Debugging at Scale – gdb4hpc, valgrind4hpc, ATP, stat

Presenter: Thierry Braconnier (HPE)

Tools discussed:

• STAT: Stack Trace Analysis Tool (module cray-stat)

• ATP: Abnormal Termination Processing (module atp)

• GDB for HPC (module gdb4hpc). Works for CPU and GPU.

• CCDB - Cray comparative Debugger

• Valgrind for HPC (module valgrid4hpc)

• Sanitizers for HPC (module sanitizers4hpc)

• CRAY_ACC_DEBUG environment variable for CCE OpenaCC/OpenMP offload

Archived materials on LUMI:

• Slides: /appl/local/training/2p3day-20250303/files/LUMI-2p3day-20250303-306-Debugging_at_Scale.pdf

• Recording: /appl/local/training/2p3day-20250303/recordings/306-Debugging_at_Scale.mp4

These materials can only be distributed to actual users of LUMI (active user account).

Q&A

1. How gdb4hpc launch call maps to srun? I mean number of ranks per node and cpus per task?

   • You can use --launcher-args="--tasks-per-node=4" for example to inject arguments that will be passed to srun in the launch command. If you start gdb4hpc and type "help launch" you can get more information on this.