A rule-based approach for improving allocation of filter structures in HLS

A rule based allocator for improving synthesis of filter systems is presented. The principles of the Enhanced AIlocation Rule Language Interpreter (EARLI) are presented. Possible transformations, optimisations and how to express them in EARLI are discussed. Experiments show that relative area gains ranging from 5 to 44%, depending on the chosen target technology, can be achieved using the designers knowledge about the design class. Experiments also indicate that employing direct mapping of CDFG subgraphs onto preoptimised RTL-level macroblocks would have resulted in a relative area gain of 500%. The macroblock had only 16% of the area produced by the HLS-tool