Closed njansson closed 1 day ago
Well, I can't really review the solver code, I know to little about how that stuff works :-). But we certainly need the new interface and I guess this is a good start.
I think a good idea for the future would be to add some tests for the solvers, i.e. letting them solve some sample linear system.
This PR adds an additional Krylov solver interface for coupled solve (x,y,z).
Currently limited to a CUDA POC for the fused kernel CG formulation