why does MS C++ add this code to assembly?

Question

i have some code(inline assembly).

void NativeLoop()
{
    int m;
    __asm
    {
        PUSH ECX
        PUSH EDX
        MOV  ECX, 100000000
NEXTLOOP:
        MOV  EDX, ECX
        AND  EDX, 0X7FFFFFFF
        MOV  DWORD PTR m, EDX
        DEC  ECX
        JNZ  NEXTLOOP
        POP  EDX
        POP  ECX
    }
}

MS C++ Automagicaly adds these codes(marked with **) to my procedure.
Why?
how to avoid it?

  **push        ebp  
  **mov         ebp,esp 
  **push        ecx  
  push        ecx  
  push        edx  
  mov         ecx,5F5E100h 
NEXTLOOP:
  mov         edx,ecx 
  and         edx,7FFFFFFFh 
  mov         dword ptr m,edx 
  dec         ecx  
  jnz         NEXTLOOP
  pop         edx  
  pop         ecx  
  **mov         esp,ebp 
  **pop         ebp  
  **ret

Answer 1

It's maintaining the call stack. If you defined the function as

int NativeLoop() { }

You would see the same assembly.

Answer 2

If you can do a search on C++ calling conventions, you'll understand better what the compiler is doing.

Answer 3

I remember that you can __declspec(naked) in MSVC++, meaning that you have to take care of the stack yourself, that means you must save every register you clobber, and restore it.

There is no the-one-rule to do that properly, as it depends on calling convention. See http://en.wikipedia.org/wiki/X86_calling_conventions .

Sidenote: In gcc, you explitly state to the compiler what you will drive invalid, so that gcc will output more optimal save/restore/stackframe-code, if any. In MSVC, asm is mostly a blackbox to the compiler, for which it will often/always the worst.

See http://www.ibiblio.org/gferg/ldp/GCC-Inline-Assembly-HOWTO.html#ss5.3 , gcc inline asm syntax is more ugly, but effectively more effective.

Answer 4

It is the standard function entry and exit code. It establishes and tears down the stack frame. If you don't want it you can use __declspec(naked). Don't forget to include the RET if you do.

However, your snippet relies on a valid stack frame, your "m" variable requires it. It is addressed at [ebp-10]. Without the preamble, the ebp register won't be set correctly and you'll corrupt the stack frame of the caller.

Answer 5

  **push        ebp  ;save EBP register
  **mov         ebp,esp  ;Save the stackframe
  **push        ecx  ; So that the variable `m` has an address
;...
  **mov         esp,ebp ;restore the stack frame to it's original address
  **pop         ebp   ;restore EBP register
  **ret ;return from function call