#include <stdio.h>
#include <emmintrin.h>

double x[4] __attribute__ ((aligned (16))) = {1.0, 2.0, 3.0, 4.0};
double y[5] __attribute__ ((aligned (16))) = {1.0, 2.0, 3.0, 4.0};

void
foo(void)
{
    __m128d x01 = _mm_load_pd(x);
    __m128d x23 = _mm_load_pd(x+2);

    __m128d y01 = _mm_load_pd(y);
    __m128d y23 = _mm_load_pd(y+2);

    y01 = _mm_add_pd(x01, y01);
    y23 = _mm_add_pd(x23, y23);

    _mm_store_pd(y+1, y01);
    _mm_store_pd(y+3, y23);
}

int
main()
{
    foo();
    printf("y = (%lf, %lf, %lf, %lf)\n", y[0], y[1], y[2], y[3]);
}