Fast Memory Size Estimation of Application Programs for System-on-Chip Signal-to-Memory Mapping

System-on-chip (SoC) is comprehensively applied in the field of multimedia information processing. The application programs of multimedia information processing are abundant in nested loops and multi-dimensional signals, which greatly affects the efficiency of data transfer and storage. Therefore, the SoC signal-to-memory mapping, which pays attention to memory size of application programs and optimizes the efficiency of memory system, tends to obtain better SoC performance and declining power consumption. Concentrating on nested loops and multi-dimension signals in multimedia processing programs, a fast memory size estimation approach is presented for SoC signal-to-memory mapping. On the basis of polyhedral model and linear bounded lattice, orthogonal linear bounded lattices are put forward to partition the data domain of the multi-dimension signals, and then the memory requirement size is computed according to the data dependency of orthogonal linear bounded lattices. Orthogonal linear bounded lattices are the minimum process unit during dependency analysis and memory size computation, which can greatly reduce the analysis time and keep the estimation accuracy. The memory requirement size is taken as heuristic information and mapping criterion by the SoC signal-to-memory mapping algorithm, which helps to explore the signal-to-memory mapping space for the purpose of efficient data transfer and storage.