Please make this function as four-way loop unrolling with four parallel accumulators (the revised one should run faster than original one with larger data length).

void inner (float *u, float *v, int length, float *dest) {

int i;

float sum = 0.0f;

for (i = 0; i < length; ++i) {

sum += u[i] * v[i];

}

*dest = sum;

}