Prof. Andrés I. Ávila / Universidad de La Frontera, Chile
Prof.Judit Gimenez Lucas / Barcelona Supercomputing Center, España
Time: Nine-hour course in three sessions
Parallel programming allows us to develop multiple variants of a single code. Doing some testing, we can select the “fastest”choice by combining core number and thread/process number. To improve performance, several tools are available based on hardware counters.
These tools help us to understand and visualize the deep relation between code and processor and give insight of code changes to speed up applications. Using the same hardware and energy, we obtain better results of HPC applications, which supports green coding.
In this advanced tutorial, we will learn about three performance tools for improving MPI and OpenMP codes: Score-P, Paraver and PAPI. We will have one session per tool corresponding to one-hour lecture about the structure of each tool and two-hour practice.
Support material will be given to each participant and a temporary account on the Soroban server at Universidad de La Frontera. Off-line support will be given during the conference. We encourage assistants to bring your own parallel code either in C++or Fortran