Function broadcast_elementwise 

pub fn broadcast_elementwise<F>(
    a: &TensorView<'_>,
    b: &TensorView<'_>,
    out: &mut TensorViewMut<'_>,
    op: F,
) -> Result<(), String>
where
    F: Fn(f64, f64) -> f64 + Copy,

Element-wise binary operation on two broadcast-compatible views into an output.

This is the core primitive for all binary tensor operators (+, -, *, /). Broadcasting is handled via the stride-zero convention: a dimension with size 1 has stride 0, so the same element is reused without copying.

Safety

To prevent undefined behavior, the caller must ensure that the output buffer window does not overlap with the backing storage buffers of a and b unless an exact in-place operation is intended and safe.

Errors

Returns Err(String) if the shapes are not broadcast-compatible or the output buffer is too small for the broadcast output shape.