Abstract:The characteristics of 2-D convolutions with big kernel in low-rank are analysed, and a fast algorithm is given. Then a systolic array implementation, which is derived by a three-stage dependence-graph-based mapping approach, is presented. It is shown that the architecture has a high efficiency for parallel processing and a nearly linear speed-up.
杨绿溪; 王保云; 何振亚. 低秩大模板二维卷积算法的脉动阵列设计[J]. 电子与信息学报, 1997, 19(1): 6-10 .
Yang Luxi; Wang Baoyun; He Zhenya. SYSTOLIC ARRAY DESIGN FOR 2-D CONVOLUTIONS WITH BIG KERNEL IN LOW-RANK. , 1997, 19(1): 6-10 .