嵌入式爪哇硬體加速器中位元碼與陣列資料預先抓取之研究

Abstract

減少資料存取時的記憶體等待(memory stall)一直是改進程式執行效能的重要課題。傳統程式常在執行時花費了許多時間在等待較低層記憶體的存取，爪哇(Java)程式亦然。為了要減少在記憶體上的等待時間，預先抓取(prefetching)所需要的資料是種可行的方案。我們觀察到爪哇程式中，位元碼(bytecode)之抓取與陣列資料之存取都有明顯的特性，可以利用這些特性去預先抓取它們。這篇論文設計了適用於嵌入式爪哇硬體加速器中，位元碼與陣列資料預先抓取的機制，以減少等待記憶體讀取資料的時間。我們分析了它們的特性，並設計了一些合適的方法。平均而言，我們的方法可降低約一半的位元碼抓取所造成的停滯；在某些以陣列存取為主的程式上，也能減少約一半因陣列存取所造成之記憶體等待。

For improving speed of program execution, it is important to reduce stalls caused by memory accesses. Traditional programs usually spend much time on memory stalls during accessing lower-level memory, and so do Java programs. In order to reduce memory stall time, prefetching is a feasible solution. We observed that there exist obvious properties of bytecode fetchings and array accesses, so we can try to prefetch them by taking advantage of their properties. This thesis proposes novel prefetching mechanisms for embedded Java hardware accelerators to prefetch bytecode and array data, so that the time spent on memory stalls can be reduced. We analyzed their properties and designed suitable approaches. Our approaches can reduce half of bytecode stall time on an average; for some array-based programs, about half of array stall time can also be eliminated.