Reconfigurable Embedded Systems For Rapid Prototyping Methods And Task Scheduling Algorithm

Conventional codesign methodology needs to be improved for reconfigurable system, especially for dynamic reconfigurable system. Transaction-level modeling has being gained considerable attention in system-level design. It places software/hardware into a unified platform for software development, design space exploration and performance verification at the early stage. Based on current SystemC transaction-level model, this paper extends tlm-rtos class for real-time operating system modeling, provides relevant simulation solution for no-peemptive task scheduling and gives analysis on problems in preemptive case. Moreover this paper presents and implements a system prototype method which supports to construct system prototype rapidly, and provides initial function verification, architecture exploration and performance estimate.Considering the architecture characteristic of reconfigurable system, codesign should put restrictions on reconfigurable resource demand and provide design supports for the run-time scheduling and management of reconfigurable resource. This paper submits a reconfigurable embedded system plat-form prototype. Based on the idea of unified multi-tasks, it conceives an operating system model for reconfigurable embedded system. This operating system gives off more models including the virtual hardware task interface, hardware task online scheduling and reconfigurable resource management, which abstract hardware modules as hardware tasks to accept its unified management with software tasks together.During the process of reconfigurable system design, aiming at how to shorten reconfiguration latency, this paper proposes a pre-configuration based scheduling algorithm. Based on the analysis to its characteristic the algorithm decomposes a hardware task, which is described in a hybrid tasks directed acyclic graph, into a configuration sub-task and a computation sub-task. Since the configuration sub-task does not have data dependency constrain, its configuration can be advanced. So algorithm employs dynamic level pre-configuration priorities to schedule the configuration sequences of hard tasks. Simulation experiment results indicate that the algorithm is able to reduce configuration overhead and shorten the total application execution time.Finally, faced CPU+FPGA hybrid computation architecture this paper presents software/hardware tasks partition and generating flow, which partitions tasks into three parts: software tasks, hardware tasks and hybrid tasks. This paper conceives an online hybrid task scheduling algorithm, which combines tasks allocation, tasks placement with tasks migration. Simulation results demonstrate the algorithm provides preferable scheduling performance and reduces the scheduling rejection rate by making use of the great flexibility of hybrid tasks.