SUIF Explorer: A Programming Assistant for Parallel Machines

This paper presents the preliminary design of SUIF Explorer, a system that uses state-ofthe -art parallelization analysis to guide a programmer in developing and optimizing efficient parallel code. It alleviates the tedious process of performance debugging by analyzing the results from both the SUIF compiler [1] and execution analyzers and presenting high-level information to users via cutting-edge visualization tools. It focuses the attention of the user on the critical slice of an application, asking pointed questions which lead even a naive user down the same path a compiler expert might take in optimizing the program. This paper illustrates the effectiveness of SUIF Explorer with a case study based on the MDG application in the PERFECT benchmark [3]. Users only need to examine a few lines of code to parallelize a loop that spans over several procedures and over a hundred lines of code