BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Equality Saturation in a Real-World Machine Learning Compiler - Al
 ex Zinenko
DTSTART:20250212T160000Z
DTEND:20250212T200000Z
UID:TALK227938@talks.cam.ac.uk
CONTACT:128239
DESCRIPTION:Join us for a relaxed chat about compilers\, while socializing
  over refreshments. Our social is open to students\, academics\, professio
 nal developers and really anyone interested in compilation. We welcome beg
 inners as well as experts. Our social is an unguided space offered for you
  to get to know people\, try out some new ideas\, get feedback on your cod
 e\, or pair-program on a difficult program. Come with just a paper noteboo
 k or bring your laptop to hack on some in-progress patches.\n\nThis social
  is traditionally organized by the LLVM community\, but is open to all (po
 tential) compiler enthusiasts. For the first time\, the next Compiler Soci
 al Talk is part of the department's Wednesday Seminar.\n\n*Equality Satura
 tion in a Real-World Machine Learning Compiler*\n\nMachine learning (ML) c
 ompilers rely on graph-level transformations to enhance the runtime perfor
 mance of ML models. However\, these program transformations are often driv
 en by manually-tuned compiler heuristics\, which are quickly rendered obso
 lete by new hardware and model architectures. Instead\, we propose the use
  of equality saturation. We replace such heuristics with a more robust glo
 bal performance model\, which accounts for downstream transformations. Whi
 le this approach still requires a global performance model to evaluate the
  profitability of transformations\,it holds significant promise for increa
 sed automation and adaptability. We address challenges in applying equalit
 y saturation on real-world ML compute graphs and state-of-the-art hardware
 \, study different cost modeling approaches to deal with fusion and layout
  optimization\, and tackle scalability issues that arise from considering 
 a very wide range of algebraic optimizations. Our implementation builds on
  and improves the XLA compilation pipeline for CPU and GPU.\n\nThis talk i
 s the second one in our compiler social\, the first talk will be given by 
 George Constantinides: https://talks.cam.ac.uk/talk/index/225220\n\nPlease
  register if you're planning to attend: https://grosser.science/compiler-s
 ocial-2025-02-12/
LOCATION:Computer Laboratory\, William Gates Building\, LT1
END:VEVENT
END:VCALENDAR