Add vmapreduce function to resolve #110.

Author: chriselrod

docs/src/vectorized_convenience_functions.md

@@ -114,6 +114,22 @@ BenchmarkTools.Trial:
   evals/sample:     193
 ```
 
+## vmapreduce
 
+Vectorized version of `mapreduce`. `vmapreduce(f, op, a, b, c)` applies `f(a[i], b[i], c[i])` for `i in eachindex(a,b,c)`, reducing the results to a scalar with `op`.
+
+```julia
+julia> using LoopVectorization, BenchmarkTools
+
+julia> x = rand(127); y = rand(127);
+
+julia> @btime vmapreduce(hypot, +, $x, $y)
+  191.420 ns (0 allocations: 0 bytes)
+96.75538300513509
+
+julia> @btime mapreduce(hypot, +, $x, $y)
+  1.777 μs (5 allocations: 1.25 KiB)
+96.75538300513509
+```
 
 

src/LoopVectorization.jl

@@ -26,7 +26,7 @@ const NativeTypes = Union{Bool, Base.HWReal}
 export LowDimArray, stridedpointer, vectorizable,
     @avx, @_avx, *ˡ, _avx_!,
     vmap, vmap!, vmapnt, vmapnt!, vmapntt, vmapntt!,
-    vfilter, vfilter!
+    vfilter, vfilter!, vmapreduce, vreduce
 
 const VECTORWIDTHSYMBOL, ELTYPESYMBOL = Symbol("##Wvecwidth##"), Symbol("##Tloopeltype##")
 
@@ -58,6 +58,7 @@ include("condense_loopset.jl")
 include("reconstruct_loopset.jl")
 include("constructors.jl")
 include("user_api_conveniences.jl")
+include("mapreduce.jl")
 
 """
 LoopVectorization provides macros and functions that combine SIMD vectorization and

test/runtests.jl

@@ -45,6 +45,8 @@ Base.IndexStyle(::Type{<:FallbackArrayWrapper}) = IndexLinear()
 
     @time include("filter.jl")
 
+    @time include("mapreduce.jl")
+
     @time include("ifelsemasks.jl")
 
     @time include("dot.jl")