GSink - A Runtime for Gamma Programs and its CGRA Mapping Proposal

Gamma programs export a naturally concurrent execution models akin to chemical reactions. To realize the potential for very efficient execution we have develop an execution scheduling algorithm that we believe is suitable for the mapping to Coarse-Grain Reconfigurable Arrays (CGRAs) as one of its key operations includes the pervasive use of low-level pattern matching. Our proposal addresses key issues of previous Gamma implementations and provides the coexistence of several reactions instances in parallel, using a mechanism based on scheduling by edge reversal. In this paper, we highlight the scheduling algorithm and its mapping proposal for CGRAs, given the potential for GSink performance as the granularity of reactions increases.