Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach

Qiao Zhang; Rabab Alomairy; Dali Wang; Zhuowei Gu; Qinglei Cao

doi:10.1007/978-3-031-97196-9_15

Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach

Qiao Zhang
, Rabab Alomairy
, Dali Wang
, Zhuowei Gu
, Qinglei Cao

Earth Systems Modeling

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research introduces an adaptive mixed-precision GEMM framework that supports different precision formats at fine-grained tile/block levels. We utilize the PaRSEC runtime system to balance workloads across various architectures. The performance scales well on ARM CPU-based Fugaku supercomputer, Nvidia GPU-based A100 DGX, and AMD GPU-based Frontier supercomputer. This research aims to enhance computational efficiency and accuracy by bridging algorithmic advancements and hardware innovations, driving transformative progress in various applications.

Original language	English
Title of host publication	Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings
Editors	Patrick Diehl, Qinglei Cao, Thomas Herault, George Bosilca
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	174-185
Number of pages	12
ISBN (Print)	9783031971952
DOIs	https://doi.org/10.1007/978-3-031-97196-9_15
State	Published - 2026
Event	3rd International Workshop on Asynchronous Many-Task Systems and Applications, WAMTA 2025 - St. Louis, United States Duration: Feb 19 2025 → Feb 21 2025

Publication series

Name	Lecture Notes in Computer Science
Volume	15690 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	3rd International Workshop on Asynchronous Many-Task Systems and Applications, WAMTA 2025
Country/Territory	United States
City	St. Louis
Period	02/19/25 → 02/21/25

Funding

This research was supported by internal awards from Saint Louis University (Grant-0001651 and PROJ-000498) and the U.S. National Science Foundation (Award OAC-2451577). For computer time, this research used the Lonestar6 cluster from Texas Advanced Computing Center, the compute node at Innovative Computing Laboratory of the University of Tennessee, Knoxville, the Fugaku supercomputer at RIKEN, and Frontier supercomputer at Oak Ridge National Laboratory.

Keywords

General matrix multiply
High-performance computing
Mixed precision
Task-based runtime

Access to Document

10.1007/978-3-031-97196-9_15

Cite this

Zhang, Q., Alomairy, R., Wang, D., Gu, Z., & Cao, Q. (2026). Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach. In P. Diehl, Q. Cao, T. Herault, & G. Bosilca (Eds.), Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings (pp. 174-185). (Lecture Notes in Computer Science; Vol. 15690 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-97196-9_15

Zhang, Qiao ; Alomairy, Rabab ; Wang, Dali et al. / Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply : A Tile-Centric Approach. Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings. editor / Patrick Diehl ; Qinglei Cao ; Thomas Herault ; George Bosilca. Springer Science and Business Media Deutschland GmbH, 2026. pp. 174-185 (Lecture Notes in Computer Science).

@inproceedings{9c8269a7b4fc42c8b28a5235f824b543,

title = "Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach",

abstract = "General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research introduces an adaptive mixed-precision GEMM framework that supports different precision formats at fine-grained tile/block levels. We utilize the PaRSEC runtime system to balance workloads across various architectures. The performance scales well on ARM CPU-based Fugaku supercomputer, Nvidia GPU-based A100 DGX, and AMD GPU-based Frontier supercomputer. This research aims to enhance computational efficiency and accuracy by bridging algorithmic advancements and hardware innovations, driving transformative progress in various applications.",

keywords = "General matrix multiply, High-performance computing, Mixed precision, Task-based runtime",

author = "Qiao Zhang and Rabab Alomairy and Dali Wang and Zhuowei Gu and Qinglei Cao",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.; 3rd International Workshop on Asynchronous Many-Task Systems and Applications, WAMTA 2025 ; Conference date: 19-02-2025 Through 21-02-2025",

year = "2026",

doi = "10.1007/978-3-031-97196-9\_15",

language = "English",

isbn = "9783031971952",

series = "Lecture Notes in Computer Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "174--185",

editor = "Patrick Diehl and Qinglei Cao and Thomas Herault and George Bosilca",

booktitle = "Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings",

}

Zhang, Q, Alomairy, R, Wang, D, Gu, Z & Cao, Q 2026, Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach. in P Diehl, Q Cao, T Herault & G Bosilca (eds), Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings. Lecture Notes in Computer Science, vol. 15690 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 174-185, 3rd International Workshop on Asynchronous Many-Task Systems and Applications, WAMTA 2025, St. Louis, United States, 02/19/25. https://doi.org/10.1007/978-3-031-97196-9_15

Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach. / Zhang, Qiao; Alomairy, Rabab; Wang, Dali et al.
Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings. ed. / Patrick Diehl; Qinglei Cao; Thomas Herault; George Bosilca. Springer Science and Business Media Deutschland GmbH, 2026. p. 174-185 (Lecture Notes in Computer Science; Vol. 15690 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply

T2 - 3rd International Workshop on Asynchronous Many-Task Systems and Applications, WAMTA 2025

AU - Zhang, Qiao

AU - Alomairy, Rabab

AU - Wang, Dali

AU - Gu, Zhuowei

AU - Cao, Qinglei

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

PY - 2026

Y1 - 2026

N2 - General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research introduces an adaptive mixed-precision GEMM framework that supports different precision formats at fine-grained tile/block levels. We utilize the PaRSEC runtime system to balance workloads across various architectures. The performance scales well on ARM CPU-based Fugaku supercomputer, Nvidia GPU-based A100 DGX, and AMD GPU-based Frontier supercomputer. This research aims to enhance computational efficiency and accuracy by bridging algorithmic advancements and hardware innovations, driving transformative progress in various applications.

AB - General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research introduces an adaptive mixed-precision GEMM framework that supports different precision formats at fine-grained tile/block levels. We utilize the PaRSEC runtime system to balance workloads across various architectures. The performance scales well on ARM CPU-based Fugaku supercomputer, Nvidia GPU-based A100 DGX, and AMD GPU-based Frontier supercomputer. This research aims to enhance computational efficiency and accuracy by bridging algorithmic advancements and hardware innovations, driving transformative progress in various applications.

KW - General matrix multiply

KW - High-performance computing

KW - Mixed precision

KW - Task-based runtime

UR - https://www.scopus.com/pages/publications/105019042986

U2 - 10.1007/978-3-031-97196-9_15

DO - 10.1007/978-3-031-97196-9_15

M3 - Conference contribution

AN - SCOPUS:105019042986

SN - 9783031971952

T3 - Lecture Notes in Computer Science

SP - 174

EP - 185

BT - Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings

A2 - Diehl, Patrick

A2 - Cao, Qinglei

A2 - Herault, Thomas

A2 - Bosilca, George

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 19 February 2025 through 21 February 2025

ER -

Zhang Q, Alomairy R, Wang D, Gu Z, Cao Q. Leveraging Hardware-Aware Computation in Mixed-Precision Matrix Multiply: A Tile-Centric Approach. In Diehl P, Cao Q, Herault T, Bosilca G, editors, Asynchronous Many-Task Systems and Applications - 3rd International Workshop, WAMTA 2025, Proceedings. Springer Science and Business Media Deutschland GmbH. 2026. p. 174-185. (Lecture Notes in Computer Science). doi: 10.1007/978-3-031-97196-9_15