Parallelism optimization with variables affinity graph for reconfigurable computing architecture

In order to parallelism, generally, compiler considers the nodes which they want them to work in parallel. Thus, the communication volume (CV) between variables is ignored. In this paper, we will make a variable affinity graph (VAG) which could imply CV. An approach which could cut VAG into optimization number pieces will be carried out. Using this approach, both CV and resource balance (RB) are considered and an optimization result could be computed. Our approach has been evaluated using a reconfigurable multi-core of four-tightly coupled processors, which can be simulated using a synthesized model or a cycle-accurate software simulator. Our results lead to an average 4-fold speed increase comparing to original systems, and an average 46% speed increase comparing DFG based optimal algorithm.