Simplify reduceAdd for ARM64

5d755e7f · Marc Marí · 83be63ce · 5d755e7f
Commit 5d755e7f authored May 02, 2020 by Marc Marí
Show whitespace changes
Inline Side-by-side

Showing with 4 additions and 0 deletions

openmmapi/include/openmm/internal/vectorize_neon.h openmmapi/include/openmm/internal/vectorize_neon.h +4 -0

No files found.
--- a/openmmapi/include/openmm/internal/vectorize_neon.h
+++ b/openmmapi/include/openmm/internal/vectorize_neon.h
@@ -340,7 +340,11 @@ static inline float dot4(fvec4 v1, fvec4 v2) {
 }

 static inline float reduceAdd(fvec4 v) {
+#ifdef __ARM64__
+    return vaddvq_f32(v);
+#else
    return dot4(v, fvec4(1.0f));
+#endif
 }

 static inline fvec4 cross(fvec4 v1, fvec4 v2) {