C++ or Swift equivalents of this python indexing/slicing code?

[dougdew64]

It's not clear to me how we can provide to C++ or Swift developers the same convenience which is provided to python developers.

For example, in this code from positional_encoding.py, there are several uses of array indexing/slicing which can be conveniently expressed in python but probably not so conveniently in C++ or Swift:

def _compute_rope(self, costheta, sintheta, x):
        x1 = x[..., : self.dims // 2]
        x2 = x[..., self.dims // 2 : self.dims]
        rx1 = x1 * costheta - x2 * sintheta
        rx2 = x1 * sintheta + x2 * costheta

        if self.dims < x.shape[-1]:
            rx = mx.concatenate([rx1, rx2, x[..., self.dims :]], axis=-1)
        else:
            rx = mx.concatenate([rx1, rx2], axis=-1)

        return rx

It seems that the python compiler provides special help in generating indexing/slicing objects from indexing/slicing expressions so that the MLX python/C++ stubs can have signatures such as:

array mlx_get_item(const array& src, const py::object& obj) { ... }

and

void mlx_set_item(array& src, const py::object& obj, const ScalarOrArray& v) { ... }

Absent similar help from the C++ compiler and the Swift compiler, it seems that developers will end up having to author calls to functions such as scatter rather than enjoy the convenience of indexing/slicing and the pile of supporting logic which is contained in indexing.cpp. That is what I'm struggling with while attempting to implement a C++ version of that python _compute_rope function.

Has anybody already figured this out? If so, I hope that they will share C++ or Swift equivalents of the _compute_rope logic above.

[dougdew64]

Maybe this is a non-issue for swift developers if there's no expectation that swift developers would ever have to index/slice arrays.

And maybe C++ developers who are implementing NN layers are accustomed to not having nice things such as python's array indexing/slicing functionality.

[dougdew64]

https://github.com/python/cpython/blob/main/Objects/sliceobject.c