Use reduced_max and reduced_min from SIMDPirates to fix #111.

Author: chriselrod

src/LoopVectorization.jl

@@ -8,7 +8,7 @@ using VectorizationBase: REGISTER_SIZE, REGISTER_COUNT, extract_data, num_vector
     AbstractColumnMajorStridedPointer, AbstractRowMajorStridedPointer, AbstractSparseStridedPointer, AbstractStaticStridedPointer,
     PackedStridedPointer, SparseStridedPointer, RowMajorStridedPointer, StaticStridedPointer, StaticStridedStruct,
     maybestaticfirst, maybestaticlast, scalar_less, scalar_greater
-using SIMDPirates: VECTOR_SYMBOLS, evadd, evsub, evmul, evfdiv, vrange, reduced_add, reduced_prod, reduce_to_add, reduce_to_prod, vsum, vprod, vmaximum, vminimum,
+using SIMDPirates: VECTOR_SYMBOLS, evadd, evsub, evmul, evfdiv, vrange, reduced_add, reduced_prod, reduced_max, reduced_min, vsum, vprod, vmaximum, vminimum,
     sizeequivalentfloat, sizeequivalentint, vadd!, vsub!, vmul!, vfdiv!, vfmadd!, vfnmadd!, vfmsub!, vfnmsub!,
     vfmadd231, vfmsub231, vfnmadd231, vfnmsub231, sizeequivalentfloat, sizeequivalentint, #prefetch,
     vmullog2, vmullog10, vdivlog2, vdivlog10, vmullog2add!, vmullog10add!, vdivlog2add!, vdivlog10add!, vfmaddaddone

test/mapreduce.jl

@@ -1,6 +1,12 @@
 
 @testset "mapreduce" begin
-
+    function maximum_avx(x)
+        s = typemin(eltype(x))
+        @avx for i in eachindex(x)
+            s = max(s, x[i])
+        end
+        s
+    end
     for T ∈ (Int32, Int64, Float32, Float64)
         if T <: Integer
             R = T(1):T(100)
@@ -19,6 +25,7 @@
         @test vmapreduce(sin, +, x7) ≈ mapreduce(sin, +, x7)
         @test vmapreduce(log, +, x) ≈ mapreduce(log, +, x)
         @test vmapreduce(abs2, +, x) ≈ mapreduce(abs2, +, x)
+        @test maximum(x) == vreduce(max, x) == maximum_avx(x)
     end
 
 end