core.resharding.nvshmem_copy_service.memory.tensor_pointer_utils#

Utilities for extracting data pointers from different tensor types.

Supports PyTorch tensors, CuPy arrays, and raw integer pointers.

Module Contents#

Classes#

TensorPointerExtractor

Extract memory pointers from various tensor types.

API#

class core.resharding.nvshmem_copy_service.memory.tensor_pointer_utils.TensorPointerExtractor#

Extract memory pointers from various tensor types.

static get_pointer(tensor: Any) int#

Extract the data pointer from a tensor.

Parameters:

tensor – Can be torch.Tensor, CuPy array, or raw int pointer

Returns:

Memory address of the tensor data

Return type:

int

.. rubric:: Examples

import torch

t = torch.zeros(100, device=’cuda’)

ptr = TensorPointerExtractor.get_pointer(t)

isinstance(ptr, int)

True