~/talk

SIMT-Step Execution: A Flexible Operational Semantics For GPU Subgroup Behavior

on YouTube 19m
Talkat PLDI 2026 · 2026-06-15..06-19· [compilers-and-language-design]

speakers

name:Zheyuan Chen
links:none on file
name:Naomi Rehman
links:none on file
name:Guido Martínez
links:none on file
name:Tyler Sorensen
links:none on file

resources

video:https://youtu.be/qAYec-YsAZg
slides:none on file

summary

Formal operational semantics defines subgroup behavior using dynamic basic blocks and a spectrum classifying instructions as independent, synchronous, or collective. An executable TLA+ model checks a suite of idiomatic tests (all verify in under one second) and a fuzzing campaign generates 10K variants per base pattern and runs across nine GPUs from seven vendors. Results show most devices behave strongly synchronously, but collective-memory semantics is not portable.

AI-written summary · about

chapters (12)

  1. 0:00SIMT-Step execution intro
  2. 1:04Motivation and problem
  3. 3:54GPU subgroup overview
  4. 6:18Subgroup behavior questions
  5. 8:44Lockstep assumptions evolution
  6. 9:50Memory model and testing
  7. 11:11Independent execution storms
  8. 12:34Synchronous control flow
  9. 13:38Subgroup sync models spectrum
  10. 14:41Empirical GPU results
  11. 15:22Summary and contributions
  12. 15:46Q&A

related talks

~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.