[+gwtc] LGTM, although I had to do some work to convince myself it was safe.
You might consider updating the comment; what I find more quickly understandable and verifiable is something like: If n<32, a[HIGH]/shiftFact is guaranteed to be an integer already. For n>32, a[HIGH]/shiftFact will have fractional bits, but we need to discard them as they shift away. (We will end up discarding all of a[LOW] in this case, as it divides out to entirely fractional.) I assume the branch to test for high-n (and work only with a[HIGH] if so) is much more expensive than the savings by avoiding the division. On Wed, Apr 22, 2009 at 3:57 PM, Lex Spoon <[email protected]> wrote: > Freeland, would you have time to review this short patch to LongLib? > > http://gwt-code-reviews.appspot.com/28801 > > -Lex > --~--~---------~--~----~------------~-------~--~----~ http://groups.google.com/group/Google-Web-Toolkit-Contributors -~----------~----~----~----~------~----~------~--~---
