Question about cholmod gpu vs cudasolversp

Hey, I’m new to this topic and trying to find solver for the linear spare system.
And I found CHOLMOD here, which as it says to be part of suitsparse lib, use sparse Cholesky factorization to solve the problem
https://developer.nvidia.com/cholmod
It supports CUDA now
And I also have found in cuSolverSP, there is also function like
cusolverSpcsrlsvchol()
to use sparse Cholesky factorization to solver the problem
Is there any documentation or material about the details of the algorithm used?
I mean I know Cholesky decomposition but It seems for the sparse system, there are some modifications like find some permutation or supernodal in cholmod.
What method is exactly used in cuSolverSP?
Any details about that?
Do you have any suggestions for choosing from these two libs?
I’m try to write an optimizer for a SLAM system.
Thank you very much