Track unison spread and pan continuously while a voice is playing (#370)

Previously the per-voice unison ratio multiplier and pan shift were
frozen at note-on, so moving the Unison Spread or Unison Pan knobs (or
automating them) only affected newly-triggered notes. Both quantities
are now derived per engine block from the smoothed mono parameter
values, so held notes follow knob moves and host automation.

Assisted-by: Claude Opus 4.7 <noreply@anthropic.com>

Author: baconpaul

src/synth/mono_values.h

@@ -81,6 +81,12 @@ struct MonoValues
 
     SRProvider sr;
 
+    // Per-engine-block hoists of mono unison params, refreshed by Synth::processInternal
+    // so each Voice::renderBlock can derive uniRatioMul / uniPanShift without per-voice
+    // table lookups. unisonSpreadFactorMinus1 = 2^unisonSpread.value - 1.
+    float unisonSpreadFactorMinus1{0.f};
+    float unisonPanScalar{0.f};
+
     float audioInBlock alignas(16)[blockSize]{}; // engine-rate audio in, mono mix
 };
 }; // namespace baconpaul::six_sines

src/synth/synth.cpp

@@ -289,6 +289,12 @@ template <bool multiOut> void Synth::processInternal(const clap_output_events_t
         loops++;
         lagHandler.process();
 
+        // Hoist mono unison params so per-voice renderBlock derives uniRatioMul / uniPanShift
+        // from the smoothed scalars without each voice repeating the twoToTheX lookup.
+        monoValues.unisonSpreadFactorMinus1 =
+            monoValues.twoToTheX.twoToThe(patch.output.unisonSpread.value) - 1.f;
+        monoValues.unisonPanScalar = patch.output.unisonPan.value;
+
         auto op1IsAudioIn =
             ((int)std::round(patch.sourceNodes[0].waveForm.value) == SinTable::AUDIO_IN);
         if (audioInResampler && op1IsAudioIn)

src/synth/synth.h

@@ -158,33 +158,18 @@ struct Synth
             int made{0};
 
             int lastStart{0};
-            auto usp = synth.patch.output.unisonSpread.value;
-            usp = synth.monoValues.twoToTheX.twoToThe(usp);
-            float uniVal[5]{1.0};
-            float uniPan[5]{0.0};
-            float uniScale[5]{0.0};
             assert(ct <= 5);
-            if (ct > 1)
-            {
-                auto dUni = 2 * (usp - 1) / (ct - 1);
-                for (int i = 0; i < ct; ++i)
-                {
-                    auto val = (1.0 - (usp - 1)) + dUni * i;
-                    if (val < 1)
-                    {
-                        val = 1.0 / ((1.0 + (usp - 1)) - dUni * i);
-                    }
-                    uniVal[i] = val;
-                    uniScale[i] = 2 * (i - 1.0 * (ct - 1) / 2) / (ct - 1);
-                    uniPan[i] =
-                        std::clamp(synth.patch.output.unisonPan.value * uniScale[i], -1.f, 1.f);
-                }
-            }
+            const bool hasCenter = (ct > 1 && (ct % 2 == 1));
 
             auto upr = synth.patch.output.uniPhaseRand.value > 0.5;
             auto prt = synth.patch.output.rephaseOnRetrigger > 0.5;
             for (int vc = 0; vc < ct; ++vc)
             {
+                // Bipolar position −1..1 across the unison field; 0 when ct==1.
+                // uniRatioMul and uniPanShift are derived from this in Voice::renderBlock
+                // each block, so they track host smoothing on unisonSpread / unisonPan.
+                const float uniScale = (ct > 1) ? (2.f * (vc - 0.5f * (ct - 1)) / (ct - 1)) : 0.f;
+
                 if (ibuf[vc].instruction !=
                     sst::voicemanager::VoiceInitInstructionsEntry<
                         baconpaul::six_sines::Synth::VMConfig>::Instruction::SKIP)
@@ -202,12 +187,12 @@ struct Synth
                             synth.voices[i].voiceValues.releaseVelocity = 0;
                             synth.voices[i].voiceValues.uniCount = ct;
                             synth.voices[i].voiceValues.uniIndex = vc;
-                            synth.voices[i].voiceValues.hasCenterVoice = (ct > 1 && (ct % 2 == 1));
+                            synth.voices[i].voiceValues.hasCenterVoice = hasCenter;
                             synth.voices[i].voiceValues.isCenterVoice =
-                                (ct > 1 && (ct % 2 == 1)) && (std::fabs(uniScale[vc]) < 1e-4);
-                            synth.voices[i].voiceValues.uniRatioMul = uniVal[vc];
-                            synth.voices[i].voiceValues.uniPanShift = uniPan[vc];
-                            synth.voices[i].voiceValues.uniPMScale = uniScale[vc];
+                                hasCenter && (std::fabs(uniScale) < 1e-4f);
+                            synth.voices[i].voiceValues.uniRatioMul = 1.f;
+                            synth.voices[i].voiceValues.uniPanShift = 0.f;
+                            synth.voices[i].voiceValues.uniPMScale = uniScale;
                             synth.voices[i].voiceValues.phaseRandom = (vc > 0 && upr);
                             synth.voices[i].voiceValues.rephaseOnRetrigger = (!upr && prt);
                             synth.voices[i].voiceValues.noteExpressionTuningInSemis = 0;

src/synth/voice.cpp

@@ -72,6 +72,17 @@ void Voice::attack()
 
 void Voice::renderBlock()
 {
+    // Refresh unison-derived per-voice scalars from the (smoothed) mono hoists so
+    // unisonSpread / unisonPan track host automation and UI knob moves mid-note.
+    if (voiceValues.uniCount > 1)
+    {
+        const float prod = monoValues.unisonSpreadFactorMinus1 * voiceValues.uniPMScale;
+        voiceValues.uniRatioMul =
+            (voiceValues.uniPMScale >= 0.f) ? (1.f + prod) : (1.f / (1.f - prod));
+        voiceValues.uniPanShift =
+            std::clamp(monoValues.unisonPanScalar * voiceValues.uniPMScale, -1.f, 1.f);
+    }
+
     float retuneKey = voiceValues.key;
     if (monoValues.mtsClient && MTS_HasMaster(monoValues.mtsClient))
     {