Research On Function-Level Dynamic Hardware-Software Partitioning Method For Reconfigurable System-on-Chip

Reconfigurable System-on-Chip(RSoC)integrates programmable microprocessor core and reconfigurable unit into one chip. As software program execute in the microprocessor core and different hardware logic can be realized in the reconfigurable unit at different moments. So designers need hardware software partitioning to map specification into the above-mentioned two types of computing components. Research on hardware software partitioning method for reconfigurable system-on-chip is very important to give full play structural advantages of reconfigurable system-on-chip and build a flexible high-performance embedded application.Presently, in the field of reconfigurable computing, most of the dynamic hardware-software partitioning is in process-level/thread-level or instruction-level granularity. Both of the above hardware-software partitioning schemes have shortcomings:In process-level, scheduling, communication, synchronization are whole managed by software, which will takes more time to complete and is not conducive to the development of embedded systems. Instruction-level scheme need to study the on-line disassemble, on-line synthesize, on-line layout and routing algorithm its implementation is very complicated. Instruction-level partitioning can't utilize existing IP Core as well. This paper does the following work to solve the shortcomings of the existing RSoC hardware-software partitioning algorithm.At first, a function-level dynamic HW/SW partitioning method is used. Using function-level partitioning granularity and then improving the performance of the partitioning algorithm by the parameters of the algorithm are adjusted according to the complexity of the specific partitioning problem.Then, Proposed adaptive ant colony algorithm supporting this partitioning method, and analyzed the parameters of ant colony algorithm to the impact of the results. Finally, solve the shortcomings of the adaptive ant colony algorithm which can easily be trapped in local optimum by parameters adaptive. The simulation experimental results show that the algorithm can avoid is trapped in local optimum well and has a high solution quality.At last, for the shortcomings of a longer running time of adaptive ant colony algorithm, proposed a hardware/software partitioning method based on the integration of genetic algorithm and adaptive ant colony algorithm. First, we use the search results of the genetic algorithm to construction the initial pheromone of the adaptive ant colony algorithm, and then look for better results by use of adaptive ant colony algorithm, so as to improve the algorithm convergence speed and avoid local optimum. The simulation experimental results show that the algorithm can speed up the convergence rate and at the same time have a high solution quality.