[Bug c/55108] New: bad compile-time evaluation of members of initialized union

[Bug c/55108] New: bad compile-time evaluation of members of initialized union

[acn1 at cam dot ac.uk]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55108

             Bug #: 55108
           Summary: bad compile-time evaluation of members of initialized
                    union
    Classification: Unclassified
           Product: gcc
           Version: 4.6.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: acn1@cam.ac.uk


Created attachment 28548
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28548
internal file "bad.i" obtained using -save-temp, as requested.

gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/arm-linux-gnueabihf/4.6/lto-wrapper
Target: arm-linux-gnueabihf
Configured with: ../src/configure -v --with-pkgversion='Debian 4.6.3-8+rpi1'
--with-bugurl=file:///usr/share/doc/gcc-4.6/README.Bugs
--enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr
--program-suffix=-4.6 --enable-shared --enable-linker-build-id
--with-system-zlib --libexecdir=/usr/lib --without-included-gettext
--enable-threads=posix --with-gxx-include-dir=/usr/include/c++/4.6
--libdir=/usr/lib --enable-nls --with-sysroot=/ --enable-clocale=gnu
--enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-gnu-unique-object
--enable-plugin --enable-objc-gc --disable-sjlj-exceptions --with-arch=armv6
--with-fpu=vfp --with-float=hard --enable-checking=release
--build=arm-linux-gnueabihf --host=arm-linux-gnueabihf
--target=arm-linux-gnueabihf
Thread model: posix
gcc version 4.6.3 (Debian 4.6.3-8+rpi1)

File bad.c contains

#include <stdio.h>

int main(int argc, char *argv[])
{
    union
    {   double d;
        unsigned char c[8];
    } u = {1.0/7.0};
    printf("Bytes from float are %x %x %x %x\n",
           u.c[0] & 0xff, u.c[1] & 0xff, u.c[2] & 0xff, u.c[3] & 0xff);
    return 0;
}

My hope is that the aliasing is OK both because the float is aliased with chars
and because I believed that GCC was kind about unions.

The output is

acn1@raspberrypi ~ $ gcc -O0 bad.c -o bad0
acn1@raspberrypi ~ $ gcc -O1 bad.c -o bad1 -Wall -Wextra
bad.c: In function ‘main’:
bad.c:3:14: warning: unused parameter ‘argc’ [-Wunused-parameter]
bad.c:3:26: warning: unused parameter ‘argv’ [-Wunused-parameter]
acn1@raspberrypi ~ $ ./bad0
Bytes from float are 92 24 49 92         <<< expected output
acn1@raspberrypi ~ $ ./bad1
Bytes from float are 92 924924 9249 92   <<< output > 0xff unexpected
acn1@raspberrypi ~ $

To my mind whatever else might be going on the "& 0xff" that I have should not
let me get such large numbers displayed!

The ".i" file is attached. gcc is evaluating the components of the union at
compile time and not respecting the "unsigned char" width or the subseqent "&
0xff".

[Bug target/55108] bad compile-time evaluation of members of initialized union

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55108

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |arm-linux-gnueabihf
          Component|c                           |target

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-29 14:44:21 UTC ---
Works for me on x86_64.

[Bug target/55108] bad compile-time evaluation of members of initialized union

[rearnsha at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55108

Richard Earnshaw <rearnsha at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-10-30
     Ever Confirmed|0                           |1
      Known to fail|                            |4.6.3, 4.7.0

--- Comment #2 from Richard Earnshaw <rearnsha at gcc dot gnu.org> 2012-10-30 14:13:15 UTC ---
Confirmed.  It seems the compiler is incorrectly eliminating some zero-extend
operations.  We have

Ra:SI = Rx:SI >> 8
Rb:QI = subreg:QI (Ra:SI 0)
Rc:SI = subreg:SI (Rb:QI 0)
Rd:SI = Rc:SI & 255

cse1 pass then decides incorrectly (and presumably because of the paradoxical
subreg) that Rd == Rc and discards the mask operation.

Also seen in 4.7, haven't been able to test trunk.

[Bug target/55108] bad compile-time evaluation of members of initialized union

[rearnsha at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55108

--- Comment #3 from Richard Earnshaw <rearnsha at gcc dot gnu.org> 2012-10-30 14:15:43 UTC ---
At armv6t2 and later we have a ubfx instruction available and that is enough to
mask this bug.

[Bug target/55108] bad compile-time evaluation of members of initialized union

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55108

--- Comment #4 from Mikael Pettersson <mikpe at it dot uu.se> 2013-02-03 23:24:37 UTC ---
On armv5tel-linux-gnueabi this bug is reproducible with gcc-4.6 but not with
gcc-4.7 or 4.8.

The wrong-code was made latent for 4.7.0 by r179556 aka PR38885, a
missed-optimization fix.  Bisecting with that fix disabled (a simple #if 0 /
#endif wrapper around the new code) showed that the wrong-code was fixed
properly for 4.8 by r187648 aka PR53352, a fix for incorrect CSE of unions on
ARM.

Backporting r187648 to 4.6 fixes this PR's test case with 4.6 on armv5tel. 
Anyone backporting r187648 to 4.7 or 4.6 should also backport the test case
fixup in r187654.

[Bug target/55108] bad compile-time evaluation of members of initialized union

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55108

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

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

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> 2013-02-04 00:05:48 UTC ---
dup then .

*** This bug has been marked as a duplicate of bug 53352 ***