透過您的圖書館登入
IP:3.22.119.251
  • 學位論文

週期性條件式資料流之管線排程演算法設計

Pipelined Schedule Synthesis for Periodic Conditional Data Flows

指導教授 : 施吉昇

摘要


由於現今多媒體程式的運算需求越來越高,因此異質性多核心平台漸漸的被廣泛使用。在異質性多核心平台上,管線設計技術常被運用以提升效能。但在多媒體程式中,常存在著許多資料相依性,這些資料相依性常使得管線設計技術無法適當的發揮。在這篇論文中,我們針對多個多媒體程式與異質性多核心平台提出了一個類似拼磚行為的演算法,目的是在建立出一個具有重複性與高效能的管線排程表,並且符合每個多媒體程式的需求。我們將我們所提出的演算法與最佳解,模擬退火法和貪婪法作比較,在一連串的實驗下我們可以印證我們所提出的設計方法的效力。

並列摘要


Heterogeneous multi-core platforms has become the trend for the high performance requirement of multimedia pplications. The pipeline techniques are widely used in the multi-core platforms to lead performance elevation, but the data dependencies of multimedia applications often make pipelined design unsatisfied. In this thesis, we target on multimedia streaming applications described as conditional data flows on heterogeneous multi-core platforms, and we design a ”Tile Piecing Algorithm” for pipelined schedule synthesis within the targeted applications and platforms. The approach gives an efficient way to construct a pipelined schedule. The performance evaluation result prove the proposed ”Tile Piecing Algorithm” could reduce the runtime overhead and derive a well-designed pipelined schedule.

參考文獻


[2] IntelCorporation, “Intel core2 quad processors.” at
Last accessed at May 2009.
[5] S. J. Vaughan-Nichols, “New approach to virtualization is a lightweight,” Computer, vol. 39, no. 11, pp. 12–14, 2006.
[6] N. Kiyanclar, G. A. Koenig, and W. Yurcik, “Maestro-vc: A paravirtualized execution environment for secure on-demand cluster computing,” Cluster Computing and the Grid, IEEE International Symposium on, vol. 2, p. 28, 2006.
[7] NewsLab, “Multimedia embedded operating system.” at

延伸閱讀