Transformer: Run-time reprogrammable heterogeneous architecture for transparent acceleration of dynamic workloads

摘要: Abstract Heterogeneous architectures face challenges regarding transparent acceleration as well the allocation of resources to cores and accelerators. The “Transformer”, a run-time reprogrammable, heterogeneous architecture consisting reconfigurable logic with support for coarse-grained dynamic, unpredictable workloads present in mobile cloud computing environments, is proposed solution. allows instantiation one or more functions, an on-chip logic, which responds demands compute-intensive software libraries. hardware controller wrapper functions are designed profile workloads, reprogram internal invoke appropriate functions. Novel heuristics derived respect accelerator function scheduling. In order optimize performance power efficiency, system parameters explored, including L1 L2 cache sizes, local buffer simulation results indicate that Transformer provides significant improvements terms performance, up 14 × single-type 2.3 dynamic energy 6.9 various workloads.