Flow Control

Introduction#

Every piece of non-trivial software needs flow-control structures to divert program flow according to conditions.

Assembly being the lowest-level programming language provides only primitives for control structures. Typically, machine operations affect flags in the CPU, and conditional branches/jumps implement the flow control. In assembly, all higher-level control structures need to be constructed from such primitives.

Trivial IF-THEN-ELSE in m68k Assembly

; IF d0 == 10 GO TO ten, ELSE GO TO other
    CMP    #10,d0        ; compare register contents to immediate value 10
                         ; instruction affects the zero flag
    BEQ    ten           ; branch if zero flag set
other:
    ; do whatever needs to be done for d0 != 10
    BRA    afterother    ; unconditionally jump across IF case 
ten:
    ; do whatever needs to be done for d0 == 10
afterother:
    ; continue normal common program flow

Which instructions are affecting which flags, and which conditional branches (that might also be based on specific combinations of flags) are available, depends very much on your chosen CPU and should be looked up in the manuals.

FOR … NEXT in Z80 Assembly

The Z80 has a specific instruction to implement loop counts: DJNZstanding for “decrement B register and jump if not zero”. So, B is the register of choice to implement loops on this processor. FOR…NEXT needs to be implemented “backwards”, because the register counts down to zero. Other CPUs (like the 8086, this CPU uses the CX register as loop counter) might have similar specific loop counter registers and instructions, some other CPUs allow loop commands with arbitrary registers (m68k has a DBRA instruction that works with any data register).

; Trivial multiplication (by repeated adding, ignores zero in factors, so 
; not recommended for general use)
;
; inputs:    A = Factor 1
;            B = Factor 2
;
; output:    A = Factor 1 * Factor 2
;
; Pseudo code
; C = A : A = 0 : FOR B = Factor 2 DOWNTO 0 : A = A + C : NEXT B

mul:
     LD    C,A        ; Save Factor 1 in C register
     XOR   A          ; Clear accumulator
mLoop:
     ADD   A,C        ; Add Factor 1 to accumulator
     DJNZ  mLoop      ; Do this Factor 2 times
     RET              ; return to caller

If-statement in Intel-syntax assembly

section .data
    msg_eq db 'Equal', 10
    len_eq equ $ - msg_eq

    msg_le db 'Less than', 10
    len_le equ $ - msg_le

    msg_gr db 'Greater than', 10
    len_gr equ $ - msg_gr ; Length of msg_gr
section .text
    global _main ; Make the _main label global for linker
_main:
    cmp 4, 5 ; Compare 4 and 5
    je _equal ; je = jump if equal
    jl _less ; jl = jump if less
    jg _greater ; jg = jump if greater
exit:
    ret ; Return
_equal:
    ; Whatever code here
    mov rax, 0x2000004 ; sys_write, 4 for linux
    mov rdi, 1 ; STDOUT
    mov rsi, msg_eq
    mov rdi, len_eq
    
    syscall
    
    jmp exit ; Exit
_less:
    ; Whatever code here
    mov rax, 0x2000004
    mov rdi, 1
    mov rsi, msg_le
    mov rdi, len_le
    
    syscall
    
    jmp exit
_greater:
    ; Whatever code here
    
    mov rax, 0x2000004
    mov rdi, 1
    mov rsi, msg_gr
    mov rdi, len_gr
    
    syscall
    jmp exit

Loop while condition is true in Intel syntax assembly

section .data
    msg db 'Hello, world!', 0xA
    len equ $ - msg
section .text
global _main
_main:
    mov rax, 0 ; This will be the current number
    mov rcx, 10 ; This will be the last number
    
_loop:
    cmp rax, rcx
    jl .loopbody ; Jump to .loopbody if rax < rcx
    jge _exit ; Jump to _exit if rax ≥ rcx
.loopbody:
    push rax ; Store the rax value for later use

    mov rax, 0x2000004 ; 4 for Linux
    mov rdi, 1 ; STDOUT
    mov rsi, msg
    mov rdx, len

    syscall

    pop rax ; Take it back to rax

    inc rax ; Add 1 to rax. This is required since the loop must have an ending.    

    jmp _loop ; Back to loop
_exit:
    ret    ; Return

This will execute .loopbody as long as rax < rcx.