tilus.ir.RegisterTensor¶

class tilus.ir.RegisterTensor[source]¶

A tensor that resides in the register memory.

dtype¶

The data type of the tensor elements.

shape¶

The shape of the tensor.

optional_layout¶

The layout of the tensor, which is optional. When not provided, the layout will be automatically inferred with compiler pass.

__add__(other)[source]¶

Perform addition with another tensor or a scalar.

Parameters:: other (RegisterTensor | int | float | Expr) – The tensor or scalar to add to this tensor.
Returns:: ret – A new tensor that is the result of the addition.
Return type:: RegisterTensor

__mul__(other)[source]¶

Perform multiplication with another tensor or a scalar.

Parameters:: other (RegisterTensor | int | float | Expr) – The tensor or scalar to multiply with this tensor.
Returns:: ret – A new tensor that is the result of the multiplication.
Return type:: RegisterTensor

__sub__(other)[source]¶

Perform subtraction with another tensor or a scalar.

Parameters:: other (RegisterTensor | int | float | Expr) – The tensor or scalar to subtract from this tensor.
Returns:: ret – A new tensor that is the result of the subtraction.
Return type:: RegisterTensor

__truediv__(other)[source]¶

Perform division with another tensor or a scalar.

Parameters:: other (RegisterTensor | int | float | Expr) – The tensor or scalar to divide this tensor by.
Returns:: ret – A new tensor that is the result of the division.
Return type:: RegisterTensor

static create(dtype, *, shape, optional_layout=None)[source]¶

Create a RegisterTensor with the given dtype, shape, and optional layout.

Parameters:

dtype (DataType) – The data type of the tensor elements.
shape (Sequence[int]) – The shape of the tensor.
optional_layout (RegisterLayout, optional) – The layout of the tensor. If not provided, the layout will be inferred later.

Returns:

ret – The created RegisterTensor instance.

Return type:

RegisterTensor

property layout: RegisterLayout¶

Get the layout of the RegisterTensor.

Returns:: ret – The layout of the RegisterTensor.
Return type:: RegisterLayout
Raises:: ValueError – If the layout of the RegisterTensor is not defined yet.

squeeze(dim)[source]¶

Squeeze the tensor by removing dimensions of size 1.

Parameters:: dim (int | Sequence[int]) – The dimension(s) to squeeze. If an integer is provided, it will squeeze that specific dimension. If a sequence of integers is provided, it will squeeze all specified dimensions.
Returns:: ret – A new tensor with the specified dimensions squeezed.
Return type:: RegisterTensor

tilus.ir.RegisterTensor