Efficient datapath merging for the overhead reduction of run-time reconfigurable systems

Abstract

High latencies in FPGA reconfiguration are known as a major overhead in run-time reconfigurable systems. This overhead can be reduced by merging multiple data flow graphs representing different kernels of the original program into a single (merged) datapath that will be configured less often compared to the separate datapaths scenario. However, the additional hardware introduced by this technique increases the kernels execution time. In this paper, we present a novel datapath merging technique that reduces both the configuration and execution times of kernels mapped on the reconfigurable fabric. Experimental results show up to 13% reduction in the configuration and execution times of kernels from media-bench workloads, compared to previous art on datapath merging. When compared to conventional high-level synthesis algorithms, our proposal reduces kernels configuration and execution times by up to 48%.

High speed merged-datapath design for run-time reconfigurable systems

“…We have implemented the proposed merging algorithms and the datapath merging algorithm in [12], as well. The algorithm in [12] is based on the maximum weighted clique, while our algorithms use the bounded-execution timeweighted clique.…”

“…So, the node a 1 ,b 1 is created in G c for this merge that its weight indicates to the 360 ns reduction in configuration time resulted from merging the vertices. This weight is the difference between the configuration time of the obtained vertex in MDP and configuration time of the original vertices before merging [12]. In this way, the compatibility graph "G c " for two DFGs G 1 and G 2 is created by using the method was explained above.…”

“…By finding the maximum weighted clique from the compatibility graph in this stage, the merged datapath which has minimal configuration time is made [12]. Fig.…”

“…Datapath merging is an efficient HLS approach that makes a multimode datapath for partially reconfigurable systems [10,11]. We showed in [12] that datapath merging is a suitable method to reduce the datapath configuration time. The method in [12] heuristically chooses a sequence of DFGs to merge together, consequently, it cannot optimize the configuration time of the merged datapath.…”

“…We showed in [12] that datapath merging is a suitable method to reduce the datapath configuration time. The method in [12] heuristically chooses a sequence of DFGs to merge together, consequently, it cannot optimize the configuration time of the merged datapath. On the other hand, datapath merging algorithms add multiplexers in the input port of the functional units in the merged datapath, and as such, increase the execution time of the loops via the merged datapath.…”