Improving Energy Efficiency of Mobile Execution Exploiting Similarity of Application Control Flow

Abstract

In this work, we propose an energy efficient scheme for application ordering and execution on modern smartphone processors. We propose to improve the branch prediction piece present inside pipelined processors, by suitable clustering and scheduling of applications that exhibit similar control flow. We expect that these applications can benefit by sharing predictor table data structures, that can do away with table initializations and predictions every time an application context switch is encountered. Preliminary experiments show promising results, and we believe this proposal will open up several interesting avenues of research.