IDaTPA:Importance Degree Based Thread Partitioning Approach in Thread Level Speculation

Abstract

As an auto-parallelization technique with the level of thread on multi-core,Thread-Level Speculation (TLS) which is also called Speculative Multithreading(SpMT), partitions programs into multiple threads and speculatively executesthem under conditions of ambiguous data and control dependence. Thread par-titioning approach plays a key role to the performance enhancement in TLS.The existing heuristic rules-based approach (HR-based approach) which is anone-size-fits-all strategy, can not guarantee to achieve the optimal thread parti-tioning. In this paper, an importance degree based thread partitioning approach(IDaTPA) is proposed to realize the partition of irregular programs into mul-tithreads. IDaTPA implements biasing partitioning for every procedure with amachine learning method. It mainly includes: constructing sample set, expres-sion of knowledge, calculation of similarity, prediction model and the partitionof the irregular programs is performed by the prediction model. Using IDaTPA,the subprocedures in unseen irregular programs can obtain their optimal parti-tion. On a generic SpMT processor (called Prophet) to perform the performanceevaluation for multithreaded programs, the IDaTPA is evaluated and averagelydelivers a speedup of 1.80 upon a 4-core processor. Furthermore, in order toobtain the portability evaluation of IDaTPA, we port IDaTPA to 8-core processorand obtain a speedup of 2.82 on average. Experiment results show that IDaTPAobtains a significant speedup increasement and Olden benchmarks respectively deliver a 5.75% performance improvement on 4-core and a 6.32% performanceimprovement on 8-core, and SPEC2020 benchmarks obtain a 38.20% performanceimprovement than the conventional HR-based approach.