Whenever GridFunction::ProjectDiscCoefficient* was called with the device being run the gridfunction didn't necessarily have the gridfunction data on the host thus causing errors when the AccumulateAndCount* functions were called.
* The parameter of Vector::UseDevice(bool) no longer has a default value. All
the calls to UseDevice() in operator.cpp bilinearform_ext.cpp gridfunc.?pp and
linearform.?pp have been replaced with UseDevice(true).
* Added shortcuts for the device flags of the Memory objects inside the Vector
and Array classes:
bool Array::UseDevice()
bool Vector::UseDevice()
* The internal device flag in Memory::FlagMask is now called USE_DEVICE (it was
previously called EXEC_FLAG). The accessor function for this flag have been
renamed:
bool Memory::GetExecFlag() -> bool Memory::UseDevice()
void Memory::SetExecFlag(bool) -> void Memory::UseDevice(bool)
* Further renamed:
Memory::SyncWith -> Memory::Sync
Memory::SyncAliasToBase -> Memory::SyncAlias
Memory::SyncAliasToBase_ -> Memory::SyncAlias_
* Replaced FIXME with TODO in general/mem_manager.cpp
and transfers.
The Memory class is now used by some MFEM classes (like Array and
Vector) which can be used on the Device. Such classes now provide
methods to access the underlying Memory object, e.g. GetMemory.
Updated ex1/ex1p and ex6/ex6p to not need to enable/disable the
Device at specific points -- the Device is now enabled just at the
start. Also, the same examples can now run on Device (e.g. -d cuda)
without the partial assembly option (-pa) -- full assembly will
be still done on CPU but the sparse matrix action and vector
operations will be done using the Device.
Reverted changes in class DenseMatrix related to using the Device.
At this point, DenseMatrix operations are only used for small matrices
and using the Device in this case is not a good option.
ZienkiewiczZhuEstimator.
This support is based on the update in ParGridFunction::ComputeFlux
which, in turn, is based on the support for GroupCommunicators in
ParFiniteElementSpaces on non-conforming meshes.
The optimization is to use the (potentially cached) inverse Jacobian
matrix from the ElementTransformation class instead of computing
it from the Jacobian.
This fixes is the same issue is in ProjectBdrCoefficient which was
fixed by the previous commit.
An outstanding issue is the handling of nonconforming meshes in these
two methods - I left 'FIXME' comments there as reminders.
workaround for nonconforming meshes.
The workaround is to disable the parallel synchronization.
Also, added a debug check that all boundary dofs are set.
class GridFunction when used in parallel.
The issue appears when the master processor for a boundary DOF does
not own any of the boundary elements adjacent to that DOF.
The method ProjectBdrCoefficientTangent is probably affected by the
same issue, however, it is not addressed by this commit.
List of changes and additions:
* In class Mesh:
* Replaced the data fields BaseGeom, BaseBdrGeom, and BaseFaceGeom
with a new data field: mesh_geoms that is a bitwise mask for all
Geometry::Types found in the (local) Mesh including entities of
lower dimensions.
* The new data field mesh_geoms is set by the method SetMeshGen.
* Removed the (protected) method InitBaseGeom.
* Add new methods: HasGeometry, GetNumGeometries, and GetGeometries
that can be used to query the content of mesh_geoms. Also, added
class Mesh::GeometryList that can simplify such queries too.
* The methods GetFaceBaseGeometry, GetElementBaseGeometry, and
GetBdrElementBaseGeometry no longer have a default value for
their input parameter (the entity index) and accept only values
within their respective ranges.
* Removed the parameter of the method GetRefinementTransforms.
* Updated the methods PrepareNodeReorder and DoNodeReorder to
support mixed meshes.
* In parallel, the data field meshgen is now required to represent
information about the global mesh - updated relevant ParMesh
methods.
* Made any required changes outside of class Mesh due to the above
changes.
* Made various tweaks and improvements in the following methods:
CheckBdrElementOrientation, Mixed2DUniformRefinement,
Mixed3DUniformRefinement, UniformRefinement, and
GeneralRefinement.
* Updated the method PrintCharacteristics to print the types of
elements, boundary elements and faces used in the mesh.
* In the method ReadInlineMesh, replace remaining "pri" strings
with "wedge".
* In the mesh reader method ReadGmshMesh, recover the term "prism"
in the comments - it was replaced, unintentionally, I assume,
by "wedge".
* Add support for refinement of curvilinear mixed meshes:
* In class FiniteElementSpace, updated the methods:
GetLocalRefinementMatrices, GetLocalDerefinementMatrices,
RefinementMatrix, RefinementMatrix_main, as well as class
RefinementOperator to support mixed meshes.
* Add two new meshes: star-mixed-p2.mesh and fichera-mixed-p2.mesh
which represent examples of quadratic mixed meshes in 2D and 3D,
respectively.
* Add sample runs in ex1/ex1p using the new quadratic mixed meshes.
* In class Geometry:
* In the enumeration Type: removed the constant MIXED and added
new constant: NUM_GEOMETRIES.
* Add new static data field: DimStart[] that indexes the Types by
dimension: the Types of dimension 'd' are given by the range
[ DimStart[d], DimStart[d+1] ).
* In class ParNCMesh, added method GetGhostFaceGeometry that,
for now, returns Geometry::SQUARE.
* Update the return type of the method DofOrderForOrientation in
class FiniteElementCollection (and all derived classes) from
'int *' to 'const int *'.
* Removed the virtual method Element::GetRefinementFlag and made
the only relevant derived version (in class Tetrahedron),
non-virtual.
* In class Tetrahedron, added method Init.
* Added FIXME comments for some outstanding issues.
* In the mesh-explorer miniapp:
* Added option to enable or disable the 'refine' parameter when
reading a mesh from file.
* In the 'jitter' menu option, ask the user if the boundary
nodes should be moved or not.
* In the 'Print sub-element stats' menu option, count and print
bad/inverted elements by their Geometry::Type as well.
William F Godoy