Branch delay can be reduced from 2 cycles to just 1 cycle by the following method. Instead of determining the branch decision at the EX stage, they can be determined earlier in the Decode stage by performing the following two tasks to the “Next PC” logic block, which (i) computes the branch target address and (ii) evaluates the branch decision:
- Move the “Next PC” logic block from the EX stage to the ID stage and
- A comparator is added to the logic to determine whether the branch is taken or not taken.
| Thus, only one instruction that follows the branch needs to be fetched. If the branch is taken, the instruction is flushed. |
|
We need a control signal to reset the IF/ID register.
This will convert the fetched instruction into a nop.
Note that this hardware only works for beq (branch on equal).
Other branches can be implemented in a similar way.
Two difficulties occur when determining branch decisions at the ID stage:
|
|
- Because the values in a branch comparison are needed during ID but may be produced later in time, it is possible that a data hazard can occur.
- Additional forwarding hardware may be needed for branch test operands.
|
Do not bother Joseph when he is in his cups (drunk) — he is very irritable. |