Abstract
This chapter focuses on IRIS Power C performance issues. The various parallel variable types incur different degrees of overhead when either accessed from within or passed to the parallel procedure created from the parallel region. The performance penalty per variable type is similar to a serial C program that makes a function call. Global variables that are on the shared or by value lists are referenced directly and have no additional overhead. The additional overhead associated with the accessing of variables inside a parallelized region can steal program performance depending on how those variables are used. The execution environment, such as the underlying system that manages the executing parallel program, is fundamentally the same for IRIS Power C and Power Fortran. A key difference is that IRIS Power C does not provide library routines for tuning the behavior of the program while it is running. The proper choices for block time and block type are critical when a parallel program competes for resources with other processes in a single computing environment.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call