Star-P capabilities can be substantially and easily extended through the use of Star-P Connect—a powerful API/SDK tool that allows addition of custom, user-defined functions and libraries. Moreover, Star-P Connect supports easy additions of unlimited variety of data—and task-parallel libraries containing user-defined, community, or commercial best-of-breed computational capabilities.

View Larger Image >

As seen in the diagram above, Star-P architecture provides for natural expansion of its parallel computing capabilities, through the addition of custom libraries.

Using Star-P Connect to Add Library or Hardware-based Functions
A user of Star-P + VHLL parallel computing environment has the ability to easily add more function for both task- and data-parallel computing. The add-on functions may be

• Application-specific, purpose-developed by the user
• Third-party (commercial, community, application-specific etc.) library functions
• Hardware-based functions (computing accelerators)

In all three cases, Star-P Connect Library API allows the user to add functions following a simple four-step process (illustrated below for adding a “foo” function):

Adding and Using Library- or Hardware-based Computing Functions

View Larger Image >

Step 1: First, a Star-P Connect package, written in C++, must be created.

Step 2: Then, the Starp-P Connect package (.cpp files) must be compiled and converted to a shared object form. At this point, the add-on function is available for computing from the Star-P + VHLL environment, almost.

Step 3: The third and last “preparation for use” step is to load the package into the Star-P + VHLL environment.

Step 4: Finally, after the package has been loaded, the user can perform computations with the add-on functions included in the package, using the appropriate Star-P compute commands to trigger the computation.

Note that with Star-P Connect, it is possible for the user to add a new parallel computing function, without using an add-on library. Simply, to achieve this, one would include the full code of the new function on the package instead of just a wrapper for the function.