Communication Optimization Algorithms based on Extend Data Flow Graph

Abstract

Reducing communication overhead is extremely important for parallelizing compiler to generate efficient codes for distributed memory machines. In this paper, a redundant parallel execution model (RPEM) is used as the model for target programs. The extend data flow graph is introduced, and optimization algorithms based on the data-flow analysis are discussed. The overhead of data flow analysis can be reduced by performing analysis on the extend dataflow graph. The analysis helps to reduce the redundant communication overhead. These optimization algorithms are able to perform inter-loop and inter-procedure analysis. Experimental results prove that these optimizations algorithms are effective in reducing both the number of communications and the communication volume.