[Bug tree-optimization/52028] New: [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[Bug tree-optimization/52028] New: [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

             Bug #: 52028
           Summary: [4.7 Regression] wrong code with -O2
                    -ftree-loop-distribution
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: zsojka@seznam.cz
              Host: x86_64-pc-linux-gnu
            Target: x86_64-pc-linux-gnu


Created attachment 26489
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26489
reduced testcase

Compiler output:
$ gcc -O2 -ftree-loop-distribution testcase4.c
$ valgrind -q ./a.out 
==32144== Conditional jump or move depends on uninitialised value(s)
==32144==    at 0x4C2D333: bcmp (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==32144==    by 0x4005F4: main (in /home/smatz/gcc-bug/1063/a.out)
==32144== 
==32144== Conditional jump or move depends on uninitialised value(s)
==32144==    at 0x4005F7: main (in /home/smatz/gcc-bug/1063/a.out)
==32144== 
Aborted

Tested revisions:
r183666 - fail
r180540 - fail
r178498 - fail
4.6 r180325 - OK

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-01-28
     Ever Confirmed|0                           |1

--- Comment #1 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-28 15:12:13 UTC ---
Reduced range:
revision 177649 - OK,
revision 178509 - fail.

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org
   Target Milestone|---                         |4.7.0

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-28 16:39:47 UTC ---
Started with http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=178028

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-30 09:37:53 UTC ---
Mine then.

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-30 13:57:30 UTC ---
loop distribution splits the

  for (;;)
    {
...
      if (K != k)
        break;
      K += 16;
    }

loop into three portions, the first writing to K[0..13], the next
writing to K[14] and the last writing to K[15].  All of which is
reasonable at a first glance.  But then the first loop is discarded
completely CDDCE2, so something is wrong.

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-30 14:16:51 UTC ---
Virtual SSA form is corrupt (and we don't detect that - bah).

<bb 3>:
  # .MEM_723 = PHI <.MEM_666(3), .MEM_661(D)(2)>
  # ivtmp.297_722 = PHI <ivtmp.297_648(3), ivtmp.297_1354(2)>
  D.2523_1353 = (void *) ivtmp.297_722;
  # .MEM_666 = VDEF <.MEM_723>
  MEM[base: D.2523_1353, offset: 0B] = 0;
  ivtmp.297_648 = ivtmp.297_722 + 4;
  if (ivtmp.297_648 != D.2525_653)
    goto <bb 3>;
  else
    goto <bb 4>;

<bb 4>:
...

<bb 5>:
  # K_1035 = PHI <&MEM[(void *)&k + 64B](5), &k(4)>
  # .MEM_703 = PHI <.MEM_562(5), .MEM_666(4)>
...
  # .MEM_562 = VDEF <.MEM_561>
  MEM[(unsigned int *)K_1035 + 52B] = pretmp.262_1186;
  D.1989_556 = D.1916_553 + 512;
  D.1990_563 = D.1989_556 * 4;
  D.1991_566 = &CAST_S_table + D.1990_563;
  # VUSE <.MEM_562>
  D.1992_567 = *D.1991_566;
  if (&k != K_1035)
    goto <bb 6>;
  else
    goto <bb 5>;

<bb 6>:
  K_1352 = &k;
  X$0_719 = 0;

<bb 7>:
  # K_1147 = PHI <&k(6), &MEM[(void *)&k + 64B](7)>
  # .MEM_1141 = PHI <.MEM_666(6), .MEM_1343(7)>
...

We have overlapping life-ranges for .MEM_666 and .MEM_562  (at least).

I have a patch.

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-30 15:52:02 UTC ---
Fixed.

[Bug tree-optimization/52028] [4.7 Regression] wrong code with -O2 -ftree-loop-distribution

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52028

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-30 15:51:34 UTC ---
Author: rguenth
Date: Mon Jan 30 15:51:23 2012
New Revision: 183730

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=183730
Log:
2012-01-30  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/52028
    * tree-loop-distribution.c (ldist_gen): Properly update
    virtual SSA form.

    * gcc.dg/torture/pr52028.c: New testcase.
    * gfortran.dg/ldist-pr45199.f: Adjust.
    * gcc.dg/tree-ssa/ldist-16.c: Likewise.
    * gcc.dg/tree-ssa/ldist-17.c: Likewise.
    * gcc.dg/tree-ssa/ldist-pr45948.c: Likewise.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr52028.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.dg/tree-ssa/ldist-16.c
    trunk/gcc/testsuite/gcc.dg/tree-ssa/ldist-17.c
    trunk/gcc/testsuite/gcc.dg/tree-ssa/ldist-pr45948.c
    trunk/gcc/testsuite/gfortran.dg/ldist-pr45199.f
    trunk/gcc/tree-loop-distribution.c