Hetero-HPL:面向异构高性能计算机的HPL测试程序

HPL（High Performance Linpack）是一套被广泛用于测评计算机性能的测试程序，几十年来学术界及产业界十分关注对HPL测试程序的定制化优化工作，以充分反应同时代新兴计算机平台的性能.面向当今主流多设备异构计算平台，本文尝试为HPL的优化工作提供一种新的解决方案：Hetero-HPL.在Hetero-HPL中，进程不再要求与（协）处理器一一对应，因此HPL算法在单节点独立运行情况下可以完全避免进程间数据传输开销，算法各个重要步骤有能力完全利用物理节点的所有资源，如内存容量，CPU核心，协处理器，PCI-e总线等.Hetero-HPL并不引入冗余计算量及通信量，并在任意设备数量下妥善应对锁页内存分配限制，确保多设备负载均衡和设备内的高效的大规模同质运算.在实验平台上，Hetero-HPL效率可以达到平台峰值性能的76.5%（其中矩阵乘函数效率为84%）；进一步的实验表明，Hetero-HPL在多节点联机运行情况下也是一种可行的方案.

Hetero-HPL: A Novel Approach for HPL Benchmark on Heterogeneous Architectures

HPL (HPL Performance Linpack) is a widely used benchmark for measuring computer performance.Over the decades,the practice of optimizing and tuning of HPL has constantly drawn great attention by both industry and academic cicle to evaluate the performace of contemporary cutting-edge computer platforms.For current heterogeneous HPC platforms with multiple accelerating co-processors,a new approach of high-performance HPL benchmark,Hetero-HPL,is proposed in this paper.In Hetero-HPL,the one-to-one match between process set and phsycal (co-)processor set becomes unnessary,so that the computation within each computing node may avoid inter-process message exchange,and each important procedure of the HPL algorithm may make full use of the hardware resourses of each computing node,such as memory,CPU cores,co-processors and PCI-e bus etc.Without redundant compution and communication,the working set of Hetero-HPL is not restricted by the limit of page-locked memory size in a single allocation,and is distributed in a way that the workload is balanced among all the co-processors and massive fine-grained parallelism can be exploited.On one experimental platform with four co-processors,Heter-HPL can reach an efficiency of 76.5% where the efficiency of dgemm is only 84% in one computing node,and further experiment suggests that Hetero-HPL is also a feasible approach in distributed environment.