CAJ | 학술논문

复杂应用领域中的一些具体计算任务不仅需要计算平台具备高效的计算能力，而且也应具有与计算任务特点相匹配的计算模式。依据NVIDIA Kepler GK110架构中Hyper-Q特性与CUDA流的关系，提出单任务并行、多任务并行与多任务流式计算三种计算模式。采用空位标记的方法对计算模式进行构建与切换，结合数据缓冲机制和计算任务加载方式，设计了众核多计算模式处理系统，实现了众核处理机多模式计算的功能。
복잡응용영역중적일사구체계산임무불부수요계산평태구비고효적계산능력，이차야응구유여계산임무특점상필배적계산모식。의거NVIDIA Kepler GK110가구중Hyper-Q특성여CUDA류적관계，제출단임무병행、다임무병행여다임무류식계산삼충계산모식。채용공위표기적방법대계산모식진행구건여절환，결합수거완충궤제화계산임무가재방식，설계료음핵다계산모식처리계통，실현료음핵처리궤다모식계산적공능。
Some specific computing tasks in complex application domains not only require that the computing platform has efficient computing capability,but also have corresponding computation modes which match with the characteristics of computing tasks. Based on the relation between hyper-Q and CUDA stream in NVIDIA Kepler GK110 architecture,three computing models are presented in this paper:single task parallel computation,multi-task parallel computation and multi-task stream-oriented com-putation. The method of marking notation on unoccupied location is adopted to construct and switch computing modes. In combi-nation with data buffering mechanism and computational task loading,the multi-mode computing system based on many-core pro-cessor was designed and the multi-mode computation function of many-core processor was implemented.