The Desing And Implementation Of Multi-Core Processor Simulation Platform

With continuous enhancement of flying vehicle's intelligent and information, flight control systems become more and more complex. Put a higher demand on embedded system's handling capacity, operating temperature range, anti-radiation capacity, power, precision of floating-point operations, reliability, stability. In missile and arrow on board computer system need to process the a lot of real-time demanding tasks. Moore's Law indicates that the chip integration is reaching its limit, single processor architecture can not significantly improve the whole performance of the computer, because of lack implicit parallelism and there are data channel and memory choke points, So the study of high-performance embedded multi-core processor has a important significance to Chinese aerospace industry.In this background, to study the chip-level high-performance parallel computing platform that based on the SPARC V8 structure SOC that applying to aircraft control system and satellite information processing and such embedded occasions. Also need to design space dedicated parallel software mathematic warehouse. Focus on resolving common serial algorithm parallel implementation in scientific computing application. To solve the critical problem that there is no hardware development platform during early algorithm design, the author designed a platform that can simulate the embedded multi-core processors which design based on the SPARC V8 architecture and the internal structure is SMP, then detailed the method of the how to implement the multi-core simulation platform.To resolve part of space dedicated parallel software mathematic warehouse design work, the author found three scientific computing problems that when dealing with space dedicated parallel computation will always encountered: massive data sort, higher-order matrix multiplication, graph theory problem, and give these three single-core algorithm in multi-core simulation platform parallel methods, finally realized them on the simulation platform. For each algorithm, author did two specific examples test, collected a lot of experimental data, in the basis of these data, author drawn a lot of the author charts, for every algorithm did a analysis based on multi-core simulation platform, analysis focused on the acceleration and the total cost these two indicators, through the analysis pointed out the each parallel algorithm's advantages: greatly shorten the running time, high acceleration, also pointed out several factors that impact on the performance of the algorithm: initialization overhead, synchronization overhead and communication overhead. The author also pointed out several facts that impact of multi-core simulation platform performance, like heterogeneous processors and the quality of communication networks, and given reasonable solutions, pending future improvement. Further work could directly start on this simulation platform, work could focus on the algorithms design and optimization algorithms, task scheduling, load balancing and so on.