一种面向DSP深度压缩指令的数据竞争检测方法  

Data hazard checking method for heavily compressed DSP instruction

作　　者：余巧艳[1] 刘鹏[1] 

机构地区：[1]浙江大学信息与电子工程学系,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2005年第10期1501-1506,共6页Journal of Zhejiang University：Engineering Science

基　　金：浙江省重点科技计划资助项目(021101559);霍英东青年教师奖优选课题资助项目(94031)

摘　　要：针对数字信号处理器(DSP)深度压缩的指令结构不归整性,难以从指令二进制码中快速提取相关信息的问题,提出分类检测DSP指令间数据竞争的方法.利用不同功能的指令涉及不同寄存器的特点,在分层译码电路和旁路电路辅助下,及时检测出数据竞争,产生流水线停顿信号,准确定位出产生数据竞争的指令类型.综合结果证明在分层译码配合下,分类检测方法比指令全译码后作检测,速度提高了18.89%.DSP常用程序证明这种方法能有效地覆盖所有出现的数据竞争,保证基于流水的DSP的正确计算.分类检测数据竞争法比指令中源操作数的比特域与目的操作数的比特域作比较的检测方法更适合于采用深度压缩指令的DSP,物理实现更简单.Heavily compressed instruction set brought extra difficulties in digital signal processor （DSP） data hazard checking. Since different type instructions used special registers respectively, which were certain as the instruction set was defined, the class-based detection method created detection mode and discrimination matrix to check out all kinds of data hazards. With the assistance of hierarchical decoder and bypass circuits, this class-based detection method worked well in the general application programs of DSP. Furthermore, the synthesis results showed that the critical path of the circuit based on this method was shortened by 18.89%. Class-based detection method compressed instruction set to check out whether write orders of registers. is more reasonable and simpler for the DSP using heavily data hazards would arise than only comparing the read and

关 键 词：数字信号处理器 数据竞争 深度压缩指令 分层译码 旁路电路 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]