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:20260625T133337Z
LOCATION:Bldg. 6 - 001 - Plenary Room
DTSTART;TZID=Europe/Stockholm:20260630T122400
DTEND;TZID=Europe/Stockholm:20260630T122500
UID:submissions.pasc-conference.org_PASC26_sess129_pos136@linklings.com
SUMMARY:P37 - Structured Reinforcement Learning for Loop Transformation in
  MLIR
DESCRIPTION:Abrar Hossain (University of Toledo)\n\nOptimizing polyhedral 
 kernels for modern multicore architectures is a high-dimensional, non-conv
 ex problem where small structural changes often yield orders-of-magnitude 
 runtime variation. While traditional compilers rely on rigid static heuris
 tics and autotuners require prohibitive search times, this poster presents
  RL Tuner, a reinforcement learning-based compilation framework that autom
 atically discovers high-performance optimization schedules within the MLIR
  ecosystem.\nUnlike prior approaches that treat code as text, RL Tuner uti
 lizes a novel, geometry-aware state representation derived from Linalg ite
 rator semantics and affine access patterns. This allows the agent to reaso
 n directly about loop hierarchy, parallelism, and data reuse. To guarantee
  semantic correctness, we implement a dependency-driven action masking mec
 hanism that restricts the agent to a subspace of legal transformations. By
  leveraging the MLIR Transform Dialect, RL Tuner applies hierarchical opti
 mizations—including tiling, interchange, and fusion—via handle-based sched
 uling.\nExperimental evaluations on PolyBench kernels demonstrate that RL 
 Tuner consistently outperforms the state-of-the-art Pluto compiler, achiev
 ing up to a 12.9× speedup on complex triangular kernels where traditional 
 heuristics stagnate. These results highlight the promise of structure-awar
 e reinforcement learning as a scalable and effective approach for automate
 d, performance-portable kernel optimization.\n\nSession Chair: Tobias Hode
 l (University of Bern, Switzerland)\n\n
END:VEVENT
END:VCALENDAR