I'm trying to efficiently compute an orthogonal transform of a real, symmetric matrix, N = Q * M * Q^{T}, where Q is orthogonal. Because M' is also symmetric I'm looking at

Code: Select all

A.noalias() = M.selfadjointview<Eigen::Lower>() * Q.transpose();
N.selfadjointview<Eigen::Lower>() = Q * A.


This works fine, but I want to avoid any temporary memory allocation in the second line. There doesn't seem to be support for mixing views with noalias, i.e.

Code: Select all

selfajointview<Eigen::Lower>().noalias()

or

Code: Select all

noalias().selfajointview<Eigen::Lower>()

; is there anyway to use the view while also avoiding a temporary?

Thanks.

-Mike

Yes, no need to add noalias here, noaliasing is the default for views.

Awesome.

On a similar note, what's the best way to access information about dense rows and columns from a partial matrix? The selfadjoint view doesn't seem to have row/column accessors, i.e.

Code: Select all

M.selfadjointView<Eigen::Lower>().row(i)


doesn't work.

Do I have to fill M explicitly? And, if so, is there a more efficient way than

Code: Select all

Mdense = M + M.triangularView<Eigen::StrictlyLower>().transpose()


?

Thanks, again!

-Mike