Jeff Law wrote: >> Btw reminds me a little bit of >> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=28417
> I wouldn't expect 28417 > to move forward without something other than Tege and Denys pushing on it. Hmm that doesn't look optimal. You can typically do a multiply with the magic constant arranged in such a way that on 32-bit targets you don't need a shift at all. There isn't much to the proof either, code like this is always self-proving: it tries several possible magic constants until a solution is found that has no error for the max input. Given there are usually several valid solutions, you can then select the best shift/magic value combination. Wilco
