I agree that the optimization isn't possible but this is for correctness reasons. The performance penalty of the extra branch is almost negligible due to branch prediction.