<br>The compilation problem that the (previously untested)<br>NonUniformBSpline class is displaying in many platforms:<br><br><a href="http://public.kitware.com/dashboard.php?name=itk">http://public.kitware.com/dashboard.php?name=itk</a><br>
<br><br>arises from the expression in line 313:<br><br>The following code:<br><br>  vnl_matrix&lt;double&gt; B = <br>    vnl_matrix_inverse&lt;double&gt;( N_matrix.transpose() * N_matrix) * N_matrix.transpose() * data_matrix;<br>
<br><br><div id=":8t" class="ii gt">is trying to solve the linear system<br><br>             X  = N * B<br><br>by doing:<br><br>        B = ( N&#39; * N )^-1  *  N&#39;  * X<br></div><br>This could be better done by using of <br>
of the linear solvers in VXL. For example<br>the LSQR one that was fixed recently.<br><br><br>So, the suggested plan to solve this<br>compilation problem is to replace that<br>line 313 with a call to the vnl linear <br>system class and then to the vnl_lsqr<br>
solver.<br><br>This should be happening later today...<br><br><br>    Luis<br><br><br>