[Bug middle-end/36296] bogus uninitialized warning (loop representation)

["manu at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org>]

------- Comment #10 from manu at gcc dot gnu dot org  2008-08-18 23:38 -------
(In reply to comment #9)
> (In reply to comment #8)
> > Please provide a preprocessed reduced testcase as similar to the original as
> > possible. 
> 
> Here's a similar testcase.

Thanks.

This is the optimized SSA dump:

f ()
{
  voidD.39 * pD.1952;
  intD.0 initD.1951;

  # BLOCK 2 freq:1
  # PRED: ENTRY [100.0%]  (fallthru,exec)
  # SUCC: 3 [100.0%]  (fallthru,exec)

  # BLOCK 3 freq:10000
  # PRED: 2 [100.0%]  (fallthru,exec) 5 [100.0%]  (fallthru,dfs_back,exec)
  # initD.1951_1 = PHI <0(2), initD.1951_2(5)>
  # pD.1952_3 = PHI <pD.1952_6(D)(2), pD.1952_4(5)>
  [/home/manuel/src/pr36296.c : 11] if (initD.1951_1 == 0)
    goto <bb 4>;
  else
    goto <bb 5>;
  # SUCC: 4 [29.0%]  (true,exec) 5 [71.0%]  (false,exec)

  # BLOCK 4 freq:2900
  # PRED: 3 [29.0%]  (true,exec)
  [/home/manuel/src/pr36296.c : 13] # SMT.10D.1967_13 = VDEF <SMT.10D.1967_10>
{ SMT.10D.1967 }
  pD.1952_7 = fooD.1945 ();
  # SUCC: 5 [100.0%]  (fallthru,exec)

  # BLOCK 5 freq:10000
  # PRED: 3 [71.0%]  (false,exec) 4 [100.0%]  (fallthru,exec)
  # initD.1951_2 = PHI <initD.1951_1(3), 2(4)>
  # pD.1952_4 = PHI <pD.1952_3(3), pD.1952_7(4)>
  [/home/manuel/src/pr36296.c : 16] # SMT.10D.1967_14 = VDEF <SMT.10D.1967_11>
{ SMT.10D.1967 }
  barD.1947 (pD.1952_4);
  [/home/manuel/src/pr36296.c : 17] goto <bb 3>;
  # SUCC: 3 [100.0%]  (fallthru,dfs_back,exec)

}

Because we create a PHI node for p in BB 3, we think that p can be used
uninitialized. Notice also that we are not able to move the 'if' and the call
to foo() out of the infinite loop. This is perhaps a missed optimization.


> This is quite strange: if I replace the value 2 by 1 or if I replace foo() by
> 0, the warning is no longer displayed.

If you replace foo() by 0, then CCP just assumes the p is always 0. In fact, it
will remove p altogether, even if you use 'if (init == 2)', thus missing a real
uninitialized use. This behaviour is known to hide warnings, both correct and
wrong warnings.

If I replace the value 2 by 1 I still get the warning in GCC 4.4, so that
really sounds strange. Are you sure about that?

Anyway, this is a confirmed bug but not easy to fix.


-- 

manu at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-08-18 23:38:16
               date|                            |
            Summary|wrong warning about         |bogus uninitialized warning
                   |potential uninitialized     |(loop representation)
                   |variable                    |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36296