> The primary increase in speed versus a simple, straightforward loop, comes from loop unwinding that reduces the number of performed branches, which are comput

> The primary increase in speed versus a simple, straightforward loop, comes from loop unwinding that reduces the number of performed branches, which are computationally expensive due to the need to flush — and hence stall — the instruction pipeline. Is this generally true, or only in the case of a branch mispredict?