/**********************************************************************
 
 Audacity: A Digital Audio Editor
 
 RealtimeEffectManager.cpp
 
 Paul Licameli split from EffectManager.cpp
 
 **********************************************************************/


#include "RealtimeEffectManager.h"

#include "tenacity/EffectInterface.h"
#include <memory>

#include <atomic>
#include <wx/time.h>

class RealtimeEffectState
{
public:
   explicit RealtimeEffectState( EffectClientInterface &effect );

   EffectClientInterface &GetEffect() const { return mEffect; }

   bool RealtimeSuspend();
   bool RealtimeResume();
   bool RealtimeAddProcessor(int group, unsigned chans, float rate);
   size_t RealtimeProcess(int group,
      unsigned chans, float **inbuf, float **outbuf, size_t numSamples);
   bool IsRealtimeActive();

private:
   EffectClientInterface &mEffect;

   std::vector<int> mGroupProcessor;
   int mCurrentProcessor;

   std::atomic<int> mRealtimeSuspendCount{ 1 };    // Effects are initially suspended
};

RealtimeEffectManager & RealtimeEffectManager::Get()
{
   static RealtimeEffectManager rem;
   return rem;
}

RealtimeEffectManager::RealtimeEffectManager()
{
   mRealtimeLock.Enter();
   mRealtimeActive = false;
   mRealtimeSuspended = true;
   mRealtimeLatency = 0;
   mRealtimeLock.Leave();
}

RealtimeEffectManager::~RealtimeEffectManager()
{
}

#if defined(EXPERIMENTAL_EFFECTS_RACK)
void RealtimeEffectManager::RealtimeSetEffects(const EffectArray & effects)
{
   // Block RealtimeProcess()
   RealtimeSuspend();

   decltype( mStates ) newStates;
   auto begin = mStates.begin(), end = mStates.end();
   for ( auto pEffect : effects ) {
      auto found = std::find_if( begin, end,
         [=]( const decltype( mStates )::value_type &state ){
            return state && &state->GetEffect() == pEffect;
         }
      );
      if ( found == end ) {
         // Tell New effect to get ready
         pEffect->RealtimeInitialize();
         newStates.emplace_back(
            std::make_unique< RealtimeEffectState >( *pEffect ) );
      }
      else {
         // Preserve state for effect that remains in the chain
         newStates.emplace_back( std::move( *found ) );
      }
   }

   // Remaining states that were not moved need to clean up
   for ( auto &state : mStates ) {
      if ( state )
         state->GetEffect().RealtimeFinalize();
   }

   // Get rid of the old chain
   // And install the NEW one
   mStates.swap( newStates );

   // Allow RealtimeProcess() to, well, process 
   RealtimeResume();
}
#endif

bool RealtimeEffectManager::RealtimeIsActive()
{
   return mStates.size() != 0;
}

bool RealtimeEffectManager::RealtimeIsSuspended()
{
   return mRealtimeSuspended;
}

void RealtimeEffectManager::RealtimeAddEffect(EffectClientInterface *effect)
{
   // Block RealtimeProcess()
   RealtimeSuspend();

   // Add to list of active effects
   mStates.emplace_back( std::make_unique< RealtimeEffectState >( *effect ) );
   auto &state = mStates.back();

   // Initialize effect if realtime is already active
   if (mRealtimeActive)
   {
      // Initialize realtime processing
      effect->RealtimeInitialize();

      // Add the required processors
      for (size_t i = 0, cnt = mRealtimeChans.size(); i < cnt; i++)
      {
         state->RealtimeAddProcessor(i, mRealtimeChans[i], mRealtimeRates[i]);
      }
   }
   

   // Allow RealtimeProcess() to, well, process 
   RealtimeResume();
}

void RealtimeEffectManager::RealtimeRemoveEffect(EffectClientInterface *effect)
{
   // Block RealtimeProcess()
   RealtimeSuspend();

   if (mRealtimeActive)
   {
      // Cleanup realtime processing
      effect->RealtimeFinalize();
   }
      
   // Remove from list of active effects
   auto end = mStates.end();
   auto found = std::find_if( mStates.begin(), end,
      [&](const decltype(mStates)::value_type &state){
         return &state->GetEffect() == effect;
      }
   );
   if (found != end)
      mStates.erase(found);

   // Allow RealtimeProcess() to, well, process 
   RealtimeResume();
}

void RealtimeEffectManager::RealtimeInitialize(double rate)
{
   // The audio thread should not be running yet, but protect anyway
   RealtimeSuspend();

   // (Re)Set processor parameters
   mRealtimeChans.clear();
   mRealtimeRates.clear();

   // RealtimeAdd/RemoveEffect() needs to know when we're active so it can
   // initialize newly added effects
   mRealtimeActive = true;

   // Tell each effect to get ready for action
   for (auto &state : mStates) {
      state->GetEffect().SetSampleRate(rate);
      state->GetEffect().RealtimeInitialize();
   }

   // Get things moving
   RealtimeResume();
}

void RealtimeEffectManager::RealtimeAddProcessor(int group, unsigned chans, float rate)
{
   for (auto &state : mStates)
      state->RealtimeAddProcessor(group, chans, rate);

   mRealtimeChans.push_back(chans);
   mRealtimeRates.push_back(rate);
}

void RealtimeEffectManager::RealtimeFinalize()
{
   // Make sure nothing is going on
   RealtimeSuspend();

   // It is now safe to clean up
   mRealtimeLatency = 0;

   // Tell each effect to clean up as well
   for (auto &state : mStates)
      state->GetEffect().RealtimeFinalize();

   // Reset processor parameters
   mRealtimeChans.clear();
   mRealtimeRates.clear();

   // No longer active
   mRealtimeActive = false;
}

void RealtimeEffectManager::RealtimeSuspend()
{
   mRealtimeLock.Enter();

   // Already suspended...bail
   if (mRealtimeSuspended)
   {
      mRealtimeLock.Leave();
      return;
   }

   // Show that we aren't going to be doing anything
   mRealtimeSuspended = true;

   // And make sure the effects don't either
   for (auto &state : mStates)
      state->RealtimeSuspend();

   mRealtimeLock.Leave();
}

void RealtimeEffectManager::RealtimeSuspendOne( EffectClientInterface &effect )
{
   auto begin = mStates.begin(), end = mStates.end();
   auto found = std::find_if( begin, end,
      [&effect]( const decltype( mStates )::value_type &state ){
         return state && &state->GetEffect() == &effect;
      }
   );
   if ( found != end )
      (*found)->RealtimeSuspend();
}

void RealtimeEffectManager::RealtimeResume()
{
   mRealtimeLock.Enter();

   // Already running...bail
   if (!mRealtimeSuspended)
   {
      mRealtimeLock.Leave();
      return;
   }

   // Tell the effects to get ready for more action
   for (auto &state : mStates)
      state->RealtimeResume();

   // And we should too
   mRealtimeSuspended = false;

   mRealtimeLock.Leave();
}

void RealtimeEffectManager::RealtimeResumeOne( EffectClientInterface &effect )
{
   auto begin = mStates.begin(), end = mStates.end();
   auto found = std::find_if( begin, end,
      [&effect]( const decltype( mStates )::value_type &state ){
         return state && &state->GetEffect() == &effect;
      }
   );
   if ( found != end )
      (*found)->RealtimeResume();
}

//
// This will be called in a different thread than the main GUI thread.
//
void RealtimeEffectManager::RealtimeProcessStart()
{
   // Protect ourselves from the main thread
   mRealtimeLock.Enter();

   // Can be suspended because of the audio stream being paused or because effects
   // have been suspended.
   if (!mRealtimeSuspended)
   {
      for (auto &state : mStates)
      {
         if (state->IsRealtimeActive())
            state->GetEffect().RealtimeProcessStart();
      }
   }

   mRealtimeLock.Leave();
}

//
// This will be called in a different thread than the main GUI thread.
//
size_t RealtimeEffectManager::RealtimeProcess(int group, unsigned chans, float **buffers, size_t numSamples)
{
   // Protect ourselves from the main thread
   mRealtimeLock.Enter();

   // Can be suspended because of the audio stream being paused or because effects
   // have been suspended, so allow the samples to pass as-is.
   if (mRealtimeSuspended || mStates.empty())
   {
      mRealtimeLock.Leave();
      return numSamples;
   }

   // Remember when we started so we can calculate the amount of latency we
   // are introducing
   wxMilliClock_t start = wxGetUTCTimeMillis();

   // Allocate the in/out buffer arrays
   float **ibuf = (float **) alloca(chans * sizeof(float *));
   float **obuf = (float **) alloca(chans * sizeof(float *));

   // And populate the input with the buffers we've been given while allocating
   // NEW output buffers
   for (unsigned int i = 0; i < chans; i++)
   {
      ibuf[i] = buffers[i];
      obuf[i] = (float *) alloca(numSamples * sizeof(float));
   }

   // Now call each effect in the chain while swapping buffer pointers to feed the
   // output of one effect as the input to the next effect
   size_t called = 0;
   for (auto &state : mStates)
   {
      if (state->IsRealtimeActive())
      {
         state->RealtimeProcess(group, chans, ibuf, obuf, numSamples);
         called++;
      }

      for (unsigned int j = 0; j < chans; j++)
      {
         float *temp;
         temp = ibuf[j];
         ibuf[j] = obuf[j];
         obuf[j] = temp;
      }
   }

   // Once we're done, we might wind up with the last effect storing its results
   // in the temporary buffers.  If that's the case, we need to copy it over to
   // the caller's buffers.  This happens when the number of effects processed
   // is odd.
   if (called & 1)
   {
      for (unsigned int i = 0; i < chans; i++)
      {
         memcpy(buffers[i], ibuf[i], numSamples * sizeof(float));
      }
   }

   // Remember the latency
   mRealtimeLatency = (int) (wxGetUTCTimeMillis() - start).GetValue();

   mRealtimeLock.Leave();

   //
   // This is wrong...needs to handle tails
   //
   return numSamples;
}

//
// This will be called in a different thread than the main GUI thread.
//
void RealtimeEffectManager::RealtimeProcessEnd()
{
   // Protect ourselves from the main thread
   mRealtimeLock.Enter();

   // Can be suspended because of the audio stream being paused or because effects
   // have been suspended.
   if (!mRealtimeSuspended)
   {
      for (auto &state : mStates)
      {
         if (state->IsRealtimeActive())
            state->GetEffect().RealtimeProcessEnd();
      }
   }

   mRealtimeLock.Leave();
}

int RealtimeEffectManager::GetRealtimeLatency()
{
   return mRealtimeLatency;
}

RealtimeEffectState::RealtimeEffectState( EffectClientInterface &effect )
   : mEffect{ effect }
{
}

bool RealtimeEffectState::RealtimeSuspend()
{
   auto result = mEffect.RealtimeSuspend();
   if ( result ) {
      mRealtimeSuspendCount++;
   }
   return result;
}

bool RealtimeEffectState::RealtimeResume()
{
   auto result = mEffect.RealtimeResume();
   if ( result ) {
      mRealtimeSuspendCount--;
   }
   return result;
}

// RealtimeAddProcessor and RealtimeProcess use the same method of
// determining the current processor index, so updates to one should
// be reflected in the other.
bool RealtimeEffectState::RealtimeAddProcessor(int group, unsigned chans, float rate)
{
   auto ichans = chans;
   auto ochans = chans;
   auto gchans = chans;

   // Reset processor index
   if (group == 0)
   {
      mCurrentProcessor = 0;
      mGroupProcessor.clear();
   }

   // Remember the processor starting index
   mGroupProcessor.push_back(mCurrentProcessor);

   const auto numAudioIn = mEffect.GetAudioInCount();
   const auto numAudioOut = mEffect.GetAudioOutCount();

   // Call the client until we run out of input or output channels
   while (ichans > 0 && ochans > 0)
   {
      // If we don't have enough input channels to accommodate the client's
      // requirements, then we replicate the input channels until the
      // client's needs are met.
      if (ichans < numAudioIn)
      {
         // All input channels have been consumed
         ichans = 0;
      }
      // Otherwise fulfill the client's needs with as many input channels as possible.
      // After calling the client with this set, we will loop back up to process more
      // of the input/output channels.
      else if (ichans >= numAudioIn)
      {
         gchans = numAudioIn;
         ichans -= gchans;
      }

      // If we don't have enough output channels to accommodate the client's
      // requirements, then we provide all of the output channels and fulfill
      // the client's needs with dummy buffers.  These will just get tossed.
      if (ochans < numAudioOut)
      {
         // All output channels have been consumed
         ochans = 0;
      }
      // Otherwise fulfill the client's needs with as many output channels as possible.
      // After calling the client with this set, we will loop back up to process more
      // of the input/output channels.
      else if (ochans >= numAudioOut)
      {
         ochans -= numAudioOut;
      }

      // Add a NEW processor
      mEffect.RealtimeAddProcessor(gchans, rate);

      // Bump to next processor
      mCurrentProcessor++;
   }

   return true;
}

// RealtimeAddProcessor and RealtimeProcess use the same method of
// determining the current processor group, so updates to one should
// be reflected in the other.
size_t RealtimeEffectState::RealtimeProcess(int group,
                                    unsigned chans,
                                    float **inbuf,
                                    float **outbuf,
                                    size_t numSamples)
{
   //
   // The caller passes the number of channels to process and specifies
   // the number of input and output buffers.  There will always be the
   // same number of output buffers as there are input buffers.
   //
   // Effects always require a certain number of input and output buffers,
   // so if the number of channels we're currently processing are different
   // than what the effect expects, then we use a few methods of satisfying
   // the effects requirements.
   const auto numAudioIn = mEffect.GetAudioInCount();
   const auto numAudioOut = mEffect.GetAudioOutCount();

   float **clientIn = (float **) alloca(numAudioIn * sizeof(float *));
   float **clientOut = (float **) alloca(numAudioOut * sizeof(float *));
   float *dummybuf = (float *) alloca(numSamples * sizeof(float));
   decltype(numSamples) len = 0;
   auto ichans = chans;
   auto ochans = chans;
   auto gchans = chans;
   unsigned indx = 0;
   unsigned ondx = 0;

   int processor = mGroupProcessor[group];

   // Call the client until we run out of input or output channels
   while (ichans > 0 && ochans > 0)
   {
      // If we don't have enough input channels to accommodate the client's
      // requirements, then we replicate the input channels until the
      // client's needs are met.
      if (ichans < numAudioIn)
      {
         for (size_t i = 0; i < numAudioIn; i++)
         {
            if (indx == ichans)
            {
               indx = 0;
            }
            clientIn[i] = inbuf[indx++];
         }

         // All input channels have been consumed
         ichans = 0;
      }
      // Otherwise fulfill the client's needs with as many input channels as possible.
      // After calling the client with this set, we will loop back up to process more
      // of the input/output channels.
      else if (ichans >= numAudioIn)
      {
         gchans = 0;
         for (size_t i = 0; i < numAudioIn; i++, ichans--, gchans++)
         {
            clientIn[i] = inbuf[indx++];
         }
      }

      // If we don't have enough output channels to accommodate the client's
      // requirements, then we provide all of the output channels and fulfill
      // the client's needs with dummy buffers.  These will just get tossed.
      if (ochans < numAudioOut)
      {
         for (size_t i = 0; i < numAudioOut; i++)
         {
            if (i < ochans)
            {
               clientOut[i] = outbuf[i];
            }
            else
            {
               clientOut[i] = dummybuf;
            }
         }

         // All output channels have been consumed
         ochans = 0;
      }
      // Otherwise fulfill the client's needs with as many output channels as possible.
      // After calling the client with this set, we will loop back up to process more
      // of the input/output channels.
      else if (ochans >= numAudioOut)
      {
         for (size_t i = 0; i < numAudioOut; i++, ochans--)
         {
            clientOut[i] = outbuf[ondx++];
         }
      }

      // Finally call the plugin to process the block
      len = 0;
      const auto blockSize = mEffect.GetBlockSize();
      for (decltype(numSamples) block = 0; block < numSamples; block += blockSize)
      {
         auto cnt = std::min(numSamples - block, blockSize);
         len += mEffect.RealtimeProcess(processor, clientIn, clientOut, cnt);

         for (size_t i = 0 ; i < numAudioIn; i++)
         {
            clientIn[i] += cnt;
         }

         for (size_t i = 0 ; i < numAudioOut; i++)
         {
            clientOut[i] += cnt;
         }
      }

      // Bump to next processor
      processor++;
   }

   return len;
}

bool RealtimeEffectState::IsRealtimeActive()
{
   return mRealtimeSuspendCount == 0;
}