GEM - International Journal on Geomathematics Modeling and Simulation

Lineal: An efficient linear algebra library

Kurt Böhm, Olaf Ippisch
Springer Nature · 2026-08-14 · DOI 10.1007/s13137-026-00284-9

Abstract

Abstract This paper introduces Lineal , a new linear algebra library optimized for runtime and especially memory efficiency to solve large sparse linear systems arising from PDE discretization on attainable hardware. Lineal supports general matrix-based linear systems using Compressed Sparse Row (CSR) matrices and matrix-free linear systems for stencil-based problems, which can store a single value per cell to minimize memory use. Both can be paired with MAMGO, Lineal ’s Algebraic Multi-Grid implementation based on DUNE ISTL’s algorithm, which can use a matrix-free linear system on the finest, most memory-intensive level and CSR matrices on coarser levels, greatly reducing memory consumption—a novel approach, to our knowledge. Lineal also offers very flexible support for mixed precision to minimize memory use and the effect of reduced precision on convergence. Almost all components are fully multithreaded and many support explicit SIMD operations and/or tiling. Tests solving the stationary diffusion problem using two sets of high-resolution scans of soil samples show Lineal ’s very good strong scaling and high performance. On the smaller dataset, Lineal outperforms hypre ’s BoomerAMG , DUNE ISTL, and Ginkgo , requiring under half the time and memory with CSR matrices only and less than a quarter with the hybrid matrix-free approach. Unlike the other libraries, Lineal can handle the much larger dataset on the test system due to its low memory requirements. On the SPE10 Model 2 benchmark, Lineal is somewhat slower than BoomerAMG but uses much less memory, and it is faster than the other libraries.

Read at publisher →