流水安全法─—一个面向软件流水技术的新的数据相关性分析方法

PIPELINING SAFE METHOD──A NEW WAY TO SUPPORT DATA DEPENDENCE ANALYSIS FOR SOFTWARE PIPELINING

软件流水是一种很有效的指令级并行优化技术，而能否进行尽可能精确的数据相关性分析是决定软件流水优化效果的一个非常重要的因素．本文通过分析软件流水技术本身的特点，从保障软件流水安全为出发点，导出了一组更严洛有效的相关方程和限制不等式，大大提高了相关性判别的能力，最后与现有工作进行了比较，并用一个例子加以验证．

Software pipelining is a very effective instruction level parallel compiling technique. It'squite important whether the data dependence analysis is capable of proving an accurate judgment onthe dependence relations among instructions of a user program. A more accurate dependenceanalysis will allow the software pipelining algorithm give a more effective parallel optimization of a givenuser program. By taking advantages of the characteristics of software pipelining technique, weobtain an array of new constraint inequalities for the dependence equation, which can greatly improvethe analyzing ability of most conventional dependence analysis algorithms. This method is only forthe dependence analysis of software pipelining.