PSDSE: Particle Swarm Driven Design Space Exploration of Architecture and Unrolling Factors for Nested Loops in High Level Synthesis

Abstract

A novel methodology for automated simultaneous design space exploration (DSE) of architecture and unrolling factors (UFs) for nested and single loops based application in high level synthesis (HLS) using particle swarm algorithm (named as 'PSDSE') is presented in this paper. The major contributions of the proposed methodology are as follows: (a) automated exploration of architecture and UFs using particle swarm intelligence that parallely maintain trade off between conflicting metrics of power -- performance and balance orthogonal issues of improving Quality of Result (Or) and reducing the exploration runtime for nested loops, (b) deriving a model which directly estimate the execution time of nested loop based on resource constraint and UFs without necessity of tediously unrolling the entire control and data flow graph (CDFG) for the specified UFs values in most cases. Results indicated an average improvement in QoR of >49 % and reduction in runtime of >97% compared to recent approaches.