Open NicoRenaud opened 4 years ago
I agree about separating the two implementations. Also, I have tried to make generic functions like DPR correction but I have not succeeded
I can look into the generic DPR. At the end of the day we only need the residues and the diagonal. So nothing of that require a free mat vec product
I Just realized that once we removed the ritz_vectors
recalculation, the DPR free and dense subroutines are basically the same
Should we separate the dense and free part of the code ? Having everything in one file makes it difficult to navigate (at least for me).
Also some subroutines have a dense and a free version while I think they should be the same (for example the DPR correction). We should simplify that.