Hermitian matrix multiplication relies on BLAS function that isn't wrapped

[kshyatt]

In base/linalg/symmetric.jl, to multiply a general matrix by a Hermitian one, we can use:
A_mul_B!{T<:BlasComplex,S<:AbstractMatrix}(y::StridedMatrix{T}, A::Hermitian{T,S}, x::StridedMatrix{T}) = BLAS.hemm!(A.uplo, one(T), A.data, B, zero(T), C)

Cool! But in blas.jl, there's no wrapper for hemm! or hemm. Also, this takes x and y as args, yet passes B and C which AFAICT don't exist.

Here is a demonstration:

julia> A = randn(3,3)
3x3 Array{Float64,2}:
 0.37118   -0.115637  -0.2626
 1.44327   -0.3389    -1.36312
 0.402348   0.465449  -0.281456

julia> B = randn(3,3)
3x3 Array{Float64,2}:
  0.63778   -0.577948   0.365766
  1.185      0.115811   1.06565
 -0.446381  -0.329414  -0.504115

julia> C = complex(A,B)
3x3 Array{Complex{Float64},2}:
  0.37118+0.63778im   -0.115637-0.577948im    -0.2626+0.365766im
  1.44327+1.185im       -0.3389+0.115811im   -1.36312+1.06565im
 0.402348-0.446381im   0.465449-0.329414im  -0.281456-0.504115im

julia> C = C + C'
3x3 Array{Complex{Float64},2}:
  0.74236+0.0im         1.32763-1.76295im   0.139748+0.812146im
  1.32763+1.76295im   -0.677799+0.0im      -0.897673+1.39507im
 0.139748-0.812146im  -0.897673-1.39507im  -0.562913+0.0im

julia> C = Hermitian(C)
3x3 Base.LinAlg.Hermitian{Complex{Float64},Array{Complex{Float64},2}}:
  0.74236-0.0im         1.32763-1.76295im   0.139748+0.812146im
  1.32763+1.76295im   -0.677799-0.0im      -0.897673+1.39507im
 0.139748-0.812146im  -0.897673-1.39507im  -0.562913-0.0im

julia> D = complex(A,B)
3x3 Array{Complex{Float64},2}:
  0.37118+0.63778im   -0.115637-0.577948im    -0.2626+0.365766im
  1.44327+1.185im       -0.3389+0.115811im   -1.36312+1.06565im
 0.402348-0.446381im   0.465449-0.329414im  -0.281456-0.504115im

julia> C*D
ERROR: UndefVarError: hemm! not defined
 in * at linalg/matmul.jl:121