tilelang.utils.sparse¶

Attributes¶

compress_util

Functions¶

compress_sm90(A, block_k, transposed)

compress_sm80(A, transposed)

compress(A, transposed[, arch])

Compress a tensor using the appropriate method based on the CUDA architecture.

Module Contents¶

tilelang.utils.sparse.compress_util¶
tilelang.utils.sparse.compress_sm90(A, block_k, transposed)¶
Parameters:

  • A (torch.Tensor)

  • block_k (int)

  • transposed (bool)

Return type:

tuple[torch.Tensor, torch.Tensor]

tilelang.utils.sparse.compress_sm80(A, transposed)¶
Parameters:

  • A (torch.Tensor)

  • transposed (bool)

Return type:

tuple[torch.Tensor, torch.Tensor]

tilelang.utils.sparse.compress(A, transposed, arch=None, **kwargs)¶

Compress a tensor using the appropriate method based on the CUDA architecture.

Parameters:

  • A (torch.Tensor)

  • transposed (bool)

  • arch (Optional[str])

Return type:

tuple[torch.Tensor, torch.Tensor]