HPC Based Algorithmic Species Extraction Tool for Automatic Parallelization of Program Code

Abstract

Without a doubt, multiple core processors have become primary stream in parallel computing. Therefore, future generations of applications pivotal role will be played by parallelism. It must be noted that, the compilers and programmers could immensely benefit from a program source code classified in a structured manner. Such a classification surely helps programmers to identify parallelization scopes or reasoning about the program code, and associate with other programmers. To address the challenge of parallel programming, we worked on source-to-source compiler Bones and developed species extraction tool extended A-Darwin to ease parallel programming. In the work done, we present ’Algorithmic Species’, a new algorithm classification, that encapsulates required information for parallelization in classes, and embeds memory transfer requirements for optimization of communication on heterogeneous platforms. The evaluation of algorithmic species and the validation of extended A-Darwin are done by testing the tool against the benchmark suit HPCC. The unique approach is developed to generate code automatically for parallel target machines.