[Bug c++/53505] New: bitfield with bool type generated broken object code

[Bug c++/53505] New: bitfield with bool type generated broken object code

[rockeet at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

             Bug #: 53505
           Summary: bitfield with bool type generated broken object code
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: critical
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: rockeet@gmail.com


compiler flag: -Wall -Wextra -O[0|1|2|3|fast]

All optimization levels: 1,2,3,fast  reproduce the bug
-O0 is OK.

The expected output is 0

//================================================
$g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/opt/libexec/gcc/x86_64-unknown-linux-gnu/4.7.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-4.7.0-RC-20120314/configure --prefix=/opt
--with-gmp=/opt --with-mpfr=/opt --with-mpc=/opt --with-ppl=/opt
--with-cloog=/opt --with-local-prefix=/opt --enable-languages=c,c++ :
(reconfigured) ../gcc-4.7.0-RC-20120314/configure --prefix=/opt --with-gmp=/opt
--with-mpfr=/opt --with-mpc=/opt --with-ppl=/opt --with-cloog=/opt
--with-local-prefix=/opt --enable-languages=c,c++,fortran
Thread model: posix
gcc version 4.7.0 20120314 (prerelease) (GCC)

//================================================
#include <string.h>
#include <stdio.h>
#include <stdlib.h>
#include <vector>

#if 1
    #define BIT bool // reproduce the bug
#else
    #define BIT unsigned // bug vanished
#endif

struct A {
    unsigned int  a;
    unsigned char c1, c2;
    BIT  b1 : 1;
    BIT  b2 : 1;
    BIT  b3 : 1;

    A() {
        memset(this, -1, sizeof(A));
        a = 0;
        c1 = c2 = 0;
        b1 = 0;
        b2 = 0;
        b3 = 0;
    }
};

void f(const std::vector<A>& v) {
    int s = 0;
    for (int i = 0, n = v.size(); i < n; ++i) {
        const A a = v[i]; // produce the bug
//    const A&a = v[i]; // bug vanished if 'a' is a const reference
        s += a.b1 ? 1 : 0;
    }
    printf("%d\n", s);
}

int main(int argc, char* argv[]) {
    size_t size = argc < 2 ? 100 : atoi(argv[1]);
    std::vector<A> v(size);
    f(v);
    return 0;
}

[Bug c++/53505] [4.7/4.8 Regression] bitfield with bool type generated broken object file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-05-28
                 CC|                            |jakub at gcc dot gnu.org
   Target Milestone|---                         |4.7.1
            Summary|bitfield with bool type     |[4.7/4.8 Regression]
                   |generated broken object     |bitfield with bool type
                   |file                        |generated broken object
                   |                            |file
     Ever Confirmed|0                           |1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-28 12:53:19 UTC ---
Started with http://gcc.gnu.org/viewcvs?view=revision&revision=184214

[Bug c++/53505] [4.7/4.8 Regression] bitfield with bool type generated broken object file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-28 13:25:34 UTC ---
SRA creates:
  <unnamed-unsigned:1> a$6;
...
  a$6_21 = MEM[(const struct A &)D.9923_20 + 6];
  D.10122_23 = (int) a$6_21;
which expansion doesn't handle as extract_bit_field (both MEM_REF and
TARGET_MEM_REF are expanded that way), so the zero extension is performed just
as QI->SImode zero extension.

I'd say that SRA shouldn't be creating such MEM_REFs and instead should keep
the BIT_FIELD_REFs.

[Bug c++/53505] [4.7/4.8 Regression] bitfield with bool type generated broken object file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-28 13:39:12 UTC ---
That revision added:
+      /* But leave bitfield accesses alone.  */
+      && (root->offset % BITS_PER_UNIT) == 0)
Here root->offset % BITS_PER_UNIT is 0, still it is a bitfield and IMHO we
should leave it alone.  The question is how to differentiate bitfields at byte
boundary offsets from e.g. non-bitfield _Bool where it should be user's
responsibility to keep just 0 or 1 in the whole byte.

[Bug c++/53505] [4.7/4.8 Regression] bitfield with bool type generated broken object file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

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

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-28 14:09:05 UTC ---
Ah, sounds like dup of PR53438.

*** This bug has been marked as a duplicate of bug 53438 ***

[Bug c++/53505] [4.7/4.8 Regression] bitfield with bool type generated broken object file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-28 14:28:06 UTC ---
Author: jakub
Date: Mon May 28 14:28:00 2012
New Revision: 187931

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187931
Log:
    PR tree-optimization/53505
    * c-c++-common/torture/pr53505.c: New test.

Added:
    trunk/gcc/testsuite/c-c++-common/torture/pr53505.c
Modified:
    trunk/gcc/testsuite/ChangeLog

[Bug c++/53505] [4.7/4.8 Regression] bitfield with bool type generated broken object file

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53505

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-28 14:28:55 UTC ---
Author: jakub
Date: Mon May 28 14:28:42 2012
New Revision: 187932

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187932
Log:
    PR tree-optimization/53505
    * c-c++-common/torture/pr53505.c: New test.

Added:
    branches/gcc-4_7-branch/gcc/testsuite/c-c++-common/torture/pr53505.c
Modified:
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog