Add vcount(::AbstractArray{Bool}), fixes #479

also add `vcount(::typeof(identity), ::AbstractArray{Bool})`
Performance was good on an x64 AVX512 system, but bad on ARM NEON.

Author: chriselrod

Project.toml

@@ -57,5 +57,5 @@ Static = "0.8.4"
 StaticArrayInterface = "1"
 ThreadingUtilities = "0.5"
 UnPack = "1"
-VectorizationBase = "0.21.60"
+VectorizationBase = "0.21.62"
 julia = "1.6"

src/simdfunctionals/count.jl

@@ -12,11 +12,12 @@ function _vcount(f::F, args::Vararg{DenseArray,M}) where {F,M}
   W = unwrap(V)
   UNROLL = 4
   LOG2UNROLL = 2
-  counts = if VERSION >= v"1.7"
+  _counts = if VERSION >= v"1.7"
     VecUnroll(ntuple(Returns(0), Val(UNROLL)))
   else
     VecUnroll(ntuple(_ -> (0), Val(UNROLL)))
   end
+  counts::typeof(_counts) = _counts
   while i < vsub_nsw(N, ((W << LOG2UNROLL) - 1))
     index = VectorizationBase.Unroll{1,W,UNROLL,1,W,zero(UInt)}((i,))
     counts += count_ones(f(VectorizationBase.fmap(vload, ptrargs, index)...))
@@ -39,3 +40,6 @@ end
   call = Expr(:call, :_vcount, :f)
   gc_preserve_call_quote(call, M::Int)
 end
+vcount(::typeof(identity), x::AbstractArray{Bool}) =
+  vcount(VectorizationBase.tomask, x)
+vcount(x::AbstractArray{Bool}) = vcount(VectorizationBase.tomask, x)