[Bug ada/43096] [4.5 regression] miscompilation of ACATS c37105a at -O2

["rguenther at suse dot de" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #7 from rguenther at suse dot de  2010-02-26 19:56 -------
Subject: Re:  [4.5 regression] miscompilation of ACATS c37105a
 at -O2

On Fri, 26 Feb 2010, ebotcazou at gcc dot gnu dot org wrote:

> 
> 
> ------- Comment #6 from ebotcazou at gcc dot gnu dot org  2010-02-26 17:46 -------
> > We might be able to save the day with the help of TYPE_CANONICAL in this case
> > since the size is fixed.
> 
> TYPE_CANONICAL is too strong, it will cause useless_type_conversion_p to return
> true for conversions between type and subtypes and, even with fixed size, these
> conversions aren't useless.  So we need this for Ada:
> 
> Index: tree-ssa-alias.c
> ===================================================================
> --- tree-ssa-alias.c    (revision 156989)
> +++ tree-ssa-alias.c    (working copy)
> @@ -544,13 +544,15 @@ same_type_for_tbaa (tree type1, tree typ
>        && TREE_CODE (type2) == ARRAY_TYPE)
>      return -1;
> 
> -  /* In Ada, an lvalue of unconstrained type can be used to access an object
> -     of one of its constrained subtypes, for example when a function with an
> -     unconstrained parameter passed by reference is called on a constrained
> -     object and inlined.  In this case, the types have the same alias set.  */
> -  if (TYPE_SIZE (type1) && TYPE_SIZE (type2)
> -      && TREE_CONSTANT (TYPE_SIZE (type1)) != TREE_CONSTANT (TYPE_SIZE
> (type2))
> -      && get_alias_set (type1) == get_alias_set (type2))
> +  /* ??? In Ada, an lvalue of an unconstrained type can be used to access an
> +     object of one of its constrained subtypes, e.g. when a function with an
> +     unconstrained parameter passed by reference is called on an object and
> +     inlined.  But, even in the case of a fixed size, type and subtypes are
> +     not equivalent enough as to share the same TYPE_CANONICAL, since this
> +     would mean that conversions between them are useless, whereas they are
> +     not (e.g. type and subtypes can have different modes).  So, in the end,
> +     they are only guaranteed to have the same alias set.  */
> +  if (get_alias_set (type1) == get_alias_set (type2))
>      return -1;
> 
>    /* The types are known to be not equal.  */
> 
> 
> The change doesn't introduce any regressions in the C/C++ testsuite on x86 so
> it might not be a big pessimization.
> 
> Richard, what do you think?

Looks reasonable.

Richard.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43096