Merge pull request #41 from tkf/literal_pow

Implement literal_pow for Vec
eschnett · Jan 7, 2019 · bb5adc9 · bb5adc9
2 parents 89ca5b4 + c6b0f84
commit bb5adc9
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 0 deletions.
diff --git a/src/SIMD.jl b/src/SIMD.jl
@@ -1022,6 +1022,12 @@ end
 @inline Base.flipsign(v1::Vec{N,T}, v2::Vec{N,T}) where {N,T<:FloatingTypes} =
     vifelse(signbit(v2), -v1, v1)
 
+# Do what Base does for HWNumber:
+@inline Base.literal_pow(::typeof(^), x::Vec, ::Val{0}) = one(typeof(x))
+@inline Base.literal_pow(::typeof(^), x::Vec, ::Val{1}) = x
+@inline Base.literal_pow(::typeof(^), x::Vec, ::Val{2}) = x*x
+@inline Base.literal_pow(::typeof(^), x::Vec, ::Val{3}) = x*x*x
+
 for op in (:fma, :muladd)
     @eval begin
         @inline function Base.$op(v1::Vec{N,T},

diff --git a/test/runtests.jl b/test/runtests.jl
@@ -122,6 +122,11 @@ llvm_ir(f, args) = sprint(code_llvm, f, Base.typesof(args...))
             @test Tuple(op(V8I32(v8i32), -3)) === map(x->op(x,-3), v8i32)
             @test Tuple(op(V8I32(v8i32), V8I32(v8i32))) === map(op, v8i32, v8i32)
         end
+
+        @test Tuple(V8I32(v8i32)^0) === v8i32.^0
+        @test Tuple(V8I32(v8i32)^1) === v8i32.^1
+        @test Tuple(V8I32(v8i32)^2) === v8i32.^2
+        @test Tuple(V8I32(v8i32)^3) === v8i32.^3
     end
 
     @testset "Floating point arithmetic functions" begin
@@ -207,6 +212,11 @@ llvm_ir(f, args) = sprint(code_llvm, f, Base.typesof(args...))
             @test Tuple(op(V4F64(v4f64), V4F64(v4f64b))) === map(op, v4f64, v4f64b)
         end
 
+        @test Tuple(V4F64(v4f64)^0) === v4f64.^0
+        @test Tuple(V4F64(v4f64)^1) === v4f64.^1
+        @test Tuple(V4F64(v4f64)^2) === v4f64.^2
+        @test Tuple(V4F64(v4f64)^3) === v4f64.^3
+
         for op in (fma, vifelsebool, muladd)
             @test Tuple(op(V4F64(v4f64), V4F64(v4f64b), V4F64(v4f64c))) ===
                 map(op, v4f64, v4f64b, v4f64c)