实现系统规模化的龙芯3号桥片设计与验证

CC-NUMA(高速缓存非均匀访存)系统应用在龙芯3A的处理器上,已得到了良好的性能体现,程序开发也有明显的简化。在维护片内和片间数据传输的稳定性和正确性的工作中,对基于一致性协议的片间互连结构进行分析和研究。为了构建更大规模的CC-NUMA系统而维护多处理器间的一致性。在片间利用互连桥片进行统一调度,从而保证跨系统访问和全局数据一致,并且实现了多核处理器在访存性能上的提升。所设计的桥片在部分结构上与龙芯3A二级缓存相似。通过定向和随机环境测试,验证了桥片在功能方面的正确性。

Implementation for large-scale system of Godson-3 bridge chip design and verification

CC-NUMA（Cache Coherent Non-Uniform Memory Access）applied on Godson-3A processors has got a good performance, programming, as well, significantly simplified. However, keeping on-chip and off-chip data transmission stable and accurate, also has some work that needs to continue analyzing and studying on the structure of chip-to-chip intercon-nection based on coherence protocol. In order to construct for larger scale expansion of CC-NUMA system, coherence should be kept among multiple processors. With the bridge chip, it gathers to control and guarantees global data coherence for crossing systems, and improves the performance of memory access. The bridge chip designed is partly similar to the Godson3A’s second level cache. The function of the bridge chip is verified through the directional and random test.