BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.devconf.info//devconf-us-2026//talk//JCUVCP
BEGIN:VTIMEZONE
TZID:EST
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10;UNTIL=20061029T070000Z
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:STANDARD
DTSTART:20071104T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000402T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=4;UNTIL=20060402T080000Z
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20070311T030000
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devconf-us-2026-JCUVCP@pretalx.devconf.info
DTSTART;TZID=EST:20260925T100000
DTEND;TZID=EST:20260925T101500
DESCRIPTION:Writing high-performance GPU kernels should be a tight develope
 r loop: edit\, run\, measure\, repeat. In Helion\, a Python DSL that lower
 s to Triton\, that loop is often dominated by autotuning. Full random sear
 ch could find excellent kernels\, but it often took 10 to 20 minutes per c
 ompile. A strategy that starts from a default kernel configuration finishe
 d in seconds\, but produced kernels 14-31% slower. Worse\, the autotune ca
 che was tied too closely to source identity: rename a variable\, move code
  around\, or change a comment\, and prior tuning work could be lost.\n\nTh
 is talk presents `FROM_BEST_AVAILABLE`\, a warm-start strategy I contribut
 ed to Helion - now the default for quick autotuning - that seeds search fr
 om configurations previously tuned for the same input characteristics and 
 device\, even when the kernel source has changed. Instead of forcing devel
 opers to choose between “slow but good” and “fast but worse\,” it 
 reuses prior tuning work to start from a promising neighborhood while stil
 l allowing the autotuner to search.\n\nThis reduced MatMul autotuning time
  from 1238 seconds to 92 seconds while matching full-search quality across
  the seven kernels we benchmarked. I’ll cover how the loose-match key wa
 s designed\, the trade-offs behind it\, and concrete examples of how this 
 approach can improve developer iteration without sacrificing kernel qualit
 y.
DTSTAMP:20260727T174640Z
LOCATION:Ladd Room (Capacity 170)
SUMMARY:Warm Starts Over Fresh Runs: Cutting GPU Kernel Autotuning from 20 
 Minutes to 90 Seconds in Helion - Alessandro Sangiorgi
URL:https://pretalx.devconf.info/devconf-us-2026/talk/JCUVCP/
END:VEVENT
END:VCALENDAR