OPTIMAL DATA SCHEDULING FOR UNIFORM MULTIDIMENSIONAL APPLICATIONS

Uniform nested loops are broadly used in scientific and multidimension al digital signal processing applications. Due to the amount of data h andled by such applications, on-chip memory is required to improve the data access and overall system performance. In this study a static da ta scheduling method, carrot-hole data scheduling, is proposed for mul tidimensional applications, in order to control the data traffic betwe en different levels of memory. Based on this data schedule, optimal pa rtitioning and scheduling are selected. Experiments show that by using this technique, on-chip memory misses are significantly reduced as co mpared to results obtained from traditional methods. The carrot-hole d ata scheduling method is proven to obtain smallest on-chip memory miss es compared with other linear scheduling and partitioning schemes.