Richard Vuduc
Lawrence Livermore National Laboratory, USA

@ "OSKI: A Library of Automatically Tuned Sparse Matrix Kernels"

Abstract:
The Optimized Sparse Kernel Interface (OSKI) is a freely
available collection of low-level primitives that provide
automatically tuned computational kernels on sparse matrices,
for use by solver libraries and applications. OSKI hides
the complexity of choosing data structure and code
transformations that yield the best performance for a given
matrix, machine, and workload of kernel operations. The interface
supports legacy applications, exposes the steps and costs of tuning,
and allows for user inspection and control of the tuning process.
This talk highlights recent work on OSKI, including distributed
memory extensions based on PETSc wrappers around OSKI.