2
votes

I have a set of instructions as follows:

loop:
        sll  $t1, $t0, 2              # t1 = (i * 4)
        add $t2, $a0, $t1             # t2 contains address of array[i]
        sw $t0, 0($t2)                # array[i] = i

        addi $t0, $t0, 1              # i = i+1
        add $t4, $t4, $t0             # sum($t4) = ($t4 + array[i])

        slt $t3, $t0,   $a1             # $t3 = ( i < array_size)
        bne $t3, $zero, loop          # if ( i < array_size ) then loop

The sll instruction has an address (program counter) of 0x18. bne has an address of 0x30. MARS Simulator interprets the bne instruction as: bne $11, $0, 0xfff9. 0xfff9 is -7, meaning the instruction will jump 7 steps back. However, sll is six steps back. Does MIPS take into account the current instruction? Or does this happen because the program counter is incremented in the fetch stage, before the instruction finishes executing?

2
I don't have the answer, but I will add "stride" as another possible reason.Brent Arias

2 Answers

2
votes

On branch mips executes two instructions -- the branch instruction itself and the one following it (so called branch delay slot).

At the time branch takes effect, PC points to the instruction that follows the branch instruction itself, so -7 is appropriate.

1
votes

On most architectures, branch targets are calculated from the instruction following the branch instruction (i.e: the PC is already advanced). IIRC ARM is the only common exception (there, the PC has advanced farther still, due to pipelining (of the original ARM implementation)).