More efficient pre-warming using multiple-sample buffers

NAM/dsp.cpp

@@ -25,14 +25,22 @@ void nam::DSP::prewarm()
   if (prewarmSamples == 0)
     return;
 
-  NAM_SAMPLE sample = 0;
-  NAM_SAMPLE* sample_ptr = &sample;
+  const size_t bufferSize = std::max(mMaxBufferSize, 1);
+  std::vector<NAM_SAMPLE> inputBuffer, outputBuffer;
+  inputBuffer.resize(bufferSize);
+  outputBuffer.resize(bufferSize);
+  for (auto it = inputBuffer.begin(); it != inputBuffer.end(); ++it)
+  {
+    (*it) = (NAM_SAMPLE)0.0;
+  }
 
-  // pre-warm the model for a model-specific number of samples
-  for (long i = 0; i < prewarmSamples; i++)
+  NAM_SAMPLE* inputPtr = inputBuffer.data();
+  NAM_SAMPLE* outputPtr = outputBuffer.data();
+  int samplesProcessed = 0;
+  while (samplesProcessed < prewarmSamples)
   {
-    this->process(sample_ptr, sample_ptr, 1);
-    sample = 0;
+    this->process(inputPtr, outputPtr, bufferSize);
+    samplesProcessed += bufferSize;
   }
 }
 

tools/benchmodel.cpp

@@ -37,7 +37,6 @@ int main(int argc, char* argv[])
       exit(1);
     }
 
-    auto t1 = high_resolution_clock::now();
 
     size_t bufferSize = AUDIO_BUFFER_SIZE;
     model->Reset(model->GetExpectedSampleRate(), bufferSize);
@@ -49,17 +48,18 @@ int main(int argc, char* argv[])
     {
       inputBuffer[i] = 0.0;
     }
+    // Run once outside to trigger any pre-allocation
+    model->process(inputBuffer, outputBuffer, AUDIO_BUFFER_SIZE);
 
     std::cout << "Running benchmark\n";
-
+    auto t1 = high_resolution_clock::now();
     for (size_t i = 0; i < numBuffers; i++)
     {
       model->process(inputBuffer, outputBuffer, AUDIO_BUFFER_SIZE);
     }
-
+    auto t2 = high_resolution_clock::now();
     std::cout << "Finished\n";
 
-    auto t2 = high_resolution_clock::now();
 
     /* Getting number of milliseconds as an integer. */
     auto ms_int = duration_cast<milliseconds>(t2 - t1);