I have known about this nicer implementation
for a while already, but forgot to actually
change the code.

Looking at the assembly, it seems that
release mode LLVM is smart enough to
optimize both versions to the same code.
So from a performance perspective, there
isn't any improvement.
Also, while ilog is used a few times,
it isn't really hot code.
Even in debug mode I didn't get any
performance impact.
However, I'd say that the code is
simpler this way.