University of Texas at Austin

News

Advancing Open Source High Performance Libraries: Highlights from BLIS 2024

By Hurley Qi

Published Oct. 22, 2024

BLIS Retreat attendees. Credit: Joanne Foote

The Science of High-Performance Computing (SHPC) Group hosted its annual BLAS-like Library Instantiation Software (BLIS) retreat from September 26-27 at the Gates Dell Complex, at the University of Texas at Austin. 

The retreat was organized by Robert van de Geijn, lead of the Science of High-Performance Computing Group at the Oden Institute for Computational Engineering and Sciences at UT, Maggie Myers, senior research fellow at the Oden Institute, and Devangi Parikh, assistant professor of instruction of computer science. The retreat focused on BLIS, an open-source framework designed for the rapid development of high-performance libraries with Basic Linear Algebra Subprograms (BLAS) functionality. 

Attendees from academia, industry, and government labs worldwide gathered in person and online to discuss recent advances in high-performance computing (HPC) and dense linear algebra libraries.

“Every year, the retreat reinvigorates us,” said van de Geijn, who is also Professor Emeritus of Computer Science. “People are central to the success of open source software, and the yearly meetings have created and nurtured a valuable community of experts. It creates new excitement and collaborations.”

block.caption

BLAS-like Library Instantiation Software (BLIS): Building new functionality. Figure courtesy of Robert van de Geijn and Maggie Myers

Initially developed at UT Austin, BLIS is now a collaborative project involving Southern Methodist University (SMU) and other contributors worldwide. Field Van Zee, originally chief architect of BLIS and former member of the Oden Institute’s SHPC Group, along with Devin Matthews, a former Oden Institute postdoctoral researcher and current associate professor of chemistry at SMU, were awarded the 2023 J.H. Wilkinson Prize for Numerical Software for their work on BLIS. BLIS has been embraced by major companies including AMD, Oracle, and NVIDIA.

The workshop covered a wide range of topics, including software optimization, accuracy, and higher-level linear and multilinear functionality. Industry professionals from Advanced RISC Machine, Nvidia, AMD, Intel, and others also discussed the commercial deployment of BLIS-related software and functions.

The retreat concluded with an informal opportunity to interact and reflect on the event's success. For more information or view the recorded talks, please visit the BLIS Retreat webpage.