cuSolverSp, how to solve sparse linear system with multiple right-hand-sides using Cholesky factorization

[Chenw831X]

Hi, I am trying to solve a sparse linear system with multiple right-hand-sides (RHS) using sparse Cholesky factorization, and my sparse matrix is symmetric positive defined. I turn to cuSparse and cuSolver, and find that:

• cusolverSpDcsrlsvchol can only handle one right-hand-side vector
• From CUDA Sample cuSolverSp_LowlevelCholesky, it seems that we can't get Cholesky factor (G) using functions like cusolverSpDcsrluExtractHost in low level LU factorization. If I could get Cholesky factor (G), then cusparseSpSM can be used to solve my problem.

Thanks for your help!

[antonande]

@Chenw831X please try cudss (cuda direct sparse solver). It supports multiple rhs: Example: https://github.com/NVIDIA/CUDALibrarySamples/blob/master/cuDSS/simple/simple.cpp Docs: https://docs.nvidia.com/cuda/cudss/index.html