Claire Foster

> By the way, can the `unsafe_store!` path support union of isbits types like `Union{Float64,Missing}`? Excellent question, I checked the data layout and I think the answer is a conclusive...

> I checked the data layout and I think the answer is a conclusive "no for now" because tuples-of-unions are not inlined Actually very recently we have https://github.com/JuliaLang/julia/pull/34126 — if...

Agreed. The reason for this is that it's dispatching to Base and `similar(::SVector)` returns an `MVector`.

> I looked into this as I wanted to learn a bit more about Julia and this library, but, oh boy, this was (to a novice like me) a surprisingly...

> not having to mess about with generated functions (a plus in I.M.H.O.) Yes it's always nice to avoid `@generated`! > `function circshift(v::SVector{L}, shift::Integer) where L` You're not using `L`...

> What happens for big numbers and other mutable things that might not work well inside an MArray? There's a WIP PR to Base (https://github.com/JuliaLang/julia/pull/34126) which puts pointers inline, so...

Highly unlikely that `muladd` was avoided on purpose (I didn't write that code but I was sitting next to @andyferris when he originally wrote the package :-) ) Given that...

Regarding unrolling, it's important to keep in mind interactions between unrolling and type inference. In particular, any type inference opportunities which are exposed by unrolling will be invisible to the...

Regarding padding - adding any would completely change the memory layout of `Vector{SVector{...}}` which is somewhat questionable in my mind. But more importantly the memory overhead would be excessive for...

@chriselrod The detailed investigation you're doing here is great! Don't let me discourage you, it would be really great if `StaticArrays` would vectorize more effectively, and I am by no...