The following is changed in this commit:
- The constructor for NewtonIterationBlackoilSimple now takes
a parameter object instead of a linear solver.
- The fully implicit black-oil simulators can now use the CPR
preconditioning strategy (by passing use_cpr=true) or the
simple strategy (the default).
Note that as of this commit, the CPR preconditioning still has
not been implemented properly, and behaves just like the simple
strategy.
Function GridManager::createGrdecl() generally allocates memory
(using std::malloc()) for the MAPAXES keyword data output of the
'grdecl' structure. Release those resources before leaving the
scope to prevent memory leak introduced in commit fdca540.
This commit switches to using the convenience method
GridManager::createGrdecl()
to construct a "struct grdecl" that can be passed directly into
CpGrid::processEclipseFormat(). The practical benefit is that we
get to leverage the opm-parser that's in active development rather
than the calcifying "EclipseGridParser" that will only become less
relevant in the future. We also don't parse the data twice.
Note: This is mostly compile tested. In the few actual tests I ran,
I got into trouble with internal assertions in processEclipseFormat.
That problem is not yet analysed.
Conflicts:
examples/sim_fibo_ad.cpp
opm/autodiff/FullyImplicitBlackoilSolver_impl.hpp
This brings the "CpGrid support" branch up to date with respect to
recent changes in opm-autodiff master.
this makes the simulator quite a bit more maintainable: setting
USE_NEW_PARSER to 0 did not even compile after the the constructor for
the wells manager which took the old deck was removed last
week. Since, according to Atgeirr, SPE-1 is now producing exactly the
same results as before, it also does no longer make too much sense to
keep that code on life support...
i.e., the simulator does not deal with any output operations
anymore. This makes sense because report steps are handled by
Opm::TimeMap and the constructor for the simulator already needs more
arguments than appropriate even without this...
Manually resolved conficts in the following files
examples/sim_fibo_ad.cpp
opm/autodiff/FullyImplicitBlackoilSolver_impl.hpp
opm/autodiff/SimulatorFullyImplicitBlackoil_impl.hpp
In additions examples/sim_fibo_ad_cp.cpp was adapted to compile again.
i.e. it now uses Opm::EclipseState. This change required to re-add the
the epoch concept in some sense, but the loop variables now call it
"episode" which sounds less ethernal IMO.
Since the SimulationTimer, EclipseWriter and WellsManager stuff has
not (yet?) been merged, the old parser is required to be around. Use
my parser-integrate-hacks_for_norne branches to test your stuff on the
Norne deck...
this required to abandon the "Epoch" stuff which the new parser does
not have (as well as the ECLIPSE file format knows nothing about
epochs). Handling well controls properly thus has to wait until Joakim
and Kristian finish refurbishing the WellsManager, so as a stop-gap
measure, only the first "epoch" is run.
This includes:
- Using the class FullyImplicitBlackoilResidual instead of
in-class definition for the residual object.
- Changing residual field name mass_balance to material_balance_eq.
- Letting the simulator and solver classes accept a
FullyImplicitSystemSolverInterface instead of a LinearSolverInterface.
- In sim_fibo_ad and test_implicit_ad, instantiate class
FullyImplicitSystemSolverSimple, replicating existing behaviour.
The most severe change probably is the removal of the AutoDiff
debugging helper functions which were useful from within a debugger
but unfortunately had to rely on a presumed linker bug in order not to
be removed in the final binary.
Also, some private attributes were unused. These have been removed and
the constructors of their respective classes have been adapted. Once
their intended functionality is actually implemented, they should be
brought back on an as-needed basis.
Thanks to @bska for the review!
