Your understanding is correct. Lemire's algorithm is not an RNG, but a way to map uniformly random bits (from some other RNG) to a non-power of two[1] range 1..

Your understanding is correct. Lemire's algorithm is not an RNG, but a way to map uniformly random bits (from some other RNG) to a non-power of two[1] range 1..N (or 0..N-1) efficiently on average. --- [1] Well, it works for power-of-two too, but you certainly wouldn't use this if you wanted a power of two range: you can just extract the required number of bits directly.