Adaptyst HPC deployability and AMD GPU support at LUMI-CSCS hackathon in Lugano

Featured image from: cscs.ch (license: CC BY-SA 3.0 Unported)
This post is published under the same CC BY-SA 3.0 Unported license.

On 1-5 December 2025, we participated in the LUMI-CSCS hackathon organised by the Swiss National Supercomputing Centre (CSCS) in Hotel de la Paix in Lugano, Switzerland. The theme of the event was porting and optimising various applications for the LUMI supercomputer in Finland.

Our team consisted of people from CERN and partner universities working on Madgraph5, Pepper, and Adaptyst. The main Adaptyst goal for the hackathon was working towards (1) the deployability of the tool to high-performance computing (HPC) infrastructures like LUMI and (2) the AMD GPU Adaptyst system module, all in the context of using Adaptyst for performance analysis of the Madgraph5 and Pepper event generators.

In the end, it was a productive week with support of high-performance computing and AMD mentors! These are the Adaptyst-related activities we completed during the hackathon:

• Learning how LUMI and Slurm work
• Making a Gentoo-based Singularity container image with the environment required to run Adaptyst (this makes us one step closer to releasing container images of the modular redesign of the tool)
• Using the aforementioned image on LUMI
• Implementing changes in Adaptyst easing its deployability to HPC infrastructures and beyond (such as more streamlined configuration path definitions when compiling and multiple search paths for Adaptyst modules: all of these updates are coming in the next development version!)
• Finding out more information about rocprofiler-sdk: AMD’s equivalent of NVIDIA CUPTI used by the nvgpu Adaptyst module

Thank you to the organisers for the event and to the mentors for their invaluable help! The work on the HPC deployability of Adaptyst and the AMD GPU module will be continued also outside of the hackathon.