Research On Hardware-software Partitioning Method For Reconfigurable Embedded System

With the technology improvement of reconfiguration and the emergence ofFPGAs, there is a tendency of using dynamically reconfigurable system in embeddedsystem. The core progress of reconfigurable computing technology is in the domainof configuring hardware circuit dynamically, the advantage of which is flexible andefficient and can structure and configure hardware circuit. At present, reconfigurableembedded system with fast computation speed of ASIC and flexibility ofmicroprocessor has been an efficient computing platform. Hardware-softwarepartitioning methods of traditional embedded system are not fit for dynamicallyreconfigurable embedded system. In dynamically reconfigurable embedded system,due to structural difference between MPU core and reconfigurable component, inorder to obtain reasonable and effective partitioning result, not only task executivecomponent of the system is taken into consideration, but also consider to divide thetask assigned on reconfigurable component into non-overlapping section on time,which makes use of reconfigurable characteristic of reconfigurable component. In theprocess of partitioning, a measure is taken to reduce configuration delay. The essenceof partitioning is a process that is to search solution space, then evaluate the solution,finally find out optimizing solution according with constraint condition.According to new features like: this paper carries on research about frequently-used hardware-software partitioning algorithm in practical application and thenproposes a new hardware-software partitioning algorithm and task schedulingalgorithm. At first in the paper a structure model is established that is composed byFPGA, microprocessor and RAM. Then DAG is used to show the restrictionrelationship of tasks. Genetic/ant search (GAMMAS) integration strategy isproposed in this paper and the application which described by DAG is mapped to thehardware and software platform. The order and start time of every task andcommunications that was assigned to each processing unit in this system is sited byused configuration prefetching and scheduling strategy. By this strategy the results of the hardware-software partitioning can meet the control and data dependencies andultimately improve the performance of the system.At last, this paper writes a test program on GAMMAS. The results of testprogram present that the solution accuracy of GAMMAS is much better than GA andMMAS, and the configuration and preparation time of the system is reduced byconfiguration prefetching and scheduling, so reducing the overall completion time oftask graph. The goal of mapping is effectively achieved, which is from applicationtask graph to a reconfigurable system.