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:20230831T095745Z
LOCATION:Seehorn
DTSTART;TZID=Europe/Stockholm:20230626T123000
DTEND;TZID=Europe/Stockholm:20230626T130000
UID:submissions.pasc-conference.org_PASC23_sess179_pap143@linklings.com
SUMMARY:Model-Based Performance Analysis of the HyTeG Finite Element Frame
 work
DESCRIPTION:Paper\n\nDominik Thönnes (Friedrich-Alexander-Universität Erla
 ngen-Nürnberg) and Ulrich Rüde (Friedrich-Alexander-Universität Erlangen-N
 ürnberg, CERFACS)\n\nIn this work, we present how code generation techniqu
 es significantly improve the performance of the computational kernels in t
 he HyTeG software framework. This HPC framework combines the performance a
 nd memory advantages of matrix-free multigrid solvers with the flexibility
  of unstructured meshes. The PYSTENCILS code generation toolbox is used to
  replace the original abstract C++ kernels with highly optimized loop nest
 s. The performance of one of those kernels (the matrix-vector multiplicati
 on) is thoroughly analyzed using the Execution-Cache-Memory (ECM) performa
 nce model. We validate these predictions by measurements on the SuperMUC-N
 G supercomputer. The experiments show that the performance mostly matches 
 the predictions. In cases where the prediction does not match, we discuss 
 the discrepancies. Additionally, we conduct a node-level scaling study whi
 ch shows the expected behavior for a memory-bound compute kernel.\n\nDomai
 n: Engineering\n\nSession Chair: Charles Moulinec (Science and Technology 
 Facilities Council)
END:VEVENT
END:VCALENDAR