Function lgamma_r_impl 

pub fn lgamma_r_impl<P, E, V, const N: usize>(
    z_in: V,
    l: &Lanczos<E, N>,
) -> (V, V)
where
    P: Policy,
    E: FloatElementWithBits,
    V: FloatVectorWithBits<Element = E> + SpecializedSpecialMath<E>,

Shared lgamma_r implementation (log-gamma with its separate sign) for all real element types, via the exp(g)-scaled Lanczos sum.

f32 short-circuits to a cheaper Pade approximant below Average precision; that branch lives at the call site and never reaches here.