自我組織與自我排程之WPU設計與實作

Abstract

傳統的主從式多處理器架構較不具彈性，必須特別去管理資源和工作的分配，當處理器的數目改變時，程式必須重新編譯。因此，具自我組織功能的WPU被提出來加速基站端的運算，有別於傳統的架構，WPU不需要由中央管理資源和工作的分配，能自行分工完成工作，在工作和資源分配上，可以更有彈性和效率。 另外，具有自我排程功能的WPU，能自行針對指令流程進行重排，加速整體執行時間。本論文規劃了WPU的內部架構，並實作了運算單元的部分，讓WPU能支援通訊上各種複雜運算的運算，以及實作了讓WPU能進行自我排程的排程器。

Traditional master-slave multiprocessor architecture is less flexible and requires management of resource and job assignments. The program should be recompiled when the number of slave processors changes. Therefore, self-organization WPU is proposed to speed up the computing of eNB. Different from the traditional architecture, WPUs do not need to be managed resource and job assignments by the center, and can complete the work by themselves. It can be more flexible and efficient for resource management and job assignments. In addition, WPU with self-scheduling can reorder instruction flow to speed up the overall execution time. This thesis plans the architecture of WPU and implement the ALU which to support the various computations in communication, and implement the scheduler to support self-scheduling.