Research On Performance Optimization Of SPM In Mpsoc

Embedded Systems has got great development these forty years. Multi-Processor Sys-tem on Chip (MPSoC) enables the embedded systems to provide more powerful features, but the gap between processor's processing power and storage systems is also growing, storage system has become a bottleneck affecting the overall performance.Through the use of appropriate task mapping algorithm to improve the parallel degree and the use of high-speed on-chip memory to enhance performance of the system are the focus of the current study. Scratch Pad Memory (SPM) is a non-cached SRAM memory integrated in micro Processor chip. because of its low power consumption, better real-time guarantees and requires less area, there are many MPSoC already have SPM integrated in them.The current research on SPM can be divided into the static and dynamic allocation strategy; the static ones,in the run time,the content of SPM will not change; but dynamic ones adjust the content continuously. Almost all these methods are for single-processor architecture. The research for multiprocessor is relatively small, the characteristics of its tasks in parallel make the management of SPM difficult.For MPSoC, this paper proposes a combination of static and dynamic SPM allocation strategy. Firstly using of ILP and an integrated task scheduling, SPM partitioning and data allocation strategy allocate global variables to SPM, because local variables can only be accessed by the corresponding processor alone, then using dynamic strategy based on greedy algorithm, the appropriate local variables are assigned to the SPM, so, the frequency of access to external memory are reduced and system performance is improved highly.The experimental data on the benchmark MiBench show that the combination of dy-namic and static allocation strategy can improve program efficiency.Our research will promote the development of SPM in the embedded system.