[Bug lto/47259] New: LTO and global register variable

[Bug lto/47259] New: LTO and global register variable

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

           Summary: LTO and global register variable
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: lto
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: d.g.gorbachev@gmail.com


Created attachment 22947
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22947
Testcase

internal compiler error: in expand_expr_addr_expr_1, at expr.c:6917

[Bug lto/47259] LTO and global register variable

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

--- Comment #1 from Dmitry Gorbachev <d.g.gorbachev at gmail dot com> 2011-01-11 18:47:49 UTC ---
Created attachment 22948
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22948
Backtrace

[Bug lto/47259] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |lto
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.01.12 11:06:55
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-01-12 11:06:55 UTC ---
Confirmed.  We probably should reject global register vars for LTO.

[Bug lto/47259] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-01-12 13:32:32 UTC ---
In 99.9999% of all cases there is a better way than using a global register
variable.

Also it is not clear to me how to merge translation units where a different
set of registers are used in global register variables.  Consider

t1.c:
int g asm("%ecx");

t2.c:
void foo(void)
{
  complex-asm-that-needs-many-regs
}

if we make ecx global in t2.c then we'd fail to assign registers to
the complex asm.  Thus one solution would be to partition the LTO
program in a way that t1.c is completely contained in its own
partition.  But then the user could have compiled t1.c without LTO
which would be my suggested workaround.  Thus, emit

sorry: translation units with global register variables need to be compiled
without LTO

[Bug lto/47259] LTO and global register variable

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

--- Comment #3 from Dmitry Gorbachev <d.g.gorbachev at gmail dot com> 2011-01-12 13:10:07 UTC ---
For example, linux uses global reg var "esp". As there is no need to reserve
this register, can GCC accept it?

[Bug lto/47259] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

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 #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-01-12 14:01:43 UTC ---
I have a patch.

[Bug lto/47259] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-01-12 15:01:16 UTC ---
Author: rguenth
Date: Wed Jan 12 15:01:09 2011
New Revision: 168713

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=168713
Log:
2011-01-12  Richard Guenther  <rguenther@suse.de>

    PR lto/47259
    * lto-streamer-out.c (output_gimple_stmt): Do not wrap
    register variables in a MEM_REF.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/lto-streamer-out.c

[Bug lto/47259] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

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> 2011-01-12 15:02:22 UTC ---
Fixed.

[Bug lto/47259] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-01-12 15:22:08 UTC ---
Author: rguenth
Date: Wed Jan 12 15:21:59 2011
New Revision: 168715

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=168715
Log:
2011-01-12  Richard Guenther  <rguenther@suse.de>

    PR lto/47259
    * gcc.dg/lto/pr47259_0.c: New testcase.
    * gcc.dg/lto/pr47259_1.c: Likewise.

Added:
    trunk/gcc/testsuite/gcc.dg/lto/pr47259_0.c
    trunk/gcc/testsuite/gcc.dg/lto/pr47259_1.c
Modified:
    trunk/gcc/testsuite/ChangeLog

[Bug lto/47259] LTO and global register variable

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

Uros Bizjak <ubizjak at gmail dot com> changed:

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

--- Comment #9 from Uros Bizjak <ubizjak at gmail dot com> 2011-12-02 11:11:43 UTC ---
(In reply to comment #7)
> Fixed.

Not really ... [1]:

Executing on host: /home/uros/gcc-build/gcc/xgcc -B/home/uros/gcc-build/gcc/
c_lto_pr47259_0.o c_lto_pr47259_1.o    -O2 -flto -w        -o
gcc-dg-lto-pr47259-01.exe    (timeout = 300)
In file included from :0:0:

/home/uros/gcc-svn/trunk/gcc/testsuite/gcc.dg/lto/pr47259_0.c:5:14: error:
invalid register name for 'r'

lto1: fatal error: errors during merging of translation units

compilation terminated.

lto-wrapper: /home/uros/gcc-build/gcc/xgcc returned 1 exit status

/usr/local/x86_64-unknown-linux-gnu/bin/ld: lto-wrapper failed

collect2: error: ld returned 1 exit status

compiler exited with status 1

[1] http://gcc.gnu.org/ml/gcc-testresults/2011-12/msg00162.html

[Bug lto/47259] [4.7 Regression] LTO and global register variable

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.7.0
            Summary|LTO and global register     |[4.7 Regression] LTO and
                   |variable                    |global register variable

--- Comment #10 from Uros Bizjak <ubizjak at gmail dot com> 2011-12-02 11:24:55 UTC ---
Testsuite failure -> 4.7 regression.

[Bug lto/47259] [4.7 Regression] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

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

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

--- Comment #11 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-02 12:14:29 UTC ---
This must be a new bug, the testcase passed previously, the register name
(esi) is rejected at link-time only.

#1  0x0000000000d4dbba in make_decl_rtl (decl=0x7ffff5a2c140)
    at /space/rguenther/src/svn/trunk/gcc/varasm.c:1207
1207            error ("invalid register name for %q+D", decl);
(gdb) l
1202          reg_number = decode_reg_name (asmspec);
1203          /* First detect errors in declaring global registers.  */
1204          if (reg_number == -1)
1205            error ("register name not specified for %q+D", decl);
1206          else if (reg_number < 0)
1207            error ("invalid register name for %q+D", decl);
(gdb) p asmspec
$2 = 0x7ffff5b58401 "esi.2318"

we mangle the register name now.  That's of course bogus.  It seems we
put the global register vars into the alias machinery now?  Honza?

[Bug lto/47259] [4.7 Regression] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

--- Comment #12 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-02 15:49:41 UTC ---
Author: rguenth
Date: Fri Dec  2 15:49:37 2011
New Revision: 181927

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=181927
Log:
2011-12-02  Richard Guenther  <rguenther@suse.de>

    PR lto/47259
    * ipa.c (varpool_externally_visible_p): Register variables
    are always externally visible.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa.c

[Bug lto/47259] [4.7 Regression] LTO and global register variable

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47259

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

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

--- Comment #13 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-05 09:30:10 UTC ---
Fixed.