Re: [vta, graphite?] propagate degenerate phi nodes into debug stmts

[Richard Guenther <richard.guenther@gmail.com>]

On Mon, Nov 16, 2009 at 9:41 PM, Alexandre Oliva <aoliva@redhat.com> wrote:
> On Nov  8, 2009, Richard Guenther <richard.guenther@gmail.com> wrote:
>
>> For the rest it would be better to re-organize the code as
>
>>    else if (gimple_code (def_stmt) == PHI_NODE)
>>      {
>>         value = degenerate_phi_result (def_stmt);
>>         if (value
>>             && walk_tree (&value, find_released_ssa_name, NULL, NULL))
>>           value = NULL_TREE;
>>      }
>
>> that looks simpler and it avoids walking PHI args uselessly.
>
> That doesn't work.  We crash deep within degenerate_phi_result given
> expressions containing released SSA names.  It's the same reason why we
> have to test for released SSA names before calling
> gimple_assign_rhs_to_tree: IIRC it has to do with testing whether the
> already-NULL type of the SSA name is a pointer type.

Well, just adjust degenerate_phi_result to do instead of calling
operand_equal_p

  else if (TREE_CODE (arg) != TREE_CODE (val)
             || (TREE_CODE (arg) == SSA_NAME
                 && arg != val)
             || !operand_equal_p (arg, val, 0))
    break;

that should fix it.  Alternatively add the SSA_NAME shortcut
to operand_equal_p.

> Regardless, I reorganized the code so as to not have to test for
> PHI_NODEs as often.  There were two unrelated code paths intermixed with
> tests every now and again.  Now they're two separate code flows.
>
>> @@ -479,6 +508,13 @@ insert_debug_temps_for_defs (gimple_stmt
>
>>    stmt = gsi_stmt (*gsi);
>
>> +  if (gimple_code (stmt) == GIMPLE_PHI)
>> +    {
>> +      tree var = gimple_phi_result (stmt);
>> +      insert_debug_temp_for_var_def (gsi, var);
>> +      return;
>> +    }
>> +
>
>> This looks odd.  SSA DEF operand iteration should walk the PHI defs
>> as well, so the change should not be necessary.
>
> I thought so, too, but by the time we get there, the operands of the PHI
> stmt have already been disconnected.

It shouldn't be.  Please try to figure out why instead.

Thanks,
Richard.

> Here's what I'm going to test now.
>
>
>
> --
> Alexandre Oliva, freedom fighter    http://FSFLA.org/~lxoliva/
> You must be the change you wish to see in the world. -- Gandhi
> Be Free! -- http://FSFLA.org/   FSF Latin America board member
> Free Software Evangelist      Red Hat Brazil Compiler Engineer
>
>