CEngine

Abstract

In 3D graphics system, only the primitives wholly or partially inside the view volume are transformed from homogeneous coordinate into screen coordinates. Then, these are sent to rasterization part. Clipping operation is to decide whether primitives are within view volume or not. The primitives wholly inside the view volume are passed to next stage. The primitives intersected by view volume are clipped. Then, inside part of the primitive is sent to next stage. In next stage, perspective division and viewport mapping operations transform homogeneous coordinate to screen coordinate. In conventional graphics hardware, these operations have three data-paths for processing three vertices of a triangle in parallel. Conventional clipping algorithm can process for all of general polygons. However, this algorithm has long time to process a non-clipped polygon. And it needs intermediate buffer to store intermediate-data generated by clipping against each clip plane. This may increase hardware cost. In this thesis, we proposed a clipping algorithm and unified architecture for embedded system. First, a clipping algorithm can reduce processing time as deciding quickly whether a triangle is clipped or not. In addition, this algorithm modified conventional clipping algorithm reduces hardware size by eliminating intermediate buffer. Then, we unify clipping stage with perspective division and viewport mapping stage of 3D graphics pipeline (Clipping Engine). This unified architecture can reduce data-path’s size and eliminate the pipeline buffers. This architecture can reduce hardware size for embedded system. But because of reduction of data-paths for perspective division and viewport mapping operation, it may drop performance of clipping, perspective division and viewport mapping stages. To supply for this performance loss, we propose reuse-method to remove redundant operations of perspective division and viewport mapping. This can improve performance of the clipping engine b...