mark two_mul as consistent (#46022)

should let the compiler prove that float64^int always gives the same result.

(cherry picked from commit fd70eab)

base/math.jl

@@ -42,15 +42,15 @@ end
 
 # non-type specific math functions
 
-@inline function two_mul(x::Float64, y::Float64)
+@assume_effects :consistent @inline function two_mul(x::Float64, y::Float64)
     if Core.Intrinsics.have_fma(Float64)
         xy = x*y
         return xy, fma(x, y, -xy)
     end
     return Base.twomul(x,y)
 end
 
-@inline function two_mul(x::T, y::T) where T<: Union{Float16, Float32}
+@assume_effects :consistent @inline function two_mul(x::T, y::T) where T<: Union{Float16, Float32}
     if Core.Intrinsics.have_fma(T)
         xy = x*y
         return xy, fma(x, y, -xy)

test/math.jl

@@ -1421,3 +1421,7 @@ end
 # the compiler ever gets good enough to figure
 # that out by itself, move this to inference).
 @test code_typed(x->Val{x^0.0}(), Tuple{Float64})[1][2] == Val{1.0}
+
+for T in (Float32, Float64)
+    @test Core.Compiler.is_foldable(Base.infer_effects(^, (T,Int)))
+end