amuse/lib/Voice.cpp

#include "amuse/Voice.hpp"

#include <cmath>
#include <cstring>

#include "amuse/AudioGroup.hpp"
#include "amuse/Common.hpp"
#include "amuse/DSPCodec.hpp"
#include "amuse/Engine.hpp"
#include "amuse/IBackendVoice.hpp"
#include "amuse/IBackendVoiceAllocator.hpp"
#include "amuse/N64MusyXCodec.hpp"
#include "amuse/Submix.hpp"
#include "amuse/VolumeTable.hpp"

namespace amuse {

float Voice::VolumeCache::getVolume(float vol, bool dls) {
  if (vol != m_curVolLUTKey || dls != m_curDLS) {
    m_curVolLUTKey = vol;
    m_curDLS = dls;
    if (dls)
      m_curVolLUTVal = LookupDLSVolume(vol);
    else
      m_curVolLUTVal = LookupVolume(vol);
  }
  return m_curVolLUTVal;
}

void Voice::_destroy() {
  Entity::_destroy();

  for (auto& vox : m_childVoices)
    vox->_destroy();

  m_studio.reset();
  m_backendVoice.reset();
  m_curSample.reset();
  m_sequencer.reset();
}

Voice::~Voice() {
  // fprintf(stderr, "DEALLOC %d\n", m_vid);
}

Voice::Voice(Engine& engine, const AudioGroup& group, GroupId groupId, int vid, bool emitter, ObjToken<Studio> studio)
: Entity(engine, group, groupId), m_vid(vid), m_emitter(emitter), m_studio(studio) {
  // fprintf(stderr, "ALLOC %d\n", m_vid);
}

Voice::Voice(Engine& engine, const AudioGroup& group, GroupId groupId, ObjectId oid, int vid, bool emitter,
             ObjToken<Studio> studio)
: Entity(engine, group, groupId, oid), m_vid(vid), m_emitter(emitter), m_studio(studio) {
  // fprintf(stderr, "ALLOC %d\n", m_vid);
}

void Voice::_macroSampleEnd() {
  if (m_sampleEndTrap.macroId != 0xffff) {
    if (m_sampleEndTrap.macroId == std::get<0>(m_state.m_pc.back())) {
      std::get<2>(m_state.m_pc.back()) = std::get<1>(m_state.m_pc.back())->assertPC(m_sampleEndTrap.macroStep);
      m_state.m_inWait = false;
    } else
      loadMacroObject(m_sampleEndTrap.macroId, m_sampleEndTrap.macroStep, m_state.m_ticksPerSec, m_state.m_initKey,
                      m_state.m_initVel, m_state.m_initMod);
  } else
    m_state.sampleEndNotify(*this);
}

bool Voice::_checkSamplePos(bool& looped) {
  looped = false;
  if (!m_curSample)
    return true;

  if (m_curSamplePos >= m_lastSamplePos) {
    if (m_curSample->isLooped()) {
      /* Turn over looped sample */
      m_curSamplePos = m_curSample->m_loopStartSample;
      if (m_curFormat == SampleFormat::DSP) {
        m_prev1 = m_curSample->m_ADPCMParms.dsp.m_hist1;
        m_prev2 = m_curSample->m_ADPCMParms.dsp.m_hist2;
      }
      looped = true;
    } else {
      /* Notify sample end */
      _macroSampleEnd();
      m_curSample = nullptr;
      return true;
    }
  }

  /* Looped samples issue sample end when ADSR envelope complete */
  if (m_volAdsr.isComplete(*this)) {
    _macroSampleEnd();
    m_curSample = nullptr;
    return true;
  }

  return false;
}

void Voice::_doKeyOff() {
  m_voxState = VoiceState::KeyOff;
  if (m_state.m_inWait && m_state.m_keyoffWait) {
    if (m_volAdsr.isAdsrSet() || m_state.m_useAdsrControllers)
      m_volAdsr.keyOff(*this);
    if (m_pitchAdsr.isAdsrSet())
      m_pitchAdsr.keyOff();
  } else {
    m_volAdsr.keyOff(*this);
    m_pitchAdsr.keyOff();
  }
  m_state.keyoffNotify(*this);
}

void Voice::_setTotalPitch(int32_t cents, bool slew) {
  // fprintf(stderr, "PITCH %d %d  \n", cents, slew);
  int32_t interval = clamp(0, cents, 12700) - m_curSample->getPitch() * 100;
  double ratio = std::exp2(interval / 1200.0) * m_dopplerRatio;
  m_sampleRate = m_curSample->m_sampleRate * ratio;
  m_backendVoice->setPitchRatio(ratio, slew);
}

bool Voice::_isRecursivelyDead() {
  if (m_voxState != VoiceState::Dead)
    return false;
  for (auto& vox : m_childVoices)
    if (!vox->_isRecursivelyDead())
      return false;
  return true;
}

void Voice::_bringOutYourDead() {
  for (auto it = m_childVoices.begin(); it != m_childVoices.end();) {
    Voice* vox = it->get();
    vox->_bringOutYourDead();
    if (vox->_isRecursivelyDead()) {
      it = _destroyVoice(it);
      continue;
    }
    ++it;
  }
}

ObjToken<Voice> Voice::_findVoice(int vid, ObjToken<Voice> thisPtr) {
  if (m_vid == vid)
    return thisPtr;

  for (ObjToken<Voice>& vox : m_childVoices) {
    ObjToken<Voice> ret = vox->_findVoice(vid, vox);
    if (ret)
      return ret;
  }

  return {};
}

std::unique_ptr<int8_t[]>& Voice::_ensureCtrlVals() {
  if (m_ctrlValsSelf)
    return m_ctrlValsSelf;
  m_ctrlValsSelf.reset(new int8_t[128]);
  memset(m_ctrlValsSelf.get(), 0, 128);
  return m_ctrlValsSelf;
}

std::list<ObjToken<Voice>>::iterator Voice::_allocateVoice(double sampleRate, bool dynamicPitch) {
  amuse::ObjToken<Voice> tok =
      MakeObj<Voice>(m_engine, m_audioGroup, m_groupId, m_engine.m_nextVid++, m_emitter, m_studio);
  auto it = m_childVoices.emplace(m_childVoices.end(), tok);
  m_childVoices.back()->m_backendVoice =
      m_engine.getBackend().allocateVoice(*m_childVoices.back(), sampleRate, dynamicPitch);
  return it;
}

std::list<ObjToken<Voice>>::iterator Voice::_destroyVoice(std::list<ObjToken<Voice>>::iterator it) {
  if ((*it)->m_destroyed)
    return m_childVoices.begin();

  (*it)->_destroy();
  return m_childVoices.erase(it);
}

template <typename T>
static T ApplyVolume(float vol, T samp) {
  return samp * vol;
}

void Voice::_procSamplePre(int16_t& samp) {
  double dt;

  /* Block linearized will use a larger `dt` for amplitude sampling;
   * significantly reducing the processing expense */
  switch (m_engine.m_ampMode) {
  case AmplitudeMode::PerSample:
    m_voiceSamples += 1;
    dt = 1.0 / m_sampleRate;
    break;
  case AmplitudeMode::BlockLinearized: {
    uint32_t rem = m_voiceSamples % 160;
    m_voiceSamples += 1;
    dt = m_sampleRate * 160;
    if (rem != 0) {
      /* Lerp within 160-sample block */
      float t = rem / 160.f;
      float l = clamp(0.f, m_lastLevel * (1.f - t) + m_nextLevel * t, 1.f);

      /* Apply total volume to sample using decibel scale */
      samp = ApplyVolume(m_lerpedCache.getVolume(l * m_engine.m_masterVolume, m_dlsVol), samp);
      return;
    }

    dt = 160.0 / m_sampleRate;
    break;
  }
  }

  m_voiceTime += dt;

  /* Process active envelope */
  if (m_envelopeTime >= 0.0) {
    m_envelopeTime += dt;
    float start = m_envelopeStart;
    float end = m_envelopeEnd;
    float t = clamp(0.f, float(m_envelopeTime / m_envelopeDur), 1.f);
    if (m_envelopeCurve && m_envelopeCurve->data.size() >= 128)
      t = m_envelopeCurve->data[t * 127.f] / 127.f;
    m_envelopeVol = clamp(0.f, (start * (1.0f - t)) + (end * t), 1.f);

    // printf("%d %f\n", m_vid, m_envelopeVol);

    /* Done with envelope */
    if (m_envelopeTime > m_envelopeDur)
      m_envelopeTime = -1.f;
  }

  /* Dynamically evaluate per-sample SoundMacro parameters */

  /* Process user volume slew */
  if (m_engine.m_ampMode == AmplitudeMode::PerSample) {
    if (m_targetUserVol != m_curUserVol) {
      float samplesPer5Ms = m_sampleRate * 5.f / 1000.f;
      if (samplesPer5Ms > 1.f) {
        float adjRate = 1.f / samplesPer5Ms;
        if (m_targetUserVol < m_curUserVol) {
          m_curUserVol -= adjRate;
          if (m_targetUserVol > m_curUserVol)
            m_curUserVol = m_targetUserVol;
        } else {
          m_curUserVol += adjRate;
          if (m_targetUserVol < m_curUserVol)
            m_curUserVol = m_targetUserVol;
        }
      } else
        m_curUserVol = m_targetUserVol;
    }
  } else
    m_curUserVol = m_targetUserVol;

  /* Factor in ADSR envelope state */
  float adsr = m_volAdsr.advance(dt, *this);
  m_lastLevel = m_nextLevel;
  m_nextLevel = m_curUserVol * m_curVol * m_envelopeVol * adsr * (m_state.m_curVel / 127.f);

  /* Apply tremolo */
  if (m_state.m_tremoloSel && (m_tremoloScale || m_tremoloModScale)) {
    float t = m_state.m_tremoloSel.evaluate(m_voiceTime, *this, m_state) / 127.f;
    if (m_tremoloScale && m_tremoloModScale) {
      float fac = (1.0f - t) + (m_tremoloScale * t);
      float modT = m_state.m_modWheelSel ? (m_state.m_modWheelSel.evaluate(m_voiceTime, *this, m_state) / 127.f)
                                         : (getCtrlValue(1) / 127.f);
      float modFac = (1.0f - modT) + (m_tremoloModScale * modT);
      m_nextLevel *= fac * modFac;
    } else if (m_tremoloScale) {
      float fac = (1.0f - t) + (m_tremoloScale * t);
      m_nextLevel *= fac;
    } else if (m_tremoloModScale) {
      float modT = m_state.m_modWheelSel ? (m_state.m_modWheelSel.evaluate(m_voiceTime, *this, m_state) / 127.f)
                                         : (getCtrlValue(1) / 127.f);
      float modFac = (1.0f - modT) + (m_tremoloModScale * modT);
      m_nextLevel *= modFac;
    }
  }

  m_nextLevel = clamp(0.f, m_nextLevel, 1.f);

  /* Apply total volume to sample using decibel scale */
  samp = ApplyVolume(m_nextLevelCache.getVolume(m_nextLevel * m_engine.m_masterVolume, m_dlsVol), samp);
}

template <typename T>
T Voice::_procSampleMaster(double time, T samp) {
  float evalVol = m_state.m_volumeSel ? (m_state.m_volumeSel.evaluate(time, *this, m_state) / 127.f) : 1.f;
  return ApplyVolume(m_masterCache.getVolume(clamp(0.f, evalVol, 1.f), m_dlsVol), samp);
}

template <typename T>
T Voice::_procSampleAuxA(double time, T samp) {
  float evalVol = m_state.m_volumeSel ? (m_state.m_volumeSel.evaluate(time, *this, m_state) / 127.f) : 1.f;
  evalVol *= m_state.m_reverbSel ? (m_state.m_reverbSel.evaluate(time, *this, m_state) / 127.f) : m_curReverbVol;
  evalVol += m_state.m_preAuxASel ? (m_state.m_preAuxASel.evaluate(time, *this, m_state) / 127.f) : 0.f;
  return ApplyVolume(m_auxACache.getVolume(clamp(0.f, evalVol, 1.f), m_dlsVol), samp);
}

template <typename T>
T Voice::_procSampleAuxB(double time, T samp) {
  float evalVol = m_state.m_volumeSel ? (m_state.m_volumeSel.evaluate(time, *this, m_state) / 127.f) : 1.f;
  evalVol *= m_state.m_postAuxB ? (m_state.m_postAuxB.evaluate(time, *this, m_state) / 127.f) : m_curAuxBVol;
  evalVol += m_state.m_preAuxBSel ? (m_state.m_preAuxBSel.evaluate(time, *this, m_state) / 127.f) : 0.f;
  return ApplyVolume(m_auxBCache.getVolume(clamp(0.f, evalVol, 1.f), m_dlsVol), samp);
}

uint32_t Voice::_GetBlockSampleCount(SampleFormat fmt) {
  switch (fmt) {
  default:
    return 1;
  case SampleFormat::DSP:
  case SampleFormat::DSP_DRUM:
    return 14;
  case SampleFormat::N64:
    return 64;
  }
}

static float TriangleWave(float t) {
  t = std::fmod(t, 1.f);
  if (t < 0.25f)
    return t / 0.25f;
  if (t >= 0.75f)
    return (t - 0.75f) / 0.25f - 1.f;
  return (t - 0.25f) / 0.5f * -2.f + 1.f;
}

void Voice::preSupplyAudio(double dt) {
  /* Process SoundMacro; bootstrapping sample if needed */
  bool dead = m_state.advance(*this, dt);

  /* Process per-block evaluators here */
  if (m_state.m_pedalSel) {
    bool pedal = m_state.m_pedalSel.evaluate(m_voiceTime, *this, m_state) >= 64.f;
    if (pedal != m_sustained)
      setPedal(pedal);
  }

  bool panDirty = false;
  if (m_state.m_panSel) {
    float evalPan = (m_state.m_panSel.evaluate(m_voiceTime, *this, m_state) - 64.f) / 63.f;
    evalPan = clamp(-1.f, evalPan, 1.f);
    if (evalPan != m_curPan) {
      m_curPan = evalPan;
      panDirty = true;
    }
  }
  if (m_state.m_spanSel) {
    float evalSpan = (m_state.m_spanSel.evaluate(m_voiceTime, *this, m_state) - 64.f) / 63.f;
    evalSpan = clamp(-1.f, evalSpan, 1.f);
    if (evalSpan != m_curSpan) {
      m_curSpan = evalSpan;
      panDirty = true;
    }
  }
  if (panDirty)
    _setPan(m_curPan);

  if (m_state.m_pitchWheelSel) {
    float evalPWheel = (m_state.m_pitchWheelSel.evaluate(m_voiceTime, *this, m_state) - 64.f) / 63.f;
    _setPitchWheel(clamp(-1.f, evalPWheel, 1.f));
  }

  /* Process active pan-sweep */
  bool refresh = false;
  if (!m_panningQueue.empty()) {
    Panning& p = m_panningQueue.front();
    p.m_time += dt;
    double start = (p.m_pos - 64) / 64.0;
    double end = (p.m_pos + p.m_width - 64) / 64.0;
    double t = clamp(0.0, p.m_time / p.m_dur, 1.0);
    _setPan(float((start * (1.0 - t)) + (end * t)));
    refresh = true;

    /* Done with panning */
    if (p.m_time > p.m_dur)
      m_panningQueue.pop();
  }

  /* Process active span-sweep */
  if (!m_spanningQueue.empty()) {
    Panning& s = m_spanningQueue.front();
    s.m_time += dt;
    double start = (s.m_pos - 64) / 64.0;
    double end = (s.m_pos + s.m_width - 64) / 64.0;
    double t = clamp(0.0, s.m_time / s.m_dur, 1.0);
    _setSurroundPan(float((start * (1.0 - t)) + (end * t)));
    refresh = true;

    /* Done with spanning */
    if (s.m_time > s.m_dur)
      m_spanningQueue.pop();
  }

  /* Calculate total pitch */
  int32_t newPitch = m_curPitch;
  refresh |= m_pitchDirty;
  m_pitchDirty = false;
  if (m_portamentoTime >= 0.f) {
    m_portamentoTime += dt;
    float t = clamp(0.f, m_portamentoTime / m_state.m_portamentoTime, 1.f);

    newPitch = (m_curPitch * (1.0f - t)) + (m_portamentoTarget * t);
    refresh = true;

    /* Done with portamento */
    if (m_portamentoTime > m_state.m_portamentoTime) {
      m_portamentoTime = -1.f;
      m_curPitch = m_portamentoTarget;
    }
  }
  if (m_pitchEnv) {
    newPitch += m_pitchAdsr.advance(dt) * m_pitchEnvRange;
    refresh = true;
  }

  /* Process vibrato */
  if (m_vibratoTime >= 0.f) {
    m_vibratoTime += dt;
    float vibrato = TriangleWave(m_vibratoTime / m_vibratoPeriod);
    if (m_vibratoModWheel) {
      int32_t range = m_vibratoModLevel ? m_vibratoModLevel : m_vibratoLevel;
      newPitch += range * vibrato * (m_state.m_curMod / 127.f);
    } else
      newPitch += m_vibratoLevel * vibrato;
    refresh = true;
  }

  /* Process pitch sweep 1 */
  if (m_pitchSweep1It < m_pitchSweep1Times) {
    ++m_pitchSweep1It;
    m_pitchSweep1 = m_pitchSweep1Add * m_pitchSweep1It;
    refresh = true;
  } else if (m_pitchSweep1Times != 0) {
    m_pitchSweep1It = 0;
    m_pitchSweep1 = 0;
    refresh = true;
  }

  /* Process pitch sweep 2 */
  if (m_pitchSweep2It < m_pitchSweep2Times) {
    ++m_pitchSweep2It;
    m_pitchSweep2 = m_pitchSweep2Add * m_pitchSweep2It;
    refresh = true;
  } else if (m_pitchSweep2Times != 0) {
    m_pitchSweep2It = 0;
    m_pitchSweep2 = 0;
    refresh = true;
  }

  if (m_curSample && refresh) {
    _setTotalPitch(newPitch + m_pitchSweep1 + m_pitchSweep2 + m_pitchWheelVal, m_needsSlew);
    m_needsSlew = true;
  }

  if (dead && (!m_curSample || m_voxState == VoiceState::KeyOff) && m_sampleEndTrap.macroId == 0xffff &&
      m_messageTrap.macroId == 0xffff && (!m_curSample || (m_curSample && m_volAdsr.isComplete(*this)))) {
    m_voxState = VoiceState::Dead;
    m_backendVoice->stop();
  }
}

size_t Voice::supplyAudio(size_t samples, int16_t* data) {
  uint32_t samplesRem = samples;

  if (m_curSample) {
    uint32_t blockSampleCount = _GetBlockSampleCount(m_curFormat);
    uint32_t block;

    bool looped = true;
    while (looped && samplesRem) {
      block = m_curSamplePos / blockSampleCount;
      uint32_t rem = m_curSamplePos % blockSampleCount;

      if (rem) {
        uint32_t remCount = std::min(samplesRem, m_lastSamplePos - block * blockSampleCount);
        uint32_t decSamples;

        switch (m_curFormat) {
        case SampleFormat::DSP: {
          decSamples =
              DSPDecompressFrameRanged(data, m_curSampleData + 8 * block, m_curSample->m_ADPCMParms.dsp.m_coefs,
                                       &m_prev1, &m_prev2, rem, remCount);
          break;
        }
        case SampleFormat::N64: {
          decSamples = N64MusyXDecompressFrameRanged(data, m_curSampleData + 256 + 40 * block,
                                                     m_curSample->m_ADPCMParms.vadpcm.m_coefs, rem, remCount);
          break;
        }
        case SampleFormat::PCM: {
          const int16_t* pcm = reinterpret_cast<const int16_t*>(m_curSampleData);
          remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos);
          for (uint32_t i = 0; i < remCount; ++i)
            data[i] = SBig(pcm[m_curSamplePos + i]);
          decSamples = remCount;
          break;
        }
        case SampleFormat::PCM_PC: {
          const int16_t* pcm = reinterpret_cast<const int16_t*>(m_curSampleData);
          remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos);
          memmove(data, pcm + m_curSamplePos, remCount * sizeof(int16_t));
          decSamples = remCount;
          break;
        }
        default:
          memset(data, 0, sizeof(int16_t) * samples);
          return samples;
        }

        /* Per-sample processing */
        for (uint32_t i = 0; i < decSamples; ++i) {
          ++m_curSamplePos;
          _procSamplePre(data[i]);
        }

        samplesRem -= decSamples;
        data += decSamples;
      }

      if (_checkSamplePos(looped)) {
        if (samplesRem)
          memset(data, 0, sizeof(int16_t) * samplesRem);
        return samples;
      }
      if (looped)
        continue;

      while (samplesRem) {
        block = m_curSamplePos / blockSampleCount;
        uint32_t remCount = std::min(samplesRem, m_lastSamplePos - block * blockSampleCount);
        uint32_t decSamples;

        switch (m_curFormat) {
        case SampleFormat::DSP: {
          decSamples = DSPDecompressFrame(data, m_curSampleData + 8 * block, m_curSample->m_ADPCMParms.dsp.m_coefs,
                                          &m_prev1, &m_prev2, remCount);
          break;
        }
        case SampleFormat::N64: {
          decSamples = N64MusyXDecompressFrame(data, m_curSampleData + 256 + 40 * block,
                                               m_curSample->m_ADPCMParms.vadpcm.m_coefs, remCount);
          break;
        }
        case SampleFormat::PCM: {
          const int16_t* pcm = reinterpret_cast<const int16_t*>(m_curSampleData);
          remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos);
          for (uint32_t i = 0; i < remCount; ++i)
            data[i] = SBig(pcm[m_curSamplePos + i]);
          decSamples = remCount;
          break;
        }
        case SampleFormat::PCM_PC: {
          const int16_t* pcm = reinterpret_cast<const int16_t*>(m_curSampleData);
          remCount = std::min(samplesRem, m_lastSamplePos - m_curSamplePos);
          memmove(data, pcm + m_curSamplePos, remCount * sizeof(int16_t));
          decSamples = remCount;
          break;
        }
        default:
          memset(data, 0, sizeof(int16_t) * samples);
          return samples;
        }

        /* Per-sample processing */
        for (uint32_t i = 0; i < decSamples; ++i) {
          ++m_curSamplePos;
          _procSamplePre(data[i]);
        }

        samplesRem -= decSamples;
        data += decSamples;

        if (_checkSamplePos(looped)) {
          if (samplesRem)
            memset(data, 0, sizeof(int16_t) * samplesRem);
          return samples;
        }
        if (looped)
          break;
      }
    }
  } else {
    _macroSampleEnd();
    memset(data, 0, sizeof(int16_t) * samples);
  }

  if (m_voxState == VoiceState::Dead)
    m_curSample.reset();

  return samples;
}

void Voice::routeAudio(size_t frames, double dt, int busId, int16_t* in, int16_t* out) {
  dt /= double(frames);

  switch (busId) {
  case 0:
  default: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleMaster(dt * i + m_voiceTime, in[i]);
    break;
  }
  case 1: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleAuxA(dt * i + m_voiceTime, in[i]);
    break;
  }
  case 2: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleAuxB(dt * i + m_voiceTime, in[i]);
    break;
  }
  }
}

void Voice::routeAudio(size_t frames, double dt, int busId, int32_t* in, int32_t* out) {
  dt /= double(frames);

  switch (busId) {
  case 0:
  default: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleMaster(dt * i + m_voiceTime, in[i]);
    break;
  }
  case 1: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleAuxA(dt * i + m_voiceTime, in[i]);
    break;
  }
  case 2: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleAuxB(dt * i + m_voiceTime, in[i]);
    break;
  }
  }
}

void Voice::routeAudio(size_t frames, double dt, int busId, float* in, float* out) {
  dt /= double(frames);

  switch (busId) {
  case 0:
  default: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleMaster(dt * i + m_voiceTime, in[i]);
    break;
  }
  case 1: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleAuxA(dt * i + m_voiceTime, in[i]);
    break;
  }
  case 2: {
    for (uint32_t i = 0; i < frames; ++i)
      out[i] = _procSampleAuxB(dt * i + m_voiceTime, in[i]);
    break;
  }
  }
}

int Voice::maxVid() const {
  int maxVid = m_vid;
  for (const ObjToken<Voice>& vox : m_childVoices)
    maxVid = std::max(maxVid, vox->maxVid());
  return maxVid;
}

ObjToken<Voice> Voice::_startChildMacro(ObjectId macroId, int macroStep, double ticksPerSec, uint8_t midiKey,
                                        uint8_t midiVel, uint8_t midiMod, bool pushPc) {
  std::list<ObjToken<Voice>>::iterator vox = _allocateVoice(NativeSampleRate, true);
  if (!(*vox)->loadMacroObject(macroId, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc)) {
    _destroyVoice(vox);
    return {};
  }
  (*vox)->setVolume(m_targetUserVol);
  (*vox)->setPan(m_curPan);
  (*vox)->setSurroundPan(m_curSpan);
  if (m_extCtrlVals)
    (*vox)->installCtrlValues(m_extCtrlVals);
  return *vox;
}

ObjToken<Voice> Voice::startChildMacro(int8_t addNote, ObjectId macroId, int macroStep) {
  return _startChildMacro(macroId, macroStep, 1000.0, m_state.m_initKey + addNote, m_state.m_initVel,
                          m_state.m_initMod);
}

bool Voice::_loadSoundMacro(SoundMacroId id, const SoundMacro* macroData, int macroStep, double ticksPerSec,
                            uint8_t midiKey, uint8_t midiVel, uint8_t midiMod, bool pushPc) {
  m_objectId = id;

  if (m_state.m_pc.empty())
    m_state.initialize(id, macroData, macroStep, ticksPerSec, midiKey, midiVel, midiMod);
  else {
    if (!pushPc)
      m_state.m_pc.clear();
    m_state.m_pc.emplace_back(id, macroData, macroData->assertPC(macroStep));
  }

  m_voxState = VoiceState::Playing;
  m_backendVoice->start();
  return true;
}

bool Voice::_loadKeymap(const Keymap* keymap, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod,
                        bool pushPc) {
  const Keymap& km = keymap[midiKey];
  midiKey += km.transpose;
  bool ret = loadMacroObject(km.macro.id, 0, ticksPerSec, midiKey, midiVel, midiMod, pushPc);
  m_curVol = 1.f;
  if (km.pan == -128) {
    _setSurroundPan(1.f);
  } else {
    _setPan((km.pan - 64) / 64.f);
    _setSurroundPan(-1.f);
  }
  return ret;
}

bool Voice::_loadLayer(const std::vector<LayerMapping>& layer, double ticksPerSec, uint8_t midiKey, uint8_t midiVel,
                       uint8_t midiMod, bool pushPc) {
  bool ret = false;
  for (const LayerMapping& mapping : layer) {
    if (midiKey >= mapping.keyLo && midiKey <= mapping.keyHi) {
      uint8_t mappingKey = midiKey + mapping.transpose;
      if (m_voxState != VoiceState::Playing) {
        ret |= loadMacroObject(mapping.macro.id, 0, ticksPerSec, mappingKey, midiVel, midiMod, pushPc);
        m_curUserVol = m_targetUserVol = mapping.volume / 127.f;
        _setPan((mapping.pan - 64) / 64.f);
        _setSurroundPan((mapping.span - 64) / 64.f);
      } else {
        ObjToken<Voice> vox = _startChildMacro(mapping.macro.id, 0, ticksPerSec, mappingKey, midiVel, midiMod, pushPc);
        if (vox) {
          vox->m_curUserVol = vox->m_targetUserVol = mapping.volume / 127.f;
          vox->_setPan((mapping.pan - 64) / 64.f);
          vox->_setSurroundPan((mapping.span - 64) / 64.f);
          ret = true;
        }
      }
    }
  }
  return ret;
}

bool Voice::loadMacroObject(SoundMacroId macroId, int macroStep, double ticksPerSec, uint8_t midiKey, uint8_t midiVel,
                            uint8_t midiMod, bool pushPc) {
  if (m_destroyed)
    return false;

  const SoundMacro* macroData = m_audioGroup.getPool().soundMacro(macroId);
  if (macroData)
    return _loadSoundMacro(macroId, macroData, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc);

  return false;
}

bool Voice::loadMacroObject(const SoundMacro* macro, int macroStep, double ticksPerSec, uint8_t midiKey,
                            uint8_t midiVel, uint8_t midiMod, bool pushPc) {
  if (m_destroyed)
    return false;

  if (macro)
    return _loadSoundMacro({}, macro, macroStep, ticksPerSec, midiKey, midiVel, midiMod, pushPc);

  return false;
}

bool Voice::loadPageObject(ObjectId objectId, double ticksPerSec, uint8_t midiKey, uint8_t midiVel, uint8_t midiMod) {
  if (m_destroyed)
    return false;

  if (objectId.id & 0x8000) {
    const std::vector<LayerMapping>* layer = m_audioGroup.getPool().layer(objectId);
    if (layer)
      return _loadLayer(*layer, ticksPerSec, midiKey, midiVel, midiMod);
  } else if (objectId.id & 0x4000) {
    const Keymap* keymap = m_audioGroup.getPool().keymap(objectId);
    if (keymap)
      return _loadKeymap(keymap, ticksPerSec, midiKey, midiVel, midiMod);
  } else {
    const SoundMacro* sm = m_audioGroup.getPool().soundMacro(objectId);
    if (sm)
      return _loadSoundMacro(objectId, sm, 0, ticksPerSec, midiKey, midiVel, midiMod);
  }

  return false;
}

void Voice::_macroKeyOff() {
  if (m_voxState == VoiceState::Playing) {
    if (m_sustained)
      m_sustainKeyOff = true;
    else
      _doKeyOff();
  }
}

void Voice::keyOff() {
  if (m_destroyed)
    return;

  if (m_keyoffTrap.macroId != 0xffff) {
    if (m_keyoffTrap.macroId == std::get<0>(m_state.m_pc.back())) {
      std::get<2>(m_state.m_pc.back()) = std::get<1>(m_state.m_pc.back())->assertPC(m_keyoffTrap.macroStep);
      m_state.m_inWait = false;
    } else
      loadMacroObject(m_keyoffTrap.macroId, m_keyoffTrap.macroStep, m_state.m_ticksPerSec, m_state.m_initKey,
                      m_state.m_initVel, m_state.m_initMod);
  } else
    _macroKeyOff();

  for (const ObjToken<Voice>& vox : m_childVoices)
    vox->keyOff();
}

void Voice::message(int32_t val) {
  if (m_destroyed)
    return;

  m_latestMessage = val;

  if (m_messageTrap.macroId != 0xffff) {
    if (m_messageTrap.macroId == std::get<0>(m_state.m_pc.back()))
      std::get<2>(m_state.m_pc.back()) = std::get<1>(m_state.m_pc.back())->assertPC(m_messageTrap.macroStep);
    else
      loadMacroObject(m_messageTrap.macroId, m_messageTrap.macroStep, m_state.m_ticksPerSec, m_state.m_initKey,
                      m_state.m_initVel, m_state.m_initMod);
  }
}

void Voice::startSample(SampleId sampId, int32_t offset) {
  if (m_destroyed)
    return;

  if (const SampleEntry* sample = m_audioGroup.getSample(sampId)) {
    std::tie(m_curSample, m_curSampleData) = m_audioGroup.getSampleData(sampId, sample);

    m_state.m_sampleEnd = false;
    m_sampleRate = m_curSample->m_sampleRate;
    m_curPitch = m_curSample->getPitch();
    m_pitchDirty = true;
    _setPitchWheel(m_curPitchWheel);
    m_backendVoice->resetSampleRate(m_curSample->m_sampleRate);
    m_needsSlew = false;

    int32_t numSamples = m_curSample->getNumSamples();
    if (offset) {
      if (m_curSample->isLooped()) {
        if (offset > int32_t(m_curSample->m_loopStartSample))
          offset = ((offset - m_curSample->m_loopStartSample) % m_curSample->m_loopLengthSamples) +
                   m_curSample->m_loopStartSample;
      } else
        offset = clamp(0, offset, numSamples);
    }
    m_curSamplePos = offset;
    m_prev1 = 0;
    m_prev2 = 0;

    m_curFormat = m_curSample->getSampleFormat();
    if (m_curFormat == SampleFormat::DSP_DRUM)
      m_curFormat = SampleFormat::DSP;

    m_lastSamplePos =
        m_curSample->isLooped() ? (m_curSample->m_loopStartSample + m_curSample->m_loopLengthSamples) : numSamples;
    if (m_lastSamplePos)
      --m_lastSamplePos;

    bool looped;
    _checkSamplePos(looped);

    /* Seek DSPADPCM state if needed */
    if (m_curSample && m_curSamplePos && m_curFormat == SampleFormat::DSP) {
      uint32_t block = m_curSamplePos / 14;
      uint32_t rem = m_curSamplePos % 14;
      for (uint32_t b = 0; b < block; ++b)
        DSPDecompressFrameStateOnly(m_curSampleData + 8 * b, m_curSample->m_ADPCMParms.dsp.m_coefs, &m_prev1, &m_prev2,
                                    14);

      if (rem)
        DSPDecompressFrameStateOnly(m_curSampleData + 8 * block, m_curSample->m_ADPCMParms.dsp.m_coefs, &m_prev1,
                                    &m_prev2, rem);
    }
  }
}

void Voice::stopSample() { m_curSample.reset(); }

void Voice::setVolume(float vol) {
  if (m_destroyed)
    return;

  m_targetUserVol = clamp(0.f, vol, 1.f);
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setVolume(vol);
}

void Voice::_panLaw(float coefs[8], float frontPan, float backPan, float totalSpan) const {
  /* -3dB panning law for various channel configs */
  switch (m_engine.m_channelSet) {
  case AudioChannelSet::Stereo:
  default:
    /* Left */
    coefs[0] = std::sqrt(-frontPan * 0.5f + 0.5f);

    /* Right */
    coefs[1] = std::sqrt(frontPan * 0.5f + 0.5f);

    break;

  case AudioChannelSet::Quad:
    /* Left */
    coefs[0] = -frontPan * 0.5f + 0.5f;
    coefs[0] *= -totalSpan * 0.5f + 0.5f;
    coefs[0] = std::sqrt(coefs[0]);

    /* Right */
    coefs[1] = frontPan * 0.5f + 0.5f;
    coefs[1] *= -totalSpan * 0.5f + 0.5f;
    coefs[1] = std::sqrt(coefs[1]);

    /* Back Left */
    coefs[2] = -backPan * 0.5f + 0.5f;
    coefs[2] *= totalSpan * 0.5f + 0.5f;
    coefs[2] = std::sqrt(coefs[2]);

    /* Back Right */
    coefs[3] = backPan * 0.5f + 0.5f;
    coefs[3] *= totalSpan * 0.5f + 0.5f;
    coefs[3] = std::sqrt(coefs[3]);

    break;

  case AudioChannelSet::Surround51:
    /* Left */
    coefs[0] = (frontPan <= 0.f) ? -frontPan : 0.f;
    coefs[0] *= -totalSpan * 0.5f + 0.5f;
    coefs[0] = std::sqrt(coefs[0]);

    /* Right */
    coefs[1] = (frontPan >= 0.f) ? frontPan : 0.f;
    coefs[1] *= -totalSpan * 0.5f + 0.5f;
    coefs[1] = std::sqrt(coefs[1]);

    /* Back Left */
    coefs[2] = -backPan * 0.5f + 0.5f;
    coefs[2] *= totalSpan * 0.5f + 0.5f;
    coefs[2] = std::sqrt(coefs[2]);

    /* Back Right */
    coefs[3] = backPan * 0.5f + 0.5f;
    coefs[3] *= totalSpan * 0.5f + 0.5f;
    coefs[3] = std::sqrt(coefs[3]);

    /* Center */
    coefs[4] = 1.f - std::fabs(frontPan);
    coefs[4] *= -totalSpan * 0.5f + 0.5f;
    coefs[4] = std::sqrt(coefs[4]);

    /* LFE */
    coefs[5] = 0.25f;

    break;

  case AudioChannelSet::Surround71:
    /* Left */
    coefs[0] = (frontPan <= 0.f) ? -frontPan : 0.f;
    coefs[0] *= (totalSpan <= 0.f) ? -totalSpan : 0.f;
    coefs[0] = std::sqrt(coefs[0]);

    /* Right */
    coefs[1] = (frontPan >= 0.f) ? frontPan : 0.f;
    coefs[1] *= (totalSpan <= 0.f) ? -totalSpan : 0.f;
    coefs[1] = std::sqrt(coefs[1]);

    /* Back Left */
    coefs[2] = -backPan * 0.5f + 0.5f;
    coefs[2] *= (totalSpan >= 0.f) ? totalSpan : 0.f;
    coefs[2] = std::sqrt(coefs[2]);

    /* Back Right */
    coefs[3] = backPan * 0.5f + 0.5f;
    coefs[3] *= (totalSpan >= 0.f) ? totalSpan : 0.f;
    coefs[3] = std::sqrt(coefs[3]);

    /* Center */
    coefs[4] = 1.f - std::fabs(frontPan);
    coefs[4] *= (totalSpan <= 0.f) ? -totalSpan : 0.f;
    coefs[4] = std::sqrt(coefs[4]);

    /* LFE */
    coefs[5] = 0.25f;

    /* Side Left */
    coefs[6] = -backPan * 0.5f + 0.5f;
    coefs[6] *= 1.f - std::fabs(totalSpan);
    coefs[6] = std::sqrt(coefs[6]);

    /* Side Right */
    coefs[7] = backPan * 0.5f + 0.5f;
    coefs[7] *= 1.f - std::fabs(totalSpan);
    coefs[7] = std::sqrt(coefs[7]);

    break;
  }
}

void Voice::_setPan(float pan) {
  if (m_destroyed || m_emitter)
    return;

  m_curPan = clamp(-1.f, pan, 1.f);
  float totalPan = clamp(-1.f, m_curPan, 1.f);
  float totalSpan = clamp(-1.f, m_curSpan, 1.f);
  float coefs[8] = {};
  _panLaw(coefs, totalPan, totalPan, totalSpan);
  _setChannelCoefs(coefs);
}

void Voice::setPan(float pan) {
  if (m_destroyed)
    return;

  _setPan(pan);
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setPan(pan);
}

void Voice::_setSurroundPan(float span) {
  m_curSpan = clamp(-1.f, span, 1.f);
  _setPan(m_curPan);
}

void Voice::setSurroundPan(float span) {
  if (m_destroyed)
    return;

  _setSurroundPan(span);
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setSurroundPan(span);
}

void Voice::_setChannelCoefs(const float coefs[8]) {
  m_backendVoice->setChannelLevels(m_studio->getMaster().m_backendSubmix.get(), coefs, true);
  m_backendVoice->setChannelLevels(m_studio->getAuxA().m_backendSubmix.get(), coefs, true);
  m_backendVoice->setChannelLevels(m_studio->getAuxB().m_backendSubmix.get(), coefs, true);
}

void Voice::setChannelCoefs(const float coefs[8]) {
  if (m_destroyed)
    return;

  _setChannelCoefs(coefs);
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setChannelCoefs(coefs);
}

void Voice::startEnvelope(double dur, float vol, const Curve* envCurve) {
  m_envelopeTime = 0.f;
  m_envelopeDur = dur;
  m_envelopeStart = clamp(0.f, m_envelopeVol, 1.f);
  m_envelopeEnd = clamp(0.f, vol, 1.f);
  m_envelopeCurve = envCurve;
}

void Voice::startFadeIn(double dur, float vol, const Curve* envCurve) {
  m_envelopeTime = 0.f;
  m_envelopeDur = dur;
  m_envelopeStart = 0.f;
  m_envelopeEnd = clamp(0.f, vol, 1.f);
  m_envelopeCurve = envCurve;
}

void Voice::startPanning(double dur, uint8_t panPos, int8_t panWidth) {
  m_panningQueue.push({0.f, dur, panPos, panWidth});
}

void Voice::startSpanning(double dur, uint8_t spanPos, int8_t spanWidth) {
  m_spanningQueue.push({0.f, dur, spanPos, spanWidth});
}

void Voice::setPitchKey(int32_t cents) {
  m_curPitch = cents;
  m_pitchDirty = true;
}

void Voice::setPedal(bool pedal) {
  if (m_destroyed)
    return;

  if (m_sustained && !pedal && m_sustainKeyOff) {
    m_sustainKeyOff = false;
    _doKeyOff();
  }
  m_sustained = pedal;

  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setPedal(pedal);
}

void Voice::setDoppler(float) {}

void Voice::setVibrato(int32_t level, bool modScale, float period) {
  m_vibratoTime = std::fabs(period) < FLT_EPSILON ? -1.f : 0.f;
  m_vibratoLevel = level;
  m_vibratoModWheel = modScale;
  m_vibratoPeriod = period;
}

void Voice::setMod2VibratoRange(int32_t modLevel) { m_vibratoModLevel = modLevel; }

void Voice::setTremolo(float tremoloScale, float tremoloModScale) {
  m_tremoloScale = tremoloScale;
  m_tremoloModScale = tremoloModScale;
}

void Voice::setPitchSweep1(uint8_t times, int16_t add) {
  m_pitchSweep1 = 0;
  m_pitchSweep1It = 0;
  m_pitchSweep1Times = times;
  m_pitchSweep1Add = add;
}

void Voice::setPitchSweep2(uint8_t times, int16_t add) {
  m_pitchSweep2 = 0;
  m_pitchSweep2It = 0;
  m_pitchSweep2Times = times;
  m_pitchSweep2Add = add;
}

void Voice::setReverbVol(float rvol) {
  if (m_destroyed)
    return;

  m_curReverbVol = clamp(0.f, rvol, 1.f);
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setReverbVol(rvol);
}

void Voice::setAuxBVol(float bvol) {
  if (m_destroyed)
    return;

  m_curAuxBVol = clamp(0.f, bvol, 1.f);
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setAuxBVol(bvol);
}

void Voice::setAdsr(ObjectId adsrId, bool dls) {
  if (m_destroyed)
    return;

  if (dls) {
    const ADSRDLS* adsr = m_audioGroup.getPool().tableAsAdsrDLS(adsrId);
    if (adsr) {
      m_volAdsr.reset(adsr, m_state.m_initKey, m_state.m_initVel);
      if (m_voxState == VoiceState::KeyOff)
        m_volAdsr.keyOff(*this);
    }
  } else {
    const ADSR* adsr = m_audioGroup.getPool().tableAsAdsr(adsrId);
    if (adsr) {
      m_volAdsr.reset(adsr);
      if (m_voxState == VoiceState::KeyOff)
        m_volAdsr.keyOff(*this);
    }
  }
}

void Voice::setPitchFrequency(uint32_t hz, uint16_t fine) {
  if (m_destroyed)
    return;

  m_sampleRate = hz + fine / 65536.0;
  m_backendVoice->setPitchRatio(1.0, false);
  m_backendVoice->resetSampleRate(m_sampleRate);
}

void Voice::setPitchAdsr(ObjectId adsrId, int32_t cents) {
  if (m_destroyed)
    return;

  const ADSRDLS* adsr = m_audioGroup.getPool().tableAsAdsrDLS(adsrId);
  if (adsr) {
    m_pitchAdsr.reset(adsr, m_state.m_initKey, m_state.m_initVel);
    m_pitchEnvRange = cents;
    m_pitchEnv = true;
  }
}

void Voice::_setPitchWheel(float pitchWheel) {
  if (m_destroyed)
    return;

  if (pitchWheel > 0.f)
    m_pitchWheelVal = m_pitchWheelUp * m_curPitchWheel;
  else if (pitchWheel < 0.f)
    m_pitchWheelVal = m_pitchWheelDown * m_curPitchWheel;
  else
    m_pitchWheelVal = 0;
  m_pitchDirty = true;
}

void Voice::setPitchWheel(float pitchWheel) {
  if (m_destroyed)
    return;

  m_curPitchWheel = amuse::clamp(-1.f, pitchWheel, 1.f);
  _setPitchWheel(m_curPitchWheel);

  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setPitchWheel(pitchWheel);
}

void Voice::setPitchWheelRange(int8_t up, int8_t down) {
  if (m_destroyed)
    return;

  m_pitchWheelUp = up * 100;
  m_pitchWheelDown = down * 100;
  _setPitchWheel(m_curPitchWheel);
}

void Voice::setAftertouch(uint8_t aftertouch) {
  if (m_destroyed)
    return;

  m_curAftertouch = aftertouch;
  for (ObjToken<Voice>& vox : m_childVoices)
    vox->setAftertouch(aftertouch);
}

bool Voice::doPortamento(uint8_t newNote) {
  bool pState;
  switch (m_state.m_portamentoMode) {
  case SoundMacro::CmdPortamento::PortState::Disable:
  default:
    pState = false;
    break;
  case SoundMacro::CmdPortamento::PortState::Enable:
    pState = true;
    break;
  case SoundMacro::CmdPortamento::PortState::MIDIControlled:
    pState = m_state.m_portamentoSel ? (m_state.m_portamentoSel.evaluate(m_voiceTime, *this, m_state) >= 64.f)
                                     : (getCtrlValue(65) >= 64);
    break;
  }

  if (!pState)
    return false;

  m_portamentoTime = 0.f;
  m_portamentoTarget = newNote * 100;
  m_state.m_initKey = newNote;
  return true;
}

void Voice::_notifyCtrlChange(uint8_t ctrl, int8_t val) {
  if (ctrl == 0x40) {
    setPedal(val >= 0x40);
  } else if (ctrl == 0x5b) {
    setReverbVol(val / 127.f);
  } else if (ctrl == 0x5d) {
    setAuxBVol(val / 127.f);
  } else if (ctrl == 0x1) {
    m_state.m_curMod = uint8_t(val);
  } else if (ctrl == 0x64) {
    // RPN LSB
    m_rpn &= ~0x7f;
    m_rpn |= val;
  } else if (ctrl == 0x65) {
    // RPN MSB
    m_rpn &= ~0x3f80;
    m_rpn |= val << 7;
  } else if (ctrl == 0x6) {
    if (m_rpn == 0)
      m_pitchWheelUp = m_pitchWheelDown = val * 100;
  } else if (ctrl == 0x60) {
    if (m_rpn == 0) {
      m_pitchWheelUp += 100;
      m_pitchWheelDown += 100;
    }
  } else if (ctrl == 0x61) {
    if (m_rpn == 0) {
      m_pitchWheelUp -= 100;
      m_pitchWheelDown -= 100;
    }
  }

  for (ObjToken<Voice>& vox : m_childVoices)
    vox->_notifyCtrlChange(ctrl, val);
}

size_t Voice::getTotalVoices() const {
  size_t ret = 1;
  for (const ObjToken<Voice>& vox : m_childVoices)
    ret += vox->getTotalVoices();
  return ret;
}

void Voice::kill() {
  if (m_destroyed)
    return;

  m_voxState = VoiceState::Dead;
  m_backendVoice->stop();
  for (const ObjToken<Voice>& vox : m_childVoices)
    vox->kill();
}
} // namespace amuse