Counting the position of ones in a binary vector

I found, that the pure combinational solution (without the registers implemented by josyb) results in a tpd of about 17 ns with Cyclone III. So 266 MHz would surely need pipelining.

I didn't try, if the synthesis tool will "pull-in" registers to the synthesis of the behavioral description. As previously reported in the Forum, it does e.g. in the case of a parallel divider. But most likely, you have to break the data path manually to achieve the speed requirement.