field_bundle

FieldBundle

Field Bundle wraps a nD array (3D + n Data Dimensions) into a complex indexing scheme that allows a dual name-based and index-based access to the underlying memory. It is paired with the FieldBundleType which provides a way to type hint parameters for stencils in the gtscript.

WARNING: The present implementation only allows for 4D array.

name = bundle_name instance-attribute

quantity property

shape property

__init__(bundle_name, quantity, mapping=None, register_type=False)

Initialize a bundle from a nD quantity.

Dev note: current implementation limits to 4D inputs.

Parameters:

Name	Type	Description	Default
bundle_name	str	name of the bundle, accessible via name.	required
quantity	Quantity	data inputs as a nD array.	required
mapping	_FieldBundleIndexer | None	sparse dict of [name, index] to be able to call tracers by name.	None
register_type	bool	boolean to register the type as part of initialization.	False

map(index, name)

Map a single index to name

groupby(name)

Not implemented

index(name)

Get index from name.

extend_3D_quantity_factory(quantity_factory, extra_dims) staticmethod

Create a nD quantity factory from a cartesian 3D factory.

Parameters:

Name	Type	Description	Default
quantity_factory	QuantityFactory	Cartesian 3D factory.	required
extra_dims	dict[str, int]	dict of [name, size] of the data dimensions to add.	required

MarkupFieldBundleType dataclass

Markup a field bundle to delay specialization.

Properties

name: name of the future type to look into the registrar.

name instance-attribute

__init__(name)

FieldBundleType

Field Bundle Types to help with static sizing of Data Dimensions.

Methods:

Name	Description
register	Register a type by sizing its data dimensions
T	access any registered types for type hinting.

register(name, data_dims, dtype=Float) classmethod

Register a name type by name by giving the size of its data dimensions.

The same type cannot be registered twice and will error out.

Parameters:

Name	Type	Description	Default
name	str	Type name, to be re-used with T.	required
data_dims	tuple[int]	tuple of int giving size of each data dimensions.	required
dtype		Inner data type, defaults to Float.	Float

T(name, do_markup=True) classmethod

Get registered type.

Dev note: The markup feature is to allow early parsing (at file import) to go ahead - while we will resolve the full type when calling the stencil.

Parameters:

Name	Type	Description	Default
name	str	name of the type as registered via register	required
do_markup	bool	if name not registered, markup for a future specialization at stencil call time	True