How to parallelize a loop using thread?

Question

C++ / C#

The len variable is always equal to 32, so I divide it into 4 threads, each of which is supposed to perform 8 calculations, but it turns out that the calculation time increases compared to one thread. What am I doing wrong?
The loop itself:

vector<thread> thds;
for (int i = 0; i < len; i = i+4) { // основной цикл умножения
    thds.emplace_back(thread (multi, x, y, ref(res), len, i));
    thds.emplace_back(thread(multi, x, y, ref(res), len, i+1));
    thds.emplace_back(thread(multi, x, y, ref(res), len, i+2));
    thds.emplace_back(thread(multi, x, y, ref(res), len, i+3));
    thds[i].join();
    thds[i+1].join();
    thds[i+2].join();
    thds[i+3].join();
    //cout << i << endl;
  }

Function:

void multi(vector<int>& x, vector<int>& y, vector<int>& res, int len, int i) {
  for (auto j = 0; j < len; ++j)
    res[i + j] += x[i] * y[j];
}

PS I ask you not to scold me much for the implementation, I work with streams for the first time.

Answer 1

Why create threads every time in a loop? It is better to first create the required number of threads, taking into account their number in the system, and give the threads a range for calculations. Maybe you should take something ready-made like en.cppreference.com/w/cpp/thread/async see the example below, there is just a parallel summation of the vector elements. Note that if there are less than 1000 elements, no threads are created at all.

Answer 2

To organize parallel computing, OpenMP or Intel tbb are usually used. There, planning, balancing, creating threads comes out of the box. Try it.