Documentation for the package

[ViralBShah]

Filing a help needed issue to put together Documenter based documentation for this package.

[ViralBShah]

@andreasnoack Could you list in the README, the current set of capabilities available in the package, and a sort of quick roadmap of the next set of things to add? This will help people pick up the project.

@alanedelman ^

[alanedelman]

i'd love for performance guidance right on top of when an svd from elemental might be beneficial --- example a statement that suggests that at least on one machine with 16 processors, we found a performance increase over lapack with mkl and open blas.....

and if there is a way others can add performance anecdotes in a useful way...

[ViralBShah]

Julia with MKL for SVD:

julia> a = randn(10^4,10^4);
julia> using LinearAlgebra
julia> @time u,s,v = svd(a)
136.560808 seconds (258.16 k allocations: 4.485 GiB, 0.16% gc time)

With Elemental using 20 processors on a single box using the example in the README:

julia> @mpi_do man Elemental.gaussian!(A, 10000, 10000);
julia> @time @mpi_do man U, s, V = svd(A);
 80.731189 seconds (2.68 k allocations: 141.296 KiB)

[andreasnoack]

I'm a little surprised by this. Would it be easy to try MKL's LAPACK.gesvd!. Our svd calls LAPACK.gesdd!.

[ViralBShah]

LAPACK+OpenBLAS (the julia default) is 170s.

Should be easy to try gesvd - just have to go through LAPACK docs and the wrappers.