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:20211207T055401Z
LOCATION:224
DTSTART;TZID=America/Chicago:20211115T113000
DTEND;TZID=America/Chicago:20211115T120000
UID:submissions.supercomputing.org_SC21_sess350_ws_ia104@linklings.com
SUMMARY:Accelerating Unstructured-Grid CFD Algorithms on NVIDIA and AMD GP
 Us
DESCRIPTION:Workshop\n\nAccelerating Unstructured-Grid CFD Algorithms on N
 VIDIA and AMD GPUs\n\nStone, Walden, Zubair, Nielson\n\nComputational perf
 ormance of the FUN3D unstructured-grid computational fluid dynamics applic
 ation on GPUs is highly dependent on the efficiency of floating-point atom
 ic updates needed to support the irregular cell-, edge-, and node-based da
 ta access patterns in massively parallel GPU environments. We examine seve
 ral optimization methods to improve GPU throughput on kernels that are dom
 inated by atomic updates on NVIDIA V100/A100 and AMD MI100 GPUs. Optimizat
 ion on the AMD MI100 GPU was of primary interest since similar hardware wi
 ll be used in the upcoming Frontier supercomputer. Techniques combining re
 gister shuffling and on-chip shared memory were used to transpose and/or a
 ggregate results amongst collaborating threads before atomically updating 
 global memory. These techniques, along with algorithmic optimizations to r
 educe the update frequency, reduced the run-time of select kernels on the 
 MI100 GPU by a factor of between 2.5 and 6.0 over atomically updating glob
 al memory directly.\n\nTag: Algorithms, Architectures, Big Data, Data Anal
 ytics, Memory Systems, Numerical Algorithms\n\nRegistration Category: Work
 shop Reg Pass
END:VEVENT
END:VCALENDAR