Feds Plot Near Human Robot Docs, Farmers, Troops
Maybe robot is the next big thing after five decades of IT revolution?
via Feds Plot Near Human Robot Docs, Farmers, Troops | News | Communications of the ACM (summary).
We presented a first approach to optimizing software for the memory architecture of the target reconfigurable computing system, in HiPEAC 2010, held in Pisa, Italy. The motivation of this work is that often in multimedia applications the performance bottleneck is in data transfer, not in computation per se. While previously we have tried only to maximize computation rate only, it may be better to sacrifice computation rate a little to increase data transfer rate if there is a trade-off between computation and data transfer as in the case of CGRA mapping. We targeted our compiler for a ADRES-like architecture, with a slightly simplified local memory subsystem: double-buffered, multi-banked, and the banks of the local memory are one-to-one mapped to the load-store units of the reconfigurable architecture.
Coarse-Grained Reconfigurable Arrays (CGRAs) are a very promising platform, providing both, up to 10-100 MOps/mW of power efficiency and are software programmable. However, this cardinal promise of CGRAs critically hinges on the effectiveness of application mapping onto CGRA platforms. While previous solutions have greatly improved the computation speed, they have largely ignored the impact of the local memory architecture on the achievable power and performance. This paper motivates the need for memory-aware application mapping for CGRAs, and proposes an effective solution for application mapping that considers the effects of various memory architecture parameters including the number of banks, local memory size, and the communication bandwidth between the local memory and the external main memory. Our proposed solution achieves 62% reduction in the energy-delay product, which factors into about 47% and 28% reduction in the energy consumption and runtime, respectively, as compared to memory-unaware mapping for realistic local memory architectures. We also show that our scheme scales across a range of applications, and memory parameters.
Read the full paper: “Memory-Aware Application Mapping on Coarse-Grained Reconfigurable Arrays,” Lecture Notes in Computer Science (HiPEAC ’10), Vol. 5952, pp. 171-185, by Yongjoo Kim, Jongeun Lee, Aviral Shrivastava, Jonghee W. Yoon and Yunheung Paek, 2010.