------- Comment #12 from pcarlini at suse dot de 2007-10-20 11:27 ------- (In reply to comment #11)
> I just tested this. It does a much better job at obfuscating the bias, which > is > still there (after all you are rescaling linearly). If you rescale a domain of > size 16 to a range of size 12, some buckets will get 2 and some will get 1. Of course. What I did is just the naive rescaling, which you can find, for example, in Numerical Recipes. It assumes, implicitly, that the source range is much bigger than the target range. In order to have a fully conforming implementation, we need something which works for *any* source range and *any* target range. Are you willing to work on that? Would be mainline-only of course, too large for 4.2.x. In that case you have also to file a Copyright assignment. Please let me know. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33815
