Concat and split operations

[clarfonthey]

I've recently found myself coding up shorthand methods for operations like:

fn concat(lhs: f64x4, rhs: f64x4) -> f64x8;
fn split(val: f64x8) -> (f64x4, f64x4);

which are shorthands for their various swizzle representations. I tried messing around with the code for the crate to make these work but got stuck with dependencies on generic_const_exprs and couldn't find a way to make it work.

Since the simd_swizzle macro can be a bit overwhelming, it would be nice to have more shorthands like these, but I'm not sure exactly how it would fit within the existing design plans, so, I figured I'd just start a discussion here.

One big problem with these methods under the existing, general implementation is checking that a size is actually supported: for example, f64x16 doesn't exist but f32x16 does.

[programmerjake]

One big problem with these methods under the existing, general implementation is checking that a size is actually supported: for example, f64x16 doesn't exist but f32x16 does.

Wether or not an element count is supported is independent of the element type, we intentionally decided on that to make generic code easier to write/use. Just the f64x16 type alias is missing, Simd<f64, 16> works:
https://play.rust-lang.org/?version=nightly&mode=release&edition=2021&gist=93d6280b54e054de94bbf17c9c1227d4

[clarfonthey]

Oh, interesting. Well, regardless, there is an upper limit to the lane count, e.g. Simd<T, 128> isn't allowed. Unless the goal is to support all powers-of-two usize, which... I'm assuming isn't the case, given the SupportedLaneCount trait.

[calebzulawski]

Not that I necessarily recommend it, but this would be possible to implement with a trait instead of generic const expressions. Either way, a function like this is probably useful.

[clarfonthey]

So, I did try and make it work with an extension trait called DoubleableLaneCount and due to limitations in the existing const generics impl, associated constants sometimes aren't allowed in generic const position. Could work on a trait that explicitly returns the Simd type, however, but I didn't see any existing ones.