Another question re asm constraints

Another question re asm constraints

[dw]

Does specifying the + operator on outputs remove the need for listing 
the variable as an input?  For example:

     __CRT_INLINE SHORT AddTest(SHORT *Addend)
     {
       SHORT ret = 1;
       __asm__ __volatile__("lock\n\txaddw %0,%1 "
     : "+r" (ret), "+m" (*Addend)
     : "0" (ret)
     : "cc");
       return ret + 1;
     }

Is the "0" constraint accomplishing anything useful here (such as 
ensuring the 1 is in the register)?  Or is it just advertising the fact 
that I don't fully understand how constraints work?

And more importantly, is the "+m" enough to ensure that any existing 
values for Addend have been written to memory?  Or do I need to add "m" 
(*Addend) as an input?

I've run some tests, and as best I can tell, neither input constraint is 
required.  But I don't rule out the possibility that that's just an 
artifact of my test code.  I can say for sure that "1" (*Addend) is the 
wrong answer (compile error).

Thanks.

Re: Another question re asm constraints

[Ian Lance Taylor]

On Mon, Mar 25, 2013 at 2:51 PM, dw <limegreensocks@yahoo.com> wrote:
> Does specifying the + operator on outputs remove the need for listing the
> variable as an input?  For example:
>
>     __CRT_INLINE SHORT AddTest(SHORT *Addend)
>     {
>       SHORT ret = 1;
>       __asm__ __volatile__("lock\n\txaddw %0,%1 "
>     : "+r" (ret), "+m" (*Addend)
>     : "0" (ret)
>     : "cc");
>       return ret + 1;
>     }
>
> Is the "0" constraint accomplishing anything useful here

No.

> And more importantly, is the "+m" enough to ensure that any existing values
> for Addend have been written to memory?

Yes, at least from the compiler's perspective (there may be pending
writes from another processor that the compiler knows nothing about).

Ian