Performance Characterization of Rasterization Algorithms for Reconfigurable Graphics Processor

For different rasterized scenes, even different areas of the same scene, the performance bottleneck of rasterization may be different, and current graphics processors cannot choose the appropriate rasterization algorithm according to the specific rendering scene. Therefore, a reconfigurable graphics processor that supports switching of algorithms to achieve the best performance is a promising choice. The existing graphics processor suffers the constraints of calculation, memory and power consumption in different rasterization application scenarios. Therefore, it is very important to determine how to schedule different rasterization algorithms with different performance according to the actual requirements in the reconfigurable graphics hardware. This paper evaluates and analyzes the performance characteristics of three main-stream rasterization algorithms (scan-line filling algorithm, edge filling algorithm, and flood filling algorithm) in different application scenarios. Pearson correlation coefficient (PCC) analysis is leveraged to analyze the relationship between performance/energy and evaluation metrics. Based on these performance characterization data, this paper puts forward some reconstruction suggestions for the reconfigurable graphics processor. We hope to contribute to reconfigurable graphics processing.