https://hgpu.org/?p=2515
Solving dense linear systems on platforms with multiple hardware accelerators