| At the beginning of the 21st century, it became apparent that the performance gains associated with continual die shrinks and the resulting increases in core central processing unit (CPU) speeds were beginning to flatten. This realization has gradually shifted the focus of CPU design away from single core speed increases and toward the idea of obtaining performance through increased concurrency. The resulting design paradigm has given us multi- and many-core CPUs, vector processing units, and more recently, programmable, massively parallel hardware coprocessors, such as graphics processing units from nVidia and Advanced Micro Devices, along with more recent general purpose devices such as Intel's "Knights Corner." One of the most significant resulting challenges in high-performance computing is to provide a framework in which the software development process is platform agnostic to its end users, while at the same time being capable of scaling efficiently on diverse hardware configurations. This thesis will present an improved approach for the analysis and scheduling of computational tasks within a heterogeneous hardware environment, while removing implementation details from end users. This will be presented within the context of the "Expressions" framework, a component within a computational fluid dynamics solver, known as "Wasatch," developed at the University of Utah. |
