Pipelined Multithreading with Clustered Communication on Commodity Multi-Core Processors

摘要

Recently proposed pipelined multithreading (PMT) techniques have shown great applicability to parallelizing general programs on multi-core processors. However, significant inter-core communication overheads limit the potential performance and prevent the commercial use. To handle this problem, this paper proposes a novel clustered pipelined multithreading (CPMT) technique, which utilizes a clustered communication mechanism to minimize the communication overheads from the average standpoint. We quantitatively demonstrate the CPMT performance can be improved through reducing average communication overheads and show its performance characteristics. The CPMT technique is evaluated on the AMD Phenom four-core processor, and experimental results show this technique has achieved speedup ranging from 116.8% to 219.8% on some typical SPEC CPU 2000 benchmark loops.