On 29/10/15 12:13, Richard Biener wrote:
> On Wed, 28 Oct 2015, Tom de Vries wrote:
>
>> >On 28/10/15 16:35, Richard Biener wrote:
>>> > >On Tue, 27 Oct 2015, Tom de Vries wrote:
>>> > >
>>>> > > >On 27/10/15 13:24, Tom de Vries wrote:
>>>>> > > > >Thinking it over a bit more, I realized the constraint handling started
>>>>> > > > >to be messy. I've reworked the patch series to simplify that first.
>>>>> > > > >
>>>>> > > > >        1    Simplify constraint handling
>>>>> > > > >        2    Rename make_restrict_var_constraints to
>>>>> > > > >make_param_constraints
>>>>> > > > >        3    Add recursion to make_param_constraints
>>>>> > > > >        4    Add handle_param parameter to create_variable_info_for_1
>>>>> > > > >        5    Handle recursive restrict pointer in
>>>>> > > > >create_variable_info_for_1
>>>>> > > > >        6    Handle restrict struct fields recursively
>>>>> > > > >
>>>>> > > > >Currently doing bootstrap and regtest on x86_64.
>>>>> > > > >
>>>>> > > > >I'll repost the patch series in reply to this message.
>>>> > > >
>>>> > > >This patch gets rid of this bit of code in intra_create_variable_infos:
>>>> > > >...
>>>> > > >        if (restrict_pointer_p)
>>>> > > >	make_constraint_from_global_restrict (p, "PARM_RESTRICT");
>>>> > > >        else
>>>> > > >..
>>>> > > >
>>>> > > >I already proposed to remove it here (
>>>> > > >https://gcc.gnu.org/ml/gcc-patches/2015-10/msg02426.html  ) but there is a
>>>> > > >problem with that approach: It can happen that restrict_pointer_p is true,
>>>> > > >but
>>>> > > >p->only_restrict_pointers is false. This happens with fipa-pta, when
>>>> > > >create_function_info_for created a varinfo for the parameter before
>>>> > > >intra_create_variable_infos was called.
>>>> > > >
>>>> > > >The patch handles that case now by setting p->only_restrict_pointers.
>>> > >
>>> > >Hmm, but ... restrict only has an effect in non-IPA mode.
>> >
>> >Indeed, I also realized that by now.
>> >
>> >I wrote attached patch to make that explicit and simplify fipa-pta.
>> >
>> >OK for trunk if bootstrap and reg-test succeeds?

First, there was an error in the patch, it tested for flag_ipa_pta (so 
it also affected ealias), but it was supposed to test for in_ipa mode. 
That is fixed in attached version.

 > I don't see the patch simplifies anything but only removes spurious
 > settings by adding IMHO redundant checks.

Consider testcase:
...
int __attribute__((noinline, noclone))
foo (int *__restrict__ a, int *__restrict__ b)
{
   *a = 1;
   *b = 2;
}

int __attribute__((noinline, noclone))
bar (int *a, int *b)
{
   foo (a, b);
}
...

The impact of this patch in the pta dump (focusing on the constraints 
bit) is:
...
  Generating constraints for foo (foo)

-foo.arg0 = &PARM_NOALIAS(20)
-PARM_NOALIAS(20) = NONLOCAL
-foo.arg1 = &PARM_NOALIAS(21)
-PARM_NOALIAS(21) = NONLOCAL
+foo.arg0 = &NONLOCAL
+foo.arg1 = &NONLOCAL
...

That's the kind of simplifications I'm trying to achieve.

Thanks,
- Tom