I have never written code in SIMD, and I wrote a simple code to test the performance and set my data as shown below however running the code does not give me superior performance as I should expect.

#include <iostream>
#include <ctime> 
#include <smmintrin.h>
#include <immintrin.h>

unsigned t0, t1;

struct VAL
{
    union
    {
        float   val[4];
        __m128  val_simd;
    };
};


int main()
{
    unsigned t0, t1;
    double time;
    VAL *a, *b, *c;
    unsigned test_num = 100000000;

    a = new VAL[test_num];
    b = new VAL[test_num];
    c = new VAL[test_num];
    
    t0 = clock();
    for (int i = 0; i < test_num; i++)
    {
        a[i].val[0] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        a[i].val[1] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        a[i].val[2] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        a[i].val[3] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        b[i].val[0] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        b[i].val[1] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        b[i].val[2] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        b[i].val[3] = static_cast <float> (rand()) / static_cast <float> (RAND_MAX);
        c[i].val[0] = 0.0f;
        c[i].val[1] = 0.0f;
        c[i].val[2] = 0.0f;
        c[i].val[3] = 0.0f;
    }
    t1 = clock();
    time = (double(t1 - t0) / CLOCKS_PER_SEC);
    std::cout << "Execution Time0: " << time << 'n';

    t0 = clock();
    for (int i = 0; i < test_num; i++)
    {
        c[i].val[0] = a[i].val[0] + b[i].val[0];
        c[i].val[1] = a[i].val[1] + b[i].val[1];
        c[i].val[2] = a[i].val[2] + b[i].val[2];
        //c[i].val[3] = a[i].val[3] + b[i].val[3];
    }
    t1 = clock();
    time = (double(t1 - t0) / CLOCKS_PER_SEC);
    std::cout << "Execution Time1: " << time << 'n';

    t0 = clock();
    for (int i = 0; i < test_num; i++)
    {
        c[i].val_simd = _mm_add_ps(a[i].val_simd, b[i].val_simd);
    }
    t1 = clock();
    time = (double(t1 - t0) / CLOCKS_PER_SEC);
    std::cout << "Execution Time2: " << time << 'n';

    delete[] a;
    delete[] b;
    delete[] c;
}

I expected to see much better performance, at least 3x the performance, however this gave the same times. On the other hand, testing the same code but multiplying SIMD gave me but performance