Create vsum

[MilesCranmer]

This creates a vectorized version of sum, as has been done with vmapreduce and vreduce already. The implementation is just an alias:

@inline vsum(f::F, A::AbstractArray{T}) where {F,T<:NativeTypes} = vmapreduce(f, +, A)
@inline vsum(A::AbstractArray{T}) where {T<:NativeTypes} = vsum(identity, A)

(There is a vsum in VectorizationBase.jl; that one is I think (?) an internal utility, as it doesn't take normal arrays.)

For example:

julia> using LoopVectorization, BenchmarkTools

julia> x = rand(127);

julia> @btime vsum(hypot, $x)
  12.095 ns (0 allocations: 0 bytes)
66.65246070098374

julia> @btime sum(hypot, $x)
  16.992 ns (0 allocations: 0 bytes)
66.65246070098372

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: -1.13 ⚠️

Comparison is base (d0c61d3) 86.78% compared to head (1063410) 85.65%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #496      +/-   ##
==========================================
- Coverage   86.78%   85.65%   -1.13%     
==========================================
  Files          39       39              
  Lines        9524     9524              
==========================================
- Hits         8265     8158     -107     
- Misses       1259     1366     +107     

Impacted Files	Coverage Δ
src/LoopVectorization.jl	100.00% <ø> (ø)
src/simdfunctionals/mapreduce.jl	92.63% <100.00%> (+0.15%)	⬆️

... and 5 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[chriselrod]

Sure.
If people want much more than this, it might also be good to point people to:
https://github.com/andrewjradcliffe/VectorizedReduction.jl
https://github.com/JuliaSIMD/VectorizedStatistics.jl

[chriselrod]

You need to export vsum if you're testing that it is exported.
The oftype failures on nightly are unrelated to this PR.

[MilesCranmer]

Fixed, thanks.

That package looks super useful, I didn’t know about it. It might be a good idea to ask and try to bring it under the JuliaSIMD organization, for consistency and long term support. I think having it seem like a more “official” package would make me more likely to use it.

[MilesCranmer]

Oh I didn’t know about VectorizedStatistics! Should I close this PR?

[chriselrod]

vsum is simple enough so that it's fine. But maybe we shouldn't export it if they do.

@andrewjradcliffe you're welcome to move your package to JuliaSIMD if you'd like.