Video: http://blip.tv/file/4196565 or http://www.sacrideo.us/~arcfide/vid/pl_wonks/nilesh_20101001.html (WEBM HTML5 capable browser required)

Abstract

Communication related optimizations form an interesting class of compiler optimizations. Data-flow analysis can be used to determine efficient placement of communication primitives (e.g. send, receive). In this talk we will see how data flow equations can be set up to perform various optimizations like communication-computation overlap, message coalescing, vector message pipelining. I will start with a brief introduction to data-flow analysis, proceed to explain the data-flow equations required for the optimizations. I will also talk about different approaches to verify these analyses. Finally I would like to discuss the efficacy of these optimizations and whether similar analyses can be applied to memory hierarchy.