适用于嵌入式应用的多线程顶点处理器设计  

Design of Multithreaded Vertex Processor for Embedded Applications

作　　者：申智波[1] 陈杰[1] 刘建[1] 王云[1] 朱玥[1,2] 

机构地区：[1]中国科学院微电子研究所,北京100029 [2]中国科学技术大学,安徽合肥230026

出　　处：《微电子学与计算机》2012年第7期76-80,共5页Microelectronics & Computer

基　　金：国家"八六三"计划项目(2009AA011700)

摘　　要：针对嵌入式应用中三维图形渲染的要求,设计了一款可编程的多线程顶点处理器.该顶点处理器采用单指令多数据结构,一条指令能够同时处理4个单精度浮点数,并采用多线程技术,支持4个线程并发执行,能够有效地减少发生数据写读冲突时的停顿周期数,提高了处理效率.相对于单线程结构,4线程顶点处理器在较小的硬件开销下,可以实现2.1～2.8倍的性能提升.该顶点处理器支持OpenGL ES 1.1和Vertex Shader Model 1.1,在90nm CMOS工艺库下可实现频率为200MHz,性能为50Mvertices/s.A programmable multithreaded Vertex Processor is designed for 3D graphics rendering in embedded applications.The Single Instruction Multiple Data（SIMD） architecture is proposed,so four single precision floating-point numbers are processed simultaneously.The multithreaded datapath supports parallel execution of four threads,so the pipeline stalls are reduced when data hazards occur and the processing efficiency rises up.The proposed four-threaded Vertex Processor achieves 2.1～2.8 times of performance improvement at a relatively small hardware cost compared to the single-threaded architecture.The Vertex Processor supports OpenGL ES 1.1 and Vertex Shader Model 1.1.Implemented in 90nm CMOS process,it achieves 50 Mvertices/s of geometry performance at the operating frequency of 200MHz.

关 键 词：图形处理器 三维图形渲染 可编程顶点处理器 多线程 单指令多数据 嵌入式应用 

分 类 号：TN391.41[电子电信—物理电子学]