Study On Task Partitioning And Task Scheduling In The Reconfigurable System

The rapid development of reconfigurable hardware technology, relational design methodology and CAD tools makes reconfigurable computation (RC) practical. The dynamic partial reconfiguration capability of FPGA changed traditional mode of hardware, making component to be the new computation method between software and hardware. Former co-design of software and hardware no longer satisfy the needs of dynamic reconfiguration system. The partition and assignment of function between software and component greatly influenced the performance of system. After analyzing and working over reconfigurable computation architectures, this thesis presented dynamic self-reconfiguration cooperate– computation architecture which was composed by a microprocessor and a FPGA chip.The Application mode of hardware was changed by the partially dynamic reconfigurable ability offered by reconfigurable hardware. And the traditional hardware-software partitioning methods can't guarantee feasible implementations for dynamically reconfigurable computing systems. Taking new features of reconfigurable hardware resource and the time cost of reconfiguration into account, this thesis bring forward the new task partitioning and task scheduling algorithm based on analyzing and summarizing every hardware-software partitioning algorithm.In the dynamic reconfiguration system, firstly we decomposed a big application into many sub-tasks which had appropriate granularity and restricted relations. Afterwards, according to the restricted relations of every sub-task executed early or late, we drew a directed acyclic graph which had the meaning of restricted relations. This thesis adopted genetic algorithm,the integration of genetic algorithm and hill-climbing algorithm and simulated algorithm to get a task partitioning result towards the task graph, and employed priority-based dynamic scheduling algorithm to evaluate the partition results, and finally efficiently implemented task partition and scheduling of the dynamic reconfiguration system.Finally, the thesis carried through performance analysis and experimentation results comparison. Obtained results indicate that the partitioning can minimize the application execution time efficiently. The thesis made a helpful explore on the study of task partitioning and task scheduling in the reconfigurable system.