Research On Real-Time Performance Based On ScratchpPad-Memory

As the rapidly development of the on-chip system's designing level, the embedded devices are becoming smaller and more powerful. Especially in real-time system, it's desirable that the execution time of the task is estimated accurately, which makes the embedded system need to be optimized. In order to improve the performance of the system, one frequently-used method is to use Cache, Cache is managed by hardware automatically, it's hard to know whether hit when visiting Cache, which increases the difficulty of the time prediction of the program and cause the WCET can't be estimated accurately. As a optional memory for modifying the real-time performance, Scratch-Pad Memory (SPM) is a on-chip SRAM,which has the superiorities of small size,low energy,fast visition,real-time performance guarantees and directly addressed with off-chip memory.This thesis focuses on studying the SPM management method for improving the system's real-time performance.Ccurrent methods that searching for the program's worst-case execution path often lead to low efficiency. Aiming at this problem,this thesis puts forward a path-based method with conflict detection for searching the program's worst-case execution path. This method not only avoids the high cost of path enmueartion, but also avoids backtracking source node of CFG to find alternative path after excluding impossible paths, they adjust the CFG and update the path when checking conflicts, which increases the speed of the searching. Current SPM management methods for improving the system's real-time performance consider only the program's code, or the program's data, also don't consider the relationship of the nodes in the CFG. Aiming at this problem, this thesis puts forward a static SPM management method that considers both code and data of the program and is based on time-density. Firstly,this method obtains the worst-case execution path through the analysis of WCET. Secondly, according to the worst-case execution path, this method transforms the application into a directed graph consisting of nodes and relationships of nodes.Lastly, it puts the node with the max time-density into SPM. Experiments show that this algorithm reports an improvement of about 42% in the real-time performance, compared to algorithm without considering the relationship of the nodes.