This online course organised in cooperation with NHR@FAU covers performance engineering approaches on the compute node level. Even application developers who are fluent in OpenMP and MPI often lack a good grasp of how much performance could at best be achieved by their code.

This is because parallelism takes us only half the way to good performance.

Even worse, slow serial code tends to scale very well, hiding the fact that resources are wasted. This course conveys the required knowledge to develop a thorough understanding of the interactions between software and hardware. This process must start at the core, socket, and node level, where the code gets executed that does the actual computational work. We introduce the basic architectural features and bottlenecks of modern processors and compute nodes.

Pipelining, SIMD, superscalarity, caches, memory interfaces, ccNUMA, etc., are covered. A cornerstone of node-level performance analysis is the Roofline model, which is introduced in due detail and applied to various examples from computational science. We also show how simple software tools can be used to acquire knowledge about the system, run code in a reproducible way, and validate hypotheses about resource consumption. Finally, once the architectural requirements of a code are understood and correlated with performance measurements, the potential benefit of code changes can often be predicted, replacing hope-for-the-best optimizations by a scientific process.

The course is a PRACE training event.

Introduction


    Basic architecture of multicore systems: pipelines, SIMD, caches, sockets, memory
    The important role of system topology


Tools: topology & affinity in multicore environments

    likwid-topology and likwid-pin, alternatives


Roofline model basics         

    Model assumptions and construction
    Simple examples
    Limitations of the Roofline model


Tools: hardware performance counters

    Why hardware performance counters?
    likwid-perfctr
    Validating performance models


Optimal use of parallel resources

    Single Instruction Multiple Data (SIMD)
    Cache-coherent Non-Uniform Memory Architecture (ccNUMA)


Roofline case studies

    Tall & skinny dense matrix-matrix multiplication
    Sparse matrix-vector multiplication
    Jacobi (stencil) smoother


Basics of performance engineering

    PE process
    Code profiling
    Proper benchmarking
    Reproducibility and documentation


Beyond Roofline: The ECM performance model (optional)

Prerequisites

You have to be able to handle a Linux command line and file editing remotely. Basic knowledge of C, C++, or Fortran programming and of OpenMP is required.

Hands-On

Exercises will be done on a cluster at NHR@FAU. For the exercises you need an SSH client on your local computer.

Language

English

Lecturers

Prof. Gerhard Wellein, Dr. habil. Georg Hager (RRZE, Uni. Erlangen)

Prices and Eligibility

The course is open and free of charge for people from academia and industry.

Withdrawal Policy

See Withdrawal

Legal Notices

This course is offered in cooperation with NHR@FAU. Some of your personal data will be transferred to NHR@FAU (title, first name, surname, institution, country, email, course) to create the course accounts. The legal basis is in accordance with Article 6(1)(b) GDPR. Please see also our data protection notice (in German: https://www.lrz.de/datenschutzerklaerung/).

https://events.prace-ri.eu/event/1249/