...

• Control hazards

  A control hazard is like a data hazard, but worse. The new PC may not be known until MEM, but the IF stage of the following instruction must know its value.

  Unfortunately, the forwarding/bypass trick is inapplicable.

  There are several different ways to address the problem:

  • delayed branch - redefined instruction set so branch takes effect one instruction later
  • predict branch taken
  • predict branch not taken
  • dynamic branch prediction

(See H&P pp.161 and pp.262)