Speed up of rendering pipeline by deferred lighting and triple queue structure
1 -7 7 7 2 7 5 0 7 3 8 7 2 7 -7 2 3 b Redundant operations and stalls prevent a rendering pipeline from full-speed operation. To speed up the rendering pipeline, a triple queue structure is proposed to smooth the pipeline and to obtain benefit from deferred lighting. The results of cycle-accurate simulation show that the proposed structure can reduce rendering cycles to 52.9y0 in small size queues. ? ! 80-60-r Introduction: The rendering pipeline is a mainstream method to implement
... onal (3D) graphics rendering. However, the rendering pipeline method has two problems. The first problem is redundant operations on invisible polygons. Deferred lighting [l] can delete the redundant operations, and save operation cycles on them. However, if not well handled, the saved cycles may become useless 'bubbles' in the pipeline. Our purpose was to design a specific architecture to obtain benefit from the save cycles. The second problem is pipeline stalls. Due to various polygon sizes, polygon and pixel rates are not in linear proportion. Stalls occur whenever the data rates mismatch the process speed of hardware. A straightforward way to avoid stalls is to insert one queue as a buffer between the polygon and pixel rate areas [2, 31. However, the queue size quickly grows if a longer queue is required, since 20 to 40 bytes are required for a single entry in this queue. To solve these two problems, we propose a triple queue structure to smooth the pipeline and to obtain benefit from deferred lighting. Since three queues surround the lighting module, bubbles and stalls in the pipeline can be avoided. Therefore, this design can obtain benefit from deferred lighting, and speed up the rendering pipeline significantly by only having small queue sizes.