Using the Roofline Model and Intel Advisor

August 16, 2017 @ 1:00 pm – 2:00 pm
Osni Marques

The IDEAS Productivity project, in partnership with the DOE Computing Facilities of the ALCF, OLCF, and NERSC and the DOE Exascale Computing Project (ECP), is resuming the webinar series on Best Practices for HPC Software Developers, which we began last year.

As part of this series, we will offer one-hour webinars on topics in scientific software development and high-performance computing, approximately once a month. Participation is free and open to the public, but registration will be required for each event.

The next webinar in the series was “Using the Roofline Model and Intel Advisor” with speaker Sam Williams of Lawrence Berkeley National Laboratory.  This webinar took place on August 16, 2017 and you can find a copy of the presentation and the video from the webinar in the Presentation Materials section below.

This webinar began by introducing the Roofline Model and its “Cache-Aware” variant.   Then, they moved to general guidelines and historical approaches to Roofline-based program analysis.  Next, they provided a short discussion of how changes in data locality and arithmetic intensity of two canonical benchmarks visually manifest in the context of these two Roofline formulations.  The speakers then conducted demonstrations of using Intel Advisor and the Roofline model within Intel Advisor.  The first demo was primarily instructive on how to compile, benchmark, and use Advisor. The second demo focused on using variants of a simple benchmark to highlight changes in the Roofline model as well as providing correlation to Advisor’s other capabilities.