標題: 多重配送處理器架構下的延伸指令集探索
Instruction Set Extension Exploration in Multiple-issue Architectures
作者: 陳志遠
單智君
資訊科學與工程研究所
關鍵字: 延伸指令集;螞蟻演算法;instruction set extension;ISE exploration;ACO;ant colony algorithm
公開日期: 2006
摘要: 為了滿足現代嵌入式裝置高效能的需求,現代的嵌入式處理器提供了延伸指令集(ISE) 供設計者定義,或是增加指令的配送寬度。通常來說這兩種方法被視為是不同的,假使我們能結合兩種方法: 執行延伸指令集並讓指令同時執行,就可以節省更多的執行時間。然而大多數的延伸指令集探索演算法,並不大適用於多發射架構下,那是由於缺乏兩個重要的考量: (1) 在多發射處理器的架構下,並不是所有指令都在關鍵路徑上,如果將不是在關鍵路徑上的指令包成ISE,那便會浪費額外的面積 (2) 在多發射處理器架構下,產生一道新的ISE後,關鍵路徑可能會改變。為了要滿足這些考量,我們提出了一個ISE Exploration的演算法。實驗結果顯示,在多發射處理器的環境下,使用一道ISE和不使用ISE相比,我們的方法可以達到17.17%, 12.9% 和14.79% (最大,最小和平均) 的執行時間縮減。再者,我們的方法在相同的面積限制下和之前的研究相比,提高了11.39%,2.87%和7.16%(最大,最小和平均)的執行速度。
To satisfy high-performance computing demands in modern embedded devices, current embedded processors architectures either provide designer with possibility to define instruction set extension (ISE) or to increase instruction issue width. In general, both approaches are regarded as different; if we can integrate both approaches to execute ISE(s) and original instruction(s) in parallel, then further execution time can be saved. Most ISE exploration algorithms, however, are unlikely to be used in the multiple-issue processor due to the lack of two important considerations: (1) for multiple-issue processor, not all operations locate on the critical path; if operations locating on the non-critical path are grouped as ISE, then it results in unnecessary waste of silicon area; (2) the critical path may change after generating a new ISE in multiple-issue processor. To conform to these considerations, in this paper, we propose an ISE exploration algorithm for multiple-issue processor.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009455599
http://hdl.handle.net/11536/82121
Appears in Collections:Thesis


Files in This Item:

  1. 559901.pdf