Resumen de Achieving high performance by targeting multiple parallelism mechanism
Recently, processor vendors have begun increasing performance by adding additional cores, rather than increasing the performance of single core. The addition of multiple cores augments several other parallel hardware mechanisms already in place on each core. These features create the potential for increased performance, but only if they are properly utilized. Programmers who disregard the underlying design of the hardware in newer processors can actually produce code that runs slower on a multicore processor. In this article, I explain what these design features are. I also discuss the underlying memory models and the impact they haveon proccesing in a multicore context. Finally, I present an example of a method of writting abstract parallel code that allows a development platform to do the heavy lifting of implementing code for different processor architectures.
