...

Where to get instructions to fill branch delay slot?

• Before branch instruction
• From the target address (only useful if branch taken)
• From fall-through (only useful if not taken)
• May need to add code to cancel effect of delay slot instructions inadvertently executed, eg at end of a loop

How well do compilers do?

• Fills about 60% of branch delay slots
• Of these, about 80% actually contribute usefully
• Overall effectiveness: 60%$\times$80% = 50%
• Effective average branch delay $\approx$ 50%

It is much harder to make effective use of multiple delay slots