cpu_neon.c
1 #ifdef _MSC_VER 2 #include <Intrin.h> 3 #endif 4 #include <arm_neon.h> 5 6 int main(int argc, char **argv) 7 { 8 // passing from untraced pointers to avoid optimizing out any constants 9 // so we can test against the linker. 10 float *src = (float*)argv[argc-1]; 11 float32x4_t v1 = vdupq_n_f32(src[0]), v2 = vdupq_n_f32(src[1]); 12 int ret = (int)vgetq_lane_f32(vmulq_f32(v1, v2), 0); 13 #ifdef __aarch64__ 14 double *src2 = (double*)argv[argc-2]; 15 float64x2_t vd1 = vdupq_n_f64(src2[0]), vd2 = vdupq_n_f64(src2[1]); 16 ret += (int)vgetq_lane_f64(vmulq_f64(vd1, vd2), 0); 17 #endif 18 return ret; 19 }