Hi all,
with the new mechanism of passing linear solver options through to
Newton, I finally get explicit stats on the linear solution steps in
each Newton step. That's great!
The revealed a possible shortcoming in my code. The lion share of the
computing time in my code is not the solution of the linear systems
(as I assumed was the case), but is spent between the two lines
============== *snip* ==============
[...]
Newton iteration 1: r (abs) = 1.414e-07 (tol = 1.000e-10) r (rel) =
1.441e-04 (tol = 0.000e+00)
Solving linear system of size 19101 x 19101 (PETSc Krylov solver).
[...]
============== *snap* ==============
in each Newton step. Am I right assuming that this is where the
Jacobian is actually assembled?
Is anyone observing a similar behavior? Anything worth caching maybe?
Cheers,
Nico
_______________________________________________
fenics-support mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics-support
