BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260202T201803Z
LOCATION:274
DTSTART;TZID=America/Chicago:20251118T153000
DTEND;TZID=America/Chicago:20251118T155200
UID:submissions.supercomputing.org_SC25_sess304_pap330@linklings.com
SUMMARY:RAPTOR: Practical Numerical Profiling of Scientific Applications
DESCRIPTION:Faveo Hoerold (ETH Zürich, RIKEN Center for Computational Scie
 nce (R-CCS)); Ivan Radanov Ivanov (Institute of Science Tokyo, RIKEN Cente
 r for Computational Science (R-CCS)); Akash Dhruv (Argonne National Labora
 tory (ANL)); William S. Moses (University of Illinois Urbana-Champaign); A
 nshu Dubey (Argonne National Laboratory (ANL)); and Mohamed Wahib and Jens
  Domke (RIKEN Center for Computational Science (R-CCS))\n\nThe proliferati
 on of low-precision units in modern high-performance architectures increas
 ingly burdens domain scientists. Historically, the choice in HPC was easy:
  Can we get away with 32-bit floating-point operations and lower bandwidth
  requirements, or is FP64 necessary? Driven by artificial intelligence, ve
 ndors introduce novel low-precision units for vector and tensor operations
 , and FP64 capabilities stagnate or are reduced. This forces scientists to
  re-evaluate their codes, but a trivial search-and-replace approach to go 
 from FP64 to FP16 will not suffice.\n\nWe introduce RAPTOR: a numerical pr
 ofiling tool to guide scientists in their search for code regions where pr
 ecision lowering is feasible. Using LLVM, we transparently replace high-pr
 ecision computations using low-precision units, or emulate a user-defined 
 precision. RAPTOR is a novel, feature-rich approach—with a focus on ease o
 f use—to change, profile, and reason about numerical requirements and inst
 abilities, which we demonstrate with four real-world multi-physics Flash-X
  applications.\n\nTag: Algorithms, Applications, Architectures & Networks\
 n\nRecording: Livestreamed, Recorded\n\nRegistration Category: Technical P
 rogram Reg Pass\n\nSession Chair: Hartwig Anzt (Technical University of Mu
 nich; University of Tennessee, Knoxville)\n\n
END:VEVENT
END:VCALENDAR
