To use multitasking successfully, the parallel functions must have computational independence. This means that no data modified by either the main task program or a parallel function is examined or modified by a parallel function that might be running simultaneously.
Figure 1 is a graphic example of hypothetical data in an array subscripted by I, J, and K. Each of the three divisions of the box represents a section of the array that can be operated on independently of the other sections. The same parallel function could be scheduled three times, with each instance of the function processing one of the three sections of the array.
Your application may not have computational independence along the same subscript axis of K, as in this picture. The divisions might have been along one of the other subscript axes, I or J. Also, the computational independence in your application may not fall into neat, box-like divisions.
It is also possible to have computational independence that is not based on sections of the same array, but rather on separate arrays (perhaps with completely different types of data), the values of which do not depend on each other. In this case, separate parallel functions could be scheduled, with each function processing its own unique data.
Computational independence also applies to input/output files. One parallel function should not use a file while another is updating it. However, different functions can successfully read the same file. No single file pointer should be used concurrently by multiple parallel functions, because the behavior is undefined in such a case.