[PATCH] Fix for PR62037

[Felix Yang]

Hi Richard,

    Yeah, the testcase is there. Sorry for my mistake.
    Thank you for back-porting this patch to 4.8 & 4.9 branch.

Cheers,
Felix


On Wed, Aug 13, 2014 at 5:30 PM, Richard Biener
<richard.guenther@gmail.com> wrote:
> On Tue, Aug 12, 2014 at 6:40 PM, Felix Yang <fei.yang0953@gmail.com> wrote:
>> Hi Richard,
>>
>>     I find that you missed the testcase with when committing the patch.
>
> I don't think so.
>
>>     GCC 4.8 & 4.9 branch also has the same issue, may be we need to
>> fix for them too?
>
> Yeah, I'll backport it.
>
> Thanks,
> Richard.
>
>> Cheers,
>> Felix
>>
>>
>> On Mon, Aug 11, 2014 at 7:24 PM, Richard Biener
>> <richard.guenther@gmail.com> wrote:
>>> On Sat, Aug 9, 2014 at 6:28 AM, Felix Yang <fei.yang0953@gmail.com> wrote:
>>>> Attached please find the patch and testcase for PR62037.
>>>>
>>>> DEF1 can be a GIMPLE_NOP and gimple_bb will be NULL then. The patch
>>>> checks for that.
>>>> Bootstrapped on x86_64-suse-linux. OK for trunk? Please commit this
>>>> patch if it's OK.
>>>>
>>>
>>> Thanks - applied.
>>>
>>> Richard.
>>>
>>>> Index: gcc/ChangeLog
>>>> ===================================================================
>>>> --- gcc/ChangeLog    (revision 213772)
>>>> +++ gcc/ChangeLog    (working copy)
>>>> @@ -1,3 +1,9 @@
>>>> +2014-08-09  Felix Yang  <fei.yang0953@gmail.com>
>>>> +
>>>> +    PR tree-optimization/62073
>>>> +    * tree-vect-loop.c (vect_is_simple_reduction_1): Check that DEF1 has
>>>> +    a basic block.
>>>> +
>>>>  2014-08-08  Guozhi Wei  <carrot@google.com>
>>>>
>>>>      * config/rs6000/rs6000.md (*movdi_internal64): Add a new constraint.
>>>> Index: gcc/testsuite/gcc.dg/vect/pr62073.c
>>>> ===================================================================
>>>> --- gcc/testsuite/gcc.dg/vect/pr62073.c    (revision 0)
>>>> +++ gcc/testsuite/gcc.dg/vect/pr62073.c    (revision 0)
>>>> @@ -0,0 +1,41 @@
>>>> +/* { dg-do compile } */
>>>> +/* { dg-options "-O1 -ftree-vectorize" } */
>>>> +
>>>> +struct S0
>>>> +{
>>>> +    int f7;
>>>> +};
>>>> +struct S0 g_50;
>>>> +int g_70;
>>>> +int g_76;
>>>> +
>>>> +
>>>> +int foo (long long p_56, int * p_57)
>>>> +{
>>>> +    int *l_77;
>>>> +    int l_101;
>>>> +
>>>> +    for (; g_70;)
>>>> +    {
>>>> +        int **l_78 = &l_77;
>>>> +        if (g_50.f7)
>>>> +            continue;
>>>> +        *l_78 = 0;
>>>> +    }
>>>> +    for (g_76 = 1; g_76 >= 0; g_76--)
>>>> +    {
>>>> +        int *l_90;
>>>> +        for (l_101 = 4; l_101 >= 0; l_101--)
>>>> +            if (l_101)
>>>> +                *l_90 = 0;
>>>> +            else
>>>> +            {
>>>> +                int **l_113 = &l_77;
>>>> +                *l_113 = p_57;
>>>> +            }
>>>> +    }
>>>> +
>>>> +    return *l_77;
>>>> +}
>>>> +
>>>> +/* { dg-final { cleanup-tree-dump "vect" } } */
>>>> Index: gcc/testsuite/ChangeLog
>>>> ===================================================================
>>>> --- gcc/testsuite/ChangeLog    (revision 213772)
>>>> +++ gcc/testsuite/ChangeLog    (working copy)
>>>> @@ -1,3 +1,8 @@
>>>> +2014-08-09  Felix Yang  <fei.yang0953@gmail.com>
>>>> +
>>>> +    PR tree-optimization/62073
>>>> +    * gcc.dg/vect/pr62073.c: New test.
>>>> +
>>>>  2014-08-08  Richard Biener  <rguenther@suse.de>
>>>>
>>>>      * gcc.dg/strlenopt-8.c: Remove XFAIL.
>>>> Index: gcc/tree-vect-loop.c
>>>> ===================================================================
>>>> --- gcc/tree-vect-loop.c    (revision 213772)
>>>> +++ gcc/tree-vect-loop.c    (working copy)
>>>> @@ -2321,7 +2321,8 @@ vect_is_simple_reduction_1 (loop_vec_info loop_inf
>>>>          }
>>>>
>>>>        def1 = SSA_NAME_DEF_STMT (op1);
>>>> -      if (flow_bb_inside_loop_p (loop, gimple_bb (def_stmt))
>>>> +      if (gimple_bb (def1)
>>>> +          && flow_bb_inside_loop_p (loop, gimple_bb (def_stmt))
>>>>            && loop->inner
>>>>            && flow_bb_inside_loop_p (loop->inner, gimple_bb (def1))
>>>>            && is_gimple_assign (def1))
>>>>
>>>>
>>>> Cheers,
>>>> Felix