loss of performance for complex valued Kronecker products

[vsaase]

Hi, here are some benchmark, seems there are inefficient allocations when using complex values:

using LazyArrays
using BenchmarkTools

spd=ones(1000,1000)
A=ones(2,2)
B=kron(spd,A)
Bi=kron(spd,im*A)
Bla = ApplyArray(kron, spd, A)
Blai = ApplyArray(kron, spd, im*A)
x=ones(2000)

@btime B*x;
# julia> @btime B*x
#   1.368 ms (1 allocation: 15.75 KiB)

@btime Bla*x;
# julia> @btime Bla*x;
#   471.707 μs (6 allocations: 31.67 KiB)

@btime Bi*x;
# julia> @btime Bi*x;
#   3.128 ms (4 allocations: 62.66 KiB)

@btime Blai*x;
# julia> @btime Blai*x;
#   32.775 ms (3112 allocations: 8.04 MiB)

thanks,
Victor