[Bug tree-optimization/45912] New: [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[Bug tree-optimization/45912] New: [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

           Summary: [4.6 regression] Patch for PR tree-optimization/44972
                    caused 2500+ FAILs on Solaris 2/SPARC
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: major
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: ebotcazou@gcc.gnu.org, jamborm@gcc.gnu.org
              Host: sparc-sun-solaris2.*
            Target: sparc-sun-solaris2.*
             Build: sparc-sun-solaris2.*


Created attachment 21973
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=21973
stage1 assembler output

The Solaris 2/SPARC port (probably every SPARC port) is in a terrible state
with
2500+ execution failures in the testsuite (mostly 64-bit).

A reghunt revealed that this was introduced by this patch:

2010-09-10  Martin Jambor  <mjambor@suse.cz>

        PR tree-optimization/44972
        * tree-sra.c: Include toplev.h.
        (build_ref_for_offset): Entirely reimplemented.
        (build_ref_for_model): New function.
        (build_user_friendly_ref_for_offset): New function.
        (analyze_access_subtree): Removed build_ref_for_offset check.
        (propagate_subaccesses_across_link): Likewise.
        (create_artificial_child_access): Use
        build_user_friendly_ref_for_offset.
[...]

The reghunt proved difficult because originally the tree didn't build due to
PR tree-optimization/45611 (fixed in r164244) and PR bootstrap/45630 (fixed in
r164163).  The problem is a miscompilation since it doesn't occur with the
stage1
compiler.

I've used the -O2 compilation of gcc.c-torture/execute/20000706-5.c as the
testcase during the reghunt.  It can be compiled with

% cc1 20000706-5.c -mptr64 -mstack-bias -mno-v8plus -mcpu=v9 -quiet -m64 -O2 -w

I'm attaching the stage1 and stage3 .s files for comparison.  The test SEGVs
here:

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 1 (LWP 1)]
0x000000010000098c in main ()
1: x/i $pc
=> 0x10000098c <main+36>:       st  %g3, [ %g4 + %g2 ]
(gdb) p/x $g4+$g2
$3 = 0x37ffff1d8

The target address is unmapped in the process, as can be seen with pmap:

19505:  /var/gcc/reghunt/102852/gcc/testsuite/gcc/stage3/20000706-5.x2
0000000100000000          8K r-x-- 
/var/gcc/reghunt/102852/gcc/testsuite/gcc/stage3/20000706-5.x2
0000000100100000          8K rwx-- 
/var/gcc/reghunt/102852/gcc/testsuite/gcc/stage3/20000706-5.x2
FFFFFFFF7F100000       1216K r-x--  /lib/sparcv9/libc.so.1
FFFFFFFF7F230000         48K r-x--  /lib/sparcv9/libc.so.1
FFFFFFFF7F300000         24K rwx--    [ anon ]
FFFFFFFF7F33C000         72K rwx--  /lib/sparcv9/libc.so.1
FFFFFFFF7F34E000          8K rwx--  /lib/sparcv9/libc.so.1
FFFFFFFF7F400000        640K r-x--  /lib/sparcv9/libm.so.2
FFFFFFFF7F500000          8K rwx--    [ anon ]
FFFFFFFF7F59E000         40K rwx--  /lib/sparcv9/libm.so.2
FFFFFFFF7F600000         64K r-x--  /lib/sparcv9/ld.so.1
FFFFFFFF7F610000         64K r-x--  /lib/sparcv9/ld.so.1
FFFFFFFF7F620000         64K r-x--  /lib/sparcv9/ld.so.1
FFFFFFFF7F630000         24K r-x--  /lib/sparcv9/ld.so.1
FFFFFFFF7F700000         64K rwx--    [ anon ]
FFFFFFFF7F736000         16K rwx--  /lib/sparcv9/ld.so.1
FFFFFFFF7F73A000          8K rwx--  /lib/sparcv9/ld.so.1
FFFFFFFF7FFF0000         64K rw---    [ stack ]
         total         2440K

Please fix asap; it is almost impossible to do any useful work on a port that
broken.

Thanks.
  Rainer

[Bug tree-optimization/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2010.10.06 17:17:47
         AssignedTo|unassigned at gcc dot       |ebotcazou at gcc dot
                   |gnu.org                     |gnu.org
     Ever Confirmed|0                           |1

--- Comment #2 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-06 17:17:47 UTC ---
Investigating.

[Bug tree-optimization/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.0

[Bug rtl-optimization/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |rtl-optimization

--- Comment #3 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-11 22:06:49 UTC ---
The new build_ref_for_offset is miscompiled at -O or above.

[Bug target/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|rtl-optimization            |target

--- Comment #4 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-12 11:11:44 UTC ---
Long-standing latent bug in the SPARC back-end.

[Bug rtl-optimization/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |rtl-optimization

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-12 15:27:27 UTC ---
But there is a much bigger problem upstream.

[Bug rtl-optimization/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

--- Comment #6 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-13 21:43:44 UTC ---
Author: ebotcazou
Date: Wed Oct 13 21:43:42 2010
New Revision: 165437

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=165437
Log:
    PR rtl-optimization/45912
    * ira-costs.c (ira_tune_allocno_costs_and_cover_classes): Test the
    regno of registers instead of their index to compute the alignment.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/20101013-1.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ira-costs.c
    trunk/gcc/testsuite/ChangeLog

[Bug rtl-optimization/45912] [4.6 regression] Patch for PR tree-optimization/44972 caused 2500+ FAILs on Solaris 2/SPARC

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45912

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

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

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2010-10-13 22:23:56 UTC ---
.