Chris Elrod

A major contributor to `StaticArray`'s overhead is that `NTuple`s aren't actually treated specially in Julia, which should eventually be fixed: https://github.com/JuliaLang/julia/pull/31681 > From some command line checks this doesn't seem...

I don't have plans, but should at least add SIMD exponential sampling. Note that you may have to change your algorithm for SIMD sampling. For example, the [ziggurat algorithm](https://en.wikipedia.org/wiki/Ziggurat_algorithm) is...

> Ah, `DiscreteNonParametric` currently does not work, since it uses the `Sampler` interface. I just added support for the `Sampler` interface. Note that this won't get full performance, just like...

> I have the poisson sampler below, but it allocates for some reason. I assumed this was due to a type instability, but haven't been able to find one. Add...

For comparison, on Cascadelake: ```julia julia> using VectorizedRNG, Random julia> x = Vector{Float64}(undef, 1024); julia> @benchmark randn!(local_rng(), $x) BenchmarkTools.Trial: 10000 samples with 10 evaluations. Range (min … max): 1.183 μs...

This is a LoopVectorization broadcasting issue. When contiguous axis are dynamically broadcasted, the behavior is undefined. That is, when an axis that would normally be contiguous in memory is of...

> Or at least I think my problem is the same?? No, actually. Fixed with LoopVectorization 0.12.163 + StrideArraysCore 0.4.17.

Sorry for dragging my feet on this. I'll take a look in a couple hours.

I've implemented a hotfix in #62 by disabling the change (`can_turbo`) that caused the regression in the first place. I do think `can_turbo` is useful with the way LV currently...

> tests have been consistently failing since last August, which isn't exactly a good look. Anyone who doesn't like the look is more than welcome to address it.