A novel approach to performance-oriented datapath allocation and floorplanning

In recent years, integrating high-level synthesis and physical design has become essential for successful timing closure. In this paper, we present a novel probabilistic approach to perform integrated datapath allocation and floorplanning. In contrast to existing methodologies that tradeoff high-level design space searched with accuracy of physical estimates, the proposed methodology performs exhaustive high-level synthesis (datapath allocation) with heuristic floorplanning. Exhaustive search guarantees optimal allocation results. Heuristic floorplanning imparts flexibility to update the floorplan while constructing the allocation solution. This improves accuracy of the floorplan estimates used. To tackle the increased design space, we propose novel probabilistic-gain measures that predict the impact of an allocation decision on the constraints, prior to the allocation decision being made. The probabilistic-gain measures are used to select allocation decisions that will most likely lead to optimal solutions early, thereby minimizing time-consuming floorplan evaluations of decisions that will most likely lead to sub-optimal solutions. Experimental results confirm the effectiveness of the proposed approach