High Performance Parallelism Pearls Volume Two

Chapter 21

Prefetch Tuning Optimizations

Diana Guttman*; Meenakshi Arunachalam^†; Vlad Calina^‡; Mahmut Taylan Kandemir*    ^* Pennsylvania State University, USA^† Intel Corporation, USA^‡ Intel Corporation, Romania

Abstract

This chapter looks at methods to improve prefetching effectiveness, and therefore increase performance of applications, through the use of the superior knowledge of the programmer. It is known that prefetching is extremely important for good performance on in-order architectures like the Intel Xeon Phi coprocessor; however, the authors surprised even themselves by being able to expose techniques which show value on out-of-order cores as well. Often simply tuning the compiler prefetching distance is an easy way for application ...