Exploiting parallelism of 3D graphics geometry using a VLIW geometry processor

Abstract

We use 4-way VLIW architecture called FLOVA (FLOating-point VLIW Architecture) as a geometry accelerator. And we propose three schemes to exploit the parallelism of the geometry accelerator using merits of the VLIW architecture and inherent parallelism in the 3D graphics geometry stage. In the first scheme, we find out ILP (Instruction Level Parallelism) on matrix multiplications in transformation. The second scheme, we are looking for the portions to be executed simultaneously in each pipeline stage of geometry, and it is called ALP (Algorithm Level Parallelism). The third scheme is communication overhead reduction by reformatting the data structure of the FGA``s context and vertex buffer. Therefore, the geometry accelerator using VLIW can improve a half and two times of the total graphics performance, and exploiting parallelism by ILP and ALP is cost effective solution with in the VLIW geometry accelerator.