BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250822T115803Z
LOCATION:Campussaal - Plenary Room
DTSTART;TZID=Europe/Stockholm:20250617T190000
DTEND;TZID=Europe/Stockholm:20250617T210000
UID:submissions.pasc-conference.org_PASC25_sess151_pos112@linklings.com
SUMMARY:P31 - Mixed Precision Customized for Discontinuous Galerkin Method
 s
DESCRIPTION:Shivam Sundriyal, Markus Büttner, and Vadym Aizinger (Universi
 ty of Bayreuth)\n\nWe present an approach to enhance storage efficiency an
 d reduce memory bandwidth utilization in modal Discontinuous Galerkin (DG)
  methods by introducing a customized mixed-precision\nrepresentation for t
 he solution vector. Our approach leverages variations in floating-point ac
 curacy requirements among the local degrees of freedom associated with dif
 ferent modal basis functions. Using a common exponent, we represent the lo
 cal solution vector compactly, optimizing storage efficiency. This approac
 h significantly reduces memory usage while preserving numerical accuracy. 
 To fully utilize this new representation, we design specialized arithmetic
  operations for the new datatype —addition, subtraction, multiplication, a
 nd division— ensuring stability and precision.\nThe findings highlight the
  potential of mixed precision to balance accuracy and performance, enablin
 g scalable and efficient implementations of DG methods on modern HPC archi
 tectures. This study provides practical insights and guidelines for integr
 ating mixed-precision strategies into high-order numerical methods, promot
 ing broader adoption in computational science.\n\n
END:VEVENT
END:VCALENDAR