Add further NEON steps to the codepath

src/sfizz/Panning.cpp

@@ -39,6 +39,8 @@ void pan(const float* panEnvelope, float* leftBuffer, float* rightBuffer, unsign
 {
 #if SFIZZ_HAVE_NEON
     uint32_t indices[4];
+    float leftPan[4];
+    float rightPan[4];
     const auto sentinel = panEnvelope + size;
     while (panEnvelope < sentinel) {
         float32x4_t mmPan = vld1q_f32(panEnvelope);
@@ -50,16 +52,18 @@ void pan(const float* panEnvelope, float* leftBuffer, float* rightBuffer, unsign
         uint32x4_t mmIdx = vcvtq_u32_f32(mmPan);
         vst1q_u32(indices, mmIdx);
 
-        #define UNROLLED_BLOCK(i)                               \
-        leftBuffer[i] *= panData[indices[i]];                \
-        rightBuffer[i] *= panData[panSize - indices[i] - 1];
 
-        UNROLLED_BLOCK(0)
-        UNROLLED_BLOCK(1)
-        UNROLLED_BLOCK(2)
-        UNROLLED_BLOCK(3)
+        leftPan[0] = panData[indices[0]];
+        rightPan[0] = panData[panSize - indices[0] - 1];
+        leftPan[1] = panData[indices[1]];
+        rightPan[1] = panData[panSize - indices[1] - 1];
+        leftPan[2] = panData[indices[2]];
+        rightPan[2] = panData[panSize - indices[2] - 1];
+        leftPan[3] = panData[indices[3]];
+        rightPan[3] = panData[panSize - indices[3] - 1];
 
-        #undef UNROLLED_BLOCK
+        vst1q_f32(leftBuffer, vmulq_f32(vld1q_f32(leftBuffer), vld1q_f32(leftPan)));
+        vst1q_f32(rightBuffer, vmulq_f32(vld1q_f32(rightBuffer), vld1q_f32(rightPan)));
 
         incrementAll<4>(panEnvelope, leftBuffer, rightBuffer);
     }
@@ -80,7 +84,8 @@ void width(const float* widthEnvelope, float* leftBuffer, float* rightBuffer, un
 {
 #if SFIZZ_HAVE_NEON
     uint32_t indices[4];
-    float coeff1, coeff2, l, r;
+    float coeff1[4];
+    float coeff2[4];
     const auto sentinel = widthEnvelope + size;
     while (widthEnvelope < sentinel) {
         float32x4_t mmWidth = vld1q_f32(widthEnvelope);
@@ -92,20 +97,22 @@ void width(const float* widthEnvelope, float* leftBuffer, float* rightBuffer, un
         uint32x4_t mmIdx = vcvtq_u32_f32(mmWidth);
         vst1q_u32(indices, mmIdx);
 
-        #define UNROLLED_BLOCK(i)                      \
-        coeff1 = panData[indices[i]];                 \
-        coeff2 = panData[panSize - indices[i] - 1];   \
-        l = leftBuffer[i];                             \
-        r = rightBuffer[i];                            \
-        rightBuffer[i] = l * coeff2 + r * coeff1;      \
-        leftBuffer[i] = l * coeff1 + r * coeff2;
-
-        UNROLLED_BLOCK(0)
-        UNROLLED_BLOCK(1)
-        UNROLLED_BLOCK(2)
-        UNROLLED_BLOCK(3)
-
-        #undef UNROLLED_BLOCK
+        coeff1[0] = panData[indices[0]];
+        coeff2[0] = panData[panSize - indices[0] - 1];
+        coeff1[1] = panData[indices[1]];
+        coeff2[1] = panData[panSize - indices[1] - 1];
+        coeff1[2] = panData[indices[2]];
+        coeff2[2] = panData[panSize - indices[2] - 1];
+        coeff1[3] = panData[indices[3]];
+        coeff2[3] = panData[panSize - indices[3] - 1];
+
+        float32x4_t mmCoeff1 = vld1q_f32(coeff1);
+        float32x4_t mmCoeff2 = vld1q_f32(coeff2);
+        float32x4_t mmLeft = vld1q_f32(leftBuffer);
+        float32x4_t mmRight = vld1q_f32(rightBuffer);
+
+        vst1q_f32(leftBuffer, vaddq_f32(vmulq_f32(mmCoeff2, mmLeft), vmulq_f32(mmCoeff1, mmRight));
+        vst1q_f32(rightBuffer, vaddq_f32(vmulq_f32(mmCoeff1, mmLeft), vmulq_f32(mmCoeff2, mmRight));
 
         incrementAll<4>(widthEnvelope, leftBuffer, rightBuffer);
     }