Physics > Computational Physics
[Submitted on 15 Aug 2022 (v1), revised 17 Jan 2023 (this version, v3), latest version 24 Sep 2023 (v5)]
Title:Fast hardware-aware matrix-free computations of higher-order finite-element discretized matrix multi-vector products
View PDFAbstract:Recent hardware-aware algorithms for higher-order finite-element (FE) discretized matrix-vector multiplications suggest that on-the-fly matrix-vector products can reduce arithmetic complexity and improve data access efficiency. These matrix-free approaches leverage the tensor-structured nature of the FE polynomial basis to evaluate the underlying integrals without explicitly constructing the global sparse matrix. Furthermore, iterative solvers for large-scale eigenvalue problems or linear systems of equations with multiple RHS vectors arising from FE discretizations necessitate efficient matrix-multivector products involving multiple vectors. However, the current state-of-the-art implementations of such matrix-free algorithms are well-suited for the action of FE-discretized matrices on a single vector and are not directly applicable to matrix-multivector products with many vectors. In this work, we propose a computationally efficient and scalable matrix-free implementation procedure for computing FE-discretized matrix-multivector products on both multi-node CPU and GPU architectures. Our implementation achieves 1.6x -- 3.1x improvement on multi-node GPU architectures and 1.6x -- 4.4x on multi-node CPU architectures for matrix-multivector products compared to the closest baseline implementation when using 1024 vectors and FE interpolating polynomial orders in the range 6 to 8.
Submission history
From: Phani Motamarri [view email][v1] Mon, 15 Aug 2022 11:44:10 UTC (4,620 KB)
[v2] Mon, 3 Oct 2022 20:16:59 UTC (4,243 KB)
[v3] Tue, 17 Jan 2023 18:12:23 UTC (3,112 KB)
[v4] Thu, 19 Jan 2023 19:02:34 UTC (3,112 KB)
[v5] Sun, 24 Sep 2023 17:15:44 UTC (8,925 KB)
Current browse context:
physics.comp-ph
Change to browse by:
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.