Dynamic Selection of Speculative Threads Based on Executed Loop Paths

点击翻译按钮获取中文摘要

Speculative multithreading is one of the most hopeful methods for speeding up the execution of programs in multicore systems. Based on our observation that only one or two paths are executed frequently in hot loops, we have established the two-path limited speculation method that speculates only the most frequent two paths based on path profiling of the whole program execution. To further improve the performance of this method, we have proposed a dynamic optimization system that changes speculative threads based on the 'phased behavior' in program execution. Our preliminary estimation results show that the dynamic optimization method performs 1.8 times speedup at the maximum.