OPTIMIZATION OF MAPPING ONTO A FLEXIBLE LOW-POWERELECTRONIC FABRIC ARCHITECTURE

A combinatorial problem that arises from a novel electronic fabric architecture designed forlow-power devices such as cellular phones and palm computers is presented. We consider theproblem of efficiently mapping a given data flow graph onto a particular implementation ofthe fabric architecture. We formulate mixed integer linear programs (MILP) and design asliding partial MILP heuristic for this problem. We highlight the modeling and algorithmicaspects that are necessary to make the MILP formulation competitive. The sliding partialMILP heuristic is developed to generate mappings faster and to find mappings for benchmarkinstances that cannot be solved by the MILP formulation.We also present a method to tune software parameters using ideas from software testingand machine learning. The method is based on the key observation that for many classes ofinstances, the software shows improved performance if a few critical parameters have good values, although which parameters are critical depends on the class of instances. Our methodattempts to find good parameter values using a relatively small number of optimization trials