RPUSM: An Effective Instruction Scheduling Method for Nested Loops

Abstract

Multi-dimensional systems are widely used to model scientific applications such as image processing, geophysical signal processing and fluid dynamics. Such systems usually contain repetitive groups of operations represented by nested loops. The optimization of such loops, considering processing resource constraints, is required in order to improve their computational time. Push_Up Scheduling Method (PUSM) is an effective technique that can get the shortest schedule table and fully utilize resources. It contains two main parts: schedule table construction and retiming base selection. In our analysis, PUSM can obtain optimal solution in the former part, but lacks a complete selecting process and uses more constrained conditions in the later part. Hence, in this paper, we propose a Relax Push_Up Scheduling Method (RPUSM) to overcome its shortcomings. In RPUSM, we inherit the former part from PUSM, and replace the later part by a complete and more relax conditions to select more appropriate retiming base. According to our analysis, RPUSM can not only get better performance than PUSM in some cases, but also preserve the advantages of PUSM, such as shortest schedule table, fully resource utilization, and polynomial scheduling time.