Just noticed that the latest version of MKL (11.2 Update 2) has these comments in the release notes:
"Added ability to free up memory used by the input matrix after the factorization step. This helps to reduce memory consumption when iterative refinement is not needed and disabled by the user."
I want to try this, How can I do it? What I need to set?
It also says that "iterative refinement should be disabled". Is setting "iparam=0" sufficient for this?
I also noticed that having " iparam" nonzero increases memory consumption significantly in "phase = 33". If I set it to "0", does it impact accuracy of results? What else it impact?
Intel® MKL PARDISO:
If pardiso doesn't use iterative refinement during solving step you can put dummy parameters instead of arrays ia, ja and a and free correspondent memory. The pardiso doesn't use iterative refinement algorithm if you set number of iteration to zero and and switched off automatic iterative refinement in case of pivots.