Energy optimization of branch-aware data variable allocation on hybrid SRAM+NVM SPM for CPS
Yixin Li, Jinyu Zhan, Wei Jiang, Jiayu Yu · 2019
With good performance, non-volatile memory (NVM) is being used increasingly in the design of cache or scratchpad memory (SPM) for cyber-physical systems (CPS). This paper presents a branch-aware data variable allocation (BADVA) approach based on hybrid SRAM+NVM SPM for low energy consumption, which consists of branch-based analysis and data variable allocation. The branch-based analysis assigns the corresponding branch prediction for conditional branches of a program. Since the branch prediction may change the Worst-Case Execution Path (WCEP), branch-based analysis is iterated for several times until WCEP is stable. After branch-based analysis, the energy consumption can be calculated according to memory access, by which the data variable allocation determines how to migrate data variables. Based on the existing benchmarks, we conduct experiments to evaluate the proposed approach. Compared with the other algorithms, the maximum and average energy consumption improvement of our approach are 39.4% and 25.1%, respectively.