Inter in-place storage size requirement estimation

Data storage is an important contributor to power dis-sipation, particularly in multi-media embedded systems. To achieve low power implementation for these systems, the storage requirement can be reduced by exploiting lim-ited data life-times and reuse of memory locations. This optimisation can only be performed quite late in the de-sign trajectory because most of the decisions on e.g. detailed memory access organisation should have been fixed. Thus an early and fast estimate of the storage requirement is crucial for the system designer. This pa-per proposes a grouping technique that identifies simul-taneously alive sets of data dependencies. The group-ing technique extends previous work on storage size re-quirement estimation for individual data dependencies, allowing the designer to obtain the global storage size requirement for an application. A prototype CAD tool has been developed and used to test the technique on real-life multi-media kernels.