Date: 12 October 2012 @ 07:10 - 16:30

This course give an overview of MPI+OpenMP hybrid programming techniques for HPC applications. OpenMP directives and a distributed memory programming paradigm can be mixed to exploit at best on a modern HPC cluster. The course aims at providing the student with the fundamental techniques for developing an hybrid code both for C/C++ and Fortran programming languages, giving also an introduction to the architecture of NUMA machines. The course summarize the performance tuning techniques, by also introducing the most common performance bottlenecks of an hybrid code.Topics

    Hybrid models for HPC

    Memory usage

    Thread safety

    False sharing

    Mixing modes

    NUMA machines

    Thread affinity

Learning outcomes
Ability to develop a simple MPI+OpenMP code. Understanding on NUMA architectures and their connection with the hybrid paradigm. Understanding on the common problems with hybrid codes.Prerequisites
Basic knowledge of MPI and OpenMP.Price: free of charge

https://events.prace-ri.eu/event/57/

Event types:

  • Workshops and courses


Activity log