Oh, I agree, I'm just saying that there is no reason in principle for floats performance to be better than integer. edit: also this might be implementable pure

Oh, I agree, I'm just saying that there is no reason in principle for floats performance to be better than integer. edit: also this might be implementable purely using bitwise vector operations. Would need to check the throughput of those.