https://hgpu.org/?p=12571
Mixed-precision orthogonalization scheme and its case studies with CA-GMRES on a GPU