Research And Implement Of A Simulator For Multi-Core Stream Architecture

The increasing importance of computing-intensive applications and the properties of modern VLSI processes have led to resurgence in the development of parallel microarchitectures with a large number of ALUs as well as extensive support for parallelism, the classical stream architecture tends to toward multi-core and multi-core stream architecture is becoming a focus of microarchitecture research. The academia and industry have designed a series of classical multi-core stream architecture and multi-core stream processors (prototype). These processors and prototype systems show great potential not only in performance, but also in the utilization of area, energe-efficiency and programmable flexibility. The author thinks that the multi-core stream architecture directs the future of architecture research and is worth studing.Architecture research depends on the architecture simulator. Multi-core stream architecture, as a novel architecture, is completely different from the current processor architecture, so the existing simulators can not be applied to the mulit-core stream architecture, and developing a new architecture simulator is an urgent need. Moreover, due to the processor getting more complex, the traditional simulation technologies show much more incompetence; thus, new simulation technologies must be explored. As a result, this paper focuses on the research and implement of a simulator for multi-core stream architecture.With full understanding of the multi-core stream architecture, this paper has implemented a cycle accurate simulator named Msim for multi-core stream architecture. The simulation results show Msim's good scalability, the ability of simulating a variety of multi-core architecture and a good performance. Compared to the Stanford's single-core stream simulator named Isim, Msim can be 4 times faster than Isim. Furthermore, this paper has conducted in-depth studies of the techniques and methods of simulation and presented a few of effective technologies such as the extensive design based on profile, the fusion technology of timing simulation and functional simulation, direct execution based on host. The experimental results show that the simulator using these technologies can achieve a speedup of 1.97 compared to that without these technologies. Moreover, this paper has studied the hardvare/software co-simulation technology and addresses a simple and effective communication model, which is based on the channel between modules.