[Bug inline-asm/67944] New: GCC emits unnecessary push/pop for callee-save reads.

[Bug inline-asm/67944] New: GCC emits unnecessary push/pop for callee-save reads.

[alex.reinking at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67944

            Bug ID: 67944
           Summary: GCC emits unnecessary push/pop for callee-save reads.
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: inline-asm
          Assignee: unassigned at gcc dot gnu.org
          Reporter: alex.reinking at gmail dot com
  Target Milestone: ---

When trying to return the current value of a register in a function via a local
register variable, the compiler will emit a useless push/pop instruction as in 
the following example:

C code:

    unsigned int readEBX(void) {
        register unsigned int reg asm("ebx");
        return reg;
    }

Assembly:

    readEBX():
        mov  eax, ebx
        push ebx
        pop  ebx
        ret

This same pattern occurs for all callee-save registers: ebx, esi, edi, ebp
The code is properly-optimized for: eax, ecx, edx, esp

See also my StackOverflow question: 
See also this Godbolt* result: https://goo.gl/bzgt4P

  * Try substituting different registers to compare

[Bug inline-asm/67944] GCC emits unnecessary push/pop for callee-save reads.

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67944

Segher Boessenkool <segher at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |segher at gcc dot gnu.org

--- Comment #1 from Segher Boessenkool <segher at gcc dot gnu.org> ---
You're returning an uninitialised variable.  What do you expect this
code to do?

[Bug inline-asm/67944] GCC emits unnecessary push/pop for callee-save reads.

[alex.reinking at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67944

--- Comment #2 from Alex Reinking <alex.reinking at gmail dot com> ---
I expect it to return whatever is currently stored in ebx. Which it does, but
suboptimally. In my particular case, ebx is populated by a system call. I
wanted to put the inline assembly for retrieving that value in its own function
so I wouldn't have to remember the local register variable syntax.

The code I posted is simply a somehow minimal example of the anomalous assembly
generated. The question here is more "Why does gcc emit a push ebx; pop ebx?"
than "How should I retrieve whatever is currently in ebx?"

[Bug inline-asm/67944] GCC emits unnecessary push/pop for callee-save reads.

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67944

--- Comment #3 from Segher Boessenkool <segher at gcc dot gnu.org> ---
Your expectation is not in line with what the compiler actually promises
you: the only guarantee is that the local register variable will be in
(in your case) ebx _where used in an asm statement_.

Other than that, it is just a local variable.

You get a save/restore sequence in the prologue and epilogue because ebx
is a non-volatile register and this function uses that register.  Scheduling
later moves the push later in the code.

If you want to simply grab what is in ebx, you can use asm, like

long reg; asm("mov %%ebx,%0" : "=r"(reg));