implementations.datatype_classes.mesh module

class MultiComponentMesh(init, *args, **kwargs)[source]

Bases: mesh

Generic mesh with multiple components.

To make a specific multi-component mesh, derive from this class and list the components as strings in the class attribute components. An example:

``` class imex_mesh(MultiComponentMesh):

components = [‘impl’, ‘expl’]

```

Instantiating such a mesh will expand the mesh along an added first dimension for each component and allow access to the components with .. Continuing the above example:

` init = ((100,), None, numpy.dtype('d')) f = imex_mesh(init) f.shape  # (2, 100) f.expl.shape  # (100,) `

Note that the components are not attributes of the mesh: "expl" in dir(f) will return False! Rather, the components are handled in __getattr__. This function is called if an attribute is not found and returns a view on to the component if appropriate. Importantly, this means that you cannot name a component like something that is already an attribute of mesh or numpy.ndarray because this will not result in calls to __getattr__.

There are a couple more things to keep in mind:

  • Because a MultiComponentMesh is just a numpy.ndarray with one more dimension, all components must have the same shape.

  • You can use the entire MultiComponentMesh like a numpy.ndarray in operations that accept arrays, but make sure that you really want to apply the same operation on all components if you do.

  • If you omit the assignment operator [:] during assignment, you will not change the mesh at all. Omitting this leads to all kinds of trouble throughout the code. But here you really cannot get away without.

components = []
class comp2_mesh(init, *args, **kwargs)[source]

Bases: MultiComponentMesh

components = ['comp1', 'comp2']
class imex_mesh(init, *args, **kwargs)[source]

Bases: MultiComponentMesh

components = ['impl', 'expl']
class mesh(init, val=0.0, offset=0, buffer=None, strides=None, order=None)[source]

Bases: ndarray

Numpy-based datatype for serial or parallel meshes. Can include a communicator and expects a dtype to allow complex data.

_comm

MPI communicator or None

bcast(root=None, comm=None)[source]

Routine for broadcasting values

Parameters:

  • root (int) – process with value to broadcast

  • comm – communicator

Returns:

broadcasted values

property comm

Getter for the communicator

irecv(source=None, tag=None, comm=None)[source]

Routine for receiving in time

Parameters:

  • source (int) – source rank

  • tag (int) – communication tag

  • comm – communicator

Returns:

None

isend(dest=None, tag=None, comm=None)[source]

Routine for sending data forward in time (non-blocking)

Parameters:

  • dest (int) – target rank

  • tag (int) – communication tag

  • comm – communicator

Returns:

request handle