Chris Elrod

LoopVectorization evaluates your code using `SVecs`. An `SVec{W,T}` contains `W` elements of type `T`. Most operations on `SVec` are defined using SIMD instructions. It's basically how LoopVectorization says "apply single...

> Is there any chance that similar cases will be handled as in the first example (i.e., naive implementation) in future releases? Yes, I'd definitely accept a PR adding support...

Thanks for the issue. I updated VectorizationBase, SIMDPirates, and LoopVectorization so that now, on the master branches of each: ```julia julia> using LoopVectorization, BenchmarkTools julia> x = rand(83); julia> x[rand(1:83,13)]...

I'll take a look at these later, but I'll need to add explicit support for `Union{T,Missing}`. That's important in Statistics, and it'd be great if `vfilter` would work on them...

VectorizationBase.jl may be a better place for this issue, but this is fine too. `erf` only gets partial credit at the moment. If someone wants to work on it, some...

However it isn't always possible, e.g. when implementations are broken up to behave differently depending on range, e.g. Taylor series for small values vs recurrences for large ones. Also of...

> If for many functions it is just a matter of broadening the allowed input types, would it make sense to just have SpecialFunctions depend on VectorizationBase? They'd also need...

No, not yet. Sometime after #17 we could add complex number support. For now, I'd follow the [dealing-with-structs](https://github.com/chriselrod/LoopVectorization.jl#dealing-with-structs) approach from the README. Note that `StructArray`s will be faster than the...

The complex support that will be added will be worse than using StructArrays, so you won't do better than the `test_avx!` implementation (barring other improvements in the library). Depending on...

Thanks @lesshaste , I'll check out those bug reports.