Overcoming Tribrid Programming with Latency Hiding Oriented Description Model

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K12008
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Software
• Research Institution: Osaka University
• Principal Investigator: Ino Fumihiko 大阪大学, 情報科学研究科, 教授 (90346172)
• Co-Investigator(Kenkyū-buntansha): 水谷 泰治 大阪工業大学, 情報科学部, 准教授 (10411414)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 並列プログラミング / 並列処理 / GPU / 性能評価

Outline of Final Research Achievements

In this work, we explored latency hiding oriented description methods that not only achieve high performance but also facilitate complicated parallel programming for parallel computers equipped with accelerators such as the graphics processing unit (GPU). In more detail, we developed a data decomposition method and a performance model required for two program description schemes: one extends a GPU-based scheme for inter-node program description and the other extends an inter-node scheme for GPU program description. The effectiveness of the proposed methods was evaluated on multi-GPU systems with practical applications such as stencil computation.

Free Research Field

高性能計算