[Bug tree-optimization/16179] useless copies not optimized away

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-06-24 18:26 -------
Subject: Re:  useless copies not optimized away

pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-24 17:57 -------
> This is not a bug, as bar could change f at any time.
> So lets examine the code:
> extern void bar(void);
> 
> int foo(const int &f)
> {
>         const int a = f; <-- load *f
>         int c = a; <-- move into c from a (all local to foo, c is not take the address of so it cannot be 
> aliased, likewise for a)
>         bar(); <-- call bar, can change *f
>         c = a; <-- move into c from a (all local to foo again)
>         return c; <-- return the value in c
> }
> 
> int foo2(const int &f)
> {
>         int c = f; <-- load *f into c
>         bar(); <-- call bar, can change *f
>         c = f;<-- load *f into c
>         return c; 
> }
> 
> So these two code cannot be equivant at all.
> 

I think they are - if bar() modifies f, the behavior is undefined as of 
6.7.3.5 (C99), 7.1.5.1.4 (C++).  Not?

If they are _not_ the same, if I want to enforce const-ness for the 
optimizers, I need to place local copies of arguments so the compiler
can tell from aliasing that modification cannot take place?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16179