For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

transpose_2d

def transpose_2d[rank: Int, dtype: DType, //](output: TileTensor[dtype, Storage=output.Storage, linear_idx_type=output.linear_idx_type], input: TileTensor[dtype, Storage=input.Storage, linear_idx_type=input.linear_idx_type], perms: Pointer[Int, _safe=False], simplified_input_shape: IndexList[rank], simplified_rank: Int, offset: Int, ctx: Optional[DeviceContext] = None)

Transposes the inner two dimensions of a simplified rank-2 tensor.

Selects a serial tiled implementation or a parallel tiled implementation based on the problem size and available parallelism.

Parameters:

rank (Int): Number of dimensions in the simplified_input_shape array (inferred).
dtype (DType): Element type of the input and output buffers (inferred).

Args:

output (TileTensor[dtype, Storage=output.Storage, linear_idx_type=output.linear_idx_type]): The output buffer with the inner two dimensions swapped.
input (TileTensor[dtype, Storage=input.Storage, linear_idx_type=input.linear_idx_type]): The input buffer.
perms (Pointer[Int, _safe=False]): Permutation of the input axes.
simplified_input_shape (IndexList[rank]): Shape of the tensor after simplification.
simplified_rank (Int): Effective rank after simplification.
offset (Int): Flat offset added to both input and output pointers.
ctx (Optional[DeviceContext]): The context to execute the work on.