BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260615T072748Z
LOCATION:Bldg. 8 - Room B 101
DTSTART;TZID=Europe/Stockholm:20260701T093000
DTEND;TZID=Europe/Stockholm:20260701T100000
UID:submissions.pasc-conference.org_PASC26_sess155_msa217@linklings.com
SUMMARY:Reactant.jl: Optimize Julia Functions for High Performance on CPU,
  GPU, TPU
DESCRIPTION:William Moses (University of Illinois Urbana-Champaign; Google
 )\n\nScientific models are today limited by compute resources, forcing app
 roximations driven by feasibility rather than theory. They consequently mi
 ss important physical processes and decision-relevant regional details. Ad
 vances in AI-driven supercomputing — specialized tensor accelerators, AI c
 ompiler stacks, and novel distributed systems — offer unprecedented comput
 ational power. Yet, scientific applications such as ocean models, often wr
 itten in Fortran, C++, or Julia and built for traditional HPC, remain larg
 ely incompatible with these technologies. This gap hampers performance por
 tability and isolates scientific computing from rapid cloud-based innovati
 on for AI workloads.\n\nIn this talk we present Reactant.jl, an open-sourc
 e optimizing compiler framework, based on MLIR and XLA. Reactant.jl preser
 ves high-level semantics (e.g. linear algebra operations), enabling aggres
 sive cross-function, high-level optimizations, and generating efficient co
 de for a variety of backends (CPU, GPU, TPU and more). Furthermore, Reacta
 nt.jl combines with Enzyme to provide high-performance multi-backend autom
 atic differentiation.\n\nWe demonstrate Reactant by compiling state of the
  art simulations (an ocean model and black hole imaging) to run on thousan
 ds of distributed accelerators. This opens a path for scientific software 
 to take full advantage of next-generation AI and cloud hardware — without 
 rewriting the codebase or sacrificing high-level expressiveness.\n\nDomain
 : Climate, Weather, and Earth Sciences, Physics, Computational Methods and
  Applied Mathematics\n\nSession Chairs: Ludovic Raess (University of Lausa
 nne, ETH Zurich) and Samuel Omlin (ETH Zurich / CSCS)\n\n
END:VEVENT
END:VCALENDAR