On Wed, 28 Oct 2015, Tom de Vries wrote:
> Richard,
>
> when compiling this testcase:
> ...
> static int __attribute__((noinline, noclone))
> foo (int *a, int *b)
> {
> *b = 1;
> *a = 2;
> return *b;
> }
>
> int __attribute__((noinline, noclone))
> bar (int *a, int *b)
> {
> return foo (a, b);
> }
> ...
>
> with -O2 -fipa-pta we find in the pta dumpfile:
> ...
> Generating constraints for bar (bar)
>
> bar.arg0 = &NONLOCAL
> bar.arg1 = &NONLOCAL
> bar.arg1 = &NONLOCAL
> ...
>
> The reason for the duplicate last two constraints is that with fipa-pta, in
> create_function_info_for we link the function arguments in a next chain.
>
> And in intra_create_variable_infos there are two iteration mechanism used:
> - the loop over the function arguments
> - the loop over the vi_next (p) for each function argument p
>
> So when processing argument a, we generate constraints for both a and it's
> next b.
> And when processing argument b, we generate constraints for b again.
>
> Is this a known issue? Should we fix this?
Didn't see that yet. Yes, we should fix it.
Index: gcc/tree-ssa-structalias.c
===================================================================
--- gcc/tree-ssa-structalias.c (revision 229481)
+++ gcc/tree-ssa-structalias.c (working copy)
@@ -5913,6 +6009,8 @@ intra_create_variable_infos (struct func
make_constraint_from_global_restrict (p, "PARM_RESTRICT");
else if (p->may_have_pointers)
make_constraint_from (p, nonlocal_id);
+ if (p->is_full_var)
+ break;
}
}
}
does for me. Pre-approved if it passes testing.
Richard.
> Thanks,
> - Tom
>
>
--
Richard Biener <[email protected]>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB
21284 (AG Nuernberg)
