Require ArrayInterface 3.1.4 to support CPUTuples

Author: chriselrod

Project.toml

@@ -18,7 +18,7 @@ UnPack = "3a884ed6-31ef-47d7-9d2a-63182c4928ed"
 VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
 
 [compat]
-ArrayInterface = "3"
+ArrayInterface = "3.1.4"
 CheapThreads = "0.1.2"
 DocStringExtensions = "0.8"
 IfElse = "0.1"

docs/make.jl

@@ -7,6 +7,7 @@ makedocs(;
         "Home" => "index.md",
         "Getting Started" => "getting_started.md",
         "Examples" => [
+            "examples/multithreading.md",
             "examples/matrix_multiplication.md",
             "examples/array_interface.md",
             "examples/matrix_vector_ops.md",

docs/src/examples/matrix_multiplication.md

@@ -5,13 +5,13 @@ One of the friendliest problems for vectorization is matrix multiplication. Give
 LoopVectorization currently doesn't do any memory-modeling or memory-based optimizations, so it will still run into problems as the size of matrices increases. But at smaller sizes, it's capable of achieving a healthy percent of potential GFLOPS.
 We can write a single function:
 ```julia
-function A_mul_B!(𝐂, 𝐀, 𝐁)
-    @avx for m ∈ axes(𝐀,1), n ∈ axes(𝐁,2)
-        𝐂mn = zero(eltype(𝐂))
-        for k ∈ axes(𝐀,2)
-            𝐂mn += 𝐀[m,k] * 𝐁[k,n]
+function A_mul_B!(C, A, B)
+	@avx for n ∈ indices((C,B), 2), m ∈ indices((C,A), 1)
+        Cmn = zero(eltype(C))
+        for k ∈ indices((A,B), (2,1))
+            Cmn += C[m,k] * B[k,n]
         end
-        𝐂[m,n] = 𝐂mn
+		C[m,n] = Cmn
     end
 end
 ```

docs/src/index.md

@@ -8,6 +8,7 @@ Please file an issue if you run into any problems.
 ```@contents
 Pages = [
     "getting_started.md",
+    "examples/multithreading.md",
     "examples/matrix_multiplication.md",
     "examples/array_interface.md",
     "examples/matrix_vector_ops.md",

src/condense_loopset.jl

@@ -395,6 +395,7 @@ function check_type(::Type{T}) where {T}
     false
 end
 @inline check_device(::ArrayInterface.CPUPointer) = true
+@inline check_device(::ArrayInterface.CPUTuple) = true
 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