Hello,

I have recently used the PardisoLU solver together with MKL 10.3. Great work.

However, is there a reason to stick to Fortan-style array indexing? Because this requires to convert outerIndex and innerIndex in PardisoImpl::compute(). This could be avoided just by setting m_iparm[34] = 1.

See also:
http://software.intel.com/sites/product ... /index.htm

I also wonder why the solver needs to store a copy of the whole sparse matrix. I think it would be sufficient to just store a reference. Clearly, this requires the caller to keep it in memory.

Hi,

The Fortan-style array indexing has been fixed as you suggest a few days ago.

Regarding the copy, it is currently always done for simplicity reasons, because when the input matrix is not in a format compatible with Pardiso we really have to make a copy. But you are right, when the input is Pardiso friendly, then this extra copy should be avoided. Patch welcome!