Creating Mapping and Schedule

Given a set of ML models and a hardware description, an application graph and hardware graph can be compiled. The next step is to feed these into the solver which then either uses an evolutionary algorithm (EA) to heuristically search for a highly optimal mapping solution or to use an integer linear program (ILP) to solve for the optimal mapping solution. Both of these solutions are then fed into an ILP formulation to solve for a schedule given the determined mapping. The figure below shows this flow.

These two approaches are outlined in the two pages ILP Mappings and Schedule and Heuristic Mapping and ILP Schedule.