Automatic Parallelization: Predicaments in Computationally Expensive Operations

Abstract

The applications developed for parallelizing are either single file based programs or algorithm structures. The
commercial large scale application pose problems when performing automatic parallelization and needs to be
addressed and hence thought to be implausible. This paper tries to surface the impediments to be addressed so that
the parallelization techniques may be applied to these applications. Benchmark suite which are specifically
developed to expose the computing requirement and are well acclaimed in the industry have been adopted.
Benchmarks used are from High Performance Group of the Standard Performance rating Corporation (SPEC), both
parallel and serial versions of applications are used.
Automatic parallel serial codes are compared with its parallel variants in this paper. Parallelizing compiler is
employed that takes language (formula based) codes and inserts Open Multi-Processing directives around loops
determined to be autonomous. Different problems were faced by an automatic parallelizing compiler when dealing
with full applications procedures, optimizations based on superficial techniques, array input output and size
variations ,Multilanguage hurdles, extreme inclination to library (system defined functions) and endowment
accumulations. The results presented in this paper shall benefit parallelizing compilers with capabilities for
handling large scale science and engineering applications.