BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000713Z
LOCATION:710
DTSTART;TZID=America/Denver:20231112T111800
DTEND;TZID=America/Denver:20231112T114200
UID:submissions.supercomputing.org_SC23_sess419_ws_prot106@linklings.com
SUMMARY:GPUscout: Locating Data Movement-Related Bottlenecks on GPUs
DESCRIPTION:Soumya Sen, Stepan Vanecek, and Martin Schulz (Technical Unive
 rsity of Munich)\n\nGPUs pose an attractive opportunity for delivering hig
 h-performance applications. However, GPU codes are often limited due to me
 mory contention, resulting in overall performance degradation. Since GPU s
 cheduling is transparent to the user, and GPU memory architectures are ver
 y complex compared to ones on CPUs, finding such bottlenecks is a very cum
 bersome process.\n\nIn this paper, we present a novel method of systematic
 ally detecting the root cause of frequent memory performance bottlenecks o
 n NVIDIA GPUs that we call GPUscout. It connects three approaches to analy
 zing performance - static CUDA SASS code analysis, sampling warp stalls, a
 nd kernel performance metrics. Connecting these approaches, GPUscout can i
 dentify the problem, locate the code segment where it originates, and asse
 ss its importance.\n\nThis paper illustrates the capabilities and the desi
 gn of our implementation of GPUscout. We show its applicability based on t
 hree commonly-used kernels, yielding promising results in terms of accurac
 y, efficiency, and usability.\n\nTag: Performance Measurement, Modeling, a
 nd Tools, Programming Frameworks and System Software\n\nRegistration Categ
 ory: Workshop Reg Pass\n\nSession Chairs: David Boehme (Lawrence Livermore
  National Laboratory (LLNL)), Anthony Danalis (AMD), and Josef Weidendorfe
 r (Dresden University of Technology)\n\n
END:VEVENT
END:VCALENDAR