void rec_swap3(int side) {
    for (int i = 0; i < side; i += 8) {
        for (int j = 0; j <= i; j += 8) {
            for (int ii = 0; ii < 8 && i + ii < side; ii++) {
                for (int jj = 0; jj < 8 && j + jj < side; jj++) {
                    int x = i + ii, y = j + jj;
                    if (x > y) {
                        swap(A[(I+x)*N+J+y], A[(I+y)*N+J+x]);
                    }
                }
            }
        }
    }
}