[Bug middle-end/46909] New: [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[Bug middle-end/46909] New: [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

           Summary: [4.6 Regression] elf32-i386.c and elf64-x86-64.c in
                    binutils are miscompiled
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com


GCC 4.6 revision 167711 miscompiles elf32-i386.c and elf64-x86-64.c
in binutils.  I got

FAIL: TLS descriptor -fpic -shared transitions
FAIL: TLS with global dynamic and descriptors

for both i386 and x86-64 TLS tests with error like:

/export/build/gnu/binutils/build-x86_64-linux/ld/../binutils/objdump -drj.text
tmpdir/libtlsdesc.so
regexp_diff match failure
regexp "^ [0-9a-f]+:    81 c3 9a 13 00 00[     ]+add    \$0x[0-9a-f]+,%ebx$"
line   " 2df:    81 c3 ae 13 00 00        add    $0x13ae,%ebx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 f8 ff ff ff[     ]+mov    -0x8\(%ebx\),%eax$"
line   " 2f5:    8b 83 f4 ff ff ff        mov    -0xc(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 c4 ff ff ff[     ]+mov    -0x3c\(%ebx\),%eax$"
line   " 301:    8b 83 bc ff ff ff        mov    -0x44(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 d4 ff ff ff[     ]+mov    -0x2c\(%ebx\),%eax$"
line   " 30d:    8b 83 cc ff ff ff        mov    -0x34(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 b0 ff ff ff[     ]+mov    -0x50\(%ebx\),%eax$"
line   " 325:    8b 83 a0 ff ff ff        mov    -0x60(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 b4 ff ff ff[     ]+mov    -0x4c\(%ebx\),%eax$"
line   " 331:    8b 83 a4 ff ff ff        mov    -0x5c(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 bc ff ff ff[     ]+mov    -0x44\(%ebx\),%eax$"
line   " 33d:    8b 83 ac ff ff ff        mov    -0x54(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 c8 ff ff ff[     ]+mov    -0x38\(%ebx\),%eax$"
line   " 361:    8b 83 c0 ff ff ff        mov    -0x40(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 e8 ff ff ff[     ]+mov    -0x18\(%ebx\),%eax$"
line   " 36d:    8b 83 e4 ff ff ff        mov    -0x1c(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 cc ff ff ff[     ]+mov    -0x34\(%ebx\),%eax$"
line   " 385:    8b 83 c4 ff ff ff        mov    -0x3c(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 ec ff ff ff[     ]+mov    -0x14\(%ebx\),%eax$"
line   " 391:    8b 83 e8 ff ff ff        mov    -0x18(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 e0 ff ff ff[     ]+mov    -0x20\(%ebx\),%eax$"
line   " 39d:    8b 83 d8 ff ff ff        mov    -0x28(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 8b f8 ff ff ff[     ]+sub    -0x8\(%ebx\),%ecx$"
line   " 3f0:    2b 8b f4 ff ff ff        sub    -0xc(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 83 d0 ff ff ff[     ]+sub    -0x30\(%ebx\),%eax$"
line   " 402:    2b 83 c8 ff ff ff        sub    -0x38(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 8b c4 ff ff ff[     ]+add    -0x3c\(%ebx\),%ecx$"
line   " 415:    03 8b bc ff ff ff        add    -0x44(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 83 d4 ff ff ff[     ]+add    -0x2c\(%ebx\),%eax$"
line   " 427:    03 83 cc ff ff ff        add    -0x34(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 8b b0 ff ff ff[     ]+sub    -0x50\(%ebx\),%ecx$"
line   " 43a:    2b 8b a0 ff ff ff        sub    -0x60(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 83 b8 ff ff ff[     ]+sub    -0x48\(%ebx\),%eax$"
line   " 44c:    2b 83 a8 ff ff ff        sub    -0x58(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 8b b4 ff ff ff[     ]+add    -0x4c\(%ebx\),%ecx$"
line   " 45f:    03 8b a4 ff ff ff        add    -0x5c(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 83 bc ff ff ff[     ]+add    -0x44\(%ebx\),%eax$"
line   " 471:    03 83 ac ff ff ff        add    -0x54(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 83 e4 ff ff ff[     ]+sub    -0x1c\(%ebx\),%eax$"
line   " 496:    2b 83 e0 ff ff ff        sub    -0x20(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 8b c8 ff ff ff[     ]+add    -0x38\(%ebx\),%ecx$"
line   " 4a9:    03 8b c0 ff ff ff        add    -0x40(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 83 e8 ff ff ff[     ]+add    -0x18\(%ebx\),%eax$"
line   " 4bb:    03 83 e4 ff ff ff        add    -0x1c(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 8b cc ff ff ff[     ]+sub    -0x34\(%ebx\),%ecx$"
line   " 4ce:    2b 8b c4 ff ff ff        sub    -0x3c(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    2b 83 dc ff ff ff[     ]+sub    -0x24\(%ebx\),%eax$"
line   " 4e0:    2b 83 d4 ff ff ff        sub    -0x2c(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 8b ec ff ff ff[     ]+add    -0x14\(%ebx\),%ecx$"
line   " 4f3:    03 8b e8 ff ff ff        add    -0x18(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    03 83 e0 ff ff ff[     ]+add    -0x20\(%ebx\),%eax$"
line   " 505:    03 83 d8 ff ff ff        add    -0x28(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 8b d8 ff ff ff[     ]+mov    -0x28\(%ebx\),%ecx$"
line   " 50f:    8b 8b d0 ff ff ff        mov    -0x30(%ebx),%ecx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 83 c0 ff ff ff[     ]+mov    -0x40\(%ebx\),%eax$"
line   " 51e:    8b 83 b0 ff ff ff        mov    -0x50(%ebx),%eax"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 93 f0 ff ff ff[     ]+mov    -0x10\(%ebx\),%edx$"
line   " 52d:    8b 93 ec ff ff ff        mov    -0x14(%ebx),%edx"
regexp_diff match failure
regexp "^ [0-9a-f]+:    8b 8b f4 ff ff ff[     ]+mov    -0xc\(%ebx\),%ecx$"
line   " 53c:    8b 8b f0 ff ff ff        mov    -0x10(%ebx),%ecx"
FAIL: TLS descriptor -fpic -shared transitions

Some bits are wrong.

[Bug middle-end/46909] [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sandra at codesourcery dot
                   |                            |com
   Target Milestone|---                         |4.6.0

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-12 17:44:39 UTC ---
It is caused by revision 160445:

http://gcc.gnu.org/ml/gcc-cvs/2010-06/msg00359.html

[Bug middle-end/46909] [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-12 18:12:47 UTC ---
elf64_x86_64_allocate_dynrelocs and elf64_x86_64_size_dynamic_sections
are miscompiled.

[Bug middle-end/46909] [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.12.12 19:22:15
     Ever Confirmed|0                           |1
           Severity|normal                      |blocker

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-12 19:22:15 UTC ---
[hjl@gnu-6 tmp]$ cat y.c
extern void abort ();
struct entry
{
  unsigned char tls_type;
} h;

int
__attribute__ ((__noinline__))
foo (struct entry *h)
{
  int tls_type = h->tls_type;
  if (! ((tls_type) == 4 || ((tls_type) == (2 | 4)))
      || ((tls_type) == 2 || ((tls_type) == (2 | 4))))
    return 1;

  return -1;
}

int
main ()
{
  struct entry h;
  h.tls_type = 6;
  if (foo (&h) != 1)
    abort ();
  return 0;
}
[hjl@gnu-6 tmp]$ /usr/gcc-4.6/bin/gcc -O2 y.c
[hjl@gnu-6 tmp]$ ./a.out 
Aborted
[hjl@gnu-6 tmp]$ gcc -O2 y.c
[hjl@gnu-6 tmp]$ ./a.out 
[hjl@gnu-6 tmp]$

[Bug middle-end/46909] [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-12 19:27:08 UTC ---
Smaller one:

[hjl@gnu-6 tmp]$ cat y.c
extern void abort ();
int
__attribute__ ((__noinline__))
foo (unsigned int tls_type)
{
  if (! (tls_type == 4 || (tls_type == 6))
      || (tls_type == 2 || (tls_type == 6)))
    return 1;
  return -1;
}
int
main ()
{
  if (foo (6) != 1)
    abort ();
  return 0;
}
[hjl@gnu-6 tmp]$ /usr/gcc-4.6/bin/gcc -O2 y.c      
[hjl@gnu-6 tmp]$ ./a.out 
Aborted
[hjl@gnu-6 tmp]$

[Bug middle-end/46909] [4.6 Regression] elf32-i386.c and elf64-x86-64.c in binutils are miscompiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

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

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

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-12-12 19:30:36 UTC ---
I'll look at it tomorrow.

[Bug middle-end/46909] [4.6 Regression] Logical OR expressions are miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

--- Comment #6 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-12 21:29:08 UTC ---
Bad patch was posted at

http://gcc.gnu.org/ml/gcc-patches/2010-06/msg00001.html

[Bug middle-end/46909] [4.6 Regression] Logical OR expressions are miscompiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

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

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

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-12-13 13:02:22 UTC ---
The fix is easy:
@@ -2441,7 +2441,7 @@ or_var_with_comparison_1 (gimple stmt,
          => (t OR inner2)
          If the partial result t is a constant, we win.  Otherwise
          continue on to try reassociating with the other inner test.  */
-      if (innercode == TRUTH_OR_EXPR)
+      if (is_or)
         {
           if (integer_onep (t))
         return boolean_true_node;
@@ -2474,7 +2474,7 @@ or_var_with_comparison_1 (gimple stmt,
       /* Handle the OR case, where we are reassociating:
          (inner1 OR inner2) OR (op2a code2 op2b)
          => (inner1 OR t)  */
-      if (innercode == TRUTH_OR_EXPR)
+      if (is_or)
         {
           if (integer_zerop (t))
         return inner1;

(the and folder does it that way correctly).  Still looking why it isn't
optimized now into x != 4.

[Bug tree-optimization/46909] [4.6 Regression] Logical OR expressions are miscompiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-12-13 14:22:39 UTC ---
Created attachment 22741
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22741
gcc46-pr46909.patch

Untested fix.

[Bug tree-optimization/46909] [4.6 Regression] Logical OR expressions are miscompiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-12-14 14:10:09 UTC ---
Author: jakub
Date: Tue Dec 14 14:09:59 2010
New Revision: 167800

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=167800
Log:
    PR tree-optimization/46909
    * gimple-fold.c (and_var_with_comparison_1): Save partial
    result even in the is_and case, if both partial results
    are the same, return it.
    (or_var_with_comparison_1): Use is_or predicate instead of
    innercode == TRUTH_OR_EXPR test.  Save partial result
    even in the is_or case, if both partial results are the
    same, return it.  In the !is_or case when both partial
    results are the same, return the partial result instead
    of boolean_true_node.

    * gcc.c-torture/execute/pr46909-1.c: New test.
    * gcc.c-torture/execute/pr46909-2.c: New test.
    * gcc.dg/pr46909.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr46909-1.c
    trunk/gcc/testsuite/gcc.c-torture/execute/pr46909-2.c
    trunk/gcc/testsuite/gcc.dg/pr46909.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/gimple-fold.c
    trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/46909] [4.6 Regression] Logical OR expressions are miscompiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46909

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

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

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-12-14 14:32:35 UTC ---
Fixed.