Here we consider the linear discrepancy of strongly unimodular matrices.
Nonsingular linear transformations that subsume the class of unimodular transformations are presented.
A matrix is totally unimodular if the determinant of every induced square submatrix is equal to - 1, 0, or +1.
It will let you do any linear one-to-one loop transformation (skewing, interchange, unimodular, or any combination thereof, etc), and all you have to do is be able to generate code from the answer it gives you (which isn't that hard at all).
The task one has to do is only to solve a linear matrix inequality consisting of the coefficients of a given unimodular matrix, which can be achieved easily by the use of numerical computation packages.