Software:BLIS
| Original author(s) | Science of High-Performance Computing (SHPC) group, UT-Austin |
|---|---|
| Developer(s) | Field Van Zee and Devin Matthews |
| Initial release | November 9, 2013 |
| Stable release | 0.9.0
/ April 5, 2022[1] |
| Operating system | Linux Microsoft Windows macOS FreeBSD |
| Platform | x86-64 ARM ARM64 ... |
| Type | Linear algebra library; implementation of BLAS |
| License | new/modified/3-clause BSD License |
In scientific computing, BLIS (BLAS-like Library Instantiation Software)[2][3][4][5] is an open-source framework for implementing a superset of BLAS (Basic Linear Algebra Subprograms) functionality for specific processor types that was recently awarded the J. H. Wilkinson Prize for Numerical Software.[6] It exposes that functionality through two traditional Application Programming Interfaces (APIs): the BLAS interface and the CBLAS interface. BLIS also includes two APIs native to the framework: a typed (BLAS-like) API and an object API. These native interfaces provide access to BLAS-like functionality that is not supported by, but closely related to, operations found in the BLAS (and CBLAS). The framework is developed and supported by the Science of High-Performance Computing (SHPC) group of the Oden Institute for Computational Engineering and Sciences at The University of Texas at Austin and the Matthews Research Group at Southern Methodist University.
BLIS yields high performance on many current CPU microarchitectures in both single-threaded and multithreaded modes of execution.[7] BLIS also offers competitive performance for some cases of matrix multiplication in which one or more matrix operands are unusually skinny and/or small.[8]
The framework achieves high performance by employing specialized kernels (typically written in GNU extended inline assembly syntax) along with cache and register blocking through matrix operands. BLIS also works on processors for which custom kernels have not yet been written; in those cases, the framework relies upon portable kernel implementations that perform at a lower rate of computation.
BLIS is sometimes described as a refactoring of GotoBLAS2, which was created by Kazushige Goto at the Texas Advanced Computing Center.[9]
See also
References
- ↑ Releases · flame/blis – GitHub
- ↑ Van Zee, Field; van de Geijn, Robert (2015). "BLIS: A Framework for Rapidly Instantiating BLAS Functionality". ACM Transactions on Mathematical Software 41 (3): 1–33. doi:10.1145/2764454.
- ↑ Van Zee, Field; Smith, Tyler; Igual, Francisco; Smelyanskiy, Mikhail; Zhang, Xiangyi; Kistler, Michael; Austel, Vernon; Gunnels, John et al. (2016). "The BLIS Framework: Experiments in Portability". ACM Transactions on Mathematical Software 42 (2): 1–19. doi:10.1145/2755561.
- ↑ Smith, Tyler M.; van de Geijn, Robert; Smelyanskiy, Mikhail; Hammond, Jeff R.; Van Zee, Field G. (2014). "Anatomy of High-Performance Many-Threaded Matrix Multiplication". 2014 IEEE 28th International Parallel and Distributed Processing Symposium. pp. 1049–1059. doi:10.1109/IPDPS.2014.110. ISBN 978-1-4799-3800-1.
- ↑ Low, Tze Meng; Igual, Francisco; Smith, Tyler; Quintana, Enrique (2016). "Analytical Modeling is Enough for High-Performance BLIS". ACM Transactions on Mathematical Software 43 (2): 1–18. doi:10.1145/2925987.
- ↑ James H. Wilkinson Prize for Numerical Software, SIAM · Prizes & Recognition · Major Prizes & Lectures.
- ↑ Performance.md, flame/blis on GitHub.
- ↑ PerformanceSmall.md, flame/blis on GitHub.
- ↑ Goto, Kazushige; van de Geijn, Robert A. (2008). "Anatomy of high-performance matrix multiplication". ACM Transactions on Mathematical Software 34 (3): 1–25. doi:10.1145/1356052.1356053.
External links
 |
