Full metadata record
DC FieldValueLanguage
dc.contributor.author蔡定谷en_US
dc.contributor.authorTsai, Ting-Kuen_US
dc.contributor.author曾憲雄en_US
dc.contributor.authorTseng Shian-Shyongen_US
dc.date.accessioned2014-12-12T02:18:46Z-
dc.date.available2014-12-12T02:18:46Z-
dc.date.issued1997en_US
dc.identifier.urihttp://140.113.39.130/cdrfb3/record/nctu/#NT860394035en_US
dc.identifier.urihttp://hdl.handle.net/11536/62863-
dc.description.abstract現代的作業系統,例如:Mach、Windows NT、OS/2、Solaris都是具有支 援多重引線的作業系統。在最近幾年,因為具有多處理器電腦的廣泛使用 ,所以多重引線的程式設計,對於應用程式設計師來說將是一個實際而且 重要的技術。一個具有多重引線的行程能同時執行多個指令,同時在相同 位址空間的引線會共享行程的記憶體,因此使得引線與引線之間的溝通和 資料共享變得更有效率。此外,多重引線的技術對許多領域是非常地重要 的,例如:平行編譯器、即時多媒體系統、分散式系統。基於上述的概念 ,我們在Windows NT上製作了一個能將單一引線行程轉換成多重引線行程 的翻譯器,並且在兩顆CPU的多處理器電腦系統上展示了十個DOALL迴圈、 兩個DOACROSS迴圈、四個DOCONSIDER迴圈的效能,這些被轉換成多重引線 的應用程式幾乎都能獲得高的Speedup。同時我們也提出改善執行時期平 行化策略的方法,即讓Inspector選擇適當的排程,接著Executor就使用 這個排程,將每個Wavefront中的迴圈分割成多重引線。經由實驗結果證 明這個新的方法能獲得更高的Speedup。 Modern operating systems like Mach, Windows NT, OS/2, and Solaris all support multiple threads of control within the same address space. With the wide availablity of multiprocessor computers in the coming years, multithreaded programming will be a viable and important technique for application programmers to master. With threads, a process can have multiple instruction streams executing simultaneously with much lower overhead than concurrent processes. Threads within the same address space inherently share that memory of process, which makes communication and data sharing among threads efficient. In addition, multithreading technology is absolutely essential for domains such as parallelizing compiler, real-time multimedia and distributed systems. Based upon the idea, we have implemented a translator with translating single-threaded process to multithreaded process on Windows NT. The performance of ten DOALL loops, two DOACROSS loops and four DOCONSIDER loops are demonstrated on 2-CPU multiprocessor system, and almost these applications translated to multithreads can achieve high speedup rates. Meantime, we have improved the run-time parallelizing strategy by using different schedules in executor phase. The inspector can choose the suitable scheduling strategy and the executor can utilize this schedule to partition loop iterations of a wavefront into multiple threads. By using our translator, the experimantial results show that the new method could achieve higher speedup on multiprocessor systems.zh_TW
dc.language.isozh_TWen_US
dc.subject多重引線zh_TW
dc.subject多處理器zh_TW
dc.subject平行編譯器zh_TW
dc.subject執行期間zh_TW
dc.subjectmultithreaden_US
dc.subjectmultiprocessoren_US
dc.subjectparallelizing compileren_US
dc.subjectrun-timeen_US
dc.title在 Windows NT 上多引線翻譯器的實作zh_TW
dc.titleImplementation of a Multithreaded Translator on Windows NTen_US
dc.typeThesisen_US
dc.contributor.department資訊科學與工程研究所zh_TW
Appears in Collections:Thesis