AMAP: A New Heuristic Communication-Aware Tasks Mapping onto 2D Mesh NoCs
Hesamedin Ziaeeziabari, Ahmad Patooghy, Midia Reshadi · 2016
This paper proposes a new heuristic communication-aware tasks mapping onto 2D mesh NoCs. The proposed mapping algorithm, called AMAP, reduces the hop count of packet delivery between tasks of an application. To do this, the AMAP algorithm divides communications of a given application graph into low and high-volume communications. AMAP algorithm bypasses the low-volume communications to map tasks with high-volume communication. AMAP algorithm fairly assigns 4-neighbor nodes of mesh topology between the most important tasks of application graph. Evaluation of the AMAP algorithm is done based on a widely used hop count parameter. Hop count parameter of AMAP algorithm is calculated and compared with those of four other mapping algorithms i.e., NMAP, DMAP, BMAP, and ONYX algorithms. Results show that AMAP algorithm offers at least 6% and at most 58% improvements with respect to four other mentioned algorithms. Evaluations reveal that improvements of AMAP algorithm grows when the number of unused cores of the chip increases. This means that the proposed algorithm offers better mapping when the mapping situation is more complicated.