Preconditioned Conjugate Gradient Solver

[bscohen1]

Has anyone implemented a preconditioned conjugate gradient solver using the incomplete Cholesky factorization? I am assuming that giving the function wrappers ic0, ico2 and sv_solve this can be implemented with CUSPARSE.jl.

[kshyatt]

As far as I know, no. That would be a great example or even library method, if you would like to take it on. Another option would be to extend something from IterativeSolvers.jl (or a similar package).

[bscohen1]

Is there a way to convert a CudaArray to a CudaSparseVector without transferring the array back to the host, calling sparse() and then constructing a new CudaSparseVector?

[kshyatt]

http://docs.nvidia.com/cuda/cusparse/#cusparse-lt-t-gt-gthr and then construct the CudaSparseVector? I can try to add this in if I find time (or if you want to open another PR 😉 ... ).