Using arbitrary precision for linear algebra