I  would like to propose minor modifications to PETScSNESSOLVER. 

Before proceeding I have two questions:

- Is it fine for everybody if I remove the support to PETSC 3.2? It complicates 
a lot the code and now PETSC 3.2 is obsolete (current version 3.4). Moreover 
SNES solvers for bound constrained problems are not reliable in 3.2.

- Currently the function PETScSNESSolver::FormJacobian uses an inefficient 
MatCopy of matrices after assembling, instead of passing them as reference. 
Neither me nor Patrick were able to find a way do it properly. May someone 
expert in shared pointers and friends help on this point?   I reported the 
details in issue #132 on bitbucket 
(https://bitbucket.org/fenics-project/dolfin/issue/132/

Best,

Corrado
_______________________________________________
fenics-support mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics-support

Reply via email to