On Fri, Dec 7, 2018 at 3:59 PM Martin Jambor <mjam...@suse.cz> wrote:
>
> Hi,
>
> ICE in PR 88214 happens because a type-mismatch in K&R C code makes
> IPA-CP analysis call ao_ref_init_from_ptr_and_size on an integer
> SSA_NAME, this function in turn constructs a temporary MEM_REF based on
> that integer SSA_NAME and then later on call_may_clobber_ref_p_1 treats
> the MEM_REF base as a pointer, gets its SSA_NAME_PTR_INFO and tries to
> work with bitmaps there.  But because the SSA_NAME is an integer, there
> is no SSA_NAME_PTR_INFO, there is range info instead and this leads to a
> crash.
>
> On a related note, would people object to adding the following assert,
> which would have made this bug much more straightforward to find?

That's fine with me.

> index 85a5de7..66cf2f2 100644
> --- a/gcc/tree-ssa-alias.c
> +++ b/gcc/tree-ssa-alias.c
> @@ -710,6 +710,7 @@ ao_ref_init_from_ptr_and_size (ao_ref *ref, tree ptr, 
> tree size)
>      }
>    else
>      {
> +      gcc_assert (POINTER_TYPE_P (TREE_TYPE (ptr)));
>        ref->base = build2 (MEM_REF, char_type_node,
>                           ptr, null_pointer_node);
>        ref->offset = 0;
>
>
> The bug itself can be fixed with the patch below.  I have verified it
> avoids the ICE on powerpc64-linux and did a full bootstrap and test on
> an x86_64-linux.  The patch is simple enough that I believe that is good
> enough.

OK.

Richard.

>
> 2018-12-06  Martin Jambor  <mjam...@suse.cz>
>
>         PR ipa/88214
>         * ipa-prop.c (determine_locally_known_aggregate_parts): Make sure
>         we check pointers against pointers.
>
>         testsuite/
>         * gcc.dg/ipa/pr88214.c: New test.
> ---
>  gcc/ipa-prop.c                     |  3 ++-
>  gcc/testsuite/gcc.dg/ipa/pr88214.c | 10 ++++++++++
>  2 files changed, 12 insertions(+), 1 deletion(-)
>  create mode 100644 gcc/testsuite/gcc.dg/ipa/pr88214.c
>
> diff --git a/gcc/ipa-prop.c b/gcc/ipa-prop.c
> index 74052350ac1..4dbe26829e3 100644
> --- a/gcc/ipa-prop.c
> +++ b/gcc/ipa-prop.c
> @@ -1569,7 +1569,8 @@ determine_locally_known_aggregate_parts (gcall *call, 
> tree arg,
>        if (TREE_CODE (arg) == SSA_NAME)
>         {
>           tree type_size;
> -          if (!tree_fits_uhwi_p (TYPE_SIZE (TREE_TYPE (arg_type))))
> +          if (!tree_fits_uhwi_p (TYPE_SIZE (TREE_TYPE (arg_type)))
> +             || !POINTER_TYPE_P (TREE_TYPE (arg)))
>              return;
>           check_ref = true;
>           arg_base = arg;
> diff --git a/gcc/testsuite/gcc.dg/ipa/pr88214.c 
> b/gcc/testsuite/gcc.dg/ipa/pr88214.c
> new file mode 100644
> index 00000000000..4daa9829e75
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/ipa/pr88214.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2" } */
> +
> +void i();
> +  short a;
> +  void b(e) char * e;
> +  {
> +    i();
> +    b(a);
> +  }
> --
> 2.19.1
>
>
>

Reply via email to