Solving a Ax = b where A is pentadaigonal block matrix

I want to solve a linear system where the coefficient matrix is pentadiagonal block matrix with many of blocks are zero matrices. My matrix is stored in the device memory in row major format with the help of cudaMallocPitch. Any help ll be appreciated.

Thank You

Did you try the cuSolver Libary?

You can do chomsky decomposition with cusolverDnpotrf() and solve this with cusolverDnpotrs(). Or if you matrix don’t have the requirements to do chomsky, you can do an LU decomposition cusolverDngetrf() and solve it with cusolverDngetrs().

row to colom major format should be possiblie to switch by a matrix transpone, which can be done with this methods.

For a sparse solver i can’t give you the commands but there should be somethig in cuSparse.