Improving Energy Efficiency of Coarse-Grained Reconfigurable Arrays through Modulo Schedule Compression/Decompression

Download paper here

In this work, we improved compressibility from our work, “A Space- and Energy- Efficient Code Compression/Decompression Technique for Coarse-Grained Reconfigurable Architecture” proposed in CGO17, by leveraging edit distance-based and bin-packing-based heuristic algorithms. First, the edit distance denotes the number of editing operations required when transforming one string into another. We calculate edit distance of the signal change patterns between hardware entities and group the entities that require a small edit distance while partitioning the configuration memory. Second, the bin-packing-based memory partitioning first creates the desired number of partitions, temporarily inserts hardware entities one by one, calculates the compressible memory size, and if it is higher than a threshold, permanentely inserts the entity into the partition. These memory partitioning policies reduce 63% of the configuration memory size and improve energy consumption by about 70% in a wide variety of application domains.