Choose whichever one agrees with normed_eigval
because that determines the numeric precision you're working with. You will be converting the ptrdiff_t
integer into a floating-point value, then taking its logarithm.
Likely you can precompute the inverse of the logarithm, if you need it to go faster.