Merge pull request #218 from DilumAluthge/patch-1

Run CI on macOS (in addition to Linux and Windows)

Author: chriselrod

.github/workflows/ci-julia-nightly.yml

@@ -28,6 +28,7 @@ jobs:
           - 'nightly'
         os:
           - ubuntu-latest
+          - macOS-latest
           - windows-latest
         arch:
           - x64

.github/workflows/ci.yml

@@ -29,6 +29,7 @@ jobs:
           - '1' # automatically expands to the latest stable 1.x release of Julia.
         os:
           - ubuntu-latest
+          - macOS-latest
           - windows-latest
         arch:
           - x64

src/condense_loopset.jl

@@ -394,20 +394,10 @@ end
 Returns true if the element type is supported.
 """
 @inline check_type(::Type{T}) where {T <: NativeTypes} = true
-@inline function check_type(::Type{T}) where {T}
-    # @info """`LoopVectorization.check_type` returned `false`, because `LoopVectorization.check_type(::$(T)) == false`.
-    #     `LoopVectorization` currently only supports `T <: $(NativeTypes)`.
-    #     Therefore compiling a probably slow `@inbounds @fastmath` fallback loop.""" maxlog=1
-    false
-end
+@inline check_type(::Type{T}) where {T} = false
 @inline check_device(::ArrayInterface.CPUPointer) = true
 @inline check_device(::ArrayInterface.CPUTuple) = true
-@inline function check_device(x)
-    # @info """`LoopVectorization.check_args` returned `false`, because `ArrayInterface.device(::$(typeof(x))) == $x`
-    #     `LoopVectorization` normally requires `ArrayInterface.CPUPointer` (exceptions include ranges, `BitVector`s, and
-    #     `BitArray`s whose number of rows is a multiple of 8). Therefore compiling a probably slow `@inbounds @fastmath` fallback loop.""" maxlog=1
-    false
-end
+@inline check_device(x) = false
 
 function check_args_call(ls::LoopSet)
     q = Expr(:call, lv(:check_args))

test/arraywrappers.jl

@@ -1,6 +1,7 @@
 using LoopVectorization, Test#, OffsetArrays
 
 @testset "Array Wrappers" begin
+	@show @__LINE__
 
     function addone!(y, x)
         @avx for i ∈ eachindex(x,y)

test/check_empty.jl

@@ -15,6 +15,7 @@ function mysum_unchecked(x)
 end
 
 @testset "Check Empty" begin
+	@show @__LINE__
     x = fill(9999, 100, 10, 10);
     xv = view(x, :, 1:0, :);
     @test mysum_checked(x) == mysum_unchecked(x) == sum(x)

test/loopinductvars.jl

@@ -1,6 +1,7 @@
 
 # testset for using them in loops
 @testset "Loop Induction Variables" begin
+	@show @__LINE__
     f(x) = cos(x) * log(x)
     function avxmax(v)
         max_x = -Inf

test/runtests.jl

@@ -30,8 +30,12 @@ const START_TIME = time()
     @time include("loopinductvars.jl")
 
     @time include("shuffleloadstores.jl")
-    
-    @time include("zygote.jl")
+
+	if (v"1.5" < VERSION < v"1.7") && Sys.iswindows()
+		println("Skipping Zygote tests.")
+	else
+    	@time include("zygote.jl")
+	end
 
     @time include("offsetarrays.jl")
 

test/shuffleloadstores.jl

@@ -173,6 +173,7 @@ function issue209_noavx(M, G, J, H, A, B, ϕ)
 end
 
 @testset "shuffles load/stores" begin
+	@show @__LINE__
     for i ∈ 1:128
         ac = rand(Complex{Float64}, i);
         bc = rand(Complex{Float64}, i);

test/threading.jl

@@ -118,10 +118,12 @@ end
         C1 = A * B; C0 = similar(C1);
         @test AmulB!(C0, A, B) ≈ C1
 
-        x = randn(Complex{Float64}, 3M-1);
-        W = randn(Complex{Float64}, 3M-1, 3M+1);
-        y = randn(Complex{Float64}, 3M+1);
-        @test dot(x,W,y) ≈ dot3(x,W,y)
+		if VERSION ≥ v"1.6"
+	        x = randn(Complex{Float64}, 3M-1);
+    	    W = randn(Complex{Float64}, 3M-1, 3M+1);
+        	y = randn(Complex{Float64}, 3M+1);
+        	@test dot(x,W,y) ≈ dot3(x,W,y)
+		end
 
         kern = OffsetArray(randn(3,3),-2,-2)
         out1 = OffsetArray(randn(size(A) .- 2), 1, 1)

test/utils.jl

@@ -2,6 +2,7 @@ using LoopVectorization
 using Test
 
 @testset "Utilities" begin
+	@show @__LINE__
     @test LoopVectorization.isscopedname(:(Base.OneTo), :Base, :OneTo)
     @test LoopVectorization.isscopedname(:(A.B.C.D), (:A, :B, :C), :D)
     @test !LoopVectorization.isscopedname(:(A.B.D),  (:A, :B, :C), :D)