EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Block-structured matrix reordering for efficient SDDMM on tensor cores

Chengxing Zou, Changwan Hong, Gordon Euhyun Moon, Jinsung Kim

The Journal of Supercomputing · 29.05.2026 · 82 (8)

DOI: 10.1007/s11227-026-08606-2 ↗

Kurz gefasst

Fachartikel in der Zeitschrift The Journal of Supercomputing, online erschienen am 29.05.2026. Verfasst von Chengxing Zou und 3 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Research Foundation of Korea.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Sampled Dense-Dense Matrix Multiplication is a fundamental operation in sparse linear algebra, widely used in graph neural networks and scientific computing. However, accelerating computations on GPUs is challenging due to data sparsity and irregular memory access, which hinder efficient use of Tensor Cores. This paper introduces a Block-Structured Matrix Reordering framework that improves Tensor Core utilization by reorganizing sparse matrices using bi-directional reordering with weighted similarity metrics. We also propose a tile-aware sparse matrix format that improves memory access and task scheduling. To enable adaptive and balanced computation, we employ a dual-path execution strategy: dense matrix blocks are assigned to Tensor Cores, while sparse blocks are handled by CUDA Cores. Experiments on the RTX 4090 demonstrate that our method achieves up to a 10.38×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10.38\times $$\end{document} speedup over the best Tensor Core baseline and 7.31×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.31\times $$\end{document} over the best CUDA Core baseline by producing denser block structures and enhancing parallelism.

Förderung

  • National Research Foundation of Korea

Zitieren

APA

Zou, C., Hong, C., Moon, G. E., & Kim, J. (2026). Block-structured matrix reordering for efficient SDDMM on tensor cores. The Journal of Supercomputing, 82(8). https://doi.org/10.1007/s11227-026-08606-2

Vancouver

Zou C, Hong C, Moon GE, Kim J. Block-structured matrix reordering for efficient SDDMM on tensor cores. The Journal of Supercomputing. 2026;82(8). doi:10.1007/s11227-026-08606-2

BibTeX
@article{Zou2026,
  author = {Zou, Chengxing and Hong, Changwan and Moon, Gordon Euhyun and Kim, Jinsung},
  title = {Block-structured matrix reordering for efficient SDDMM on tensor cores},
  journal = {The Journal of Supercomputing},
  year = {2026},
  volume = {82},
  number = {8},
  doi = {10.1007/s11227-026-08606-2}
}
Kostenlos starten →