ABSTRACT
In Chapter 5, the strategy for approaching the optimization of a large application for an MPP is given. If the majority of the runtime indicates that the compute time is the bottleneck, this is the right chapter to look at. We look at some simple but informative kernels that illustrate the right and wrong way to program for multicore processors. In this chapter we are only concerned with optimizing on a single core and cover parallelization over multiple cores later.
