Area and performance optimizations in path-based scheduling

Abstract

This paper describes the area and performance optimizations implemented in the As-Fast-As-Possible (AFAP) scheduling algorithm. The AFAP scheduling algorithm is a path-based technique that finds the minimum number of control steps for all possible sequences of operations in the control-flow graph, under given constraints. Area requirements for functional units, such as their number and type, are translated into constraints which are then met exactly. The number of registers is also minimized. The performance optimizations included in this paper are concerned mainly with the scheduling of loops.