切换执行； 臂; 汇编器 aarch64; 臂64

Question

I'm interesting in the ways to implement 'switch' operator on aarch64 assembler. On arm32 platforms I used something like

    ldr         pc,         [pc, ta, LSL#2]
    nop                                         // alignment
    .int        .L.case1
    .int        .L.case2
    ...
    .int        .L.caseN

But since 64bit version has a lot of restriction on 'pc' register usage, such implementation doesn't work any more.

It seems that the easiest way is to use pair of compare and branch operations, like

cmp ta, #1
b.eq .L.case1
cmp ta, #2
b.eq .L.case2
...

But sometimes there are up to dozen cases and it would lead to a significant delay before reaching the last 'case'.

Could you share your ideas how to implement fast switch on aarch64, please.

Thanks :)

Answer 1

I don't have an 64-bit ARM assembler to test this with but I believe you would do something like the following to implement a jump table:

    adr  x0, jmp_table
    ldr  x0, [x0, x1, LSL#3]
    br   x0

jmp_table:
    .quad .L.case1
    .quad .L.case2
    .quad .L.case3

The first instruction, ADR , loads the address of a label into a register. The last instruction, BR , jumps to the address stored in the register.

If you're creating a shared library or a position independent executable you can try something like following:

    adr  x0, jmp_table
    add  x0, x0, x1, LSL#2
    br   x0

jmp_table:
    b .L.case1
    b .L.case2
    b .L.case3

Alternate PIC example

    adr  x0, jmp_table
    ldr  w1, [x0, x1, LSL#2]
    add  x0, x0, x1
    br   x0

jmp_table:
    .int  .L.case1 - jmp_table
    .int  .L.case2 - jmp_table
    .int  .L.case3 - jmp_table