Several CPU usage optimizations

CMakeLists.txt

@@ -12,6 +12,7 @@ set(SOURCES
     lib/SoundMacroState.cpp
     lib/SongState.cpp
     lib/Voice.cpp
+    lib/VolumeLUT.cpp
     lib/Submix.cpp
     lib/EffectBase.cpp
     lib/EffectReverb.cpp

driver/main.cpp

@@ -778,7 +778,7 @@ struct AppCallback : boo::IApplicationCallback
         /* Build voice engine */
         std::unique_ptr<boo::IAudioVoiceEngine> voxEngine = boo::NewAudioVoiceEngine();
         amuse::BooBackendVoiceAllocator booBackend(*voxEngine);
-        m_engine.emplace(booBackend);
+        m_engine.emplace(booBackend, amuse::AmplitudeMode::BlockLinearized);
 
         /* Load group into engine */
         const amuse::AudioGroup* group = m_engine->addAudioGroup(data);

include/amuse/Engine.hpp

@@ -20,6 +20,12 @@ class AudioGroup;
 class AudioGroupData;
 class IMIDIReader;
 
+enum class AmplitudeMode
+{
+    PerSample, /**< Per-sample amplitude evaluation (dt = 1.0 / sampleRate, rather CPU demanding) */
+    BlockLinearized /**< Per-block lerp amplitude evaluation (dt = 160.0 / sampleRate) */
+};
+
 /** Main audio playback system for a single audio output */
 class Engine
 {
@@ -29,6 +35,7 @@ class Engine
     friend struct Sequencer::ChannelState;
 
     IBackendVoiceAllocator& m_backend;
+    AmplitudeMode m_ampMode;
     std::unique_ptr<IMIDIReader> m_midiReader;
     std::unordered_map<const AudioGroupData*, std::unique_ptr<AudioGroup>> m_audioGroups;
     std::list<std::shared_ptr<Voice>> m_activeVoices;
@@ -55,7 +62,7 @@ class Engine
     void _5MsCallback(double dt);
 public:
     ~Engine();
-    Engine(IBackendVoiceAllocator& backend);
+    Engine(IBackendVoiceAllocator& backend, AmplitudeMode ampMode=AmplitudeMode::PerSample);
 
     /** Access voice backend of engine */
     IBackendVoiceAllocator& getBackend() {return m_backend;}

include/amuse/Envelope.hpp

@@ -30,7 +30,7 @@ public:
     void reset(const ADSR* adsr);
     void reset(const ADSRDLS* adsr, int8_t note, int8_t vel);
     void keyOff();
-    float nextSample(double sampleRate);
+    float advance(double dt);
     bool isComplete() const {return m_phase == State::Complete;}
 };
 

include/amuse/Voice.hpp

@@ -61,6 +61,9 @@ class Voice : public Entity
     int16_t m_prev2 = 0; /**< DSPADPCM prev-prev sample */
     double m_sampleRate = 32000.0; /**< Current sample rate computed from relative sample key or SETPITCH */
     double m_voiceTime = 0.0; /**< Current seconds of voice playback (per-sample resolution) */
+    uint32_t m_voiceSamples = 0; /**< Count of samples processed over voice's lifetime */
+    float m_lastLevel = 0.f; /**< Last computed level ([0,1] mapped to [-10,0] clamped decibels) */
+    float m_nextLevel = 0.f; /**< Next computed level used for lerp-mode amplitude */
 
     VoiceState m_voxState = VoiceState::Dead; /**< Current high-level state of voice */
     bool m_sustained = false; /**< Sustain pedal pressed for this voice */

lib/Engine.cpp

@@ -23,8 +23,8 @@ Engine::~Engine()
         vox->_destroy();
 }
 
-Engine::Engine(IBackendVoiceAllocator& backend)
+Engine::Engine(IBackendVoiceAllocator& backend, AmplitudeMode ampMode)
-: m_backend(backend)
+: m_backend(backend), m_ampMode(ampMode)
 {
     backend.register5MsCallback(std::bind(&Engine::_5MsCallback, this, std::placeholders::_1));
     m_midiReader = backend.allocateMIDIReader(*this);

lib/Envelope.cpp

@@ -31,9 +31,9 @@ void Envelope::keyOff()
     m_curTime = 0.0;
 }
 
-float Envelope::nextSample(double sampleRate)
+float Envelope::advance(double dt)
 {
-    m_curTime += 1.0 / sampleRate;
+    m_curTime += dt;
 
     switch (m_phase)
     {

lib/Sequencer.cpp

@@ -211,10 +211,16 @@ void Sequencer::ChannelState::setCtrlValue(uint8_t ctrl, int8_t val)
     for (const auto& vox : m_keyoffVoxs)
         vox->_notifyCtrlChange(ctrl, val);
 
-    if (ctrl == 7)
+    switch (ctrl)
+    {
+    case 7:
         setVolume(val / 127.f);
-    else if (ctrl == 8 || ctrl == 10)
+        break;
+    case 10:
         setPan(val / 64.f - 1.f);
+        break;
+    default: break;
+    }
 }
 
 bool Sequencer::ChannelState::programChange(int8_t prog)

lib/Voice.cpp

@@ -11,6 +11,7 @@
 
 namespace amuse
 {
+extern "C" const float VolumeLUT[];
 
 void Voice::_destroy()
 {
@@ -201,14 +202,41 @@ std::list<std::shared_ptr<Voice>>::iterator Voice::_destroyVoice(Voice* voice)
 static void ApplyVolume(float vol, int16_t& samp)
 {
     /* -10dB to 0dB mapped to full volume range */
-    float factor = std::sqrt(std::pow(10.f, (vol - 1.f))) - 0.317f;
+    samp *= VolumeLUT[int(vol * 65536)];
-    if (factor < 0.f) factor = 0.f;
-    samp *= factor;
 }
 
 bool Voice::_advanceSample(int16_t& samp, int32_t& newPitch)
 {
-    double dt = 1.0 / m_sampleRate;
+    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;
+        if (rem != 0)
+        {
+            /* Lerp within 160-sample block */
+            float t = rem / 160.f;
+            float l = m_lastLevel * (1.f - t) + m_nextLevel * t;
+
+            /* Apply total volume to sample using decibel scale */
+            ApplyVolume(l, samp);
+            return false;
+        }
+
+        dt = 160.0 / m_sampleRate;
+        break;
+    }
+    }
+
     m_voiceTime += dt;
     bool refresh = false;
 
@@ -229,8 +257,9 @@ bool Voice::_advanceSample(int16_t& samp, int32_t& newPitch)
     }
 
     /* Factor in ADSR envelope state */
-    float adsr = m_volAdsr.nextSample(m_sampleRate);
+    float adsr = m_volAdsr.advance(dt);
-    float totalVol = m_userVol * m_curVol * adsr * (m_state.m_curVel / 127.f);
+    m_lastLevel = m_nextLevel;
+    m_nextLevel = m_userVol * m_curVol * adsr * (m_state.m_curVel / 127.f);
 
     /* Apply tremolo */
     if (m_state.m_tremoloSel && (m_tremoloScale || m_tremoloModScale))
@@ -241,23 +270,25 @@ bool Voice::_advanceSample(int16_t& samp, int32_t& newPitch)
             float fac = (1.0f - t) + (m_tremoloScale * t);
             float modT = getModWheel() / 127.f;
             float modFac = (1.0f - modT) + (m_tremoloModScale * modT);
-            totalVol *= fac * modFac;
+            m_nextLevel *= fac * modFac;
         }
         else if (m_tremoloScale)
         {
             float fac = (1.0f - t) + (m_tremoloScale * t);
-            totalVol *= fac;
+            m_nextLevel *= fac;
         }
         else if (m_tremoloModScale)
         {
             float modT = getModWheel() / 127.f;
             float modFac = (1.0f - modT) + (m_tremoloModScale * modT);
-            totalVol *= modFac;
+            m_nextLevel *= modFac;
         }
     }
 
-    /* Apple total volume to sample using decibel scale */
+    m_nextLevel = ClampFull<float>(m_nextLevel);
-    ApplyVolume(ClampFull<float>(totalVol), samp);
+
+    /* Apply total volume to sample using decibel scale */
+    ApplyVolume(m_nextLevel, samp);
 
     /* Process active pan-sweep */
     if (m_panningTime >= 0.f)
@@ -295,7 +326,7 @@ bool Voice::_advanceSample(int16_t& samp, int32_t& newPitch)
     m_pitchDirty = false;
     if (m_pitchEnv)
     {
-        newPitch = m_curPitch * m_pitchAdsr.nextSample(m_sampleRate);
+        newPitch = m_curPitch * m_pitchAdsr.advance(dt);
         refresh = true;
     }
 
@@ -452,7 +483,9 @@ size_t Voice::supplyAudio(size_t samples, int16_t* data)
         m_messageTrap.macroId == 0xffff &&
         (!m_curSample || (m_curSample && m_volAdsr.isComplete())))
     {
+        m_curSample = nullptr;
         m_voxState = VoiceState::Dead;
+        m_backendVoice->stop();
     }
     return samples;
 }