Asterix A Graphical Automated Program Transformation and Parallelization Framework

Abstract

Parallel Processing of programs is a process where independent parts of a sequential program are newlineidentified and run concurrently on multiple processors with the sole aim of reducing the runtime newlineof the targeted program. The simplest way to speed up programs is to execute them on faster newlinehardware. This method is limited by the availability of such faster hardware. In the absence of newlinesuch faster hardware the only resort is Parallel Processing of programs. Earlier efforts included newlinemanually identifying concurrent parts of a program and achieve Parallel Processing through newlineexplicit programming. The next step in the evolution of Parallel Processing is the use of newlinespecialized parallel languages to identify parallel parts and achieve the same result. Of late newlineresearchers have tried to automate the entire Parallel Processing step. Compilers can now newlineautomate the various steps of Parallel Processing of programs using the sequential program as newlineinput and using the results of analyzing the given program. However the automated parallel newlineprocessing method is still evolutionary and offers ample opportunities for research and newlineinnovation. Till now researchers have focused on specific type of program mostly numerical in newlinenature at the cost of ignoring other kinds of applications. Researchers have targeted specific newlinetypes of hardware such as Shared Memory Processors and almost ignored other architectures newlinesuch as Distributed Memory Processors. Also missing at this juncture is a methodology to newlineestimate and predict the benefits of Parallel Processing. It is important for programmers to have a newlineperformance estimate upfront so that they can embark on Parallel Processing projects with newlineconfidence instead of resorting to runtime measurements at the end after completing the Parallel newlineConversions. Researchers now use multiple models of programs to solve the Parallel Processing newlineproblems.