create_distributed_tensor_descriptor#

cuquantum.bindings.cutensornet.create_distributed_tensor_descriptor( intptr_t handle, int32_t num_modes, extents, element_strides, block_sizes, block_strides, nranks_per_mode, mode_labels, int data_type, ) → intptr_t[source]#

Creates a distributed tensor descriptor.

Parameters:

handle (intptr_t) – Opaque handle holding cuTensorNet’s library context and active distributed communicator.
num_modes (int32_t) – Number of tensor modes; must be greater than zero.
extents (object) –
Array of size num_modes containing the full tensor extents before distribution. It can be:
- an int as the pointer address to the array, or
- a Python sequence of int64_t.
element_strides (object) –
Optional array of size num_modes describing the local element layout. It can be:
- an int as the pointer address to the array, or
- a Python sequence of int64_t.
block_sizes (object) –
Optional array of size num_modes describing block sizes. It can be:
- an int as the pointer address to the array, or
- a Python sequence of int64_t.
block_strides (object) –
Optional array of size num_modes describing local physical spacing between blocks. It can be:
- an int as the pointer address to the array, or
- a Python sequence of int64_t.
nranks_per_mode (object) –
Array of size num_modes describing the process grid. It can be:
- an int as the pointer address to the array, or
- a Python sequence of int64_t.
mode_labels (object) –
Array of size num_modes; mode_labels[j] denotes the label of the j-th mode of the tensor. It can be:
- an int as the pointer address to the array, or
- a Python sequence of int32_t.
data_type (int) – Tensor element data type.

Returns:

On return, holds the created tensor descriptor.

Return type:

intptr_t