Study Of Mixed-Precision Modeling Technology For Heterogeneous Multi-core System

With the hardware computing efficiency and the flexibility of software programming, heterogeneous multi-core systems combine the reconfigurable technology and multi-core system and also increase the complexity of the system at the same time, bringing new challenges for software and hardware system design. In the early stages of the design, through the system modeling, the technology of Electronic design on system level not only achieves the coordination software/hardware design, but also can reduce the design cost, shorten the design cycle. For system model, compared with Float model, the hybrid Double model can improve the operation rate and the overall performance on the premise of guarantee the model accuracy. Based on C++ as the modeling language, this paper researches on instruction set simulator modeling technology and hybrid precision system model modeling. In this paper, the main work is as follows:First of all, this paper design the two kinds of simulator in different precision to simulate the programmable co-processor. They are the cycle accurate simulator and the instruction precisely simulator. The cycle accurate simulator simulates the variation of the registers, memory, etc which indicate the state of the processor on working. The instruction precisely simulator just needs to do the accurate instructions and completes computing tasks according to the corresponding instruction.Secondly, this paper generally describes the design process of cycle-accurate RTL-level model of NoC and different precision models of RCU.Finally, the hybrid precision model of the heterogeneous multi-core system is constructed. Through the interface encapsulation technology, the different precision models of the reconfigurable computing units and programmable co-processor are hanged on the RTL-level NoC. At the same time, the scale of the NoC, the mount, type, precision and location of the IP core can be controlled by parametric configuration. The model has good extensibility, reconfiguration, reusability, good observation of high precision models and high operation rate of low accuracy models. Compared with the high precision model, this model saves 83.4% of time in simulation.