Energy-aware Allocation of Dynamic Variables in Partitioned Memory Architectures

This paper addresses the problem of minimizing the energy consumption by the memory subsystem in an embedded system. While significant work in the literature has addressed compiler driven energy aware allocation of global static variables, no general solutions have been proposed for the corresponding problem of allocating run-time dynamic variables. This paper proposes an approach, and the corresponding framework, to compiler driven energy aware allocation of dynamic variables. The approach is driven by the use of execution profile information and the concept of heap segmentation. Our framework and solution have been integrated into a compiler backend tool. Experimental results with some benchmark applications indicate that our techniques result in up to 40 % memory energy reduction and 20-25 % overall system energy reduction