A memory allocation technique for low-energy embedded DSP software

Embedded DSP applications are characterized by the presence of a dedicated DSP processor and application specific software that runs on it. Energy consumption is increasingly becoming a critical design constraint for such applications. This necessitates the analysis and optimization of the energy consumption of the software component of these systems. This paper investigates an energy optimization technique in this regard. It involves the problem of allocating memory to variables in embedded DSP software to maximize simultaneous data transfers from different memory banks to registers. A memory allocation technique based on simulated annealing is proposed. Experimental results for example DSP programs demonstrate that energy reductions of upto 47% can be achieved by this approach.