complex_op

→TensorImplPtr

int complex_op(const int & re, const int & im)

source edit

Python wrappers

pylucid.ops.complexfunction

Combine two real tensors into a complex tensor.

The result has dtype C64 and the same shape and device as the inputs. Both inputs must be real-floating (F16 / F32 / F64); the resulting interleaved storage holds $a + b\,i$ element-wise.

Math

z_k = a_k + b_k\,i

See Also

real_op, imag_op, conj_op

Parameters

reTensorImplPtr

Real part. Must be a real-floating dtype.

imTensorImplPtr

Imaginary part. Same shape, device, and dtype family as re.

Returns

TensorImplPtr

Complex (C64) tensor with the same shape and device as the inputs.

Raises

DtypeMismatch

If either input is not a real-floating dtype.

ShapeMismatch

If re and im have different shapes.

DeviceMismatch

If re and im live on different devices.

Notes

Backward is supplied by the Python autograd layer via real / imag extraction of the incoming complex gradient; this entry point only implements the forward.