width, etc1_uint32 height) {
+    return (((width + 3) & ~3) * ((height + 3) & ~3)) >> 1;
+}
+
+// Encode an entire image.
+// pIn - pointer to the image data. Formatted such that the Red component of
+//       pixel (x,y) is at pIn + pixelSize * x + stride * y + redOffset;
+// pOut - pointer to encoded data. Must be large enough to store entire encoded image.
+
+int etc1_encode_image(const etc1_byte* pIn, etc1_uint32 width, etc1_uint32 height,
+        etc1_uint32 pixelSize, etc1_uint32 stride, etc1_byte* pOut) {
+    if (pixelSize < 2 || pixelSize > 3) {
+        return -1;
+    }
+    static const unsigned short kYMask[] = { 0x0, 0xf, 0xff, 0xfff, 0xffff };
+    static const unsigned short kXMask[] = { 0x0, 0x1111, 0x3333, 0x7777,
+            0xffff };
+    etc1_byte block[ETC1_DECODED_BLOCK_SIZE];
+    etc1_byte encoded[ETC1_ENCODED_BLOCK_SIZE];
+
+    etc1_uint32 encodedWidth = (width + 3) & ~3;
+    etc1_uint32 encodedHeight = (height + 3) & ~3;
+
+    for (etc1_uint32 y = 0; y < encodedHeight; y += 4) {
+        etc1_uint32 yEnd = height - y;
+        if (yEnd > 4) {
+            yEnd = 4;
+        }
+        int ymask = kYMask[yEnd];
+        for (etc1_uint32 x = 0; x < encodedWidth; x += 4) {
+            etc1_uint32 xEnd = width - x;
+            if (xEnd > 4) {
+                xEnd = 4;
+            }
+            int mask = ymask & kXMask[xEnd];
+            for (etc1_uint32 cy = 0; cy < yEnd; cy++) {
+                etc1_byte* q = block + (cy * 4) * 3;
+                const etc1_byte* p = pIn + pixelSize * x + stride * (y + cy);
+                if (pixelSize == 3) {
+                    memcpy(q, p, xEnd * 3);
+                } else {
+                    for (etc1_uint32 cx = 0; cx < xEnd; cx++) {
+                        int pixel = (p[1] << 8) | p[0];
+                        *q++ = convert5To8(pixel >> 11);
+                        *q++ = convert6To8(pixel >> 5);
+                        *q++ = convert5To8(pixel);
+                        p += pixelSize;
+                    }
+                }
+            }
+            etc1_encode_block(block, mask, encoded);
+            memcpy(pOut, encoded, sizeof(encoded));
+            pOut += sizeof(encoded);
+        }
+    }
+    return 0;
+}
+
+// Decode an entire image.
+// pIn - pointer to encoded data.
+// pOut - pointer to the image data. Will be written such that the Red component of
+//       pixel (x,y) is at pIn + pixelSize * x + stride * y + redOffset. Must be
+//        large enough to store entire image.
+
+
+int etc1_decode_image(const etc1_byte* pIn, etc1_byte* pOut,
+        etc1_uint32 width, etc1_uint32 height,
+        etc1_uint32 pixelSize, etc1_uint32 stride) {
+    if (pixelSize < 2 || pixelSize > 3) {
+        return -1;
+    }
+    etc1_byte block[ETC1_DECODED_BLOCK_SIZE];
+
+    etc1_uint32 encodedWidth = (width + 3) & ~3;
+    etc1_uint32 encodedHeight = (height + 3) & ~3;
+
+    for (etc1_uint32 y = 0; y < encodedHeight; y += 4) {
+        etc1_uint32 yEnd = height - y;
+        if (yEnd > 4) {
+            yEnd = 4;
+        }
+        for (etc1_uint32 x = 0; x < encodedWidth; x += 4) {
+            etc1_uint32 xEnd = width - x;
+            if (xEnd > 4) {
+                xEnd = 4;
+            }
+            etc1_decode_block(pIn, block);
+            pIn += ETC1_ENCODED_BLOCK_SIZE;
+            for (etc1_uint32 cy = 0; cy < yEnd; cy++) {
+                const etc1_byte* q = block + (cy * 4) * 3;
+                etc1_byte* p = pOut + pixelSize * x + stride * (y + cy);
+                if (pixelSize == 3) {
+                    memcpy(p, q, xEnd * 3);
+                } else {
+                    for (etc1_uint32 cx = 0; cx < xEnd; cx++) {
+                        etc1_byte r = *q++;
+                        etc1_byte g = *q++;
+                        etc1_byte b = *q++;
+                        etc1_uint32 pixel = ((r >> 3) << 11) | ((g >> 2) << 5) | (b >> 3);
+                        *p++ = (etc1_byte) pixel;
+                        *p++ = (etc1_byte) (pixel >> 8);
+                    }
+                }
+            }
+        }
+    }
+    return 0;
+}
+
+static const char kMagic[] = { 'P', 'K', 'M', ' ', '1', '0' };
+
+static const etc1_uint32 ETC1_PKM_FORMAT_OFFSET = 6;
+static const etc1_uint32 ETC1_PKM_ENCODED_WIDTH_OFFSET = 8;
+static const etc1_uint32 ETC1_PKM_ENCODED_HEIGHT_OFFSET = 10;
+static const etc1_uint32 ETC1_PKM_WIDTH_OFFSET = 12;
+static const etc1_uint32 ETC1_PKM_HEIGHT_OFFSET = 14;
+
+static const etc1_uint32 ETC1_RGB_NO_MIPMAPS = 0;
+
+static void writeBEUint16(etc1_byte* pOut, etc1_uint32 data) {
+    pOut[0] = (etc1_byte) (data >> 8);
+    pOut[1] = (etc1_byte) data;
+}
+
+static etc1_uint32 readBEUint16(const etc1_byte* pIn) {
+    return (pIn[0] << 8) | pIn[1];
+}
+
+// Format a PKM header
+
+void etc1_pkm_format_header(etc1_byte* pHeader, etc1_uint32 width, etc1_uint32 height) {
+    memcpy(pHeader, kMagic, sizeof(kMagic));
+    etc1_uint32 encodedWidth = (width + 3) & ~3;
+    etc1_uint32 encodedHeight = (height + 3) & ~3;
+    writeBEUint16(pHeader + ETC1_PKM_FORMAT_OFFSET, ETC1_RGB_NO_MIPMAPS);
+    writeBEUint16(pHeader + ETC1_PKM_ENCODED_WIDTH_OFFSET, encodedWidth);
+    writeBEUint16(pHeader + ETC1_PKM_ENCODED_HEIGHT_OFFSET, encodedHeight);
+    writeBEUint16(pHeader + ETC1_PKM_WIDTH_OFFSET, width);
+    writeBEUint16(pHeader + ETC1_PKM_HEIGHT_OFFSET, height);
+}
+
+// Check if a PKM header is correctly formatted.
+
+etc1_bool etc1_pkm_is_valid(const etc1_byte* pHeader) {
+    if (memcmp(pHeader, kMagic, sizeof(kMagic))) {
+        return false;
+    }
+    etc1_uint32 format = readBEUint16(pHeader + ETC1_PKM_FORMAT_OFFSET);
+    etc1_uint32 encodedWidth = readBEUint16(pHeader + ETC1_PKM_ENCODED_WIDTH_OFFSET);
+    etc1_uint32 encodedHeight = readBEUint16(pHeader + ETC1_PKM_ENCODED_HEIGHT_OFFSET);
+    etc1_uint32 width = readBEUint16(pHeader + ETC1_PKM_WIDTH_OFFSET);
+    etc1_uint32 height = readBEUint16(pHeader + ETC1_PKM_HEIGHT_OFFSET);
+    return format == ETC1_RGB_NO_MIPMAPS &&
+            encodedWidth >= width && encodedWidth - width < 4 &&
+            encodedHeight >= height && encodedHeight - height < 4;
+}
+
+// Read the image width from a PKM header
+
+etc1_uint32 etc1_pkm_get_width(const etc1_byte* pHeader) {
+    return readBEUint16(pHeader + ETC1_PKM_WIDTH_OFFSET);
+}
+
+// Read the image height from a PKM header
+
+etc1_uint32 etc1_pkm_get_height(const etc1_byte* pHeader){
+    return readBEUint16(pHeader + ETC1_PKM_HEIGHT_OFFSET);
+}