Suppose I have a large matrix `M`

composed of homogeneous Unitful data, and a vector `v`

, which contains homogeneus Unitful data as well. (By Unitful data I mean the data with the types from Unitful.jl)

Does the calculation of matrix-vector product `M*v`

suffers from the fact that the arrays are Unitful?

For example, can it use fast BLAS implementations, or does it fall back to generic implementation instead?