SIAM-CSE19 Minisymposium

Tuesday February 26, 2019 in Spokane, WA, USA

The Basic Linear Algebra Subprograms (BLAS) are the most widely accepted standard in high performance dense matrix computation. The unified BLAS API, which is implemented by many vendors and research groups, enables performance portability for many applications across different architectures and platforms. The past few years have witnessed a continuously growing interest in optimizing BLAS for a batch of small independent problems, hence the name “Batched BLAS”. Such interest is driven by numerous applications, including tensor contractions, sparse solvers, astrophysics, quantum chemistry, and many others.

This minisymposium (MS) covers a wide range of ongoing research activities about the Batched BLAS. The MS features three main categories of research. The first one covers standardization of the Batched BLAS API, as well as the next-generation BLAS to support reproducibility and extended precisions. The second category features many libraries, from different vendors and research groups, that provide optimized Batched BLAS routines on different hardware architectures. The third category highlights several scientific applications where optimized Batched BLAS has a great impact as a critical building block. At the end of the MS, the audience will be aware of the latest developments in Batched BLAS. The MS also provides an excellent collaboration opportunity between the Batched BLAS research community and potential application developers.