Chris Elrod

> Are those the intended results? No. > Also, should ArrayInterface.size produce an error if given a Type as an argument? That's what it does for `Matrix{Float64}`, for example. I'm...

It should probably be defined like this: ```julia julia> itersize(::Type{ itersize(SArray{Tuple{3,4}}) Base.HasShape{2}() julia> itersize(SArray{Tuple{3,4,5}}) Base.HasShape{3}() ``` So that they only need the first parameter.

> or otherwise throws an "indeterminate size" error? I have something like this in my own code. I'd prefer to return `nothing`s and let the user decide how to handle...

Both `1`s above should be static, otherwise performance with `LoopVectorization.jl` will be garbage. `static(1)` is needed for avoiding the extremely slow gather and scatter instructions.

Why is `D`'s first stride `1`? I'll have to look at this later. Does base have special casing based on array size changing how cartesian indices are mapped to linear?...

Interesting, those strides are indeed correct. So, when you reshape an axis, the reshape is in column-major order. Hence ```julia julia> C = randn(1,60); julia> D = reshape(C', (3,4,5)); ```...

> Does this break anything. How many libraries are currently using `ArrayInterface.jl`? It's easy enough to test a small handful locally, like `DifferentialEquations.jl`. Barring breaking a bunch of stuff, I...

FWIW, my experimentation didn't show llvm actually use `invariant` information to optimize in the cases I wanted it to/thought it should. Although there is still a lot of higher level...

> I just seem to get segfaults though. Yikes. I'll look into it. EDIT Travis says: ``` signal (11): Segmentation fault in expression starting at /home/travis/build/SciML/StochasticDiffEq.jl/test/first_rand_test.jl:23 macro expansion at /home/travis/.julia/packages/SIMDPirates/ShWZm/src/memory.jl:131...

What is `_seed`? My fault for not documenting ```julia help?> local_rng() No documentation found. ``` But `local_rng`'s argument is the thread id - 1. If you pass it an integer...