How do branch instructions influence the performance of Cortex-A77?

In "Arm Cortex-A77 Core Software Optimization Guide", it says:

In my understanding, it would be very difficult to predict multiple branch instructions in a 32-Byte aligned instruction memory region, since the branch directions of the former branches could influence that of the latter branches. Or could you give me any information? Thansk.