MPIArray

class caput.mpiarray.MPIArray(shape, dtype=float, buffer=None, offset=0, strides=None, order=None)

Bases: numpy.ndarray

A numpy-like array distributed across multiple processes.

Parameters:

global_shapeSequence[int]

The global shape of the array.

axisint, optional

The axis over which the array is distributed. Default is 0.

commMPI.Comm | None, optional

The communicator over which the array is distributed. Default is None, which will use MPI.COMM_WORLD.

argsAny, optional

Additional positional arguments to pass to the ndarray constructor.

**kwargsAny, optional

Additional keyword arguments to pass to the ndarray constructor.

property axis: int

Distributed axis.

Returns:

distributed_axisint

property comm: mpi4py.MPI.Comm

The communicator over which the array is distributed.

Returns:

commMPI.Comm

property global_shape: tuple[int, Ellipsis]

Global array shape.

Returns:

global_shapetuple[int, …]

property global_slice: _global_resolver

Return an objects that presents a view of the array with global slicing.

Returns:

global_slice_resolver_global_resolver

_global_resolver instance.

property local_array: numpy.ndarray

rank-local numpy.ndarray view of the array.

Returns:

rank_local_arrayndarray

property local_bounds: slice

Global bounds of the local array along the distributed axis.

Returns:

rank_local_boundsslice

property local_offset: tuple[int, Ellipsis]

Offset into global array.

This is equivalent to the global-index of the [0, 0, …] element of the local section.

Returns:

rank_local_offsettuple[int, …]

property local_shape: tuple[int, Ellipsis]

Shape of local section.

Returns:

rank_local_shapeint

Methods

__array_finalize__(→ None)	Finalizes the creation of the MPIArray, when viewed.
__array_ufunc__(→ MPIArray | tuple[MPIArray, ...)	Handles ufunc operations for MPIArray.
__getitem__(→ numpy.ndarray)	Returned a slice view of this array.
allgather(→ numpy.ndarray)	Gather a full copy onto each rank.
allreduce(→ numpy.number)	Perform MPI reduction op within memory buffer.
copy(→ MPIArray)	Return a copy of the MPIArray.
enumerate(→ collections.abc.Iterator[tuple[int, int]])	Helper for enumerating over a given axis.
from_file(→ MPIArray)	Read MPIArray from an HDF5 dataset or Zarr array on disk in parallel.
from_hdf5(→ MPIArray)	Read MPIArray from an HDF5 dataset in parallel.
gather(→ numpy.ndarray | None)	Gather a full copy onto a specific rank.
ravel(→ None)	Method is explicitly not implemented.
redistribute(→ MPIArray)	Change the rank distribution axis of the array.
reshape(→ MPIArray)	Reshape the array.
to_file(→ None)	Parallel write into a contiguous HDF5/Zarr dataset.
to_hdf5(→ None)	Parallel write into a contiguous HDF5 dataset.
to_zarr(→ None)	Parallel write into a contiguous Zarr dataset.
transpose(→ MPIArray)	Transpose the array axes.
wrap(→ MPIArray)	Turn ndarray`s spread across ranks into a distributed :py:class:.MPIArray` object.