Turns out that I need to call potrf before potri.
The documentation
Computes the inverse of positive-definite matrix
Aafter callingpotrf!to find its (upper ifuplo = U, lower ifuplo = L) Cholesky decomposition.
made me think that potri would call potrf internally. But that is not the case.
So
LinearAlgebra.LAPACK.potrf!('U', V)
LinearAlgebra.LAPACK.potri!('U', V)
LinearAlgebra.copytri!(V, 'U')
V * Vcopy
would give back identity.