Finding and fixing bugs in parallel and GPU applications on Hunter

A debugger shows you what your code is really doing, not what you think it is doing. Learn to productively use gdb4hpc, ROCgdb, and Linaro DDT in hands-on sessions led by experts from HPE, AMD, and Linaro.

The online course consists of two days of lectures interwoven with hands-on exercises run directly on Hunter. The course is aimed at people new to debugging tools and does not require prior experience with the tools. On the second day, time is also reserved for applying Linaro DDT to your own code.

gdb, the GNU debugger, is the standard debugging tool on Linux. It lets you pause a program, step through it line by line, and inspect variables and memory.

ROCgdb, developed by AMD, extends gdb to AMD GPUs and APUs. It keeps the familiar gdb workflow while adding what GPU code needs: stepping through kernels, switching between GPU threads, and inspecting GPU registers and memory.

gdb4hpc, developed by HPE, brings gdb-style debugging to large parallel applications. Instead of launching a separate debugger per process, it lets you inspect and control a job across many processes and nodes from a single interface.

Linaro DDT is a graphical debugger built for HPC. It connects to your running compute job and shows the whole parallel execution visually, making it easy to set breakpoints, inspect variables, and follow control flow across many processes at once.

Veranstaltungsort

Online course
Organizer: HLRS, University of Stuttgart, Germany

Anmeldung

Veranstaltungsbeginn

14. Okt. 2026
10:00

Verstaltungsende

15. Okt. 2026
16:00

Sprache

Englisch

Einstiegslevel

Basis

Themenbereiche

HPC-Cluster: Nutzung und Betrieb

Performance-Optimierung & Debugging

Themen

Debugging

Zurück zur Liste

Prerequisites

An account on Hunter. HLRS users can use their existing account; participants from other centers can be given a guest account after submitting a completed user form, sent as a PDF and as a signed original by post.
Basic Linux command-line familiarity (navigating the shell, editing files, compiling and submitting jobs). No prior debugger experience is required; the course starts with gdb basics.
For the Linaro DDT "bring your own code" session: a code that builds and runs on Hunter, compiled with debug information (-g, and -O0 recommended). Without a debug build you can follow along but won't be able to debug your own application.

Content levels

Beginners: 9.5 hours

Learn more about course curricula and content levels.

Learning outcomes

After this course, participants will be able to debug parallel and GPU applications on Hunter. Specifically, they will be able to:

find and fix bugs in serial code with gdb: set breakpoints, step through code, and inspect variables
debug GPU kernels on the AMD MI300A APUs with ROCgdb, including stepping through kernel code and inspecting GPU threads
debug MPI applications at scale with gdb4hpc, controlling and inspecting all processes of a job from a single command line
use Linaro DDT to debug parallel applications visually, inspecting and comparing state across many processes at once

Instructors

Juha Jaykka (AMD)
John Vogt (HPE)
Rudy Shand (Linaro)

Agenda

Preliminary agenda, the course date might still shift by one day.

Day 1: TBD, 10:00–16:00

10:00–13:00 — Introduction to gdb and GPU debugging with ROCgdb (AMD), including hands-on exercises
13:00–14:00 — Lunch break
14:00–16:00 — Parallel debugging with gdb4hpc (HPE)

Day 2: TBD, 10:00–16:00

10:00–11:00 — Debugging with Linaro DDT (lecture)
11:00–13:00 — DDT hands-on exercises
13:00–14:00 — Lunch break
14:00–15:00 — Applying DDT to your own code

Recording

Please be aware that the talks will be recorded. You declare that you are aware of and consent to the recording by registering.

Registration information

Apply for this course via the button at the top of this page.

Registration closes on 26 September, 2026.

Late applications after the registration phase are still possible according to the course capacity.

Fees

This course is free of charge.

Target audience

This course is for users of the Hunter system at HLRS who develop or maintain their own applications in C, C++, or Fortran. It is suitable for beginners: no prior experience with debuggers is required.

Contact

Lucas Jordan phone 0711 685 87206, training(at)hlrs.de

HLRS Training Collaborations in HPC and AI

HLRS is part of the Gauss Centre for Supercomputing (GCS), together with JSC in Jülich and LRZ in Garching near Munich. SIDE is the German National Competence Centre (NCC) for High-Performance Computing. HLRS is also a member of the Baden-Württemberg initiative bwHPC.
Since 2025, HLRS has been coordinating one of the AI Factories of the EuroHPC JU: HammerHAI.