Parallel analysis of TACLeBench kernel benchmark's loop and procedure level speculation

• Published in: 2021 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Big Data & Cloud Computing, Sustainable Computing & Communications, Social Computing & Networking (ISPA/BDCloud/SocialCom/SustainCom) pp. 1035 - 1040
• Main Authors: Meng, Huiling, Wang, Yaobin, Li, Ling, Musariri, Manasah, Wang, Xinyi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2021
• Subjects: Benchmark testing; Codes; Instruction sets; Kernel; multi-core; Parallel processing; Runtime; TACLeBench; Thread level speculation (TLS)
• DOI: 10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00144

Thread level speculation (TLS) technology has gained substantial international recognition due to its unique parallel program execution. It uses the potential parallel execution of programs to improve the utilization of multi-core resources. However, the kernel benchmark of TACLeBench has not effectively been analyzed using TLS parallelization. In response to this problem, we select 7 programs in the TACLeBench kernel benchmark to analyze its loop and procedure level speculation executions to measure their maximum potential parallelism. Furthermore, we discuss its runtime characteristics (thread size, speculative parallelism coverage, dependency feature) and the influence of program source code on speedup. Our experimental results illustrate that: 1) Most applications achieve impressive results. Bsort reaches 20.79x speedup in loop level speculation, and lms has 9.51x speedup in procedure level speculation; 2) By using TLS to accelerate the TACLeBench kernel benchmark, most applications effectively utilize computing resources from 4 to 16 cores; 3) The kernel benchmark is more suitable for developing parallelism in loop level speculation.