超純量多處理機系統編譯器中有關引線平行化技術之研究

Abstract

本論文提出在多處理機系統編譯器中引線平行化的方法。我們所提出的引 線平行化方法乃是對迴圈中循環 (Iteration) 間的平行度加以分析，進 而改變迴圈的架構，使其成為可以平行執行的引線 (Thread) ，而可在數 個處理機上同時地執行，以增進程式的執行效率。在產生平行引線的同時 ，我們還提出了簡單而有效率的引線自我排程 (Self- scheduling) 方法 與迅速且節省記憶體空間的引線同步 (Synchronization) 技巧。本文中 將對這些方法、原理做詳細的說明。目前這些方法已建立在我們所發展 的 CSSD (Compilation System for Superscalar Design) 中，同時也 在 IBM RS/6000 工作站上完成初步之測試工作。經由若干標竿程式 (Benchmark programs) 之初步測試與評估，均有良好的效果。這對超純 量多處理機系統編譯器的設計提供相當重要的參考。 The multi-threaded system design has become more and more important trends in various kinds of high performance computer system design. In this thesis, we have proposed an efficient method to generate parallel threads. In our method, we analyze a loop structure first, and then construct it to parallel threads running on many processors simutaneously. At the same time, we also proposed a simple and efficient self-scheduling method and synchronization technique using a few memory space. We have implemented these method in our CSSD (Compilation System for Superscalar Design) and running on IBM RS/6000 workstation. We have also evaluated our method by using several benchmark programs. The results show that some reasonable execution speedup can be obtained.