On Sun, 13 Oct 2024 16:03:09 GMT, Raffaello Giulietti <rgiulie...@openjdk.org> wrote:
>> @rgiulietti Good, but I would not put the inequality `maxPowsOf5 <= m + 1` >> and say that `maxPowsOf5` is never off by more than 1 from the theoretical >> `m`, because it is not crucial as `m <= maxPowsOf5`, and because `m` is in >> function of `intVal`, while `maxPowsOf5` is in function of `2^b`, so it is >> not obvious that there's no power of five `pow` such that `intVal < pow < >> 2^b`... > > What's really crucial for _correctness_ is to ensure maxPowsOf5 >= k. > > But for performance you also want maxPowsOf5 to be as small as possible. So, > the fact that it turns out that maxPowsOf5 <= m + 1 guarantees that > maxPowsOf5 is the best value that can be computed very efficiently. It's more > a "quality of service" guarantee than anything fundamental. Perhaps leave m <= maxPowsOf5 <= m + 1 and maxPowsOf5 >= k and drop the note "and is never off by more than 1 from the theoretical m" ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/21323#discussion_r1798430974
