SVD computation fails with Lapack when m < n

[Fabien Spindler]

Using the Lapack backend to calculate the SVD of a rectangular matrix A(m, n) where m the number of rows is less than n the number of columns leads to a poor SVD decomposition.