We have found several issues with MKL PARDISO for improved two-level factorization algorithm (iparm=10), namely:
1. CNR functionality works incorrectly, solutions could be incorrect;
2. pardiso_getdiag - segmentation faulted in some cases;
3. user permutation vector (iparm) works incorrectly in multi-threading mode.
Major version: 2021
Minor version: 0
Update version: 1
We have prepared reproducers with specific matrices, how can we share them for confirmation of the issues?
All of the issues you reported were reproduced with MKL 2021.3 and will be escalated against the Sparse Solver Team. We will keep this thread informed of the status of these issues.