mul_op

→TensorImplPtr

int mul_op(const int & a, const int & b)

Public entry point — element-wise multiplication with autograd support.

Math

$y = a \cdot b$

See Also

add_op, sub_op, div_op

a, bconst TensorImplPtr&

Operands of any rank. Shapes must be NumPy-broadcast-compatible.

TensorImplPtr

Output tensor of shape broadcast_shapes(a.shape, b.shape) and promoted dtype. Participates in autograd via MulBackward.

LucidError

If a and b have incompatible shapes under NumPy broadcasting rules, or if their devices differ.

>>> auto c = mul_op(a, b);    // c = a * b (Hadamard / elementwise)