fix neon off build

Tencent · Sep 3, 2024 · f269b20 · f269b20
1 parent 464da66
commit f269b20
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/src/layer/arm/rmsnorm_arm.cpp b/src/layer/arm/rmsnorm_arm.cpp
@@ -90,13 +90,15 @@ static void rmsnorm(float* ptr, const float* gamma_ptr, float eps, int elemcount
     float a;
     if (elempack == 1)
     {
+#if __ARM_NEON
 #if __aarch64__
         sqsum += vaddvq_f32(_sqsum);
 #else
         float32x2_t _s2 = vadd_f32(vget_low_f32(_sqsum), vget_high_f32(_sqsum));
         _s2 = vpadd_f32(_s2, _s2);
         sqsum += vget_lane_f32(_s2, 0);
 #endif
+#endif // __ARM_NEON
 
         a = 1.f / sqrtf(sqsum / elemcount + eps);
 #if __ARM_NEON
@@ -286,13 +288,15 @@ static void rmsnorm_bf16s(unsigned short* ptr, const float* gamma_ptr, float eps
     float a;
     if (elempack == 1)
     {
+#if __ARM_NEON
 #if __aarch64__
         sqsum += vaddvq_f32(_sqsum);
 #else
         float32x2_t _s2 = vadd_f32(vget_low_f32(_sqsum), vget_high_f32(_sqsum));
         _s2 = vpadd_f32(_s2, _s2);
         sqsum += vget_lane_f32(_s2, 0);
 #endif
+#endif // __ARM_NEON
 
         a = 1.f / sqrtf(sqsum / elemcount + eps);
 #if __ARM_NEON