I am looking for a software package to compute incomplete LU factorisations in parallel. Further considerations are:
- The package must allow for arbitrary level-of-fill or threshold-based truncation. ILU(0) is not good enough.
- The package must be "direct", i.e. based on the same recursion formulae as the exact factorisation and not on some iterative procedure. The approach by Chow and Patel is not suitable for our purposes.
- Both multi-threaded and distributed parallelism is acceptable.
As far as I can tell, no such package seems to exist, but maybe I missed something?