Ultra-Fast CGRA Scheduling to Enable Run Time, Programmable CGRAs
TimeThursday, December 9th4:30pm - 5:00pm PST
SoC, Heterogeneous, and Reconfigurable Architectures
DescriptionCoarse-Grained Reconfigurable Arrays (CGRAs) offer the potential for both energy-efficiency and high-throughput in embedded systems today. One limitation of CGRAs is the extremely long mapping time that can take many hours to complete. This, coupled with static programming, limits the number of kernels that can be accelerated.
In this work, we propose a solution that aims to improve performance through a new low-complexity CGRA mapping algorithm that still achieves comparable performance through key instruction placement restrictions. This results in mapping times in the milliseconds; run time programming, and therefore higher overall performance and efficiency, is now within reach for CGRAs.