A Novel Low Energy Scheduling Algorithm for Clustered Very Long Instruction Word Architectures

This work presents a novel approach for improving the energy consumption for instruction scheduling of clustered VLIW architectures. The proposed scheme is based on a preliminary cluster assignment phase implemented through partitioning and analyzing of Data Dependence Graph (DDG) and a scheduling phase using different heuristics to find an optimized schedule. The preliminary cluster assign scheme is shown to be very effective due to its global view of the whole DDG. We have implemented and evaluated the proposed scheme with UTDSP benchmarks. Results show a significant reduction in energy consumption when compared with previously proposed techniques, while also maintaining a high performance level. Energy consumption reductions are up to 41%, with average energy consumption improving ranging from 26% (2-Cluster) to 30% (4-Cluster). While performance speedup can up to 33% compared with result of list scheduling, with average speedup ranging from 25% (2-Cluster) to 30% (4-Cluster).