/**
 * Use of AVX vector units 
 * unroll by a factor of 2
 * perform independant calculations
 */
f32 dp_avx_vec(f32 *x, f32 *y, u32 size) {
   // use ymm0 to store v_sum and ymm2 for w_sum
   f32 v_sum[8], w_sum[8];
    u32 i;
    
    // use xorps ymm0, ymm0
    v_sum[0] = v_sum[1] = v_sum[2] = v_sum[3] = 0.0;
    v_sum[4] = v_sum[5] = v_sum[6] = v_sum[7] = 0.0;
    
    // use xorps ymm2, ymm2
    w_sum[0] = w_sum[1] = w_sum[2] = w_sum[3] = 0.0;
    w_sum[4] = w_sum[5] = w_sum[6] = w_sum[7] = 0.0;
    
   
    // loop unrolling by a factor of 16
    // use ymm1 to store x[i:i+7] and ymm3 to store y[i+8:i+15]
    for (i = 0; i < (size & ~15); i += 16) {
        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];
        v_sum[4] += x[i + 4] * y[i + 4];
        v_sum[5] += x[i + 5] * y[i + 5];
        v_sum[6] += x[i + 6] * y[i + 6];
        v_sum[7] += x[i + 7] * y[i + 7];
        
        w_sum[0] += x[i + 8]  * y[i + 8];
        w_sum[1] += x[i + 9]  * y[i + 9];
        w_sum[2] += x[i + 10] * y[i + 10];
        w_sum[3] += x[i + 11] * y[i + 11];
        w_sum[4] += x[i + 12] * y[i + 12];
        w_sum[5] += x[i + 13] * y[i + 13];
        w_sum[6] += x[i + 14] * y[i + 14];
        w_sum[7] += x[i + 15] * y[i + 15];
        
    }
    
    // sum of partial sums
    float v_sum_total = v_sum[0] + v_sum[1] + v_sum[2] + v_sum[3] +;
    	v_sum[4] + v_sum[5] + v_sum[6] + v_sum[7];
    float w_sum_total = w_sum[0] + w_sum[1] + w_sum[2] + w_sum[3] +;
    	w_sum[4] + w_sum[5] + w_sum[6] + w_sum[7];
    	
    float sum = v_sum_total + w_sum_total;	
    
    // last iterations	
    while (i < size) {
    	sum += x[i] * y[i];
    	++i;
    }
   
    return sum;
}