標題: 在PlayStation 3上的即時多媒體處理
Real-time Multimedia Processing on PlayStation 3
作者: 陳慶至
Ching-Chih Chen
劉志尉
Chih-Wei Liu
電子研究所
關鍵字: 多媒體;PlayStation;Multimedia
公開日期: 2007
摘要: 現今的多媒體發展非常迅速,為了提升資料速率(data rate)和壓縮比率(compression ratio),使得演算法的複雜度不斷增加。所以通常會使用可程式化處理器而非特定功能積體電路(ASIC),以因應各種新穎的多媒體標準。在處理器為主的平台上,只需要作軟體的更新,就可以讓新的多媒體應用能夠順利運行,延長產品在市場的壽命。在多媒體應用中,除了傳統處理器所利用到的指令層級的平行度(ILP)和資料層級的平行度(DLP)之外,多執行緒(multi-thread)或多核心(multi-core)架構進一步利用執行緒(thread)層級的平行度(TLP)。多核心處理器能提供很高的運算能力去處理複雜的應用,但是多核心處理器的程式撰寫非常困難。多核心程式撰寫有兩個最主要的問題是,由工作分派所產生的同步花費和工作之間資料傳遞所產生的溝通花費。本論文中採用以幅為基礎(frame-based)的資料切割方式,去消除工作之間的資料依靠關係(data dependency),減少同步的花費。溝通的問題則是利用資料流動最佳化來解決。在一個著名的多核心架構—PlayStation 3上,可以使用這些技巧對一個1080p多媒體影片解碼器作最佳化。由實驗結果可知,這些最佳化的技巧能使解碼器加速20倍,最後超過即時(real-time)的限制以60Hz的頻率解碼出1080p的影片。
Today’s multimedia applications evolve very fast. In order to improve the data rate and compression ratio, the complexity of algorithm is enhanced. Instead of ASIC, programmable processors are usually used to deal with the variety of new multimedia standards. Processor-based architectures can use software patches to keep up with new multimedia applications, thus extend products’ time-in-market. However, conventional processor architectures are unable to provide sufficient computing power for real-time constraints. Beyond instruction-level-parallelism (ILP) and data-level-parallelism (DLP) used in conventional processors, multithreaded/multi-core architecture further acquire thread-level-parallelism (TLP) to exploit parallelism in multimedia applications. Multi-core processor can provide high computing power to support complex applications, but multi-core programming is much more difficult. There are two main issues in multi-core programming, synchronization overhead introduced by task parathion and communication overhead arising from data communications between tasks. In this thesis, we adopt frame-based data partition to eliminate the data dependency between tasks, thus minimize synchronization overhead. Communication problem is solved using dataflow optimization. Using these techniques, a multimedia video decoder processing 1080p frames is then optimized on a famous multi-core architecture – PlayStation3. The experimental results show that the optimization techniques make the decoder run 20-time faster, which finally exceed real-time constraint to decode 1080p frame at 60 Hz.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009411600
http://hdl.handle.net/11536/80513
Appears in Collections:Thesis


Files in This Item:

  1. 160001.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.