Game applications nowadays run on platforms that contain multiple and different types of processors. Benefiting from the available processing power requires: (1) Code to be designed to run in parallel; and (2) Efficient mapping techniques to schedule tasks optimally on appropriate processors for increasing performance. The work in this thesis is directed towards improving the execution time performance of game applications on host platforms. The methodology adopted consisted of the following different stages: (1) Studying the main features of game applications; (2) Investigating parallelization opportunities within these applications, (3) Identifying/designing the programming structures needed to realize those opportunities; and (4) Designing, implementing, verifying and testing the parallel programming structures developed to support heterogeneous processing and maximize use of available processing power.

The research reported in this thesis presents a new framework for parallel programming of games. This framework consists of two major components which form the two main contributions of our work. These are (1) Sayl, a new design pattern to enable parallel programming of game applications and to reduce the amount of programmer work when writing parallel code; and (2) Arbiter Work Stealing, a new task scheduler capable of efficiently and automatically scheduling dynamically generated tasks to processors in a heterogeneous processing and memory environment. The framework was used to successfully parallelize the serial code of a sample game application to work in a heterogeneous processing environment. The parallel version shows far superior performance as compared to the serial version.