[PR46761] graphite: fix testing of boundary wrapping

[Richard Guenther]

On Wed, Dec 15, 2010 at 4:46 PM, Alexander Monakov <amonakov@ispras.ru> wrote:
>
>
> On Mon, 13 Dec 2010, Richard Guenther wrote:
>
>> 2010/12/13 Sebastian Pop <sebpop@gmail.com>:
>> > On Mon, Dec 13, 2010 at 07:40, Alexander Monakov <amonakov@ispras.ru> wrote:
>> >> Hi,
>> >>
>> >> As indicated by the testcase, sometimes Graphite would generate wrong region
>> >> guards if UB_ONE overflows, but does not become zero (i.e. it is signed).
>> >> The patch changes the corresponding test to use TREE_OVERFLOW flag.
>> >>
>> >> Bootstrapped and regtested on x86_64-linux, OK for trunk?
>> >>
>> >> 2010-12-13  Alexander Monakov  <amonakov@ispras.ru>
>> >>
>> >>        PR middle-end/46761
>> >>        * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Use
>> >>        TREE_OVERFLOW_P to test overflow.
>> >>
>> >> testsuite:
>> >>        * gcc.dg/graphite/pr46761.c: New.
>> >>
>> >
>> > Ok.  Thanks for fixing this,
>>
>> Hmm, I don't like new uses of TREE_OVERFLOW checking.  And for
>> unsigned types it won't be set anyway.
>
> Thanks.  The following patch changes guard generation so that using unadjusted
> UB is preferred (if it's an integer constant or an SSA_NAME).  If it is
> neither, it uses the previous behaviour of bumping UB by one and changing the
> test.  Bootstrapped and tested with make -k check RUNTESTFLAGS="graphite.exp",
> OK for trunk?

That looks better.

Thanks,
Richard.

> 2010-12-15  Alexander Monakov  <amonakov@ispras.ru>
>
>         PR middle-end/46761
>         * graphite-clast-to-gimple.c (graphite_create_new_loop_guard): Prefer
>         to use unadjusted UB.
>
> diff --git a/gcc/graphite-clast-to-gimple.c b/gcc/graphite-clast-to-gimple.c
> index 4894b52..4725608 100644
> --- a/gcc/graphite-clast-to-gimple.c
> +++ b/gcc/graphite-clast-to-gimple.c
> @@ -975,20 +975,24 @@ graphite_create_new_loop_guard (sese region, edge entry_edge,
>                                     newivs_index, params_index);
>   tree ub = clast_to_gcc_expression (type, stmt->UB, region, newivs,
>                                     newivs_index, params_index);
> -  tree one = POINTER_TYPE_P (type) ? size_one_node
> -    : fold_convert (type, integer_one_node);
> -  /* Adding +1 and using LT_EXPR helps with loop latches that have a
> -     loop iteration count of "PARAMETER - 1".  For PARAMETER == 0 this becomes
> -     2^{32|64}, and the condition lb <= ub is true, even if we do not want this.
> -     However lb < ub + 1 is false, as expected.  */
> -  tree ub_one = fold_build2 (POINTER_TYPE_P (type) ? POINTER_PLUS_EXPR
> -                            : PLUS_EXPR, type, ub, one);
> -
> -  /* When ub + 1 wraps around, use lb <= ub.  */
> -  if (integer_zerop (ub_one))
> +  /* When ub is simply a constant or a parameter, use lb <= ub.  */
> +  if (TREE_CODE (ub) == INTEGER_CST || TREE_CODE (ub) == SSA_NAME)
>     cond_expr = fold_build2 (LE_EXPR, boolean_type_node, lb, ub);
>   else
> -    cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
> +    {
> +      tree one = (POINTER_TYPE_P (type)
> +                 ? size_one_node
> +                 : fold_convert (type, integer_one_node));
> +      /* Adding +1 and using LT_EXPR helps with loop latches that have a
> +        loop iteration count of "PARAMETER - 1".  For PARAMETER == 0 this becomes
> +        2^k-1 due to unsigned overflow, and the condition lb <= ub is true,
> +        even if we do not want this.  However lb < ub + 1 is false, as
> +        expected.  */
> +      tree ub_one = fold_build2 (POINTER_TYPE_P (type) ? POINTER_PLUS_EXPR
> +                                : PLUS_EXPR, type, ub, one);
> +
> +      cond_expr = fold_build2 (LT_EXPR, boolean_type_node, lb, ub_one);
> +    }
>
>   exit_edge = create_empty_if_region_on_edge (entry_edge, cond_expr);
>
>