多项式数据通路的高层次综合方法

The Method of High Level Synthesis for Polynomial Datapaths

为了实现多项式数据通路的高层次综合,采用有序、简化和正则的带权值广义表模型表达该多项式.首先对该带权值广义表进行线性化处理;然后提出了基于带权值广义表的多项式数据通路的高层次优化方法.该方法以自底向上的方式遍历带权值广义表中的节点,并迭代地析取该带权值广义表中的乘法和加法项,进而将该带权值广义表转换为不可简化的有层次的带权值广义表集合,最终将该集合转换为更适于高层次综合的可调度数据流图.实验结果表明,与传统的方法相比,采用该方法得到的数据流图转化为寄存器传输级结构具有更小的延迟和面积.

In order to implement high level synthesis for polynomial datapaths, the ordered, reduced and canonical weighted generalized list was used to represent for the polynomials. The weighted generalized list was linearized firstly. And based on the weighted generalized list, a high level optimization method for polynomial datapaths was given, which traverses the nodes of the weighted generalized list in a bottom-up fashion and extracted the product terms and sum terms from the weighted generalized list iteratively, and then transformes weighted generalized list into a set of irreducible hierarchical weighted generalized lists, and finally transformes the set of irreducible hierarchical weighted generalized lists into corresponding schedulable data flow graphs which are better suited for high level synthesis. Experiments show that the register transfer level structure which obtained from the sehedulable data flow graphs generated by the pro- posed method had smaller latency and less datapth area than those obtained using traditional methods.