Dynamic Template Generation for Resource Sharing in Control and Data Flow Graphs

David Zaretsky, Gaurav Mittal, Robert P. Dick, Prith Banerjee

Abstract: High-level synthesis compilers often produce reoccurring patterns in intermediate CDFGs during translation. By identifying large reoccurring patterns, one may reduce area and communication overhead by efficiently reusing hardware for multiple operations. This paper presents an algorithm for dynamically generating templates of reoccurring patterns for resource sharing in CDFGs. Results show 40-80% resource reduction using small, incremental template growth, and variations within a 5% margin among varying look-ahead depths.