Simultaneous exploration of optimal datapath and loop based high level transformation during area-delay tradeoff in architectural synthesis using swarm intelligence

Abstract

This paper presents the first transformation methodology of a swarm intelligence algorithm into design space exploration (DSE) of optimal datapath and unrolling factor during area-performance tradeoff in high level synthesis (HLS). To the best of the authors' belief, in the literature, the above transformation has not been performed so far for developing a fully automated integrated exploration methodology of optimal datapath and unrolling factor. DSE problem being intractable and complex in the presence of auxiliary variable such as loop unrolling factor requires administration of intelligent decision making strategies at multiple stages to yield optimal results. The major sub-contributions of this proposed algorithm includes: a) Deriving a model for execution delay computation of a loop unrolled control data flow graph (CDFG) based on available resources, b) handling area-execution delay tradeoff during our integrated DSE process; c) Sensitivity analysis of particle swarm optimization (PSO) parameters used in assisting the designer in pre-tuning the baseline parameters during final area-performance tradeoff. Results of the proposed approach indicated an average improvement in Quality of Results (QoR) of > 23% and reduction in runtime of > 92% compared to recent approaches.