Configuration Sharing to Reduce Reconfiguration Overhead Using Static Partial Reconfiguration

Abstract

SUMMARY Reconfigurable architectures are one of the most promising solutions satisfying both performance and flexibility. However, reconfiguration overhead in those architectures makes them inappropriate for repetitive reconfigurations. In this paper, we introduce a configuration sharing technique to reduce reconfiguration overhead between similar applications using static partial reconfiguration. Compared to the traditional resource sharing that configures multiple temporal partitions simultaneously and employs a time-multiplexing technique, the proposed configuration sharing reconfigures a device incrementally as an application changes and requires a backend adaptation to reuse configurations between applications. Adopting a data-flow intermediate representation, our compiler framework extends a min-cut placer and a negotiation-based router to deal with the configuration sharing. The results report that the framework could reduce 20% of configuration time at the expense of 1.9% of computation time on average.

An Operation and Interconnection Sharing Algorithm for Reconfiguration Overhead Reduction Using Static Partial Reconfiguration

“…This means a synthesis process including placement and routing has to be enhanced to deal with resource sharing. Our recent research introduces configuration sharing that is the resource sharing problem in a configuration level [12]. In the work, we extend a traditional min-cut placer with a method to move and swap nodes between spatial partitions by which nets cut is reduced and spatial resource constraints throughout temporal partitions are satisfied.…”