Lookahead Memory Prefetching for CGRAs Using Partial Loop Unrolling

摘要

Coarse Grained Reconfigurable Arrays have become an established approach to provide high computational performance in various environments. Several researchers have found that the achievable performance highly depends on the interface between memory and CGRA. In this contribution we show that a smart prefetching mechanism can increase the performance of the CGRA. At the same time it consumes less hardware resources and energy as state of the art prefetching mechanisms.