Add NEON-enhanced FIR loops for convolution and UHJ

1 files changed, 15 insertions, 0 deletions

diff --git a/alc/effects/convolution.cpp b/alc/effects/convolution.cpp
index 64c05172..4e10b62a 100644
--- a/alc/effects/convolution.cpp
+++ b/alc/effects/convolution.cpp
@@ -5,6 +5,8 @@
 
 #ifdef HAVE_SSE_INTRINSICS
 #include <xmmintrin.h>
+#elif defined(HAVE_NEON)
+#include <arm_neon.h>
 #endif
 
 #include "alcmain.h"
@@ -126,6 +128,19 @@ void apply_fir(al::span<float> dst, const float *RESTRICT src, const float *REST
         ++src;
     }
 
+#elif defined(HAVE_NEON)
+
+    for(float &output : dst)
+    {
+        float32x4_t r4{vdupq_n_f32(0.0f)};
+        for(size_t j{0};j < ConvolveUpdateSamples;j+=4)
+            r4 = vmlaq_f32(r4, vld1q_f32(&src[j]), vld1q_f32(&filter[j]));
+        r4 = vaddq_f32(r4, vrev64q_f32(r4));
+        output = vget_lane_f32(vadd_f32(vget_low_f32(r4), vget_high_f32(r4)), 0);
+
+        ++src;
+    }
+
 #else
 
     for(float &output : dst)