Ginkgo
High-performance linear algebra library for manycore systems, with a focus on sparse solution of linear systems
Area: Mathematical libraries
CASS member: FASTMath
Description
Ginkgo is a modern C++ library focusing on efficient and portable solutions for sparse linear systems. It features backends for NVIDIA, AMD, and Intel GPUs, as well as an OpenMP backend for CPUs, all available from a common interface. Ginkgo includes advanced preconditioners like Incomplete Sparse Approximate Inverse (ISAI) and the iterative factorization-based Parallel IC/ILU, as well as standard ILU(0)/IC(0); it also has support for algebraic multigrid (AMG). Ginkgo includes a variety of iterative solvers and batched iterative solvers. Ginkgo can be built with or without MPI; most solvers support distributed matrices.
Target audience
Anyone needing to solve large, sparse linear systems, particularly those wanting platform portability and GPU support.