exanauts/CUDSS.jl

Refactorization example

IdRatherBeCoding opened this issue · 3 comments

IdRatherBeCoding commented

I've found the library really useful, thank you. However, I cannot work out how to do a refactorization. How can I update the solver by factorizing a matrix with the same sparsity pattern but different values? Are you available to provide an example of this?

EDIT: I can see that the examples do indeed show how to reuse the symbolic analysis and refactor. I expected to see use of CUDSS_PHASE_REFACTORIZATION for phase "refactorization". Is that not required?

amontoison commented

Hi @IdRatherBeCoding,
You can find examples in the README.
It's in-place LU, LDL' and Cholesky.

amontoison commented

CUDSS_PHASE_REFACTORIZATION is only relevant for a specific algorithm of the LU decomposition, otherwise you have the same bahavior with CUDSS_PHASE_FACTORIZATION.

amontoison commented

@IdRatherBeCoding
I checked the documentation and you're right that using "refactorization" is more relevant.
In many cases it's equivalent to "factorization" but it could change in future releases:
https://docs.nvidia.com/cuda/cudss/types.html#cudssphase-t