Create vsum (#496)

* Create `vsum`

* Add docs and unittests for `vsum`

* Export `vsum`

Author: MilesCranmer

docs/src/vectorized_convenience_functions.md

@@ -132,4 +132,22 @@ julia> @btime mapreduce(hypot, +, $x, $y)
 96.75538300513509
 ```
 
+## vsum
+
+Vectorized version of `sum`. `vsum(f, a)` applies `f(a[i])` for `i in eachindex(a)`, then sums the results.
+
+```julia
+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
+```
+
 

src/LoopVectorization.jl

@@ -196,6 +196,7 @@ export LowDimArray,
   vfilter,
   vfilter!,
   vmapreduce,
+  vsum,
   vreduce,
   vcount
 
@@ -245,6 +246,7 @@ loop-reordering so as to improve performance:
   - [`@turbo`](@ref): transform `for`-loops and broadcasting
   - [`vmapreduce`](@ref): vectorized version of `mapreduce`
   - [`vreduce`](@ref): vectorized version of `reduce`
+  - [`vsum`](@ref): vectorized version of `sum`
   - [`vmap`](@ref) and `vmap!`: vectorized version of `map` and `map!`
   - [`vmapnt`](@ref) and `vmapnt!`: non-temporal variants of `vmap` and `vmap!`
   - [`vmapntt`](@ref) and `vmapntt!`: threaded variants of `vmapnt` and `vmapnt!`

src/simdfunctionals/mapreduce.jl

@@ -1,3 +1,4 @@
+import VectorizationBase: vsum
 
 @inline vreduce(::typeof(+), v::VectorizationBase.AbstractSIMDVector) = vsum(v)
 @inline vreduce(::typeof(*), v::VectorizationBase.AbstractSIMDVector) = vprod(v)
@@ -107,6 +108,16 @@ end
 end
 @inline vmapreduce(f, op, args...) = mapreduce(f, op, args...)
 
+"""
+    vsum(A::DenseArray)
+    vsum(f, A::DenseArray)
+
+Vectorized version of `sum`. Providing a function as the first argument
+will apply the function to each element of `A` before summing.
+"""
+@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)
+
 length_one_axis(::Base.OneTo) = Base.OneTo(1)
 length_one_axis(::Any) = 1:1
 

test/mapreduce.jl

@@ -60,6 +60,9 @@
     end
     @test vmapreduce(log, +, x) ≈ sum(log, x)
     @test vmapreduce(abs2, +, x) ≈ sum(abs2, x)
+    @test vsum(log, x) ≈ sum(log, x)
+    @test vsum(abs2, x) ≈ sum(abs2, x)
+    @test vsum(x) ≈ sum(x)
     @test maximum(x) == vreduce(max, x) == maximum_avx(x)
     @test minimum(x) == vreduce(min, x) == minimum_avx(x)