Julia High Performance - Second Edition

Adjoints

Sometimes, you may have an algorithm that is naturally expressed as row-first. Converting the algorithm to column-first iteration can be a difficult process, prone to errors. In such cases, Julia provides an Adjoint type that can store a transpose of a matrix as a view into the original matrix. Row-wise iteration will be the fast option. We illustrate this in the following code. 

julia> b=a'1000×1000 LinearAlgebra.Adjoint{Float64,Array{Float64,2}}:...julia> @btime col_iter($b)  3.521 ms (0 allocations: 0 bytes)julia> @btime row_iter($b)  1.160 ms (0 allocations: 0 bytes)

We see here that the timings of the function have been inverted from those in the previous section. Column iteration is now slower than row iteration. You will also ...