一种高效纹理映射单元的硬件体系结构设计  

An Efficient Hardware Architecture Design of Texture Mapping Unit

作　　者：赵国宇[1] 郭炜[2] 常轶松[2] 魏继增[2] 

机构地区：[1]天津大学电子信息工程学院,天津300072 [2]天津大学计算机科学与技术学院,天津300072

出　　处：《计算机工程》2013年第5期92-95,105,共5页Computer Engineering

基　　金：国家自然科学基金资助项目(61070136);教育部博士点专项科研基金资助项目(20110032120037;20100032110041)

摘　　要：针对嵌入式纹理映射过程中处理速度慢和存储带宽对系统性能制约等问题,提出一种专用纹理映射单元体系结构,对单精度浮点除法器和纹理Cache进行优化设计。采用较小查找表结合二次多项式逼近算法实现浮点除法运算,根据纹理采样的不同方式,对纹理Cache的映射方式在直接映射、两路组关联和四路组关联之间进行动态配置。对纹理映射单元在SMIC0.13μm CMOS工艺下进行仿真,结果表明,纹理映射模块工作主频可达313 MHz,对于480×640像素,帧率可达51 f/s,能够满足移动设备对实时渲染的需求。This paper presents a dedicated architecture of texture mapping unit with an optimized design of single-precision floating-point divider and texture Cache, against the processing speed and memory bandwidth problem. It uses a small look-up table with quadratic polynomial approximation algorithm to achieve floating-point division. According to different sampling methods, it configures Cache mapping strategy among direct mapping, two-way set associative mapping and four-way set associative mapping. The design of the texture mapping unit is synthesized under SMIC0.13 pm CMOS process technology. Experimental results show that the design can achieve frame rate up to 51 f/s for 480x640 resolution at frequency up to 313 MHz, it is efficient for real time rendering for mobile devices.

关 键 词：纹理映射 透视投影 纹理Cache 纹理采样 单精度浮点除法器 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]