Advection data storage

[timspainNERSC]

Advection data storage

Fixes #782

---

Change Description

Share data between the column physics and dynamics parts of the module without copying. Tidy up some related classes and code that are no longer used and became a problem as these changes were implemented.

• Create a distinction between ModelArray::Components (all the DG components at a particular physical location) and a ModelArray::Component (a particular DG component, especially DG0, at all grid locations).
  • A set of Components map to an Eigen::Array::row.
  • A Component maps to a an Eigen::Array::col.
• Add functions to access a set of Components.
• Add functions to access a specific Component.
• Allow a ModelArray to be cast to and from ModelArray::DataType, that is Eigen::Array.
• Make the ModelComponent ocean mask into an array held in a static function.
• Move the thickness and concentration arrays with all DG components from the dynamics kernel to PrognosticData.
• Add a function to PrognosticData to copy all components of one ModelArray to the correct components of another.
• Update PrognosticData::updatePrognosticData to work correctly with the new full DG arrays.
• The output from PrognositcData is now all DG components in all cases.
• Add a complementary function to update the arrays after the dynamics runs.
  • Eventually this function will do nothing and should be removed.
• Create a map between a type with multiple components and the equivalent type with only one component.
  • For example ModelArray::Type::DG maps to ModelArray::Type::H.
• Add a const pseudo-Type that is the type that should be used for advected fields. Defined in the relevant ModelArrayDetails.hpp.
• Add a using typedef that defines an AdvectionField, which is just ModelArray.
• Remove the function ModelArrayRef::data(). The underlying data array is no longer accessible (this way).
• Add the four basic arithmetic operators to ModelArrayRef for double and ModelArray arguments.
• Make ModelArrayRef work only on the 0 component of the array it references.
  • Except for the allComponents() member function which returns a reference to the full array, with all components (also known as the second Eigen dimension).
• Add TextTags for read-write access to the cell-averaged DG fields.
• Adjust the dynamics implementations (core/src/modules/DynamicsModule/*.cpp) to work with an external source of data for thickness and concentration.
• It's not necessary to use ModelArrayRef::data() when passing ModelArrayRefs to the dynamics kernels' setData() functions.
• Create DGVectorHolder class which can point to a DG ModelArray (or at least its underlying Eigen::Array) and can decay to a DGVector&.
  • Use DGVectorHolders for ice thickness and concentration in the dynamics.
  • Add exception errors if the setData() function is used to try to set thickness or concentration.
  • Add a function that set the DGVectorHolder data.
• Added a setVelocites function to DGTransport
• Clean up module functions and features that have never really been used.
  • The Module registry in ModelComponent.
  • The advectionStore in ModelComponent. The pre-existing store works just fine.
  • The hFields (&c.) lists of field names from ModelComponent and derived classes.
  • Deleted from the dynamics source tree:
    • cgParametricMomentum.hpp
    • cgParametricMomentum.cpp
    • BBM.hpp
    • MEB.hpp
    • mevp.hpp

---

Test Description

• Added a test for the new DGVectorHolder class.
• Test the access of ModelArrayRef to only the 0 component.
• Test the access of ModelArrayRef::allComponents to all components.
• Test the component(s) functions of ModelArray.
• Test the direct access of a ModelArray to the underlying data type.
• Reconfigure the ModelComponent test since there is no longer a Module registry.
• Add a ModelArray::Type with components to the test ModelArrayDetails.
• Adjusted the advection tests to use the setVelocities function of DGTransport.
• Test creating a DGVector directly from Eigen::Matrix and Eigen::Array.
• The ThermoIntegration test now only checks the DG0 component of the read ice thickness.

---

Documentation Impact

Added docs/data_storage.rst, which documents how data is shared (after applying this PR) between the column physics and dynamics parts of the model.

[timspainNERSC]

@winzerle I've added you as a reviewer as this PR changes how the data is passed through the dynamics, and I would like a check that it doesn't break anything. The output of the tests and usual example runs looks fine, at least to me.

[winzerle]

Looks good although it's a bit difficult since many files are touched. But interface to transport seems ok.

It might be good to do a simple speed-check before and after applying such spaces to see, if there are any efficiency issues.

[einola]

Not an in-depth review, but this looks good to me.

I suspect you forgot to update the VP and freedrift modules, and we need to do something about the advection of snow, temperature and damage.

[timspainNERSC]

I suspect you forgot to update the VP and freedrift modules

Yes, I did!

and we need to do something about the advection of snow, temperature and damage.

Coming soon, as a separate issue and PR.

[timspainNERSC]

Looks good although it's a bit difficult since many files are touched. But interface to transport seems ok.

It might be good to do a simple speed-check before and after applying such spaces to see, if there are any efficiency issues.

It should be (slightly) faster, as there is no longer a copy in each direction when going to and from the dynamics.