https://hgpu.org/?p=11334
Taking advantage of hybrid systems for sparse direct solvers via task-based runtimes