I need to use PARDISO on multiple (millions) structurally-symmetric systems with the same sparsity pattern. These systems are relatively small (only a few hundred unknowns per system) but tend to be wildly different in conditioning. I use the default parameters (and mtype=11 because mtype=1 gives bad results) and solve the systems with a single phase=11 run and multiple phase=23 runs. I want to know if the same nonsymmetric permutation and scaling matrices are used for all systems and, if that is the case, whether that will has any impact on numerical accuracy.
For more complete information about compiler optimizations, see our Optimization Notice.