int i = 0;
    for (int row = 0 ; row < first_rows ; row++) {
        mat_a[row] = (int *) malloc(shared_dim * sizeof(int));
        for (int col = 0 ; col < shared_dim ; col++) {
            mat_a[row][col] = ++i;
        }
    }
    for (int row = 0 ; row < shared_dim ; row++) {
        mat_b[row] = (int *) malloc(second_cols * sizeof(int));
        for (int col = 0 ; col < second_cols ; col++) {
            mat_b[row][col] = ++i;
        }
    }