Description
@andyferris suggested that the above methods should be implemented in the review of #770, I am planning to do this and just wanted to open an issue for discussion.
LinearAlgebra uses Bunch-Kaufman if possible, then falls back to LU. I imagine we would fall back to this above a certain matrix size, to be determined empirically, like everywhere else in the package.
I would appreciate advice on what to do for small matrices. Copying the triangle and then LU may be the simplest solution. Unless there are better suggestions, I would just do this. Incidentally, this would implement LinearAlgebra.lu for Hermitian/Symmetric, which currently errors.
I have never programmer B-K but if it is not too much work and someone has a reference to get me started I would also consider that.
@andreasnoack, I would also appreciate your advice.