An implementation of tensor product patch smoothers on GPU

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cui, Cu, Grosse-Bley, Paul, Kanschat, Guido, Strzodka, Robert
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913821488054272
author Cui, Cu
Grosse-Bley, Paul
Kanschat, Guido
Strzodka, Robert
author_facet Cui, Cu
Grosse-Bley, Paul
Kanschat, Guido
Strzodka, Robert
contents We present a GPU implementation of vertex-patch smoothers for higher order finite element methods in two and three dimensions. Analysis shows that they are not memory bound with respect to GPU DRAM, but with respect to on-chip scratchpad memory. Multigrid operations are optimized through localization and reorganized local operations in on-chip memory, achieving minimal global data transfer and a conflict free memory access pattern. Performance tests demonstrate that the optimized kernel is at least 2 times faster than the straightforward implementation for the Poisson problem, across various polynomial degrees in 2D and 3D, achieving up to 36% of the peak performance in both single and double precision on Nvidia A100 GPU.
format Preprint
id arxiv_https___arxiv_org_abs_2405_19004
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle An implementation of tensor product patch smoothers on GPU
Cui, Cu
Grosse-Bley, Paul
Kanschat, Guido
Strzodka, Robert
Numerical Analysis
65N55, 65Y20
We present a GPU implementation of vertex-patch smoothers for higher order finite element methods in two and three dimensions. Analysis shows that they are not memory bound with respect to GPU DRAM, but with respect to on-chip scratchpad memory. Multigrid operations are optimized through localization and reorganized local operations in on-chip memory, achieving minimal global data transfer and a conflict free memory access pattern. Performance tests demonstrate that the optimized kernel is at least 2 times faster than the straightforward implementation for the Poisson problem, across various polynomial degrees in 2D and 3D, achieving up to 36% of the peak performance in both single and double precision on Nvidia A100 GPU.
title An implementation of tensor product patch smoothers on GPU
topic Numerical Analysis
65N55, 65Y20
url https://arxiv.org/abs/2405.19004