Fix VecUnroll promotion with shifts

Author: chriselrod

Diff for: .gitignore

@@ -9,4 +9,7 @@ tests/#*#
 *~
 *.mem
 *#*
+Manifest.toml
+test/Manifest.toml
+
 

Diff for: src/vecunroll/fmap.jl

@@ -1,4 +1,3 @@
-
 # unary # 2^2 - 2 = 2 definitions
 @inline fmap(f::F, x::Tuple{X}) where {F,X} = (f(first(x)),)
 @inline fmap(f::F, x::NTuple) where {F} = (f(first(x)), fmap(f, Base.tail(x))...)
@@ -90,9 +89,9 @@ for op ∈ [:vgt,:vge,:vlt,:vle,:veq,:vne,:vmax,:vmax_fast,:vmin,:vmin_fast]
 end
 
 for op ∈ [:vrem, :vshl, :vashr, :vlshr, :vdiv, :vfdiv, :vrem_fast, :vfdiv_fast]
-    @eval begin
-        @inline $op(v1::VecUnroll, v2::VecUnroll) = VecUnroll(fmap($op, getfield(v1, :data), getfield(v2, :data)))
-    end
+  @eval begin
+    @inline $op(v1::VecUnroll, v2::VecUnroll) = VecUnroll(fmap($op, getfield(v1, :data), getfield(v2, :data)))
+  end
 end
 for op ∈ [:vrem, :vand, :vor, :vxor, :vshl, :vashr, :vlshr, :vlt, :vle,:vgt,:vge,:veq,:vne]
     @eval begin
@@ -101,7 +100,11 @@ for op ∈ [:vrem, :vand, :vor, :vxor, :vshl, :vashr, :vlshr, :vlt, :vle,:vgt,:v
     end
 end
 for op ∈ [:vshl, :vashr, :vlshr]
-    @eval @inline $op(m::AbstractMask, vu::VecUnroll) = $op(Vec(m), vu)
+  @eval begin
+    @inline $op(m::AbstractMask, vu::VecUnroll) = $op(Vec(m), vu)
+    @inline $op(v1::AbstractSIMDVector, v2::VecUnroll) = VecUnroll(fmap($op, v1, getfield(v2, :data)))
+    @inline $op(v1::VecUnroll, v2::AbstractSIMDVector) = VecUnroll(fmap($op, getfield(v1, :data), v2))
+  end
 end
 for op ∈ [:rotate_left,:rotate_right,:funnel_shift_left,:funnel_shift_right]
     @eval begin
@@ -238,4 +241,3 @@ end
 # @inline vminimum(vu::VecUnroll) = vminimum(collapse_min(vu))
 # @inline vall(vu::VecUnroll) = vall(collapse_and(vu))
 # @inline vany(vu::VecUnroll) = vany(collapse_or(vu))
-

Diff for: test/runtests.jl

@@ -531,7 +531,8 @@ include("testsetup.jl")
         At = ai ? A : (similar(A')');
         Bt = bi ? B : (similar(B')');
         Ct = ci ? C : (similar(C')');
-        gsp, pres = @inferred(VectorizationBase.grouped_strided_pointer((At,Bt,Ct), Val{(((1,1),(3,1)),((1,2),(2,1)),((2,2),(3,2)))}()))
+        spdw = VectorizationBase.DensePointerWrapper{(true,true)}(VectorizationBase.stridedpointer(At))
+        gsp, pres = @inferred(VectorizationBase.grouped_strided_pointer((spdw,Bt,Ct), Val{(((1,1),(3,1)),((1,2),(2,1)),((2,2),(3,2)))}()))
         if ai === ci
           @test sizeof(gsp.strides) == 2sizeof(Int)
         end
@@ -821,6 +822,9 @@ include("testsetup.jl")
             @test gcd(Vec(42,64,0,-37), Vec(18,96,-38,0)) === Vec(6,32,38,37)
             @test lcm(Vec(24,16,42,0),Vec(18,12,18,17)) === Vec(72, 48, 126, 0)
         end
+        if VectorizationBase.simd_integer_register_size() ≥ 16
+          @test VecUnroll((Vec(ntuple(Int32,Val(4))...),Vec(ntuple(Int32 ∘ Base.Fix2(+,4), Val(4))...))) << Vec(0x01,0x02,0x03,0x04) === VecUnroll((Vec(map(Int32,(2,8,24,64))...), Vec(map(Int32,(10,24,56,128))...)))
+        end
     end
     println("Ternary Functions")
     @time @testset "Ternary Functions" begin