S4 Tier 0: the bank plays — VST3 marshals MIDI to the S3 core, reads the live bank + resolves WAV the M4 way, mono downmix, lock-free load handoff, LICE sample-pick

src/vst/sampler_core.cpp

@@ -289,18 +289,29 @@ void VoiceEngine::noteOff(int note) {
     if (target != kNoVoice) voices_[target].release();
 }
 
-void VoiceEngine::render(std::vector<AudioSample>& out, std::size_t frameCount) {
-    const std::size_t base = out.size();
-    out.resize(base + frameCount, 0.0f);  // S4: caller must pre-reserve — no allocation allowed under the VST3 process callback.
+void VoiceEngine::render(AudioSample* out, std::size_t frameCount) {
+    // Real-time safe: no allocation, no resize — mix straight into the caller's buffer.
+    // The VST3 process callback hands us the host's output channel buffer here, so the
+    // audio thread never touches the heap (S4 real-time discipline).
+    if (out == nullptr || frameCount == 0) return;
     for (Voice& voice : voices_) {
         if (!voice.active()) continue;
         for (std::size_t f = 0; f < frameCount; ++f) {
             if (!voice.active()) break;
-            out[base + f] += voice.renderFrame();
+            out[f] += voice.renderFrame();
         }
     }
 }
 
+void VoiceEngine::render(std::vector<AudioSample>& out, std::size_t frameCount) {
+    // Off-thread / test path: grow the buffer (this allocates — never call under
+    // process), zero-fill the appended span, then delegate to the RT mix loop so both
+    // overloads share exactly one summation path.
+    const std::size_t base = out.size();
+    out.resize(base + frameCount, 0.0f);
+    render(out.data() + base, frameCount);
+}
+
 std::size_t VoiceEngine::activeVoiceCount() const {
     std::size_t n = 0;
     for (const Voice& v : voices_) {