Introduction to Node-Level Performance Engineering @ Torino

Topic outline

• 

  General

  Performance engineering does not identify the search of hotspots and bottlenecks of computer software solely; instead, it covers the process of understanding the deeper relations between software and hardware, to exploit the underlying architecture improving the efficiency of computational codes.
  
  At the end of this all-day course, participants will have a clearer picture of the basic architecture of a computational node, and how to measure code performances on it. Moreover, a way to predict how to gain benefits from code optimisation will be introduced.

  
  

  Lecturer: Dr. Georg Hager, Erlangen Regional Computing Center, University of Erlangen-Nuremberg, Germany

  Date and time: May 31, 2019, 10:00-18:00

  Location: DISMA - Politecnico di Torino, Aula Buzano (3rd floor)

  
  

  • Agenda File
• 

  Slides

  • Introduction File
  • Introduction to computer architecture File
  • A quick introduction to OpenMP File
  • Topology and affinity File
  • Microbenchmarking File
  • Working with hardware performance counters File
  • Introduction to the Roofline model File
  • Roofline examples File
• 

  Additional Material

  • Divide microbenchmark File
  • The Bandwidth Benchmark File

    Documentation and current source code at https://github.com/RRZE-HPC/TheBandwidthBenchmark