面向异构体系结构的GA模型拓展

发布时间：2018-04-28 21:28

本文选题：至强融核 + 全局数组　；参考：《计算机研究与发展》2017年04期

【摘要】：科学与工程应用对计算性能要求的不断增加使得异构计算得到了迅速发展,然而CPU与加速单元之间没有共享内存的特点增加了异构编程难度,编程人员必须显式地指定数据在不同设备之间的传递情况.全局数组(global arrays,GA)模型基于聚合远程内存拷贝接口(ARMCI)为分布式存储系统提供异步单边通信、共享内存的编程环境,但ARMCI接口拓展的复杂性使得GA不能根据特定计算平台的特点迅速在该平台上实现.CoGA模型是对GA模型的异构拓展,旨在为CPU+英特尔至强融核(MIC)的异构系统提供全局数组结构,隐藏数据传输细节从而简化异构编程难度.CoGA基于MIC上的对称传输接口(SCIF)实现对CPU和MIC的内存管理,并结合SCIF远程内存访问特点优化CPU与MIC间的数据传输性能.最后,通过数据传输带宽、通信延迟和稀疏矩阵乘问题的测试,证明了CoGA简化编程并优化数据传输性能的有效性和实用性.
[Abstract]:The increasing demand for computing performance in scientific and engineering applications has led to rapid development of heterogeneous computing. However, the fact that there is no shared memory between CPU and accelerator increases the difficulty of heterogeneous programming. Programmers must explicitly specify how data is passed between different devices. The global array model based on the aggregate remote memory copy interface (ARMCI) provides asynchronous unilateral communication and shared memory programming environment for distributed storage systems. However, because of the complexity of ARMCI interface expansion, GA can not rapidly implement the .CoGA model on the platform according to the characteristics of the specific computing platform. It is a heterogeneous extension of the GA model, aiming at providing a global array structure for the heterogeneous system of CPU Intel to strong core. Hiding the details of data transmission simplifies heterogeneous programming. CoGA implements memory management of CPU and MIC based on symmetric transport interface on MIC, and optimizes the data transfer performance between CPU and MIC in combination with the characteristics of SCIF remote memory access. Finally, the validity and practicability of CoGA to simplify programming and optimize data transmission performance are proved by testing data transmission bandwidth, communication delay and sparse matrix multiplication problem.
【作者单位】：高性能计算国家重点实验室(国防科学技术大学);国防科学技术大学计算机学院;
【基金】：国家自然科学基金项目(61120106005)~~
【分类号】：TP333

【相似文献】