I just uploaded a fixed version of my old radix sort code. A bug was introduced in january 2006 when I added some optimizations sent to me by a user. Apparently I didn’t check them out enough: they made the code crash when negative zeros (binary 0×80000000) are involved. So I reverted to the previous code. Grab the fixed version here.
Big thanks to Ignacio Castano for reporting this!
Note that this only affects the latest radix code. The original code didn’t have the bug.