The Batched Blas

Abstract. In this talk we shall motivate and describe the proposed Batched Basic Linear Algebra Subprograms (BBLAS). The BBLAS are intended to independently perform a large number of a specific BLAS operation, such as matrix multiplication, on small matrices. As with the existing BLAS, the aim is to agree a specification for the BBLAS so that code which requires the BBLAS can be portable, but at the same time utilise efficient versions of the BBLAS produced by vendors and other developers. It is hoped that by the time of the SIAM conference, the proposed specification should be nearing completion.


  • Sven J. Hammarling, The University of Manchester, UK,
  • Mawussi Zounon, The University of Manchester, UK,