This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] Enable saliasing for structs with arrays

On Fri, 2005-07-22 at 09:58 +0200, Richard Guenther wrote:
> On Thu, 21 Jul 2005, Jan Hubicka wrote:
> > > For unions we don't create subvars, as is, I'm looking for an exact
> > Has that changed with your patch?
> > I quite definitly was debugging some issues with subvars on unions ;)
> Well, I have that information from existing code:
> static unsigned int
> create_variable_info_for (tree decl, const char *name)
> {
> ...
>   hasunion = TREE_CODE (decltype) == UNION_TYPE
>              || TREE_CODE (decltype) == QUAL_UNION_TYPE;
>   if (var_can_have_subvars (decl) && use_field_sensitive && !hasunion)
>     {
>       push_fields_onto_fieldstack (decltype, &fieldstack, 0, &hasunion);
>       if (hasunion)
>         {
>           VEC_free (fieldoff_s, heap, fieldstack);
>           notokay = true;
> at least it creates at most one subvar for the union itself, but
> doesn't decompose it.

No, you've got it wrong.

You are confusing the structure var creation code with the structalias

The points-to analyzer globs unions to a single variable.

The structure var creation does not, and handles unions just fine.

Look at tree-ssa-alias.c:create_overlap_variables

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]