Advances in configurable logic technology have permitted the development of low-cost, high-speed configurable devices, allowing one or more soft processor cores to be introduced into a configurable computing system. Soft processor cores offer logic-area savings and reduced configuration times when compared to the hardware-only implementations typically used for application specific acceleration. Programs for a soft processor core are small and simple compared to the design of a hardware core, but can leverage custom hardware within the processor core to provide greater acceleration for specific applications. This thesis presents several configurable system models, and implements one such model on a Nios Embedded Processor Development Board. A software programmable and hardware configurable lightweight processor core known as the FAST CPU is introduced. The configurable system implementation attaches several FAST CPUs to a standard Nios processor to create a system for experimentation with application specific acceleration. This system incorporating the FAST CPUs was tested for bus utilization behaviour, computing performance, and execution times for a minheap application. Experimental results are compared to the performance of a software-only solution, and also with previous research results. Experimental results verify that the theory and models used to predict bus utilization are correct. Performance testing shows that the FAST CPU is approximately 25% slower than a general purpose processor, which is expected. The FAST CPU, however, is 31% smaller in terms of logic area than the general purpose processor, and is 8% smaller than the design of a hardware-only implementation of a minheap for application specific acceleration. The results verify that it is possible to move functionality from a general purpose processor to a lightweight processor, and further, to realize an increase in performance when a task is parallelized across multiple FAST CPUs. The experimentation uses a procedure by which a set of equations can be derived for predicting bus utilization and deriving a cost-benefit curve for a coprocessing entity. They are applied to a specific system in this research, but the methods are generalizable to any coprocessing entity.

Advances in configurable logic technology have permitted the development of low-cost, high-speed configurable devices, allowing one or more soft processor cores to be introduced into a configurable computing system. Soft processor cores offer logic-area savings and reduced configuration times when compared to the hardware-only implementations typically used for application specific acceleration. Programs for a soft processor core are small and simple compared to the design of a hardware core, but can leverage custom hardware within the processor core to provide greater acceleration for specific applications. This thesis presents several configurable system models, and implements one such model on a Nios Embedded Processor Development Board. A software programmable and hardware configurable lightweight processor core known as the FAST CPU is introduced. The configurable system implementation attaches several FAST CPUs to a standard Nios processor to create a system for experimentation with application specific acceleration. This system incorporating the FAST CPUs was tested for bus utilization behaviour, computing performance, and execution times for a minheap application. Experimental results are compared to the performance of a software-only solution, and also with previous research results. Experimental results verify that the theory and models used to predict bus utilization are correct. Performance testing shows that the FAST CPU is approximately 25% slower than a general purpose processor, which is expected. The FAST CPU, however, is 31% smaller in terms of logic area than the general purpose processor, and is 8% smaller than the design of a hardware-only implementation of a minheap for application specific acceleration. The results verify that it is possible to move functionality from a general purpose processor to a lightweight processor, and further, to realize an increase in performance when a task is parallelized across multiple FAST CPUs. The experimentation uses a procedure by which a set of equations can be derived for predicting bus utilization and deriving a cost-benefit curve for a coprocessing entity. They are applied to a specific system in this research, but the methods are generalizable to any coprocessing entity.