The matrix is always written in PETSc ordering, but DA vectors are
written in natural order.  This means that loading a problem with ex10
produces a genuinely different RHS than the user had initially, and the
solution is truly different (not just different by a permutation) when
they write the problem out on a different number of procs.  Should
-ksp_view_binary push PETSC_VIEWER_NATIVE so that the vector is always
consistent with the matrix ordering?

Jed

Reply via email to