Introduction to Performance Tuning with Cray Perftools

Presenter: Galen Arnold, Senior Systems Engineer, National Center for Supercomputing Application, University of Illinois

Date: October 24, 2018

Gprof slides: https://uofi.box.com/s/9sgz6sgv39h4ptq5e7thbj6bwjhooi8u

Cray Reveal slides: https://uofi.box.com/s/g3pthzchqews0ddg5xgfvxj0szqfjzck

Python Profiling slides: https://uofi.box.com/s/uael8fwpsmrcjaq1omqpqw8piu0cabft

Video: https://www.youtube.com/watch?v=T7FjIhwFZU8

Abstract

Application optimization is important when running at scale because small increases in performance can add up to more time for your science and more system availability for others on the shared resource. Many teams stop working on their code once they see the correct result, but much more can be done to optimize your throughput. In this webinar we’ll cover classic performance profiling for HPC applications using the Cray perftools.

In this webinar, we’ll look at two components provided by Cray perftools:

  • Perftools-lite, for performing basic initial profiling via time sampling
  • Cray Reveal, for adding OpenMP directives to candidate loops based on perftools data

We will present two case studies and do a demo with each as time permits.

Target audience: Researchers.
Prerequisites: The audience should be familiar with one of the following programming languages (C, C++, Fortran).
Reference materials:

Biography

Galen is experienced with high-performance computing (C, C++, Fortran) and performance analysis/tuning from both application and system perspectives.