katgpucbf.dsim.shared_array module

Shared memory arrays.

class katgpucbf.dsim.shared_array.SharedArray(fd: int, shape: tuple[int, ...], dtype: dtype[Any] | None | type[Any] | _SupportsDType[dtype[Any]] | str | tuple[Any, int] | tuple[Any, SupportsIndex | Sequence[SupportsIndex]] | list[Any] | _DTypeDict | tuple[Any, Any])

Bases: object

An array that can be passed to another process.

Unlike multiprocessing.shared_memory, the shared memory used for this is not backed by a file, and so is guaranteed to be cleaned up when the processes involved die off, without the need for a manager process.

This is UNIX (probably Linux) specific.

Do not construct directly. Instead, either use create() to allocate a new array, or multiprocessing.connection.Connection.recv() to construct a new reference to an existing array in another process.

close() → None

Close the reference shared array and release the mapping.

Accessing the array after this will most likely crash. It is safe to call twice.

classmethod create(name: str, shape: tuple[int, ...], dtype: dtype[Any] | None | type[Any] | _SupportsDType[dtype[Any]] | str | tuple[Any, int] | tuple[Any, SupportsIndex | Sequence[SupportsIndex]] | list[Any] | _DTypeDict | tuple[Any, Any]) → SharedArray

Create a new array from scratch.

Parameters:

• name – An arbitrary name to associate with the array. See os.memfd_create().

• shape – Shape of the array. To simplify this function, it requires a tuple (a scalar cannot be used).

• dtype – The type of the array.