Sami Farin wrote: > jhash seems to be okay with idutils, at least on my CPU. It's faster than > the hack in my previous post, due to fewer collisions. It could cache result > of strlen's somewhere, but that might be micro-optimizing. > Note that I have not thought about portability or coding style of this patch, > but it works for me. > > I also took the hash_ptr from Linux 2.6.36.
Thanks for working on idutils. Before we can use contributions by you, you'll have to fill out some paperwork, assuming any school/employer is ok with that. Here's the procedure I wrote up for coreutils. For idutils it's the same: http://git.savannah.gnu.org/cgit/coreutils.git/tree/HACKING#n429 There might be copyright complications with adding jhash (unless it's GPLv3) or code from the kernel. If you're inclined, I'd like to see if you get similar gains by switching to the use of the hash_pjw function from gnulib (lib/hash-pjw.c). Using that would require changing hash_init to take hashing functions with different signatures: this new hash function has to know the size of the hash table. But that would be a good thing regardless, since it permits better use of the limited number of bits in each key. _______________________________________________ bug-idutils mailing list bug-idutils@gnu.org http://lists.gnu.org/mailman/listinfo/bug-idutils
