BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:Performance Engineering with a Coding Agent: A Real HPC Case Stu dy DTSTART:20260830T090000Z DTEND:20260830T160000Z DTSTAMP:20260924T145800Z UID:indico-event-157@events.plgrid.pl CONTACT:training@cyfronet.pl\;events-noreply@plgrid.pl DESCRIPTION:Speakers: Leszek Grzanka\, Klemens Noga (ACC Cyfronet AGH)\n\n Abstract\n"Vibe coding" — letting an AI agent produce software mostly on its own\, calling tools\, testing\, and fixing its own mistakes in a loop \, with little human interaction along the way — is everywhere right now . This tutorial looks at a different use of the same AI agents: hands-on p ractice in data analysis and exploration\, and in performance engineering under real hardware constraints on an actual HPC system.\nPartial differen tial equations (PDEs) are a common way to model reality across physics and other fields\, and getting a PDE solver to run correctly and effectively on real hardware is one of the most demanding engineering tasks. This tuto rial works through one such example: our own implementation of a solver — computing the recombination correction factor for a pulsed proton beam in an ionization chamber — with a physically validated simulation\, an analytic correctness check\, and a slow\, naive implementation to start fr om.\nParticipants work on that naive implementation using opencode\, an AI coding agent that can use several models hosted directly on Cyfronet's in frastructure — including GLM 5.2\, whose coding ability is similar to Cl aude Sonnet's — at no cost to the user. All data and tasks stay within C yfronet’s infrastructure. Three skills matter here: using the agent as a data-analysis tool (profile\, plot\, and explain a run instead of hand-ty ping matplotlib)\, directing it toward optimization that is both algorithm ic and hardware-aware (using Athena's real core count\, cache size\, and N UMA layout as constraints\, not guesses)\, and using it safely — permiss ions\, review discipline\, and version control\, so it can't do real damag e.\nLevel\n\nTarget audience\nTraining is intended for users who want to u se the AI agents for their work on HPC machines.\nAgenda\n11:00–13:30 — Infrastructure\, safety\, and diagnosis\n\n\nIntroduction to the domai n problem\, and a walkthrough of the naive implementation of the PDE solve r. \n\n\nCyfronet-hosted LLMs: why running GLM 5.2 on-premises matters — free at this scale\, data never leaves the cluster\, coding ability si milar to Claude Sonnet's\n\n\nUsing an agent safely: permission modes\, re viewing diffs before accepting them\, git commits as checkpoints\, never a pproving destructive commands or job submissions blindly — the rules for the rest of the day\n\n\nSetup: Athena account\, opencode + GLM 5.2 token \, a quick "hello agent" test — and the day's submission rule: the agent drafts job scripts\, participants review and submit them by hand\n\n\nOri entation: the problem we're solving\, how the starting code works\, and th e materials everyone begins with — the code itself\, a way to measure pe rformance\, and a way to check the results are still correct\n\n\nHands-on : treat the agent as a lab assistant\, not an author — "run the sweep\, plot wall time and speedup vs. thread count\, explain the shape" — and w rite a short diagnosis based on the output\n\n\n14:20–16:00 — Optimiza tion: algorithmic and hardware-aware\n\n\nReframing the question: instead of "make it faster\," have the agent check the hardware it's running on — cores\, cache\, NUMA layout.\n\n\nHands-on: algorithmic optimization — restructuring the hot loop\n\n\nHands-on: hardware-aware optimization — deciding how\, and whether\, to parallelize\n\n\nRe-run the timing har ness and the correctness check after every change\n\n\n16:30–18:00 — S olution reveal\, security debrief\, and wrap-up\n\n\nSynthesis: compare ea ch participant's findings against the optimized reference\n\n\nDiscussion: where the agent's hardware reasoning held up under measurement\, where it didn't\, and how participants told the difference\n\n\nSecurity retrospec tive: LLMs executing risky commands\, why the rules held\, and what to tak e away for using coding agents on shared HPC accounts\n\n\nWrap-up: agent as an analysis-and-optimization tool\, not autocomplete\; the case for on- prem LLMs\; pointers to the reference repo\n\n\nRequirements\nNo recommend ed training courses before this one.\nNo account required.\nWorking knowle dge of Python and basic experience with AI agents will be benefitial.\nVen ue\nThe workshop will be conducted during PPAM 2026 conference.\nLanguage\ nEnglish\nDuration\n7 hours\nRegistration\nThe Registration via PPAM2026 c onference site.\nOrganized by\n\nFunded by\n\nAcknowledgements\nThis event is partially funded by the EuroCC 3 project.\nThe project has received fu nding from the European High-Performance Computing Joint Undertaking (JU) under grant agreement No 101101903. The JU receives support from the Digit al Europe Programme and Austria\, Belgium\, Bulgaria\, Croatia\, Cyprus\, Czech Republic\, Denmark\, Estonia\, Finland\, France\, Germany\, Greece\, Hungary\, Iceland\, Ireland\, Italy\, Latvia\, Lithuania\, Luxembourg\, M ontenegro\, Netherlands\, North Macedonia\, Norway\, Poland\, Portugal\, R omania\, Serbia\, Slovakia\, Slovenia\, Spain\, Sweden\, and Türkiye.\n\n https://events.plgrid.pl/event/157/ LOCATION:Centrum Wykładowe Politechniki Poznańskiej URL:https://events.plgrid.pl/event/157/ END:VEVENT END:VCALENDAR