/**
 * Use of SSE vector units 
 * unroll by a factor of 2
 * perform independant calculations
 */
f32 dp_sse_vec(f32 *x, f32 *y, u32 size) {
    u32 i;
    
    // for independant calculcations we use two vectors
    // for example xmm0 for v_sum and xmm2 for w_sum
    f32 v_sum[4], w_w_sum[4];
    
    v_sum[0] = v_sum[1] = v_sum[2] = v_sum[3] = 0.0;
    w_sum[0] = w_sum[1] = w_sum[2] = w_sum[3] = 0.0; 
   
    // loop unrolling by a factor of 2x4
    // for example
    // we load x[i:i+3] in xmm1 and x[i+4:i+7] in xmm3
    for (i = 0; i < (size & ~7); i += 8) {
        v_sum[0] += x[i + 0] * y[i + 0];
        v_sum[1] += x[i + 1] * y[i + 1];
        v_sum[2] += x[i + 2] * y[i + 2];
        v_sum[3] += x[i + 3] * y[i + 3];
        
        w_sum[0] += x[i + 4] * y[i + 4];
        w_sum[1] += x[i + 5] * y[i + 5];
        w_sum[2] += x[i + 6] * y[i + 6];
        w_sum[3] += x[i + 7] * y[i + 7];
        
    }
    
    // partial sums of both vectors
    float v_sum_total = v_sum[0] + v_sum[1] + v_sum[2] + v_sum[3];
    float w_sum_total = w_sum[0] + w_sum[1] + w_sum[2] + w_sum[3];
    
    float sum = v_sum_total + w_sum_total;
    
    // last iterations
    while (i < size) {
    	sum += x[i] * y[i];
    	++i;
    }
   
    return sum;
}