Evolutionary algorithm based exploration of software schedules for digital signal processors
Eckart Zitzler, Jürgen Teich, Shuvra S. Bhattacharyya · 1999
The simultaneous exploration of tradeoffs between program memory, data memory and execution time requirements (3D) for DSP (digital signal processing) algorithms in embedded computing environments is a demanding application and example par excellence of a multi-objective optimization problem. In order to solve this problem, two evolutionary algorithms are shown to be successfully applicable for exploring Pareto-optimal solutions. For different well-known target DSP processors, the trade-off fronts are analyzed. The two approaches are quantitatively compared. 1 Introduction Starting with a data flow graph specification to be implemented on a digital signal processor, we study the effects between instantiating code by inlining or subroutine calls as well as the effect of loop nesting and context switching on a target processor (DSP) that is used as a component in a memory and cost-critical environment, e.g., a single-chip solution. For such applications, a careful exploration of the ...