bitwise_xor

→Tensor

bitwise_xor(input: Tensor, other: Tensor | Scalar)

source edit

Implementing kernel

C++bitwise_xor_opfree fn

Element-wise bitwise XOR.

Element-wise bitwise operation on integer (or boolean) operands. For boolean inputs this reduces to the corresponding logical operation. Floating-point inputs are not supported and will raise.

Parameters

inputTensor

Left operand. Must have integer or boolean dtype.

otherTensor or scalar

Right operand. Must have integer or boolean dtype. Broadcasts against input.

Returns

Tensor

Tensor with shape broadcast(input.shape, other.shape) and dtype given by the usual integer type-promotion rules.

Notes

Bit-level definition (for each bit position k of every element):

\text{out}_i[k] = \text{input}_i[k] \oplus \text{other}_i[k]

For boolean tensors this is the same as logical_xor. Bitwise operations are not differentiable; gradients are dropped.

Examples

>>> import lucid
>>> a = lucid.tensor([0b1100, 0b1010], dtype=lucid.int32)
>>> b = lucid.tensor([0b1010, 0b0110], dtype=lucid.int32)
>>> lucid.bitwise_xor(a, b)
Tensor([...])