[Bug tree-optimization/32328] [4.2/4.3 Regression] -fstrict-aliasing causes skipped code

["dberlin at dberlin dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #12 from dberlin at gcc dot gnu dot org  2007-06-20 20:03 -------
Subject: Re:  [4.2 Regression] -fstrict-aliasing causes skipped code

On 20 Jun 2007 15:12:53 -0000, rguenth at gcc dot gnu dot org
<gcc-bugzilla@gcc.gnu.org> wrote:
>
>
> ------- Comment #9 from rguenth at gcc dot gnu dot org  2007-06-20 15:12 -------
> Confirmed.
>
> struct barstruct { char const* some_string; };
>
> void changethepointer(struct barstruct**);
>
> void baz()
> {
>         struct barstruct bar1;
>         struct barstruct* barptr = &bar1;
>         changethepointer(&barptr);
>         barptr->some_string = "Everything's OK!";
> }
>
>
> We end up with
>
> baz ()
> {
>   struct barstruct * barptr;
>   struct barstruct bar1;
>   struct barstruct * barptr.0;
>
> <bb 2>:
>   #   barptr_2 = V_MUST_DEF <barptr_1>;
>   barptr = &bar1;
>   #   barptr_6 = V_MAY_DEF <barptr_2>;
>   #   SFT.1_7 = V_MAY_DEF <SFT.1_4>;
>   #   NONLOCAL.7_8 = V_MAY_DEF <NONLOCAL.7_5>;
>   changethepointer (&barptr);
>   #   VUSE <barptr_6>;
>   barptr.0_3 = barptr;
>   #   SFT.1_9 = V_MAY_DEF <SFT.1_7>;
>   barptr.0_3->some_string = &"Everything\'s OK!"[0];
>   return;
>
> }
>
> because of a wrong points-to set again:

Why do you believe this is wrong?
It looks exactly right

>
> barptr.0_3 = { bar1 }
>
> The constraints are
>
> Constraints:
>
> ANYTHING = &ANYTHING
> READONLY = &ANYTHING
> INTEGER = &ANYTHING
> ESCAPED_VARS = *ESCAPED_VARS
> NONLOCAL.7 = ESCAPED_VARS
> ESCAPED_VARS = &NONLOCAL.7
> ESCAPED_VARS = &NONLOCAL.7
> barptr = &bar1
> ESCAPED_VARS = &bar1
> ESCAPED_VARS = &barptr
> barptr.0_3 = barptr
>
> where ESCAPED_VARS = &barptr looks wrong?  I'd say it needs to be
> ESCAPED_VARS = barptr instead.

No, it's right on that part.  It says ESCAPED_VARS points to barptr.
ESCAPED_VARS = *ESCAPED_VARS will take care of making it point what
barptr points to.

What is missing here is that we should be doing barptr = ESCAPED_VARS
when we see &barptr passed to the call.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32328