JuliaGPU#362: More tests and documentation; simplify the dims

src/mapreduce.jl

@@ -97,11 +97,11 @@ function Base._mapreducedim!(f, op, R::CuArray{T}, A::CuArray{T}) where {T}
 end
 
 import Base.minimum, Base.maximum, Base.reduce
-_initarray(x::CuArray{T}, ::Colon, init) where {T} = fill!(similar(x, T, Base.reduced_indices(x, 1:ndims(x))), init)
-_initarray(x::CuArray{T}, dims, init) where {T} = fill!(similar(x, T, Base.reduced_indices(x, dims)), init)
 
+_reduced_dims(x::CuArray, ::Colon) = Tuple(ones(Int, ndims(x)))
+_reduced_dims(x::CuArray, dims) = Base.reduced_indices(x, dims)
 
-reduce(op, x::CuArray; dims=:, init) where {T} = _reduce(op, x, init, dims)
+_initarray(x::CuArray{T}, dims, init) where {T} = fill!(similar(x, T,     _reduced_dims(x, dims)), init)
 
 function _reduce(op, x::CuArray, init, ::Colon)
     mx = _initarray(x, :, init)
@@ -114,5 +114,12 @@ function _reduce(op, x::CuArray, init, dims)
     Base._mapreducedim!(identity, op, mx, x)
 end
 
+"""
+    reduce(op, x::CuArray; dims=:, init)
+
+The initial value `init` is mandatory for `reduce` on `CuArray`'s. It must be a neutral element for `op`.
+"""
+reduce(op, x::CuArray; dims=:, init) = _reduce(op, x, init, dims)
+
 minimum(x::CuArray{T}; dims=:) where {T} = _reduce(min, x, typemax(T), dims)
 maximum(x::CuArray{T}; dims=:) where {T} = _reduce(max, x, typemin(T), dims)

test/base.jl

@@ -143,16 +143,25 @@ end
 
   @test testf(x -> minimum(x), rand(2, 3))
   @test testf(x -> minimum(x, dims=2), rand(2, 3))
-  @test testf(x -> minimum(x, dims=(2,3)), rand(2, 3, 4))
+  @test testf(x -> minimum(x, dims=(2, 3)), rand(2, 3, 4))
 
   @test testf(x -> maximum(x), rand(2, 3))
   @test testf(x -> maximum(x, dims=2), rand(2, 3))
-  @test testf(x -> maximum(x, dims=(2,3)), rand(2, 3, 4))
+  @test testf(x -> maximum(x, dims=(2, 3)), rand(2, 3, 4))
 
-  myreducer(x1, x2) = x1+x2 # bypass optimisations for sum()
-  @test testf(x -> reduce(myreducer, x, dims=(2,3), init=0.0), rand(2, 3, 4))
+  myreducer(x1, x2) = x1 + x2 # bypass optimisations for sum()
+  @test testf(x -> reduce(myreducer, x, dims=(2, 3), init=0.0), rand(2, 3, 4))
   @test testf(x -> reduce(myreducer, x, init=0.0), rand(2, 3))
   @test testf(x -> reduce(myreducer, x, dims=2, init=42.0), rand(2, 3))
+
+  ex = ErrorException("Please supply a neutral element for &. E.g: mapreduce(f, &, A; init = 1)")
+  @test_throws ex mapreduce(t -> t > 0.5, &, cu(rand(2, 3)))
+  @test testf(x -> mapreduce(t -> t > 0.5, &, x, init=true), rand(2, 3)) 
+
+  ex = UndefKeywordError(:init)
+  cub = map(t -> t > 0.5, cu(rand(2, 3)))
+  @test_throws ex reduce(|, cub)
+  @test testf(x -> reduce(|, x, init=false), map(t -> t > 0.5, cu(rand(2, 3)))) 
 end
 
 @testset "0D" begin