ICL Research Profile
Batched BLAS
Batched Basic Linear Algebra Subroutines
The Batched Basic Linear Algebra Subprograms (BBLAS) effort, an international collaboration between INRIA, Rutherford Appleton Laboratory, UmeƄ University, the University of Manchester, and UTK, created an API for numerical computing routines that process batches of either uniformly sized or varying-size matrices or vectors. This goes beyond the original Basic Linear Algebra Subprogram (BLAS) standard by specifying a programming interface for modern scientific applications, which produce large numbers of small matrices simultaneously.
Individually, the small sizes of the inputs obviate the potential benefits of using BLAS but are a perfect fit for BBLAS. The BBLAS project also serves as a working forum for establishing the consensus for the next official standard that will serve the scientific community and ensure support from hardware vendors.
Sponsored by
- National Science Foundation
With Support From
- AMD
- Intel
- NVIDIA