The error checking macro makes it harder to read and harder to write, so
instead we now only check for functions that can contain errors. Bounds and
range checks are handled by PETSc and not OPM.
The previous implementation set plenty of values in the initialization list and
immediately overwrote these values with values looked up from the param group.
This patch makes it look up the parameteres from the param group argument,
making the constructor simpler.
Petsc only supports initialisation through the ParameterGroup constructor.
Calling the default, non-arg constructor is a static error, and not
implementing it makes using it break compiles.
call_petsc.c was really a thin C wrapper around the call to petsc itself and
turns out was mostly unnecessary C++ emulation. This removes the file entirely
and ports its functionality into LinearSolverPetsc.cpp.
All features from the file should now be more readable as well as properly
utilising modern C++ features.
The patch uses the CHKERRXX macro from petsc to handle errors reported by
petsc, and currently does not handle this and give the control back to OPM's
error/throw system.
because if that is done, the integration for the production total is
wrong because the time step length is days and not seconds.
Note that the conversion now uses Opm::unit::convert::to(<>,
1/conversionFactor) instead of simply multiplying with the conversion
factor. I still think this obfuscates things more than it helps, but
[at] bska will hopefully want it this way...
Using "simple", the Norne deck aborts with an exception.
Note: SPE1 and SPE9 still seem to work fine with this but there might
be small differences so that the reference solutions of the Jenkins
server might need to be updated...
Commit 96cf137 introduced support for Peaceman index calculation
that honoured general completion directions (X,Y,Z). This was
accomplished through a permutation index that reordered the
permeability and geometric extent components according to a local
coordinate system along the completion.
In a complete breakdown of logic, however, the d-component extent
vector was indexed as though it were a d-by-d matrix. This commit
restores sanity to the processing.
Pointy hat: @bska.
This commit extends the feature set of the WellsManager to support
horizontal ("X" and "Y") completions and include the net-to-gross
ratio in the Peaceman index ("Completion Transmissibility Factor,
CTF") of a well completion. The NTG factor is included if present
in the input deck represented by the "eclipseState".
There are two separate, though related, parts to this commit. The
first part splits the calculation of Peaceman's "effective radius"
out to a separate utility function, effectiveRadius(), and
generalises WellsManagerDetail::computeWellIndex() to account for
arbitrary directions and NTG factors. The second part uses
GridPropertyAccess::Compressed<> to extract the NTG vector from the
input if present while providing a fall-back value of 1.0 if no such
vector is available.
Note: We may wish to make the extraction policy configurable at some
point in the future.
This commit tightens the function header of method
WellsManager::createWellsFromSpecs()
to accept a reference-to-const 'cartesian_to_compressed' map. It
used to be a complete, copy-constructed object, so this is a slight
performance enhancement as we no longer need to copy a (somewhat)
large object on every call to the method.
This commit generalises the implementation of utility function
'getCubeDim' to support arbitrary number of space dimensions. In
actual practice there's no change in features as we only really use
a compile-time constant (= 3) to specify the number of space
dimensions.
This is a demonstration of using the
GridPropertyAccess::Compressed<>
class template. We save (some) memory by not creating the zero
fall-back vector in assignPermeability(), preferring instead to use
the fall-back/default mechanism of ArrayPolicy::ExtractFromDeck<>.
While here, adjust vector<PermComponent>::reserve() capacity to
reflect actual requirements.
This commit introduces a fairly general mechanism for accessing the
active subset of a global grid (property) array. Essentially, this
takes on the role of translating the active cell index through the
"global_cell" mapping when accessing, e.g., the net-to-gross data
value.
The primary component is class template
Opm::GridPropertyAccess::Compressed<DataArray,Tag>
which implements a read-only
value_type operator[](const int c)
that encapsulates and performs the compressed-to-global cell index
translation. Template parameter "DataArray" is intended as a policy
parameter that for instance wraps access to a "GridProperty<T>" from
module opm-parser (with a fall-back default value if the data is not
specified on input). The "Tag" parameter is a provision for type
safety--e.g., to prevent passing a region ID into a function that
requires a porosity value.
Clients expect column-major (Fortran) ordering of the contiguous
"permeability_" array so that's what we create despite "tensor"
being row-major.
Suggested by: [at] atgeirr
This commit switches the assignment
diagonal = max(diagonal, minval)
to using a reference in the "diagonal" expression. This guarantees
that the indexing is done once which eases maintainability. While
here, replace the hard-coded dimension stride ('3') with the current
run-time dimension. This is mostly for symmetry because the overall
code is only really supported in three space dimension.
Calling code relies on permeability tensors being stored in column
major order (row index cycling the most rapidly). Honour that
requirement. The previous assignment implied row major ordering
(column index cycling the most rapidly). This, however, is a
pedantic rather than visible change because the surrounding code
enforces symmetric tensors whence both orderings produce the same
results when the array is viewed contiguously.
This commit increases the precision of the (floating-point) vector
output (e.g., permeabilities) from method writeGrdecl(). This
reduces the impact of rounding errors when the sub-set sample is
output to disk for subsequent processing (e.g., property upscaling).
This is in preparation of increasing the vector output precision to
reduce the impact of rounding errors in subsequent upscaling. The
change impacts floating-point vectors only.
Background: The ECL simulator stipulates an upper bound on the
number of characters in a record (i.e., one line) and while we don't
have such limits, there's no reason to emit lines that can't be
input by other tools.
The summary files now always features all wells which ever appear in
the deck in every timestep (even if they are not specified for the
time step). This _should_ fix the crashes when writing Eclipse output
in the Norne deck, but I have no idea if the resulting summary file is
correct. More testing would be thus highly appreciated...
For constant capillar pressure function the saturation is
determined by cell depths:
Sg_max, Sw_min
----- goc ----
Sg_min, Sw_min
----- woc ----
Sg_min, Sw_max
This commit removes several instances of EOL whitespace in function
'swatInitScaling()'. Aesthetic only. No functional changes.
Suggested by: [at] atgeirr
The constant 1.0e-8 was used as a threshold to distinguish "low"
from "high" capillary pressure values. Introduce acutual constant
"pc_low_threshold" to clarify that role.
Suggested by: [at] atgeirr
This commit renames the 'np' parameter used to allocate small arrays
for saturations and capillary pressures to 'max_np' to better
reflect its purpose.
Suggested by: [at] atgeirr
instead, use Opm::EclipseState. This requires to pass the PhaseUsage
object to the EclipseWriter, as this one cannot be recovered from
EclipseState (yet?).
copying the EclipseGrid object is required as the final set of active
cells requires knowledge of the grid used by the simulator which is
not available in opm-parser. In turn, this requires to call
EclipseGrid::resetACTNUM() which is non-const.
instead, the wrapper classes now call ERT directly, so it's easy to
see what a given class does. interestingly the amount of additional
code required is neglectible (or even negative).
ERT has been a requirement for opm-parser for a while now and
opm-parser is a prerequisite of opm-core, so ERT is always
available. Thus remove the stub code to avoid bitrot...
New function well_controls_clone(), implemented in terms of the
public API only, mirrors the objective of function clone_wells(),
only for well control sets. Add a basic test to demonstrate the
function too.
The 'cpty' field is for internal memory management purposes only.
No client can know of its existence, let alone inspect or directly
change the value, so it should not be used to adjudicate control set
equality. This was useful during the refactoring work to introduce
the opm-parser support, but its utility has since ceased.
Okay'ed by: [at] atgeirr and [at] joakim-hove
Current version executes reordered solve once for each tracer. The benefit
is a simpler code and the ability to use MDU with tracers. The cost is
potentially higher runtime, compared to doing a single sweep for all
tracers (and tof).
When this boolean parameter is true (the default), tracer solutions
will be normalized so that the tracer averages will sum to one in each cell.
This behaviour is the same as before, the change is that it can now be
turned off.
this basically means using Opm::EclipseState instead of the raw deck
for these keywords.
with this, property modifiers like ADD, MULT, COPY and friends are
supported for at least the PERM* keywords. If additional keywords are
required these can be added relatively easily as well.
no ctest regressions have been observed with this patch on my machine.
The queried keywords are unknown to the parser and after manual
inspection also to the Eclipse RM. There might be wrong keywords some
left, but these were to ones needed to get SPE9 started using
sim_fibo_ad...
the largest change is that all classes below opm/core/props/pvt take
the PVT region index as an argument, the higher-level ones (i.e.,
BlackoilProps*) take cell indices.
