Using call-preserved registers vs just the stack for saving values around a function call

Asked Dec 25 '22 at 19:10

Active Dec 25 '22 at 20:01

Viewed 36 times

Take:

void callee(void);
long call_and_ret_saved(long ToSave){
    callee();
    return ToSave;
}

where the generated assembly will need to save the value of the ToSave argument in order to be able to return it after the call to callee.

Both gcc and clang generate:

call_and_ret_saved:
    pushq   %rbx
    movq    %rdi, %rbx
    call    callee
    movq    %rbx, %rax
    popq    %rbx
    ret

for it. Why not simply push %rdi; call callee; pop %rax; ret; ? Does using the call-preserved register make the return value available sooner or something?

edited Dec 25 '22 at 20:01

Michael Petch

46,082
8
107
198

asked Dec 25 '22 at 19:10

Petr Skocik

58,047
6
95
142

1

What optimization level are you using? Try with -O3 – Dario Petrillo Dec 25 '22 at 19:25
@DarioPetrillo Their codegen for this is the same at all optimizing (!=-O0) levels https://godbolt.org/z/TWfP8ecbo. – Petr Skocik Dec 25 '22 at 19:31
Looks like a missed optimization to me -- or an unwarranted optimization choice to use a call-preserved register for that. – Erik Eidt Dec 25 '22 at 21:23
@JosephSible-ReinstateMonica Yup. That's an quintessential duplicate. Thanks. – Petr Skocik Dec 25 '22 at 21:34

Using call-preserved registers vs just the stack for saving values around a function call

0 Answers0