These articles are written by Codalogic empowerees as a way of sharing knowledge with the programming community. They do not necessarily reflect the opinions of Codalogic.

Experiments with Aarch64 Assembler Part 6 - strlen

By: Pete, October 2022

strlen

(Code for this mini-series can be downloaded from Github)

Having to specify a length in order to display a string is tedious. Instead it is easier to use null terminated strings.

You can specify null-terminated strings in the assembler using the .asciz directive instead of the .ascii directive. For example:

.data

msg:    .asciz "Hello, Aarch64 World!\n"

But before we can display a null-terminated string we need to work out its length. That's what these subroutines do. I have defined two subroutines here, strnlen_s and strlen. strnlen_s will work out the length of a string up to a specified maximum. The intent is to be safer if it is accidently asked to work out the length of a string that turns out not to be null-terminated. strlen doesn't require the maximum length to be specified. However, rather than allow any length string as C's strlen() does, it uses strnlen_s and specifies a maximum length of 1000 characters.

This makes the code for strlen very simple. We only have to load x1 with the second parameter for strnlen_s (1000) and then branch to the strnlen_s code.

strlen:
    // int strlen( const char * x0 /* s */ )

    // Set a reasonable maximum and branch to strnlen_s
    mov     x1, #1000
    b       strnlen_s

strnlen_s is more involved.

Because we are using a number of registers it's a good idea to document their use at the start of the subroutine.

strnlen_s:
    // int strnlen_s( const char * x0 /* s */, int x1 /* max_len */ )

    // Register usage:
    //  x0 - Input pointer to string
    //  x1 - Max length we want to search
    // Intermediates:
    //  x2 - Running count of non-null bytes in string
    //  x3 - Place for testing if byte in string is null

We're not calling any subroutines and we're not using any registers that we are not allowed to corrupt. Therefore we don't need any stack preamble.

x2 will count the number of characters in the string so we initialise it to 0:

    mov     x2, #0

Now we start the main loop.

We load the byte pointed to by x0 into w3, post-incrementing the value in x0 by 1 and then testing the loaded byte to see if it is zero using the cbz (Compare and Branch on Zero) instruction. If it is we branch to the exit code.

.L_strnlen_s_main_loop:
    // Load byte pointed to by x0 ready and test if it is zero
    // Branch to exit if it is 0
    ldrb    w3, [x0], #+1
    cbz     x3, .L_strnlen_s_exit

In the above the cbz instruction (Compare and Branch on Zero) is equivalent to:

    cmp     x3, #0
    b.eq    .L_strnlen_s_exit

If the byte is not zero we increment the byte count by 1.

    // Record we have another byte
    add     x2, x2, 1

We then decrement the count of the maximum number of bytes we're prepared to look at by 1 and test if it is zero. Note the use of subs instead of sub here to update the status registers. If the count is zero we go to the exit code.

    // Decrement our maximum string length counter
    // Branch to exit if it is zero
    subs    x1, x1, 1
    b.eq    .L_strnlen_s_exit

Now we just repeat this for the next byte by going back to the start of the loop:

    // Repeat
    b       .L_strnlen_s_main_loop

In our exit section we move the count from x2 into the x0 register. x0 is the register specified to be used for subroutine return values in the Arm Procedure Call Standard.

.L_strnlen_s_exit:
    // Return the discovered length in x0
    mov     x0, x2

Then we can invoke ret.

    // Return
    ret

The complete code is:

strlen:
    // int strlen( const char * x0 /* s */ )

    // Set a reasonable maximum and branch to strnlen_s
    mov     x1, #1000
    b       strnlen_s

strnlen_s:
    // int strnlen_s( const char * x0 /* s */, int x1 /* max_len */ )

    // Register usage:
    //  x0 - Input pointer to string
    //  x1 - Max length we want to search
    // Intermediates:
    //  x2 - Running count of non-null bytes in string
    //  x3 - Place for testing if byte in string is null

    // We're not calling any subroutines.  No need for any stack work

    // Initialise our count
    mov     x2, #0

.L_strnlen_s_main_loop:
    // Load byte pointed to by x0 ready and test if it is zero
    // Branch to exit if it is 0
    ldrb    w3, [x0], #+1
    cmp     x3, #0
    b.eq    .L_strnlen_s_exit

    // Record we have another byte
    add     x2, x2, 1

    // Decrement our maximum string length counter
    // Branch to exit if it is zero
    subs    x1, x1, 1
    b.eq    .L_strnlen_s_exit

    // Repeat
    b       .L_strnlen_s_main_loop

.L_strnlen_s_exit:
    // Return the discovered length in x0
    mov     x0, x2

    // Return
    ret

It is called using:

_start:
    ldr     x0, =msg
    mov     x1, #7
    bl      strnlen_s
    bl      puthex
    bl      putnl
    ldr     x0, =msg
    bl      strlen
    bl      puthex
    bl      putnl
    bl      exit

The output is:

0x07
0x16

As usual, using the script, this can be assembled and run using:

aarch64 strlen
Keywords