Dr. James Quinlan, Chair and Assistant Professor of the Department of Computer Science has had his article titled “Evaluation of Bfloat16, Posit, and Takum Arithmetics in Sparse Linear Solvers” accepted for presentation at ARITH 2025, the 32nd IEEE International Symposium on Computer Arithmetic, taking place at the University of Texas, El Paso, from May 4–7.
The article, co-authored with Laslo Hunhold from the Parallel and Distributed Systems Group at the University of Cologne (Germany), is an experimental comparison of the impact of several arithmetics (IEEE, BFloat16, posits, takums) on the accuracy of four sparse matrix solvers. Specifically, it focuses on solving $Ax = b$ (with $A$ sparse) via LU, QR, MPIR (mixed precision iterative refinement), and GMRES. To facilitate their experiments, they developed a software library called “Multi-Format Linear Algebra Benchmarks” (or MuFoLab) for short.