n2 loops

struct Array {
    double * restrict data;
    long n;
};


#define X   0
#define Y   1
#define Z   2


void __attribute__((noinline))
func1(double * restrict d, double * restrict out, const long n)
{
    for (long i = 0; i < n; ++ i) {
        for (long j = i + 1; j < n; ++ j) {

            const double rx = d[i * 3 + 0] * d[j * 3 + 0];
            const double ry = d[i * 3 + 1] * d[j * 3 + 1];
            const double rz = d[i * 3 + 2] * d[j * 3 + 2];

            out[i * 3 + X] += rx;
            out[i * 3 + Y] += ry;
            out[i * 3 + Z] += rz;

            out[j * 3 + X] -= rx;
            out[j * 3 + Y] -= ry;
            out[j * 3 + Z] -= rz;
        }
    }
}


void __attribute__((noinline))
func2(struct Array * restrict d, struct Array * restrict out, const long n)
{
    for (long i = 0; i < n; ++ i) {
        for (long j = i + 1; j < n; ++ j) {

            const double rx = d->data[i * 3 + X] * d->data[j * 3 + X];
            const double ry = d->data[i * 3 + Y] * d->data[j * 3 + Y];
            const double rz = d->data[i * 3 + Z] * d->data[j * 3 + Z];

            out->data[i * 3 + X] += rx;
            out->data[i * 3 + Y] += ry;
            out->data[i * 3 + Z] += rz;

            out->data[j * 3 + X] -= rx;
            out->data[j * 3 + Y] -= ry;
            out->data[j * 3 + Z] -= rz;
        }
    }
}