利用Stencil建模及评估Intel IMCI vgather指令

计算机工程与科学

• 论文 • 下一篇

利用Stencil建模及评估Intel IMCI vgather指令

林新华1,2，王一超1，秦强1，李硕3，文敏华1，松岡聡2

（1.上海交通大学高性能计算中心,上海 200240;
2.东京工业大学,日本东京 1528550;3.Intel公司,美国波特兰 OR97124）

收稿日期:2015-12-11 修回日期:2016-03-21 出版日期:2016-09-25 发布日期:2016-09-25
基金资助:
国家863计划（2014AA01A302）;日本学术振兴会RONPAKU Fellowship资助

Modeling and evaluating Intel IMCI vgather instruction using stencilsJames

Lin1,2,WANG Yi chao1,QIN Qiang1,LI Shuo3,WEN Min hua1,Satoshi Matsuoka2

（1.Center for High Performance Computing,Shanghai Jiao Tong University,Shanghai 200240,China;
2.Tokyo Institute of Technology,Tokyo 1528550,Japan;
3.Intel Corporation,Portland OR97124,USA）

Received:2015-12-11 Revised:2016-03-21 Online:2016-09-25 Published:2016-09-25

摘要/Abstract

摘要：

Intel Xeon Phi协处理器的指令集IMCI引入了硬件实现的vgather指令，旨在帮助512位SIMD寄存器访问非连续内存地址上的数据。然而实验结果显示，vgather很有可能成为应用在Xeon Phi协处理器上关键的性能瓶颈之一。基于以上结论，针对vgather的性能建模可以帮助用户深入地掌握和理解Xeon Phi协处理器的性能特性。在实验方法上，本文方法与现存的通过程序段内嵌入汇编代码进行数据统计不同，使用PAPI等性能分析工具直接收集硬件计数器的统计结果，作为模型的实验数据。本文的性能模型基于AGI事件次数和根据VPU_DATA_READ次数估算得出的vgather所导致的平均延迟构建而成。该模型能够对Xeon Phi应用代码中由vgather所导致的总延迟进行预测。最终，为了验证模型预测的准确性，将该模型应用在三维7点stencil应用代码上，预测结果显示，vgather耗时占计算总耗时的约40%。再将该结果与利用intrinsics指令去除vgather后的计算耗时进行了对比验证，结果显示模型预测准确。基于上述结论，采用硬件计数器的统计结果在Xeon Phi协处理器上针对vgather构建了性能模型。同时，通过与其他平台的vgather对比，认为该模型也可以应用在同样具备vgather的Intel CPU处理器平台上。

关键词: 性能建模, vgather, Xeon Phi, 硬件计数器

Abstract:

Vgather is a hardwareimplemented vector instruction introduced by Intel Initial ManyCore Instructions (IMCI) for Xeon Phi. Its target is to help SIMD registers access data from noncontiguous memory locations. However, experimental results show that it can also be one of the key performance bottlenecks on Xeon Phi. We model the performance of Vgather by using the profiling tool PAPI to directly collect the results of hardware performance counters. Address Generation Interlock (AGI) events are profiled as the number of Vgather and the average latency of Vgather are estimated with VPU_DATA_READ events based on which we model the total latencies of Vgather instructions. 3D7P stencils are used to evaluate our model and the results show that Vgather spents nearly 40% of total kernel time. We implement a Vgatherfree version with intrinsic instruction to validate this model. Our contribution includes modeling Intel IMCI vgather instruction with hardware counters and validating it by stencils. The model can also be applicable on CPUs.

Key words: performance modeling, vgather, Xeon Phi, hardware performance counters

林新华1,2，王一超1，秦强1，李硕3，文敏华1，松岡聡2. 利用Stencil建模及评估Intel IMCI vgather指令[J]. 计算机工程与科学.

Lin1,2,WANG Yi chao1,QIN Qiang1,LI Shuo3,WEN Min hua1,Satoshi Matsuoka2. Modeling and evaluating Intel IMCI vgather instruction using stencilsJames [J]. Computer Engineering & Science.

参考文献

［1］Xue W,Yang C,Fu H,et al. Enabling and scaling a global shallowwater atmospheric model on Tianhe2［C］∥Proc of the 2014 IEEE 28th International Parallel and Distributed Processing Symposium,2014:745754.
［2］Pennycook S J,Hughes C J,Smelyanskiy M,et al. Exploring SIMD for molecular dynamics,using Intel Xeon processors and Intel Xeon Phi coprocessors［C］∥Proc of IPDPS ’13,2013:10851097.
［3］Hofmann J. Performance evaluation of the Intel many integrated core architecture for 3D image reconstruction in computed tomography［D］. ErlangenNuremberg:FriedrichAlexanderUniversity ErlangenNuremberg,2013.
［4］Hofmann J, Treibig J, Hager G,et al. Performance engineering for a medical imaging application on the Intel Xeon Phi accelerator［C］∥Proc of the 27th International Conference on Architecture of Computing Systems (ARCS2014),2014:18.
［5］Krishnaiyer R,Kultursay E,Chawla P,et al. Compilerbased data prefetching and streaming nontemporal store generation for the Intel(R) Xeon Phi(TM) Coprocessor［C］∥Proc of the 2013 IEEE 27th International Symposium on Parallel and Distributed Processing Workshops and PhD Forum,2013:15751586.
［6］Peraza J,Tiwari A,Laurenzano M,et al. Understanding the performance of stencil computations on Intel’s Xeon Phi［C］∥Proc of 2013 IEEE International Conference on Uster Computing (USTER13),2013:15.
［7］Lin J,Nukada A,Matsuoka S. Modeling gather and scatter with hardware performance counters for Xeon Phi［J］. ACM/IEEE International Symposium on Cluster, Cloud & Grid Computing, 2015:713716.
［8］Henretty T,Stock K,Pouchet LN,et al. Data layout transformation for stencil computations on shortvector SIMD architectures［C］∥Proc of International Conference on Compiler Construction,2011:225245.
［9］Lin Xinhua,Li Shuo,Zhao Jiaming,et al. Nodelevel memory access optimization on Intel Knights Corner［J］. Computer Science,2015,42(11):3742. (in Chinese)
［10］Treibig J, Hager G. Introducing a performance model for bandwidthlimited loop kernels［C］∥Proc of the 18th International Conference on Parallel Processing and Applied Mathematics,2010:615524.
［11］Ramos S,Hoefler T. Modeling communication in cachecoherent SMP systems:A case study with Xeon Phi［C］∥Proc of the 22nd International Symposium on Highperformance Parallel and Distributed Computing,2013:97.
［12］Zhang Y, Owens J D. A quantitative performance analysis model for GPU architectures［C］∥Proc of the 17th IEEE International Symposium on High Performance Computer Architecture (HPCA 2011),2011:382393.
［13］Hong S,Kim H. An analytical model for a GPU architecture with memorylevel and threadlevel parallelism awareness［C］∥Proc of the 36th Annual International Symposium on Computer Architecture,2009:152163.
［14］Browne S, Dongarra J, Garner N,et al. A scalable crossplatform infrastructure for application performance tuning using hardware counters［C］∥Proc of ACM/IEEE 2000 Supercomputing Conference(SC00),2000:42.
附中文参考文献：
［12］林新华,李硕,赵嘉明,等. Intel Knights Corner 的结点级内存访问优化［J］. 计算机科学,2015,42(11):3742.

[1]	杨祥森1，金君2，王鹏1，马召贵1，亢永敢1. 基于Xeon Phi平台的波动方程叠前深度偏移[J]. J4, 2015, 37(05): 907-913.
[2]	齐金, 李宽, 杨灿群, 杜云飞. Xeon Phi平台上基于模板优化的3D-GVF场计算加速[J]. 计算机工程与科学, 2014, 36(08): 1435-1440.
[3]	姚文科1，杜云飞2，吴强1，杨灿群1. 基于Intel Xeon Phi的激光等离子体粒子模拟研究[J]. J4, 2014, 36(05): 809-813.

利用Stencil建模及评估Intel IMCI vgather指令

Modeling and evaluating Intel IMCI vgather instruction using stencilsJames

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 3

编辑推荐

Metrics

本文评价