Now that #131 is merged, it's faster to construct a PDMat with an SMatrix than it is with Cholesky{T, SMatrix}. I believe this is because the constructor here creates a Matrix then here converts it to an SMatrix.
Naively, it seems like this could be improved simply by changing
PDMat(fac::Cholesky) = PDMat(Matrix(fac), fac)
to
PDMat(fac::Cholesky) = PDMat(AbstractMatrix(fac), fac)
in pdmat.jl
Now that #131 is merged, it's faster to construct a
PDMat
with anSMatrix
than it is withCholesky{T, SMatrix}
. I believe this is because the constructor here creates aMatrix
then here converts it to anSMatrix
.Naively, it seems like this could be improved simply by changing
PDMat(fac::Cholesky) = PDMat(Matrix(fac), fac)
toPDMat(fac::Cholesky) = PDMat(AbstractMatrix(fac), fac)
inpdmat.jl