Accelerating floating-point 3D graphics for vector microprocessors

We present performance results and discuss optimization techniques for floating-point 3D graphics primitives on the ARM vector floating-point (VFP) coprocessor. Much of the work in 3D graphics is performed by a surprisingly small number of primitives. Handheld devices cannot always afford the power and cost of dedicated graphics processors, so there is considerable interest in performing these primitives on the handheld's own processor. In this paper, we examine the 3D graphics pipeline, break out the important primitives, and then show how to vectorize and optimize these primitives on the VFP.