The count wasn't clamped above zero in some newly optimized code. Instead of adding it there, move the clamping to the field network, similar to some other nodes. That makes it so the rest of the code doesn't have to deal with the clamping, and should be faster in the single-value case.