ipa: Check cst type when propagating controled uses info (PR 105639)

Hi,

PR 105639 shows that code with type-mismatches can trigger an assert
after runnning into a branch that was inteded only for references to
variables - as opposed to references to functions.  Fixed by moving
the condition from the assert to the guarding if statement.

Bootstrapped and tested on x86_64.  OK for trunk and then gcc 12?

Thanks,

Martin



gcc/ChangeLog:

2022-05-25  Martin Jambor  <mjambor@suse.cz>

	PR ipa/105639
	* ipa-prop.cc (propagate_controlled_uses): Check type of the
	constant before adding a LOAD reference.

gcc/testsuite/ChangeLog:

2022-05-25  Martin Jambor  <mjambor@suse.cz>

	PR ipa/105639
	* gcc.dg/ipa/pr105639.c: New test.
---
 gcc/ipa-prop.cc                     | 10 ++++------
 gcc/testsuite/gcc.dg/ipa/pr105639.c | 16 ++++++++++++++++
 2 files changed, 20 insertions(+), 6 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/ipa/pr105639.c

On Thu, May 26, 2022 at 4:47 PM Martin Jambor <mjambor@suse.cz> wrote:
>
> Hi,
>
> PR 105639 shows that code with type-mismatches can trigger an assert
> after runnning into a branch that was inteded only for references to
> variables - as opposed to references to functions.  Fixed by moving
> the condition from the assert to the guarding if statement.
>
> Bootstrapped and tested on x86_64.  OK for trunk and then gcc 12?

OK.  There's the same assert below - I guess that doesn't need the same
treatment?

Thanks,
Richard.

> Thanks,
>
> Martin
>
>
>
> gcc/ChangeLog:
>
> 2022-05-25  Martin Jambor  <mjambor@suse.cz>
>
>         PR ipa/105639
>         * ipa-prop.cc (propagate_controlled_uses): Check type of the
>         constant before adding a LOAD reference.
>
> gcc/testsuite/ChangeLog:
>
> 2022-05-25  Martin Jambor  <mjambor@suse.cz>
>
>         PR ipa/105639
>         * gcc.dg/ipa/pr105639.c: New test.
> ---
>  gcc/ipa-prop.cc                     | 10 ++++------
>  gcc/testsuite/gcc.dg/ipa/pr105639.c | 16 ++++++++++++++++
>  2 files changed, 20 insertions(+), 6 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.dg/ipa/pr105639.c
>
> diff --git a/gcc/ipa-prop.cc b/gcc/ipa-prop.cc
> index c6c745f84a0..afd9222b5a2 100644
> --- a/gcc/ipa-prop.cc
> +++ b/gcc/ipa-prop.cc
> @@ -4187,14 +4187,13 @@ propagate_controlled_uses (struct cgraph_edge *cs)
>         {
>           int d = ipa_get_controlled_uses (old_root_info, i);
>           int c = rdesc->refcount;
> +         tree cst = ipa_get_jf_constant (jf);
>           rdesc->refcount = combine_controlled_uses_counters (c, d);
>           if (rdesc->refcount != IPA_UNDESCRIBED_USE
> -             && ipa_get_param_load_dereferenced (old_root_info, i))
> +             && ipa_get_param_load_dereferenced (old_root_info, i)
> +             && TREE_CODE (cst) == ADDR_EXPR
> +             && TREE_CODE (TREE_OPERAND (cst, 0)) == VAR_DECL)
>             {
> -             tree cst = ipa_get_jf_constant (jf);
> -             gcc_checking_assert (TREE_CODE (cst) == ADDR_EXPR
> -                                  && (TREE_CODE (TREE_OPERAND (cst, 0))
> -                                      == VAR_DECL));
>               symtab_node *n = symtab_node::get (TREE_OPERAND (cst, 0));
>               new_root->create_reference (n, IPA_REF_LOAD, NULL);
>               if (dump_file)
> @@ -4204,7 +4203,6 @@ propagate_controlled_uses (struct cgraph_edge *cs)
>             }
>           if (rdesc->refcount == 0)
>             {
> -             tree cst = ipa_get_jf_constant (jf);
>               gcc_checking_assert (TREE_CODE (cst) == ADDR_EXPR
>                                    && ((TREE_CODE (TREE_OPERAND (cst, 0))
>                                         == FUNCTION_DECL)
> diff --git a/gcc/testsuite/gcc.dg/ipa/pr105639.c b/gcc/testsuite/gcc.dg/ipa/pr105639.c
> new file mode 100644
> index 00000000000..5534fe93fbf
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/ipa/pr105639.c
> @@ -0,0 +1,16 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O -w" } */
> +
> +void typedef (*cb) (void);
> +
> +static void
> +bar (cb *fp)
> +{
> +  (*fp) ();
> +}
> +
> +void
> +foo (void)
> +{
> +  bar (foo);
> +}
> --
> 2.36.0
>

Hi,

On Fri, May 27 2022, Richard Biener wrote:
> On Thu, May 26, 2022 at 4:47 PM Martin Jambor <mjambor@suse.cz> wrote:
>>
>> Hi,
>>
>> PR 105639 shows that code with type-mismatches can trigger an assert
>> after runnning into a branch that was inteded only for references to
>> variables - as opposed to references to functions.  Fixed by moving
>> the condition from the assert to the guarding if statement.
>>
>> Bootstrapped and tested on x86_64.  OK for trunk and then gcc 12?
>
> OK.  There's the same assert below - I guess that doesn't need the same
> treatment?

Thanks.  The assert below also allows the case when cst is an ADDR_EXPR of
a function.  And the whole reference-counting is initiated only for
these two cases.  So in all other cases the condition

  rdesc->refcount != IPA_UNDESCRIBED_USE

will be false.

Martin

>>
>>
>> gcc/ChangeLog:
>>
>> 2022-05-25  Martin Jambor  <mjambor@suse.cz>
>>
>>         PR ipa/105639
>>         * ipa-prop.cc (propagate_controlled_uses): Check type of the
>>         constant before adding a LOAD reference.
>>
>> gcc/testsuite/ChangeLog:
>>
>> 2022-05-25  Martin Jambor  <mjambor@suse.cz>
>>
>>         PR ipa/105639
>>         * gcc.dg/ipa/pr105639.c: New test.