[Bug middle-end/54985] New: Dom optimization erroneous remove conditional goto.

[Bug middle-end/54985] New: Dom optimization erroneous remove conditional goto.

[vbyakovl23 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

             Bug #: 54985
           Summary: Dom optimization erroneous remove conditional goto.
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: vbyakovl23@gmail.com


Attached test case fails if compiled with -O1 and higher.

gcc -O0 -m32  q.c qm.c  ; ./a.out && echo "pass" || echo "FAIL"                 
pass
gcc -g3 -O1 -m32  q.c qm.c  ; ./a.out && echo "pass" || echo "FAIL"             
FAIL

gcc -v
Using built-in specs.
COLLECT_GCC=/export/users/vbyakovl/workspaces/gcc/install-ref/bin/gcc
COLLECT_LTO_WRAPPER=/export/users/vbyakovl/workspaces/gcc/install-ref/libexec/gcc/x86_64-unknown-linux-gnu/4.8.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc/configure CFLAGS='-O0 -g3'
--prefix=/export/users/vbyakovl/workspaces/gcc/install-ref --disable-bootstrap
--enable-languages=c,c++,fortran,lto CXXFLAGS='-O0 -g3' : (reconfigured)
../gcc/configure CFLAGS='-O0 -g3'
--prefix=/export/users/vbyakovl/workspaces/gcc/install-ref --disable-bootstrap
CXXFLAGS='-O0 -g3' --enable-languages=c,c++,fortran,lto --no-create
--no-recursion : (reconfigured) ../gcc/configure CFLAGS='-O0 -g3'
--prefix=/export/users/vbyakovl/workspaces/gcc/install-ref --disable-bootstrap
CXXFLAGS='-O0 -g3' --enable-languages=c,c++,fortran,lto --no-create
--no-recursion
Thread model: posix
gcc version 4.8.0 20121015 (experimental) (GCC)

Wrong compilation of a routine

int foo(ST *s, int c)
{
    int first = 1;
    int count = c;
    ST *item = s;
    int a = s->a;
    int x;

    while (count--)
    {  
        x = item->a;
        if (first)
            first = 0;
        else if (x >= a)
            return 1;
        a = x;
        item++;
    }
    return 0;
}

Compiler sets equivalence between ‘x’ and ‘a’ (routine tree-ssa-threadedge.c
/record_temporary_equivalences_from_phis() ) and folds comparison x >= a to
true.

[Bug middle-end/54985] Dom optimization erroneous remove conditional goto.

[vbyakovl23 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #1 from Vladimir Yakovlev <vbyakovl23 at gmail dot com> 2012-10-19 10:58:26 UTC ---
Created attachment 28489
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28489
Test case

[Bug middle-end/54985] Dom optimization erroneous remove conditional goto.

[vbyakovl23 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #2 from Vladimir Yakovlev <vbyakovl23 at gmail dot com> 2012-10-19 10:59:15 UTC ---
Created attachment 28490
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28490
Main routine

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
      Known to work|                            |4.6.3
           Keywords|                            |wrong-code
   Last reconfirmed|                            |2012-10-19
          Component|middle-end                  |tree-optimization
     Ever Confirmed|0                           |1
            Summary|Dom optimization erroneous  |[4.7/4.8 Regression] dom
                   |remove conditional goto.    |optimization erroneous
                   |                            |remove conditional goto.
   Target Milestone|---                         |4.7.3
      Known to fail|                            |4.7.2, 4.8.0

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-19 11:19:38 UTC ---
Confirmed.  Single-file testcase, fails with 4.7 and 4.8.

typedef struct st {
    int a;
} ST;

int __attribute__((noinline,noclone))
foo(ST *s, int c)
{
  int first = 1;
  int count = c;
  ST *item = s;
  int a = s->a;
  int x;

  while (count--)
    {
      x = item->a;
      if (first)
        first = 0;
      else if (x >= a)
        return 1;
      a = x;
      item++;
    }
  return 0;
}

extern void abort (void);

int main ()
{
  ST _1[2] = {{2}, {1}};
  if (foo(_1, 2) != 0)
    abort ();
  return 0;
}

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at gcc dot gnu.org

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-19 11:26:15 UTC ---
Jeff - you updated jump-threading code in DOM quite a bit for 4.7.

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com

--- Comment #5 from Jeffrey A. Law <law at redhat dot com> 2012-10-19 13:27:02 UTC ---
Thanks.  I'll take a looksie.

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #6 from Jeffrey A. Law <law at redhat dot com> 2012-10-19 22:00:45 UTC ---
This is definitely something introduced when I extended the threading code last
year.

When threading across a backedge in the CFG we run the risk of simplifying a
conditional using equivalences which are valid for one loop iteration, but not
the next iteration.

There's a bit of special casing done to handle this. We either need to add
another case or actually invalidate the temporary equivalences we made for the
statements in the loop.  The first is pretty trivial, but I want to ponder the
special case code over the weekend for it's overall safety/correctness.

I don't see any difficulty wrapping this up early next week.

Jeff

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[law at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #7 from Jeffrey A. Law <law at gcc dot gnu.org> 2012-10-23 20:33:52 UTC ---
Author: law
Date: Tue Oct 23 20:33:49 2012
New Revision: 192745

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192745
Log:
        PR tree-optimization/54985
        * tree-ssa-threadedge.c (cond_arg_set_in_bb): New function
        * extracted
        from thread_across_edge.
        (thread_across_edge): Use it in all cases where we might thread
        across a back edge.

        * gcc.c-torture/execute/pr54985.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr54985.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-threadedge.c

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #8 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-10-23 21:23:10 UTC ---
After revision 192745 bootstrap fails with

...
../../work/gcc/tree-ssa-threadedge.c:583:1: error: unused parameter 'n'
[-Werror=unused-parameter]
 cond_arg_set_in_bb (edge e, basic_block bb, int n)
 ^
cc1plus: all warnings being treated as errors
...

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #9 from Jeffrey A. Law <law at redhat dot com> 2012-10-23 21:25:43 UTC ---
debugging cruft.  I'll deal with it

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #10 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-10-23 21:35:09 UTC ---
Jeffrey are you sure that r192746 is enough? cond_arg_set_in_bb is used with 3
arguments elsewhere in the file.

[Bug tree-optimization/54985] [4.7/4.8 Regression] dom optimization erroneous remove conditional goto.

[izamyatin at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #11 from Igor Zamyatin <izamyatin at gmail dot com> 2012-10-30 08:32:01 UTC ---
Are there any plans to backport this to 4.7?

[Bug tree-optimization/54985] [4.7 Regression] dom optimization erroneous remove conditional goto.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.7/4.8 Regression] dom    |[4.7 Regression] dom
                   |optimization erroneous      |optimization erroneous
                   |remove conditional goto.    |remove conditional goto.

--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-30 15:12:59 UTC ---
Fixed on trunk sofar.

[Bug tree-optimization/54985] [4.7 Regression] dom optimization erroneous remove conditional goto.

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #13 from Jeffrey A. Law <law at redhat dot com> 2012-10-31 04:09:59 UTC ---
Yes we expect to backport to 4.7.x (I've already discussed with Jakub).  I'm
letting it simmer on the trunk for a bit before backporting the bits.

[Bug tree-optimization/54985] [4.7 Regression] dom optimization erroneous remove conditional goto.

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |konstantin.vladimirov at
                   |                            |gmail dot com

--- Comment #14 from Jeffrey A. Law <law at redhat dot com> 2012-10-31 04:11:38 UTC ---
*** Bug 55138 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/54985] [4.7 Regression] dom optimization erroneous remove conditional goto.

[law at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

--- Comment #15 from Jeffrey A. Law <law at gcc dot gnu.org> 2012-11-02 20:19:23 UTC ---
Author: law
Date: Fri Nov  2 20:19:16 2012
New Revision: 193108

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=193108
Log:
       PR tree-optimization/54985
        * tree-ssa-threadedge.c (cond_arg_set_in_bb): New function extracted
        from thread_across_edge.
        (thread_across_edge): Use it in all cases where we might thread
        across a back edge.

        * gcc.c-torture/execute/pr54985.c: New test.



Added:
    branches/gcc-4_7-branch/gcc/testsuite/gcc.c-torture/execute/pr54985.c
Modified:
    branches/gcc-4_7-branch/gcc/ChangeLog
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_7-branch/gcc/tree-ssa-threadedge.c

[Bug tree-optimization/54985] [4.7 Regression] dom optimization erroneous remove conditional goto.

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54985

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #16 from Jeffrey A. Law <law at redhat dot com> 2012-11-02 20:24:54 UTC ---
Fixed on 4-7 branch and trunk.