EoCoE-II Performance Evaluation Workshop

Topic outline

• 

  General

  The focus of this workshop is performance engineering. Performance engineering is more than employing tools to identify hotspots and bottlenecks. It is about developing a thorough understanding of the interactions between software and hardware. In the first part of the workshop we deal with the core, socket, and node level, where the code gets executed that does the actual computational work. Once the architectural requirements of a code are understood and correlated with performance measurements, the potential benefit of optimizations can often be predicted. We introduce a “holistic” node-level performance engineering strategy centered around the roofline performance model and apply it to different algorithms from computational science. The second part of the workshop (kindly provided in collaboration with the POP project) deals with identifying parallelization and efficiency issues beyond the node level. A set of metrics is introduced which allows the developer to assess quickly the behavior of their code in view of parallelism, and develop appropriate optimization strategies. Attendees will have the opportunity to work with their own codes while being guided by performance experts.

  Course info and schedule: https://indico.math.cnrs.fr/event/4587/
  

  Time: Monday, October 7, 1:30 p.m. - Thursday, October 10, 12:30 p.m.

  Place: University of Erlangen-Nuremberg, Department of Mathematics
  Cauerstr. 11
  91058 Erlangen
  Room number 00.325-128 (ground floor)
  GPS: 49.574092N, 11.030351E
  https://goo.gl/maps/dt36BzisdpJQbmCP9

  This course is supported by the EoCoE-II project, which is funded by the European Union’s Horizon 2020 research and innovation programme under Grant Agreement 824158.
  

  
  

  Download all slides and code in a single ZIP
• 

  Day 1

  • Agenda File
  • General introduction including quiz answers File
  • Introduction to computer architecture File
  • Topology, affinity, clock speed: LIKWID tools File
  • Microbenchmarking for architectural exploration (and more) File
  • Exercise 1: Fun with divides Page
• 

  Day 2

  • Introduction to the Roofline model File
  • Hardware performance counters: likwid-perfctr File
  • Exercise: Dense matrix-vector multiplication Page
  • SIMD and ccNUMA File
  • Roofline case studies File
  • Exercise: Matrix-free CG solver Page
  • 19:30 Night out at "Mein Lieber Schwan" in Erlangen

    GPS: 49.6027046N,11.0040878E
    https://goo.gl/maps/xbQyF95fAACywBiCA
    

• 

  Day 3

  • Understanding parallel applications File
  • Hands-On File
  • Places to get stuff to take home for your loved ones:

    • Tegut supermarket in the basement of "Arkaden": https://goo.gl/maps/qx663tRu9QTQGJpS6 (might have gingerbread) - closes 8 p.m.
      
    • Gift shop ("Töpferladen") at Hauptmarkt Nürnberg (near "Schöner Brunnen"): https://goo.gl/maps/ssiG25qK1zKkLwDY7
      - closes 6 p.m.
    • Lebkuchen Schmidt at Nürnberg Airport Terminal 2: https://goo.gl/maps/jr2Ri41rPQUrc7BM8 (for local gingerbread) - closes 6 p.m.
      
• 

  Day 4

  • Exercise: code optimization Page
• 

  Additional material

  • Getting started with the Emmy cluster at RRZE Page
  • • OSACA: Predicting in-core execution time of loops using automatic throughput and critical path analysis on x86 and ARM architectures http://tiny.cc/OSACA
    • Kerncraft: Loop kernel performance analysis and modeling toolkit http://tiny.cc/kerncraft

    
    

  • Download all slides and code in a single ZIP