src/engines/EngineBase.h

/***************************************************************************
 *                                                                         *
 *   LinuxSampler - modular, streaming capable sampler                     *
 *                                                                         *
 *   Copyright (C) 2003,2004 by Benno Senoner and Christian Schoenebeck    *
 *   Copyright (C) 2005-2008 Christian Schoenebeck                         *
 *   Copyright (C) 2009-2013 Christian Schoenebeck and Grigor Iliev        *
 *                                                                         *
 *   This program is free software; you can redistribute it and/or modify  *
 *   it under the terms of the GNU General Public License as published by  *
 *   the Free Software Foundation; either version 2 of the License, or     *
 *   (at your option) any later version.                                   *
 *                                                                         *
 *   This program is distributed in the hope that it will be useful,       *
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
 *   GNU General Public License for more details.                          *
 *                                                                         *
 *   You should have received a copy of the GNU General Public License     *
 *   along with this program; if not, write to the Free Software           *
 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston,                 *
 *   MA  02111-1307  USA                                                   *
 ***************************************************************************/

#ifndef __LS_ENGINEBASE_H__
#define __LS_ENGINEBASE_H__

#include "AbstractEngine.h"
#include "EngineChannelBase.h"
#include "common/DiskThreadBase.h"
#include "common/MidiKeyboardManager.h"
#include "InstrumentManager.h"
#include "../common/global_private.h"


namespace LinuxSampler {

    class AbstractEngineChannel;

    template <
        class V   /* Voice */,
        class RR  /* Root Region */,
        class R   /* Region */,
        class D   /* Disk Thread */,
        class IM  /* Instrument Manager */,
        class I   /* Instrument */
    >
    class EngineBase: public AbstractEngine, public RegionPools<R>, public VoicePool<V> {

        public:
            typedef typename RTList<V>::Iterator VoiceIterator;
            typedef typename Pool<V>::Iterator PoolVoiceIterator;
            typedef typename RTList<RR*>::Iterator RootRegionIterator;
            typedef typename MidiKeyboardManager<V>::MidiKey MidiKey;
            
            EngineBase() : SuspendedRegions(128) {
                pDiskThread          = NULL;
                pVoicePool           = new Pool<V>(GLOBAL_MAX_VOICES);
                pRegionPool[0]       = new Pool<R*>(GLOBAL_MAX_VOICES);
                pRegionPool[1]       = new Pool<R*>(GLOBAL_MAX_VOICES);
                pVoiceStealingQueue  = new RTList<Event>(pEventPool);
                iMaxDiskStreams      = GLOBAL_MAX_STREAMS;

                for (VoiceIterator iterVoice = pVoicePool->allocAppend(); iterVoice == pVoicePool->last(); iterVoice = pVoicePool->allocAppend()) {
                    iterVoice->SetEngine(this);
                }
                pVoicePool->clear();

                ResetInternal();
                ResetScaleTuning();
                ResetSuspendedRegions();
            }

            virtual ~EngineBase() {
                if (pDiskThread) {
                    dmsg(1,("Stopping disk thread..."));
                    pDiskThread->StopThread();
                    delete pDiskThread;
                    dmsg(1,("OK\n"));
                }

                if (pVoicePool) {
                    pVoicePool->clear();
                    delete pVoicePool;
                }

                if (pVoiceStealingQueue) delete pVoiceStealingQueue;

                if (pRegionPool[0]) delete pRegionPool[0];
                if (pRegionPool[1]) delete pRegionPool[1];
                ResetSuspendedRegions();
            }

            // implementation of abstract methods derived from class 'LinuxSampler::Engine'

            /**
             * Let this engine proceed to render the given amount of sample points.
             * The engine will iterate through all engine channels and render audio
             * for each engine channel independently. The calculated audio data of
             * all voices of each engine channel will be placed into the audio sum
             * buffers of the respective audio output device, connected to the
             * respective engine channel.
             *
             *  @param Samples - number of sample points to be rendered
             *  @returns       0 on success
             */
            virtual int RenderAudio(uint Samples) {
                dmsg(8,("RenderAudio(Samples=%d)\n", Samples));

                // return if engine disabled
                if (EngineDisabled.Pop()) {
                    dmsg(5,("EngineBase: engine disabled (val=%d)\n",EngineDisabled.GetUnsafe()));
                    EngineDisabled.RttDone();
                    return 0;
                }

                // process requests for suspending / resuming regions (i.e. to avoid
                // crashes while these regions are modified by an instrument editor)
                ProcessSuspensionsChanges();

                // update time of start and end of this audio fragment (as events' time stamps relate to this)
                pEventGenerator->UpdateFragmentTime(Samples);

                // We only allow the given maximum number of voices to be spawned
                // in each audio fragment. All subsequent request for spawning new
                // voices in the same audio fragment will be ignored.
                VoiceSpawnsLeft = MaxVoices();

                // get all events from the engine's global input event queue which belong to the current fragment
                // (these are usually just SysEx messages)
                ImportEvents(Samples);

                // process engine global events (these are currently only MIDI System Exclusive messages)
                {
                    RTList<Event>::Iterator itEvent = pGlobalEvents->first();
                    RTList<Event>::Iterator end     = pGlobalEvents->end();
                    for (; itEvent != end; ++itEvent) {
                        switch (itEvent->Type) {
                            case Event::type_sysex:
                                dmsg(5,("Engine: Sysex received\n"));
                                ProcessSysex(itEvent);
                                break;
                        }
                    }
                }

                // reset internal voice counter (just for statistic of active voices)
                ActiveVoiceCountTemp = 0;

                HandleInstrumentChanges();

                // handle events on all engine channels
                for (int i = 0; i < engineChannels.size(); i++) {
                    ProcessEvents(engineChannels[i], Samples);
                }

                // render all 'normal', active voices on all engine channels
                for (int i = 0; i < engineChannels.size(); i++) {
                    RenderActiveVoices(engineChannels[i], Samples);
                }

                // now that all ordinary voices on ALL engine channels are rendered, render new stolen voices
                RenderStolenVoices(Samples);

                // handle audio routing for engine channels with FX sends
                for (int i = 0; i < engineChannels.size(); i++) {
                    AbstractEngineChannel* pChannel = static_cast<AbstractEngineChannel*>(engineChannels[i]);
                    if (pChannel->fxSends.empty()) continue; // ignore if no FX sends
                    RouteAudio(engineChannels[i], Samples);
                }

                // handle cleanup on all engine channels for the next audio fragment
                for (int i = 0; i < engineChannels.size(); i++) {
                    PostProcess(engineChannels[i]);
                }


                // empty the engine's event list for the next audio fragment
                ClearEventLists();

                // reset voice stealing for the next audio fragment
                pVoiceStealingQueue->clear();

                // just some statistics about this engine instance
                SetVoiceCount(ActiveVoiceCountTemp);
                if (VoiceCount() > ActiveVoiceCountMax) ActiveVoiceCountMax = VoiceCount();

                // in case regions were previously suspended and we killed voices
                // with disk streams due to that, check if those streams have finally
                // been deleted by the disk thread
                if (iPendingStreamDeletions) ProcessPendingStreamDeletions();

                // Release the instrument change command. (This has to
                // be done after all voices have been rendered and not
                // in HandleInstrumentChanges, as the RegionsInUse
                // list has been built up by the voice renderers.)
                for (int i = 0; i < engineChannels.size(); i++) {
                    EngineChannelBase<V, R, I>* channel =
                        static_cast<EngineChannelBase<V, R, I>*>(engineChannels[i]);
                    channel->InstrumentChangeCommandReader.Unlock();
                }
                FrameTime += Samples;

                EngineDisabled.RttDone();
                return 0;
            }

            virtual int MaxVoices() { return pVoicePool->poolSize(); }

            virtual void SetMaxVoices(int iVoices) throw (Exception) {
                if (iVoices < 1)
                    throw Exception("Maximum voices for an engine cannot be set lower than 1");

                SuspendAll();

                // NOTE: we need to clear pRegionsInUse before deleting pDimRegionPool,
                // otherwise memory corruption will occur if there are active voices (see bug #118)
                for (int iChannel = 0; iChannel < engineChannels.size(); iChannel++) {
                    EngineChannelBase<V, R, I>* pChannel = static_cast<EngineChannelBase<V, R, I>*>(engineChannels[iChannel]);
                    pChannel->ClearRegionsInUse();
                }

                if (pRegionPool[0]) delete pRegionPool[0];
                if (pRegionPool[1]) delete pRegionPool[1];

                pRegionPool[0] = new Pool<R*>(iVoices);
                pRegionPool[1] = new Pool<R*>(iVoices);

                for (int iChannel = 0; iChannel < engineChannels.size(); iChannel++) {
                    EngineChannelBase<V, R, I>* pChannel = static_cast<EngineChannelBase<V, R, I>*>(engineChannels[iChannel]);
                    pChannel->ResetRegionsInUse(pRegionPool);
                }

                try {
                    pVoicePool->resizePool(iVoices);
                } catch (...) {
                    throw Exception("FATAL: Could not resize voice pool!");
                }

                for (VoiceIterator iterVoice = pVoicePool->allocAppend(); iterVoice == pVoicePool->last(); iterVoice = pVoicePool->allocAppend()) {
                    iterVoice->SetEngine(this);
                    iterVoice->pDiskThread = this->pDiskThread;
                }
                pVoicePool->clear();

                PostSetMaxVoices(iVoices);
                ResumeAll();
            }
            
            /** Called after the new max number of voices is set and before resuming the engine. */
            virtual void PostSetMaxVoices(int iVoices) { }

            virtual uint DiskStreamCount() { return (pDiskThread) ? pDiskThread->GetActiveStreamCount() : 0; }
            virtual uint DiskStreamCountMax() { return (pDiskThread) ? pDiskThread->ActiveStreamCountMax : 0; }
            virtual int  MaxDiskStreams() { return iMaxDiskStreams; }

            virtual void SetMaxDiskStreams(int iStreams) throw (Exception) {
                if (iStreams < 0)
                    throw Exception("Maximum disk streams for an engine cannot be set lower than 0");

                SuspendAll();

                iMaxDiskStreams = iStreams;

                // reconnect to audio output device, because that will automatically
                // recreate the disk thread with the required amount of streams
                if (pAudioOutputDevice) Connect(pAudioOutputDevice);

                ResumeAll();
            }

            virtual String DiskStreamBufferFillBytes() { return (pDiskThread) ? pDiskThread->GetBufferFillBytes() : ""; }
            virtual String DiskStreamBufferFillPercentage() { return (pDiskThread) ? pDiskThread->GetBufferFillPercentage() : ""; }
            virtual InstrumentManager* GetInstrumentManager() { return &instruments; }

            /**
             * Connect this engine instance with the given audio output device.
             * This method will be called when an Engine instance is created.
             * All of the engine's data structures which are dependant to the used
             * audio output device / driver will be (re)allocated and / or
             * adjusted appropriately.
             *
             * @param pAudioOut - audio output device to connect to
             */
            virtual void Connect(AudioOutputDevice* pAudioOut) {
                // caution: don't ignore if connecting to the same device here,
                // because otherwise SetMaxDiskStreams() implementation won't work anymore!

                pAudioOutputDevice = pAudioOut;

                ResetInternal();

                // inform audio driver for the need of two channels
                try {
                    pAudioOutputDevice->AcquireChannels(2); // default stereo
                }
                catch (AudioOutputException e) {
                    String msg = "Audio output device unable to provide 2 audio channels, cause: " + e.Message();
                    throw Exception(msg);
                }

                this->MaxSamplesPerCycle = pAudioOutputDevice->MaxSamplesPerCycle();
                this->SampleRate         = pAudioOutputDevice->SampleRate();

                MinFadeOutSamples = int(double(SampleRate) * CONFIG_EG_MIN_RELEASE_TIME) - 1;
                if (MaxSamplesPerCycle < MinFadeOutSamples) {
                    std::cerr << "EngineBase: WARNING, CONFIG_EG_MIN_RELEASE_TIME "
                              << "too big for current audio fragment size & sampling rate! "
                              << "May lead to click sounds if voice stealing chimes in!\n" << std::flush;
                    // force volume ramp downs at the beginning of each fragment
                    MinFadeOutSamples = MaxSamplesPerCycle;
                    // lower minimum release time
                    const float minReleaseTime = (float) MaxSamplesPerCycle / (float) SampleRate;
                    for (VoiceIterator iterVoice = pVoicePool->allocAppend(); iterVoice == pVoicePool->last(); iterVoice = pVoicePool->allocAppend()) {
                        iterVoice->CalculateFadeOutCoeff(minReleaseTime, SampleRate);
                    }
                    pVoicePool->clear();
                }

                // (re)create disk thread
                if (this->pDiskThread) {
                    dmsg(1,("Stopping disk thread..."));
                    this->pDiskThread->StopThread();
                    delete this->pDiskThread;
                    dmsg(1,("OK\n"));
                }
                this->pDiskThread = CreateDiskThread();

                if (!pDiskThread) {
                    dmsg(0,("EngineBase  new diskthread = NULL\n"));
                    exit(EXIT_FAILURE);
                }

                for (VoiceIterator iterVoice = pVoicePool->allocAppend(); iterVoice == pVoicePool->last(); iterVoice = pVoicePool->allocAppend()) {
                    iterVoice->pDiskThread = this->pDiskThread;
                    dmsg(3,("d"));
                }
                pVoicePool->clear();

                // (re)create event generator
                if (pEventGenerator) delete pEventGenerator;
                pEventGenerator = new EventGenerator(pAudioOut->SampleRate());

                dmsg(1,("Starting disk thread..."));
                pDiskThread->StartThread();
                dmsg(1,("OK\n"));

                bool printEqInfo = true;
                for (VoiceIterator iterVoice = pVoicePool->allocAppend(); iterVoice == pVoicePool->last(); iterVoice = pVoicePool->allocAppend()) {
                    if (!iterVoice->pDiskThread) {
                        dmsg(0,("Engine -> voice::trigger: !pDiskThread\n"));
                        exit(EXIT_FAILURE);
                    }
                    
                    iterVoice->CreateEq();
                    
                    if(printEqInfo) {
                        iterVoice->PrintEqInfo();
                        printEqInfo = false;
                    }
                }
                pVoicePool->clear();
                
                // (re)create dedicated voice audio buffers
                //TODO: we could optimize resource usage a bit by just allocating these dedicated voice buffers when there is at least one engine channel with FX sends, because only in this case those special buffers are used actually, but since it would usually only save couple bytes in total, its probably not worth it
                if (pDedicatedVoiceChannelLeft)  delete pDedicatedVoiceChannelLeft;
                if (pDedicatedVoiceChannelRight) delete pDedicatedVoiceChannelRight;
                pDedicatedVoiceChannelLeft  = new AudioChannel(0, MaxSamplesPerCycle);
                pDedicatedVoiceChannelRight = new AudioChannel(1, MaxSamplesPerCycle);
            }
        
            // Implementattion for abstract method derived from Engine.
            virtual void ReconnectAudioOutputDevice() {
                SuspendAll();
                if (pAudioOutputDevice) Connect(pAudioOutputDevice);
                ResumeAll();
            }

            /**
             * Similar to @c Disable() but this method additionally kills all voices
             * and disk streams and blocks until all voices and disk streams are actually
             * killed / deleted.
             *
             * @e Note: only the original calling thread is able to re-enable the
             * engine afterwards by calling @c ResumeAll() later on!
             */
            virtual void SuspendAll() {
                dmsg(2,("Engine: Suspending all ...\n"));
                // stop the engine, so we can safely modify the engine's
                // data structures from this foreign thread
                DisableAndLock();
                // we could also use the respective class member variable here,
                // but this is probably safer and cleaner
                int iPendingStreamDeletions = 0;
                // kill all voices on all engine channels the *die hard* way
                for (int iChannel = 0; iChannel < engineChannels.size(); iChannel++) {
                    EngineChannelBase<V, R, I>* pEngineChannel =
                            static_cast<EngineChannelBase<V, R, I>*>(engineChannels[iChannel]);

                    iPendingStreamDeletions += pEngineChannel->KillAllVoicesImmediately();
                }
                // wait until all streams were actually deleted by the disk thread
                while (iPendingStreamDeletions) {
                    while (
                        iPendingStreamDeletions &&
                        pDiskThread->AskForDeletedStream() != Stream::INVALID_HANDLE
                    ) iPendingStreamDeletions--;
                    if (!iPendingStreamDeletions) break;
                    usleep(10000); // sleep for 10ms
                }
                dmsg(2,("EngineBase: Everything suspended.\n"));
            }

            /**
             * At the moment same as calling @c Enable() directly, but this might
             * change in future, so better call this method as counterpart to
             * @c SuspendAll() instead of @c Enable() !
             */
            virtual void ResumeAll() { Enable(); }

            /**
             * Order the engine to stop rendering audio for the given region.
             * Additionally this method will block until all voices and their disk
             * streams associated with that region are actually killed / deleted, so
             * one can i.e. safely modify the region with an instrument editor after
             * returning from this method.
             *
             * @param pRegion - region the engine shall stop using
             */
            virtual void Suspend(RR* pRegion) {
                dmsg(2,("EngineBase: Suspending Region %x ...\n",pRegion));
                {
                    LockGuard lock(SuspendedRegionsMutex);
                    SuspensionChangeOngoing.Set(true);
                    pPendingRegionSuspension = pRegion;
                    SuspensionChangeOngoing.WaitAndUnlockIf(true);
                }
                dmsg(2,("EngineBase: Region %x suspended.",pRegion));
            }

            /**
             * Orders the engine to resume playing back the given region, previously
             * suspended with @c Suspend() .
             *
             * @param pRegion - region the engine shall be allowed to use again
             */
            virtual void Resume(RR* pRegion) {
                dmsg(2,("EngineBase: Resuming Region %x ...\n",pRegion));
                {
                    LockGuard lock(SuspendedRegionsMutex);
                    SuspensionChangeOngoing.Set(true);
                    pPendingRegionResumption = pRegion;
                    SuspensionChangeOngoing.WaitAndUnlockIf(true);
                }
                dmsg(2,("EngineBase: Region %x resumed.\n",pRegion));
            }

            virtual void ResetSuspendedRegions() {
                SuspendedRegions.clear();
                iPendingStreamDeletions = 0;
                pPendingRegionSuspension = pPendingRegionResumption = NULL;
                SuspensionChangeOngoing.Set(false);
            }

            /**
             * Called by the engine's (audio) thread once per cycle to process requests
             * from the outer world to suspend or resume a given @c gig::Region .
             */
            virtual void ProcessSuspensionsChanges() {
                // process request for suspending one region
                if (pPendingRegionSuspension) {
                    // kill all voices on all engine channels that use this region
                    for (int iChannel = 0; iChannel < engineChannels.size(); iChannel++) {
                        EngineChannelBase<V, R, I>* pEngineChannel =
                                static_cast<EngineChannelBase<V, R, I>*>(engineChannels[iChannel]);
                        SuspensionVoiceHandler handler(pPendingRegionSuspension);
                        pEngineChannel->ProcessActiveVoices(&handler);
                        iPendingStreamDeletions += handler.PendingStreamDeletions;
                    }
                    // make sure the region is not yet on the list
                    bool bAlreadySuspended = false;
                    RootRegionIterator iter = SuspendedRegions.first();
                    RootRegionIterator end  = SuspendedRegions.end();
                    for (; iter != end; ++iter) { // iterate through all suspended regions
                        if (*iter == pPendingRegionSuspension) { // found
                            bAlreadySuspended = true;
                            dmsg(1,("EngineBase: attempt to suspend an already suspended region !!!\n"));
                            break;
                        }
                    }
                    if (!bAlreadySuspended) {
                        // put the region on the list of suspended regions
                        RootRegionIterator iter = SuspendedRegions.allocAppend();
                        if (iter) {
                            *iter = pPendingRegionSuspension;
                        } else std::cerr << "EngineBase: Could not suspend Region, list is full. This is a bug!!!\n" << std::flush;
                    }
                    // free request slot for next caller (and to make sure that
                    // we're not going to process the same request in the next cycle)
                    pPendingRegionSuspension = NULL;
                    // if no disk stream deletions are pending, awaken other side, as
                    // we're done in this case
                    if (!iPendingStreamDeletions) SuspensionChangeOngoing.Set(false);
                }

                // process request for resuming one region
                if (pPendingRegionResumption) {
                    // remove region from the list of suspended regions
                    RootRegionIterator iter = SuspendedRegions.first();
                    RootRegionIterator end  = SuspendedRegions.end();
                    for (; iter != end; ++iter) { // iterate through all suspended regions
                        if (*iter == pPendingRegionResumption) { // found
                            SuspendedRegions.free(iter);
                            break; // done
                        }
                    }
                    // free request slot for next caller
                    pPendingRegionResumption = NULL;
                    // awake other side as we're done
                    SuspensionChangeOngoing.Set(false);
                }
            }

            /**
             * Called by the engine's (audio) thread once per cycle to check if
             * streams of voices that were killed due to suspension request have
             * finally really been deleted by the disk thread.
             */
            virtual void ProcessPendingStreamDeletions() {
                if (!iPendingStreamDeletions) return;
                //TODO: or shall we better store a list with stream handles instead of a scalar amount of streams to be deleted? might be safer
                while (
                    iPendingStreamDeletions &&
                    pDiskThread->AskForDeletedStream() != Stream::INVALID_HANDLE
                ) iPendingStreamDeletions--;
                // just for safety ...
                while (pDiskThread->AskForDeletedStream() != Stream::INVALID_HANDLE);
                // now that all disk streams are deleted, awake other side as
                // we're finally done with suspending the requested region
                if (!iPendingStreamDeletions) SuspensionChangeOngoing.Set(false);
            }

            /**
             * Returns @c true if the given region is currently set to be suspended
             * from being used, @c false otherwise.
             */
            virtual bool RegionSuspended(RR* pRegion) {
                if (SuspendedRegions.isEmpty()) return false;
                //TODO: or shall we use a sorted container instead of the RTList? might be faster ... or trivial ;-)
                RootRegionIterator iter = SuspendedRegions.first();
                RootRegionIterator end  = SuspendedRegions.end();
                for (; iter != end; ++iter)  // iterate through all suspended regions
                    if (*iter == pRegion) return true;
                return false;
            }

            // implementation of abstract method derived from class 'LinuxSampler::RegionPools'
            virtual Pool<R*>* GetRegionPool(int index) {
                if (index < 0 || index > 1) throw Exception("Index out of bounds");
                return pRegionPool[index];
            }

            // implementation of abstract method derived from class 'LinuxSampler::VoicePool'
            virtual Pool<V>* GetVoicePool() { return pVoicePool; }

            D* GetDiskThread() { return pDiskThread; }

            //friend class EngineChannelBase<V, R, I>;

            static IM instruments;

        protected:
            class SuspensionVoiceHandler : public MidiKeyboardManager<V>::VoiceHandler {
            public:
                int PendingStreamDeletions;
                RR* pPendingRegionSuspension;
                SuspensionVoiceHandler(RR* pPendingRegionSuspension) {
                    PendingStreamDeletions = 0;
                    this->pPendingRegionSuspension = pPendingRegionSuspension;
                }

                virtual bool Process(MidiKey* pMidiKey) {
                    VoiceIterator itVoice = pMidiKey->pActiveVoices->first();
                    // if current key is not associated with this region, skip this key
                    if (itVoice->GetRegion()->GetParent() != pPendingRegionSuspension) return false;

                    return true;
                }

                virtual void Process(VoiceIterator& itVoice) {
                    // request a notification from disk thread side for stream deletion
                    const Stream::Handle hStream = itVoice->KillImmediately(true);
                    if (hStream != Stream::INVALID_HANDLE) { // voice actually used a stream
                        PendingStreamDeletions++;
                    }
                    //NOTE: maybe we should call FreeVoice() here, shouldn't cause a harm though I think, since the voices should be freed by RenderActiveVoices() in the render loop, they are probably just freed a bit later than they could/should be
                }
            };

            Pool<R*>* pRegionPool[2]; ///< Double buffered pool, used by the engine channels to keep track of regions in use.
            int       MinFadeOutSamples;     ///< The number of samples needed to make an instant fade out (e.g. for voice stealing) without leading to clicks.
            D*        pDiskThread;

            int                          ActiveVoiceCountTemp;  ///< number of currently active voices (for internal usage, will be used for incrementation)
            VoiceIterator                itLastStolenVoice;     ///< Only for voice stealing: points to the last voice which was theft in current audio fragment, NULL otherwise.
            RTList<uint>::Iterator       iuiLastStolenKey;      ///< Only for voice stealing: key number of last key on which the last voice was theft in current audio fragment, NULL otherwise.
            EngineChannelBase<V, R, I>*  pLastStolenChannel;    ///< Only for voice stealing: points to the engine channel on which the previous voice was stolen in this audio fragment.
            VoiceIterator                itLastStolenVoiceGlobally; ///< Same as itLastStolenVoice, but engine globally
            RTList<uint>::Iterator       iuiLastStolenKeyGlobally;  ///< Same as iuiLastStolenKey, but engine globally
            RTList<Event>*               pVoiceStealingQueue;   ///< All voice-launching events which had to be postponed due to free voice shortage.
            Mutex                        ResetInternalMutex;    ///< Mutex to protect the ResetInternal function for concurrent usage (e.g. by the lscp and instrument loader threads).
            int iMaxDiskStreams;

            /**
             * Dispatch and handle all events in this audio fragment for the given
             * engine channel.
             *
             * @param pEngineChannel - engine channel on which events should be
             *                         processed
             * @param Samples        - amount of sample points to be processed in
             *                         this audio fragment cycle
             */
            void ProcessEvents(EngineChannel* pEngineChannel, uint Samples) {
                // get all events from the engine channels's input event queue which belong to the current fragment
                // (these are the common events like NoteOn, NoteOff, ControlChange, etc.)
                AbstractEngineChannel* pChannel = static_cast<AbstractEngineChannel*>(pEngineChannel);
                pChannel->ImportEvents(Samples);

                // process events
                {
                    RTList<Event>::Iterator itEvent = pChannel->pEvents->first();
                    RTList<Event>::Iterator end     = pChannel->pEvents->end();
                    for (; itEvent != end; ++itEvent) {
                        switch (itEvent->Type) {
                            case Event::type_note_on:
                                dmsg(5,("Engine: Note on received\n"));
                                ProcessNoteOn((EngineChannel*)itEvent->pEngineChannel, itEvent);
                                break;
                            case Event::type_note_off:
                                dmsg(5,("Engine: Note off received\n"));
                                ProcessNoteOff((EngineChannel*)itEvent->pEngineChannel, itEvent);
                                break;
                            case Event::type_control_change:
                                dmsg(5,("Engine: MIDI CC received\n"));
                                ProcessControlChange((EngineChannel*)itEvent->pEngineChannel, itEvent);
                                break;
                            case Event::type_pitchbend:
                                dmsg(5,("Engine: Pitchbend received\n"));
                                ProcessPitchbend(static_cast<AbstractEngineChannel*>(itEvent->pEngineChannel), itEvent);
                                break;
                        }
                    }
                }

                // reset voice stealing for the next engine channel (or next audio fragment)
                itLastStolenVoice         = VoiceIterator();
                itLastStolenVoiceGlobally = VoiceIterator();
                iuiLastStolenKey          = RTList<uint>::Iterator();
                iuiLastStolenKeyGlobally  = RTList<uint>::Iterator();
                pLastStolenChannel        = NULL;
            }

            /**
             *  Will be called by LaunchVoice() method in case there are no free
             *  voices left. This method will select and kill one old voice for
             *  voice stealing and postpone the note-on event until the selected
             *  voice actually died.
             *
             *  @param pEngineChannel - engine channel on which this event occured on
             *  @param itNoteOnEvent - key, velocity and time stamp of the event
             *  @returns 0 on success, a value < 0 if no active voice could be picked for voice stealing
             */
            int  StealVoice(EngineChannel* pEngineChannel, Pool<Event>::Iterator& itNoteOnEvent) {
                if (VoiceSpawnsLeft <= 0) {
                    dmsg(1,("Max. voice thefts per audio fragment reached (you may raise CONFIG_MAX_VOICES).\n"));
                    return -1;
                }

                EngineChannelBase<V, R, I>* pEngineChn = static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);

                if (!pEventPool->poolIsEmpty()) {

                    if(!pEngineChn->StealVoice(itNoteOnEvent, &itLastStolenVoice, &iuiLastStolenKey)) {
                        --VoiceSpawnsLeft;
                        return 0;
                    }

                    // if we couldn't steal a voice from the same engine channel then
                    // steal oldest voice on the oldest key from any other engine channel
                    // (the smaller engine channel number, the higher priority)
                    EngineChannelBase<V, R, I>*  pSelectedChannel;
                    int                       iChannelIndex;
                    VoiceIterator             itSelectedVoice;

                    // select engine channel
                    if (pLastStolenChannel) {
                        pSelectedChannel = pLastStolenChannel;
                        iChannelIndex    = pSelectedChannel->iEngineIndexSelf;
                    } else { // pick the engine channel followed by this engine channel
                        iChannelIndex    = (pEngineChn->iEngineIndexSelf + 1) % engineChannels.size();
                        pSelectedChannel = static_cast<EngineChannelBase<V, R, I>*>(engineChannels[iChannelIndex]);
                    }

                    // if we already stole in this fragment, try to proceed on same key
                    if (this->itLastStolenVoiceGlobally) {
                        itSelectedVoice = this->itLastStolenVoiceGlobally;
                        do {
                            ++itSelectedVoice;
                        } while (itSelectedVoice && !itSelectedVoice->IsStealable()); // proceed iterating if voice was created in this fragment cycle
                    }

                    #if CONFIG_DEVMODE
                    EngineChannel* pBegin = pSelectedChannel; // to detect endless loop
                    #endif // CONFIG_DEVMODE

                    // did we find a 'stealable' voice?
                    if (itSelectedVoice && itSelectedVoice->IsStealable()) {
                        // remember which voice we stole, so we can simply proceed on next voice stealing
                        this->itLastStolenVoiceGlobally = itSelectedVoice;
                    } else while (true) { // iterate through engine channels
                        // get (next) oldest key
                        RTList<uint>::Iterator iuiSelectedKey = (this->iuiLastStolenKeyGlobally) ? ++this->iuiLastStolenKeyGlobally : pSelectedChannel->pActiveKeys->first();
                        this->iuiLastStolenKeyGlobally = RTList<uint>::Iterator(); // to prevent endless loop (see line above)
                        while (iuiSelectedKey) {
                            MidiKey* pSelectedKey = &pSelectedChannel->pMIDIKeyInfo[*iuiSelectedKey];
                            itSelectedVoice = pSelectedKey->pActiveVoices->first();
                            // proceed iterating if voice was created in this fragment cycle
                            while (itSelectedVoice && !itSelectedVoice->IsStealable()) ++itSelectedVoice;
                            // found a "stealable" voice ?
                            if (itSelectedVoice && itSelectedVoice->IsStealable()) {
                                // remember which voice on which key on which engine channel we stole, so we can simply proceed on next voice stealing
                                this->iuiLastStolenKeyGlobally  = iuiSelectedKey;
                                this->itLastStolenVoiceGlobally = itSelectedVoice;
                                this->pLastStolenChannel        = pSelectedChannel;
                                goto stealable_voice_found; // selection succeeded
                            }
                            ++iuiSelectedKey; // get next key on current engine channel
                        }
                        // get next engine channel
                        iChannelIndex    = (iChannelIndex + 1) % engineChannels.size();
                        pSelectedChannel = static_cast<EngineChannelBase<V, R, I>*>(engineChannels[iChannelIndex]);

                        #if CONFIG_DEVMODE
                        if (pSelectedChannel == pBegin) {
                            dmsg(1,("FATAL ERROR: voice stealing endless loop!\n"));
                            dmsg(1,("VoiceSpawnsLeft=%d.\n", VoiceSpawnsLeft));
                            dmsg(1,("Exiting.\n"));
                            exit(-1);
                        }
                        #endif // CONFIG_DEVMODE
                    }

                    // jump point if a 'stealable' voice was found
                    stealable_voice_found:

                    #if CONFIG_DEVMODE
                    if (!itSelectedVoice->IsActive()) {
                        dmsg(1,("EngineBase: ERROR, tried to steal a voice which was not active !!!\n"));
                        return -1;
                    }
                    #endif // CONFIG_DEVMODE

                    // now kill the selected voice
                    itSelectedVoice->Kill(itNoteOnEvent);

                    --VoiceSpawnsLeft;

                    return 0; // success
                }
                else {
                    dmsg(1,("Event pool emtpy!\n"));
                    return -1;
                }
            }

            void HandleInstrumentChanges() {
                bool instrumentChanged = false;
                for (int i = 0; i < engineChannels.size(); i++) {
                    EngineChannelBase<V, R, I>* pEngineChannel =
                        static_cast<EngineChannelBase<V, R, I>*>(engineChannels[i]);

                    // as we're going to (carefully) write some status to the
                    // synchronized struct, we cast away the const
                    InstrumentChangeCmd<R, I>& cmd =
                        const_cast<InstrumentChangeCmd<R, I>&>(pEngineChannel->InstrumentChangeCommandReader.Lock());

                    pEngineChannel->pRegionsInUse = cmd.pRegionsInUse;
                    pEngineChannel->pRegionsInUse->clear();

                    if (cmd.bChangeInstrument) {
                        // change instrument
                        dmsg(5,("Engine: instrument change command received\n"));
                        cmd.bChangeInstrument = false;
                        pEngineChannel->pInstrument = cmd.pInstrument;
                        instrumentChanged = true;

                        pEngineChannel->MarkAllActiveVoicesAsOrphans();
                    }
                }

                if (instrumentChanged) {
                    //TODO: this is a lazy solution ATM and not safe in case somebody is currently editing the instrument we're currently switching to (we should store all suspended regions on instrument manager side and when switching to another instrument copy that list to the engine's local list of suspensions
                    ResetSuspendedRegions();
                }
            }

            /**
             * Render all 'normal' voices (that is voices which were not stolen in
             * this fragment) on the given engine channel.
             *
             * @param pEngineChannel - engine channel on which audio should be
             *                         rendered
             * @param Samples        - amount of sample points to be rendered in
             *                         this audio fragment cycle
             */
            void RenderActiveVoices(EngineChannel* pEngineChannel, uint Samples) {
                #if !CONFIG_PROCESS_MUTED_CHANNELS
                if (pEngineChannel->GetMute()) return; // skip if sampler channel is muted
                #endif

                EngineChannelBase<V, R, I>* pChannel =
                    static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);
                pChannel->RenderActiveVoices(Samples);

                ActiveVoiceCountTemp += pEngineChannel->GetVoiceCount();
            }

            /**
             * Render all stolen voices (only voices which were stolen in this
             * fragment) on the given engine channel. Stolen voices are rendered
             * after all normal voices have been rendered; this is needed to render
             * audio of those voices which were selected for voice stealing until
             * the point were the stealing (that is the take over of the voice)
             * actually happened.
             *
             * @param pEngineChannel - engine channel on which audio should be
             *                         rendered
             * @param Samples        - amount of sample points to be rendered in
             *                         this audio fragment cycle
             */
            void RenderStolenVoices(uint Samples) {
                RTList<Event>::Iterator itVoiceStealEvent = pVoiceStealingQueue->first();
                RTList<Event>::Iterator end               = pVoiceStealingQueue->end();
                for (; itVoiceStealEvent != end; ++itVoiceStealEvent) {
                    EngineChannelBase<V, R, I>* pEngineChannel =
                        static_cast<EngineChannelBase<V, R, I>*>(itVoiceStealEvent->pEngineChannel);;
                    if (!pEngineChannel->pInstrument) continue; // ignore if no instrument loaded
                    PoolVoiceIterator itNewVoice =
                        LaunchVoice(pEngineChannel, itVoiceStealEvent, itVoiceStealEvent->Param.Note.Layer, itVoiceStealEvent->Param.Note.ReleaseTrigger, false, false);
                    if (itNewVoice) {
                        itNewVoice->Render(Samples);
                        if (itNewVoice->IsActive()) { // still active
                            *(pEngineChannel->pRegionsInUse->allocAppend()) = itNewVoice->GetRegion();
                            ActiveVoiceCountTemp++;
                            pEngineChannel->SetVoiceCount(pEngineChannel->GetVoiceCount() + 1);

                            if (itNewVoice->PlaybackState == Voice::playback_state_disk) {
                                if (itNewVoice->DiskStreamRef.State != Stream::state_unused) {
                                    pEngineChannel->SetDiskStreamCount(pEngineChannel->GetDiskStreamCount() + 1);
                                }
                            }
                        } else { // voice reached end, is now inactive
                            pEngineChannel->FreeVoice(itNewVoice); // remove voice from the list of active voices
                        }
                    }
                    else dmsg(1,("EngineBase: ERROR, voice stealing didn't work out!\n"));

                    // we need to clear the key's event list explicitly here in case key was never active
                    MidiKey* pKey = &pEngineChannel->pMIDIKeyInfo[itVoiceStealEvent->Param.Note.Key];
                    pKey->VoiceTheftsQueued--;
                    if (!pKey->Active && !pKey->VoiceTheftsQueued) pKey->pEvents->clear();
                }
            }

            /**
             * Free all keys which have turned inactive in this audio fragment, from
             * the list of active keys and clear all event lists on that engine
             * channel.
             *
             * @param pEngineChannel - engine channel to cleanup
             */
            void PostProcess(EngineChannel* pEngineChannel) {
                EngineChannelBase<V, R, I>* pChannel =
                    static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);
                 pChannel->FreeAllInactiveKyes();

                // empty the engine channel's own event lists
                pChannel->ClearEventLists();
            }

            /**
             * Process MIDI control change events with hard coded behavior,
             * that is controllers whose behavior is defined independently
             * of the actual sampler engine type and instrument.
             *
             * @param pEngineChannel - engine channel on which the MIDI CC event was received
             * @param itControlChangeEvent - the actual MIDI CC event
             */
            void ProcessHardcodedControllers (
                EngineChannel*          pEngineChannel,
                Pool<Event>::Iterator&  itControlChangeEvent
            ) {
                EngineChannelBase<V, R, I>* pChannel =
                    static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);

                switch (itControlChangeEvent->Param.CC.Controller) {
                    case 5: { // portamento time
                        pChannel->PortamentoTime = (float) itControlChangeEvent->Param.CC.Value / 127.0f * (float) CONFIG_PORTAMENTO_TIME_MAX + (float) CONFIG_PORTAMENTO_TIME_MIN;
                        break;
                    }
                    case 6: { // data entry (currently only used for RPN and NRPN controllers)
                        //dmsg(1,("DATA ENTRY %d\n", itControlChangeEvent->Param.CC.Value));
                        if (pChannel->GetMidiRpnController() >= 0) { // RPN controller number was sent previously ...
                            dmsg(4,("Guess it's an RPN ...\n"));
                            if (pChannel->GetMidiRpnController() == 2) { // coarse tuning in half tones
                                int transpose = (int) itControlChangeEvent->Param.CC.Value - 64;
                                // limit to +- two octaves for now
                                transpose = RTMath::Min(transpose,  24);
                                transpose = RTMath::Max(transpose, -24);
                                pChannel->GlobalTranspose = transpose;
                                // workaround, so we won't have hanging notes
                                pChannel->ReleaseAllVoices(itControlChangeEvent);
                            }
                            // to prevent other MIDI CC #6 messages to be misenterpreted as RPN controller data
                            pChannel->ResetMidiRpnController();
                        } else if (pChannel->GetMidiNrpnController() >= 0) { // NRPN controller number was sent previously ...
                            dmsg(4,("Guess it's an NRPN ...\n"));
                            const int NrpnCtrlMSB = pChannel->GetMidiNrpnController() >> 8; 
                            const int NrpnCtrlLSB = pChannel->GetMidiNrpnController() & 0xff;
                            dmsg(4,("NRPN MSB=%d LSB=%d Data=%d\n", NrpnCtrlMSB, NrpnCtrlLSB, itControlChangeEvent->Param.CC.Value));
                            switch (NrpnCtrlMSB) {
                                case 0x1a: { // volume level of note (Roland GS NRPN)
                                    const uint note = NrpnCtrlLSB;
                                    const uint vol  = itControlChangeEvent->Param.CC.Value;
                                    dmsg(4,("Note Volume NRPN received (note=%d,vol=%d).\n", note, vol));
                                    if (note < 128 && vol < 128)
                                        pChannel->pMIDIKeyInfo[note].Volume = VolumeCurve[vol];
                                    break;
                                }
                                case 0x1c: { // panpot of note (Roland GS NRPN)
                                    const uint note = NrpnCtrlLSB;
                                    const uint pan  = itControlChangeEvent->Param.CC.Value;
                                    dmsg(4,("Note Pan NRPN received (note=%d,pan=%d).\n", note, pan));
                                    if (note < 128 && pan < 128) {
                                        pChannel->pMIDIKeyInfo[note].PanLeft  = PanCurve[128 - pan];
                                        pChannel->pMIDIKeyInfo[note].PanRight = PanCurve[pan];
                                    }
                                    break;
                                }
                                case 0x1d: { // reverb send of note (Roland GS NRPN)
                                    const uint note = NrpnCtrlLSB;
                                    const float reverb = float(itControlChangeEvent->Param.CC.Value) / 127.0f;
                                    dmsg(4,("Note Reverb Send NRPN received (note=%d,send=%d).\n", note, reverb));
                                    if (note < 128)
                                        pChannel->pMIDIKeyInfo[note].ReverbSend = reverb;
                                    break;
                                }
                                case 0x1e: { // chorus send of note (Roland GS NRPN)
                                    const uint note = NrpnCtrlLSB;
                                    const float chorus = float(itControlChangeEvent->Param.CC.Value) / 127.0f;
                                    dmsg(4,("Note Chorus Send NRPN received (note=%d,send=%d).\n", note, chorus));
                                    if (note < 128)
                                        pChannel->pMIDIKeyInfo[note].ChorusSend = chorus;
                                    break;
                                }
                            }
                            // to prevent other MIDI CC #6 messages to be misenterpreted as NRPN controller data
                            pChannel->ResetMidiNrpnController();
                        }
                        break;
                    }
                    case 7: { // volume
                        //TODO: not sample accurate yet
                        pChannel->MidiVolume = VolumeCurve[itControlChangeEvent->Param.CC.Value];
                        pChannel->bStatusChanged = true; // engine channel status has changed, so set notify flag
                        break;
                    }
                    case 10: { // panpot
                        //TODO: not sample accurate yet
                        pChannel->iLastPanRequest = itControlChangeEvent->Param.CC.Value;
                        break;
                    }
                    case 64: { // sustain
                        if (itControlChangeEvent->Param.CC.Value >= 64 && !pChannel->SustainPedal) {
                            dmsg(4,("DAMPER (RIGHT) PEDAL DOWN\n"));
                            pChannel->SustainPedal = true;
                            pChannel->listeners.PreProcessSustainPedalDown();

                            #if !CONFIG_PROCESS_MUTED_CHANNELS
                            if (pEngineChannel->GetMute()) { // skip if sampler channel is muted
                                pChannel->listeners.PostProcessSustainPedalDown();
                                return;
                            }
                            #endif

                            pChannel->ProcessSustainPedalDown(itControlChangeEvent);
                            pChannel->listeners.PostProcessSustainPedalDown();
                        }
                        if (itControlChangeEvent->Param.CC.Value < 64 && pChannel->SustainPedal) {
                            dmsg(4,("DAMPER (RIGHT) PEDAL UP\n"));
                            pChannel->SustainPedal = false;
                            pChannel->listeners.PreProcessSustainPedalUp();

                            #if !CONFIG_PROCESS_MUTED_CHANNELS
                            if (pChannel->GetMute()) { // skip if sampler channel is muted
                                pChannel->listeners.PostProcessSustainPedalUp();
                                return;
                            }
                            #endif

                            pChannel->ProcessSustainPedalUp(itControlChangeEvent);
                            pChannel->listeners.PostProcessSustainPedalUp();
                        }
                        break;
                    }
                    case 65: { // portamento on / off
                        const bool bPortamento = itControlChangeEvent->Param.CC.Value >= 64;
                        if (bPortamento != pChannel->PortamentoMode)
                            KillAllVoices(pChannel, itControlChangeEvent);
                        pChannel->PortamentoMode = bPortamento;
                        break;
                    }
                    case 66: { // sostenuto
                        if (itControlChangeEvent->Param.CC.Value >= 64 && !pChannel->SostenutoPedal) {
                            dmsg(4,("SOSTENUTO (CENTER) PEDAL DOWN\n"));
                            pChannel->SostenutoPedal = true;
                            pChannel->listeners.PreProcessSostenutoPedalDown();

                            #if !CONFIG_PROCESS_MUTED_CHANNELS
                            if (pEngineChannel->GetMute()) { // skip if sampler channel is muted
                                pChannel->listeners.PostProcessSostenutoPedalDown();
                                return;
                            }
                            #endif

                            pChannel->ProcessSostenutoPedalDown();
                            pChannel->listeners.PostProcessSostenutoPedalDown();
                        }
                        if (itControlChangeEvent->Param.CC.Value < 64 && pChannel->SostenutoPedal) {
                            dmsg(4,("SOSTENUTO (CENTER) PEDAL UP\n"));
                            pChannel->SostenutoPedal = false;
                            pChannel->listeners.PreProcessSostenutoPedalUp();

                            #if !CONFIG_PROCESS_MUTED_CHANNELS
                            if (pEngineChannel->GetMute()) { // skip if sampler channel is muted
                                pChannel->listeners.PostProcessSostenutoPedalUp();
                                return;
                            }
                            #endif

                            pChannel->ProcessSostenutoPedalUp(itControlChangeEvent);
                            pChannel->listeners.PostProcessSostenutoPedalUp();
                        }
                        break;
                    }
                    case 98: { // NRPN controller LSB
                        dmsg(4,("NRPN LSB %d\n", itControlChangeEvent->Param.CC.Value));
                        pEngineChannel->SetMidiNrpnControllerLsb(itControlChangeEvent->Param.CC.Value);
                        break;
                    }
                    case 99: { // NRPN controller MSB
                        dmsg(4,("NRPN MSB %d\n", itControlChangeEvent->Param.CC.Value));
                        pEngineChannel->SetMidiNrpnControllerMsb(itControlChangeEvent->Param.CC.Value);
                        break;
                    }
                    case 100: { // RPN controller LSB
                        dmsg(4,("RPN LSB %d\n", itControlChangeEvent->Param.CC.Value));
                        pEngineChannel->SetMidiRpnControllerLsb(itControlChangeEvent->Param.CC.Value);
                        break;
                    }
                    case 101: { // RPN controller MSB
                        dmsg(4,("RPN MSB %d\n", itControlChangeEvent->Param.CC.Value));
                        pEngineChannel->SetMidiRpnControllerMsb(itControlChangeEvent->Param.CC.Value);
                        break;
                    }


                    // Channel Mode Messages

                    case 120: { // all sound off
                        KillAllVoices(pEngineChannel, itControlChangeEvent);
                        break;
                    }
                    case 121: { // reset all controllers
                        pChannel->ResetControllers();
                        break;
                    }
                    case 123: { // all notes off
                        #if CONFIG_PROCESS_ALL_NOTES_OFF
                        pChannel->ReleaseAllVoices(itControlChangeEvent);
                        #endif // CONFIG_PROCESS_ALL_NOTES_OFF
                        break;
                    }
                    case 126: { // mono mode on
                        if (!pChannel->SoloMode)
                            KillAllVoices(pEngineChannel, itControlChangeEvent);
                        pChannel->SoloMode = true;
                        break;
                    }
                    case 127: { // poly mode on
                        if (pChannel->SoloMode)
                            KillAllVoices(pEngineChannel, itControlChangeEvent);
                        pChannel->SoloMode = false;
                        break;
                    }
                }
            }

            virtual D* CreateDiskThread() = 0;

            /**
             *  Assigns and triggers a new voice for the respective MIDI key.
             *
             *  @param pEngineChannel - engine channel on which this event occured on
             *  @param itNoteOnEvent - key, velocity and time stamp of the event
             */
            virtual void ProcessNoteOn(EngineChannel* pEngineChannel, Pool<Event>::Iterator& itNoteOnEvent) {
                EngineChannelBase<V, R, I>* pChannel =
                        static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);

                //HACK: we should better add the transpose value only to the most mandatory places (like for retrieving the region and calculating the tuning), because otherwise voices will unintendedly survive when changing transpose while playing
                int k = itNoteOnEvent->Param.Note.Key + pChannel->GlobalTranspose;
                if (k < 0 || k > 127) return; //ignore keys outside the key range

                itNoteOnEvent->Param.Note.Key += pChannel->GlobalTranspose;
                int vel = itNoteOnEvent->Param.Note.Velocity;

                const int key = itNoteOnEvent->Param.Note.Key;
                MidiKey* pKey = &pChannel->pMIDIKeyInfo[key];

                pChannel->listeners.PreProcessNoteOn(key, vel);
                #if !CONFIG_PROCESS_MUTED_CHANNELS
                if (pEngineChannel->GetMute()) { // skip if sampler channel is muted
                    pChannel->listeners.PostProcessNoteOn(key, vel);
                    return;
                }
                #endif

                if (!pChannel->pInstrument) {
                    pChannel->listeners.PostProcessNoteOn(key, vel);
                    return; // ignore if no instrument loaded
                }

                // move note on event to the key's own event list
                RTList<Event>::Iterator itNoteOnEventOnKeyList = itNoteOnEvent.moveToEndOf(pKey->pEvents);

                // if Solo Mode then kill all already active voices
                if (pChannel->SoloMode) {
                    Pool<uint>::Iterator itYoungestKey = pChannel->pActiveKeys->last();
                    if (itYoungestKey) {
                        const int iYoungestKey = *itYoungestKey;
                        const MidiKey* pOtherKey = &pChannel->pMIDIKeyInfo[iYoungestKey];
                        if (pOtherKey->Active) {
                            // get final portamento position of currently active voice
                            if (pChannel->PortamentoMode) {
                                VoiceIterator itVoice = pOtherKey->pActiveVoices->last();
                                if (itVoice) itVoice->UpdatePortamentoPos(itNoteOnEventOnKeyList);
                            }
                            // kill all voices on the (other) key
                            VoiceIterator itVoiceToBeKilled = pOtherKey->pActiveVoices->first();
                            VoiceIterator end               = pOtherKey->pActiveVoices->end();
                            for (; itVoiceToBeKilled != end; ++itVoiceToBeKilled) {
                                if (!(itVoiceToBeKilled->Type & Voice::type_release_trigger))
                                    itVoiceToBeKilled->Kill(itNoteOnEventOnKeyList);
                            }
                        }
                    }
                    // set this key as 'currently active solo key'
                    pChannel->SoloKey = key;
                }

                pChannel->ProcessKeySwitchChange(key);

                pKey->KeyPressed = true; // the MIDI key was now pressed down
                pKey->Velocity   = itNoteOnEventOnKeyList->Param.Note.Velocity;
                pKey->NoteOnTime = FrameTime + itNoteOnEventOnKeyList->FragmentPos(); // will be used to calculate note length

                // cancel release process of voices on this key if needed
                if (pKey->Active && !pChannel->SustainPedal) {
                    RTList<Event>::Iterator itCancelReleaseEvent = pKey->pEvents->allocAppend();
                    if (itCancelReleaseEvent) {
                        *itCancelReleaseEvent = *itNoteOnEventOnKeyList;         // copy event
                        itCancelReleaseEvent->Type = Event::type_cancel_release; // transform event type
                    }
                    else dmsg(1,("Event pool emtpy!\n"));
                }

                TriggerNewVoices(pEngineChannel, itNoteOnEventOnKeyList);

                // if neither a voice was spawned or postponed then remove note on event from key again
                if (!pKey->Active && !pKey->VoiceTheftsQueued)
                    pKey->pEvents->free(itNoteOnEventOnKeyList);

                if (!pChannel->SoloMode || pChannel->PortamentoPos < 0.0f) pChannel->PortamentoPos = (float) key;
                if (pKey->pRoundRobinIndex) {
                    (*pKey->pRoundRobinIndex)++; // counter specific for the key or region
                    pChannel->RoundRobinIndex++; // common counter for the channel
                }
                pChannel->listeners.PostProcessNoteOn(key, vel);
            }

            /**
             * Allocate and trigger new voice(s) for the key.
             */
            virtual void TriggerNewVoices (
                EngineChannel*            pEngineChannel,
                RTList<Event>::Iterator&  itNoteOnEvent,
                bool                      HandleKeyGroupConflicts = true
            ) = 0;

            /**
             * Allocate and trigger release voice(s) for the key.
             */
            virtual void TriggerReleaseVoices (
                EngineChannel*            pEngineChannel,
                RTList<Event>::Iterator&  itNoteOffEvent
            ) = 0;

            /**
             *  Releases the voices on the given key if sustain pedal is not pressed.
             *  If sustain is pressed, the release of the note will be postponed until
             *  sustain pedal will be released or voice turned inactive by itself (e.g.
             *  due to completion of sample playback).
             *
             *  @param pEngineChannel - engine channel on which this event occured on
             *  @param itNoteOffEvent - key, velocity and time stamp of the event
             */
            virtual void ProcessNoteOff(EngineChannel* pEngineChannel, Pool<Event>::Iterator& itNoteOffEvent) {
                EngineChannelBase<V, R, I>* pChannel = static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);

                int k = itNoteOffEvent->Param.Note.Key + pChannel->GlobalTranspose;
                if (k < 0 || k > 127) return; //ignore keys outside the key range

                //HACK: we should better add the transpose value only to the most mandatory places (like for retrieving the region and calculating the tuning), because otherwise voices will unintendedly survive when changing transpose while playing
                itNoteOffEvent->Param.Note.Key += pChannel->GlobalTranspose;
                int vel = itNoteOffEvent->Param.Note.Velocity;

                const int iKey = itNoteOffEvent->Param.Note.Key;
                MidiKey* pKey = &pChannel->pMIDIKeyInfo[iKey];

                pChannel->listeners.PreProcessNoteOff(iKey, vel);

                #if !CONFIG_PROCESS_MUTED_CHANNELS
                if (pEngineChannel->GetMute()) { // skip if sampler channel is muted
                    pChannel->listeners.PostProcessNoteOff(iKey, vel);
                    return;
                }
                #endif

                pKey->KeyPressed = false; // the MIDI key was now released

                // move event to the key's own event list
                RTList<Event>::Iterator itNoteOffEventOnKeyList = itNoteOffEvent.moveToEndOf(pKey->pEvents);

                bool bShouldRelease = pKey->Active && pChannel->ShouldReleaseVoice(itNoteOffEventOnKeyList->Param.Note.Key);

                // in case Solo Mode is enabled, kill all voices on this key and respawn a voice on the highest pressed key (if any)
                if (pChannel->SoloMode && pChannel->pInstrument) { //TODO: this feels like too much code just for handling solo mode :P
                    bool bOtherKeysPressed = false;
                    if (iKey == pChannel->SoloKey) {
                        pChannel->SoloKey = -1;
                        // if there's still a key pressed down, respawn a voice (group) on the highest key
                        for (int i = 127; i > 0; i--) {
                            MidiKey* pOtherKey = &pChannel->pMIDIKeyInfo[i];
                            if (pOtherKey->KeyPressed) {
                                bOtherKeysPressed = true;
                                // make the other key the new 'currently active solo key'
                                pChannel->SoloKey = i;
                                // get final portamento position of currently active voice
                                if (pChannel->PortamentoMode) {
                                    VoiceIterator itVoice = pKey->pActiveVoices->first();
                                    if (itVoice) itVoice->UpdatePortamentoPos(itNoteOffEventOnKeyList);
                                }
                                // create a pseudo note on event
                                RTList<Event>::Iterator itPseudoNoteOnEvent = pOtherKey->pEvents->allocAppend();
                                if (itPseudoNoteOnEvent) {
                                    // copy event
                                    *itPseudoNoteOnEvent = *itNoteOffEventOnKeyList;
                                    // transform event to a note on event
                                    itPseudoNoteOnEvent->Type                = Event::type_note_on;
                                    itPseudoNoteOnEvent->Param.Note.Key      = i;
                                    itPseudoNoteOnEvent->Param.Note.Velocity = pOtherKey->Velocity;
                                    // allocate and trigger new voice(s) for the other key
                                    TriggerNewVoices(pChannel, itPseudoNoteOnEvent, false);
                                    // if neither a voice was spawned or postponed then remove note on event from key again
                                    if (!pOtherKey->Active && !pOtherKey->VoiceTheftsQueued)
                                        pOtherKey->pEvents->free(itPseudoNoteOnEvent);

                                } else dmsg(1,("Could not respawn voice, no free event left\n"));
                                break; // done
                            }
                        }
                    }
                    if (bOtherKeysPressed) {
                        if (pKey->Active) { // kill all voices on this key
                            bShouldRelease = false; // no need to release, as we kill it here
                            VoiceIterator itVoiceToBeKilled = pKey->pActiveVoices->first();
                            VoiceIterator end               = pKey->pActiveVoices->end();
                            for (; itVoiceToBeKilled != end; ++itVoiceToBeKilled) {
                                if (!(itVoiceToBeKilled->Type & Voice::type_release_trigger))
                                    itVoiceToBeKilled->Kill(itNoteOffEventOnKeyList);
                            }
                        }
                    } else pChannel->PortamentoPos = -1.0f;
                }

                // if no solo mode (the usual case) or if solo mode and no other key pressed, then release voices on this key if needed
                if (bShouldRelease) {
                    itNoteOffEventOnKeyList->Type = Event::type_release; // transform event type

                    // spawn release triggered voice(s) if needed
                    if (pKey->ReleaseTrigger && pChannel->pInstrument) {
                        TriggerReleaseVoices(pChannel, itNoteOffEventOnKeyList);
                        pKey->ReleaseTrigger = false;
                    }
                }

                // if neither a voice was spawned or postponed on this key then remove note off event from key again
                if (!pKey->Active && !pKey->VoiceTheftsQueued)
                    pKey->pEvents->free(itNoteOffEventOnKeyList);

                pChannel->listeners.PostProcessNoteOff(iKey, vel);
            }

            /**
             *  Reset all voices and disk thread and clear input event queue and all
             *  control and status variables. This method is protected by a mutex.
             */
            virtual void ResetInternal() {
                LockGuard lock(ResetInternalMutex);

                // make sure that the engine does not get any sysex messages
                // while it's reseting
                bool sysexDisabled = MidiInputPort::RemoveSysexListener(this);
                SetVoiceCount(0);
                ActiveVoiceCountMax = 0;

                // reset voice stealing parameters
                pVoiceStealingQueue->clear();
                itLastStolenVoice          = VoiceIterator();
                itLastStolenVoiceGlobally  = VoiceIterator();
                iuiLastStolenKey           = RTList<uint>::Iterator();
                iuiLastStolenKeyGlobally   = RTList<uint>::Iterator();
                pLastStolenChannel         = NULL;

                // reset all voices
                for (VoiceIterator iterVoice = pVoicePool->allocAppend(); iterVoice == pVoicePool->last(); iterVoice = pVoicePool->allocAppend()) {
                    iterVoice->Reset();
                }
                pVoicePool->clear();

                // reset disk thread
                if (pDiskThread) pDiskThread->Reset();

                // delete all input events
                pEventQueue->init();
                pSysexBuffer->init();
                if (sysexDisabled) MidiInputPort::AddSysexListener(this);
            }

            /**
             * Kills all voices on an engine channel as soon as possible. Voices
             * won't get into release state, their volume level will be ramped down
             * as fast as possible.
             *
             * @param pEngineChannel - engine channel on which all voices should be killed
             * @param itKillEvent    - event which caused this killing of all voices
             */
            virtual void KillAllVoices(EngineChannel* pEngineChannel, Pool<Event>::Iterator& itKillEvent) {
                EngineChannelBase<V, R, I>* pChannel = static_cast<EngineChannelBase<V, R, I>*>(pEngineChannel);
                int count = pChannel->KillAllVoices(itKillEvent);
                VoiceSpawnsLeft -= count; //FIXME: just a temporary workaround, we should check the cause in StealVoice() instead
            }

            /**
             *  Allocates and triggers a new voice. This method will usually be
             *  called by the ProcessNoteOn() method and by the voices itself
             *  (e.g. to spawn further voices on the same key for layered sounds).
             *
             *  @param pEngineChannel      - engine channel on which this event occured on
             *  @param itNoteOnEvent       - key, velocity and time stamp of the event
             *  @param iLayer              - layer index for the new voice (optional - only
             *                               in case of layered sounds of course)
             *  @param ReleaseTriggerVoice - if new voice is a release triggered voice
             *                               (optional, default = false)
             *  @param VoiceStealing       - if voice stealing should be performed
             *                               when there is no free voice
             *                               (optional, default = true)
             *  @param HandleKeyGroupConflicts - if voices should be killed due to a
             *                                   key group conflict
             *  @returns pointer to new voice or NULL if there was no free voice or
             *           if the voice wasn't triggered (for example when no region is
             *           defined for the given key).
             */
            virtual PoolVoiceIterator LaunchVoice (
                EngineChannel*          pEngineChannel,
                Pool<Event>::Iterator&  itNoteOnEvent,
                int                     iLayer,
                bool                    ReleaseTriggerVoice,
                bool                    VoiceStealing,
                bool                    HandleKeyGroupConflicts
            ) = 0;

            virtual int GetMinFadeOutSamples() { return MinFadeOutSamples; }

            int InitNewVoice (
                EngineChannelBase<V, R, I>*  pChannel,
                R*                           pRegion,
                Pool<Event>::Iterator&       itNoteOnEvent,
                Voice::type_t                VoiceType,
                int                          iLayer,
                int                          iKeyGroup,
                bool                         ReleaseTriggerVoice,
                bool                         VoiceStealing,
                typename Pool<V>::Iterator&  itNewVoice
            ) {
                int key = itNoteOnEvent->Param.Note.Key;
                typename MidiKeyboardManager<V>::MidiKey* pKey = &pChannel->pMIDIKeyInfo[key];
                if (itNewVoice) {
                    // launch the new voice
                    if (itNewVoice->Trigger(pChannel, itNoteOnEvent, pChannel->Pitch, pRegion, VoiceType, iKeyGroup) < 0) {
                        dmsg(4,("Voice not triggered\n"));
                        pKey->pActiveVoices->free(itNewVoice);
                    }
                    else { // on success
                        --VoiceSpawnsLeft;
                        if (!pKey->Active) { // mark as active key
                            pKey->Active = true;
                            pKey->itSelf = pChannel->pActiveKeys->allocAppend();
                            *pKey->itSelf = itNoteOnEvent->Param.Note.Key;
                        }
                        if (itNewVoice->Type & Voice::type_release_trigger_required) pKey->ReleaseTrigger = true; // mark key for the need of release triggered voice(s)
                        return 0; // success
                    }
                }
                else if (VoiceStealing) {
                    // try to steal one voice
                    int result = StealVoice(pChannel, itNoteOnEvent);
                    if (!result) { // voice stolen successfully
                        // put note-on event into voice-stealing queue, so it will be reprocessed after killed voice died
                        RTList<Event>::Iterator itStealEvent = pVoiceStealingQueue->allocAppend();
                        if (itStealEvent) {
                            *itStealEvent = *itNoteOnEvent; // copy event
                            itStealEvent->Param.Note.Layer = iLayer;
                            itStealEvent->Param.Note.ReleaseTrigger = ReleaseTriggerVoice;
                            pKey->VoiceTheftsQueued++;
                        }
                        else dmsg(1,("Voice stealing queue full!\n"));
                    }
                }

                return -1;
            }

        private:
            Pool<V>*    pVoicePool;            ///< Contains all voices that can be activated.
            Pool<RR*>   SuspendedRegions;
            Mutex       SuspendedRegionsMutex;
            Condition   SuspensionChangeOngoing;
            RR*         pPendingRegionSuspension;
            RR*         pPendingRegionResumption;
            int         iPendingStreamDeletions;
    };

    template <class V, class RR, class R, class D, class IM, class I>
    IM EngineBase<V, RR, R, D, IM, I>::instruments;

} // namespace LinuxSampler

#endif  /* __LS_ENGINEBASE_H__ */
