[Bug bootstrap/48000] New: [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[Bug bootstrap/48000] New: [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

           Summary: [4.6 Regression] LTO bootstrap failed with
                    bootstrap-profiled
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: jakub@redhat.com


On Linux/x86-64, revision 170699 failed to LTO bootstrap
with bootstrap-profiled:

http://gcc.gnu.org/ml/gcc-regression/2011-03/msg00030.html

../../src-trunk/gcc/reginfo.c: In function 'init_reg_modes_target':
../../src-trunk/gcc/reginfo.c:586:1: internal compiler error: in
dfs_enumerate_from, at cfganal.c:1209
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Revision 170697 is OK. It is caused by revision 170699:

http://gcc.gnu.org/ml/gcc-cvs/2011-03/msg00120.html

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

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

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

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-05 23:07:47 UTC ---
Created attachment 23555
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=23555
reginfo.i

./cc1 reginfo.i -quiet -O2 -fprofile-use
reproduces this.

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-05 23:08:27 UTC ---
Created attachment 23556
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=23556
reginfo.gcda

Corresponding reginfo.gcda.

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.03.06 09:26:02
                 CC|                            |hubicka at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-03-06 09:26:02 UTC ---
Confirmed. Most porbably fallout from Zdenek's patch. The loop body size is
wrong.

Honza

(gdb) bt
#0  fancy_abort (file=0x310ef83 "../../gcc/cfganal.c", line=1209,
function=0x310f1a0 "dfs_enumerate_from") at ../../gcc/diagnostic.c:893
#1  0x000000000087e887 in dfs_enumerate_from (bb=0x7ffff6619618, reverse=1,
predicate=0x924030 <glb_enum_p>, rslt=0x466d2a0, rslt_max=1,
data=0x7ffff69a0880)
    at ../../gcc/cfganal.c:1209
#2  0x00000000009240c1 in get_loop_body_with_size (loop=0x7ffff69a0880,
body=0x466d2a0, max_size=1) at ../../gcc/cfgloop.c:800
#3  0x0000000000924202 in get_loop_body (loop=0x7ffff69a0880) at
../../gcc/cfgloop.c:829
#4  0x0000000000924b6f in get_loop_exit_edges (loop=0x7ffff69a0880) at
../../gcc/cfgloop.c:1146
#5  0x0000000000962c3b in fix_loop_placement (loop=0x7ffff69a0880) at
../../gcc/cfgloopmanip.c:129
#6  0x0000000000962eb8 in fix_bb_placements (from=0x7ffff6619618,
irred_invalidated=0x7fffffffe10f "") at ../../gcc/cfgloopmanip.c:215
#7  0x000000000096350a in remove_path (e=0x7ffff6870840) at
../../gcc/cfgloopmanip.c:383
#8  0x000000000106b1d2 in unswitch_loop (loop=0x7ffff7361990,
unswitch_on=0x7ffff660d7b8, cond=0x7ffff64feba0, cinsn=0x0) at
../../gcc/loop-unswitch.c:458
#9  0x000000000106a6d6 in unswitch_single_loop (loop=0x7ffff7361990,
cond_checked=0x0, num=0) at ../../gcc/loop-unswitch.c:370
#10 0x0000000001069c75 in unswitch_loops () at ../../gcc/loop-unswitch.c:150
#11 0x0000000001007332 in rtl_unswitch () at ../../gcc/loop-init.c:285
#12 0x000000000121f258 in execute_one_pass (pass=0x43cfe20) at
../../gcc/passes.c:1556
#13 0x000000000121f432 in execute_pass_list (pass=0x43cfe20) at
../../gcc/passes.c:1611
#14 0x000000000121f450 in execute_pass_list (pass=0x43cfca0) at
../../gcc/passes.c:1612
#15 0x000000000121f450 in execute_pass_list (pass=0x43d00c0) at
../../gcc/passes.c:1612
#16 0x00000000019a333b in tree_rest_of_compilation (fndecl=0x7ffff7007300) at
../../gcc/tree-optimize.c:422
#17 0x00000000025a1146 in cgraph_expand_function (node=0x7ffff69a86e0) at
../../gcc/cgraphunit.c:1576
#18 0x00000000025a12f6 in cgraph_expand_all_functions () at
../../gcc/cgraphunit.c:1635
#19 0x00000000025a1943 in cgraph_optimize () at ../../gcc/cgraphunit.c:1899
#20 0x000000000259f575 in cgraph_finalize_compilation_unit () at
../../gcc/cgraphunit.c:1096
#21 0x00000000004fd529 in c_write_global_declarations () at
../../gcc/c-decl.c:9872
#22 0x0000000001644bfd in compile_file () at ../../gcc/toplev.c:591
#23 0x0000000001646e94 in do_compile () at ../../gcc/toplev.c:1900
#24 0x0000000001646fbf in toplev_main (argc=4, argv=0x7fffffffe7c8) at
../../gcc/toplev.c:1963
#25 0x00000000007109ff in main (argc=4, argv=0x7fffffffe7c8) at
../../gcc/main.c:36

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[rakdver at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

--- Comment #4 from Zdenek Dvorak <rakdver at gcc dot gnu.org> 2011-03-06 10:43:41 UTC ---
Created attachment 23559
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=23559
A proposed patch

This fixes the problem uncovered by the previous fix.  In fix_bb_placements, we
restrict the placement changes to the basic blocks that originally were in the
base loop (the loop containing the FROM basic block).  This is achieved by
marking the header of the loop visited, so that we never enter it to the
processing queue.  However, if FROM itself is the header of the loop, this
fails, as we enter FROM to the queue unconditionally.  Fixed by ending
immediatelly if FROM is a loop header.

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

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> 2011-03-06 17:38:11 UTC ---
(In reply to comment #4)
> Created attachment 23559 [details]
> A proposed patch
> 
> This fixes the problem uncovered by the previous fix.  In fix_bb_placements, we
> restrict the placement changes to the basic blocks that originally were in the
> base loop (the loop containing the FROM basic block).  This is achieved by
> marking the header of the loop visited, so that we never enter it to the
> processing queue.  However, if FROM itself is the header of the loop, this
> fails, as we enter FROM to the queue unconditionally.  Fixed by ending
> immediatelly if FROM is a loop header.

Successfully bootstrapped/regtested on x86_64-linux and i686-linux as well as
successfully profiledbootstrapped/regtested on those two targets (the latter is
what was failing before), no regressions.

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

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

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

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-07 10:13:47 UTC ---
Looks good to me, quite obvious in fact.

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-07 16:41:57 UTC ---
Author: jakub
Date: Mon Mar  7 16:41:53 2011
New Revision: 170740

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=170740
Log:
    PR bootstrap/48000
    * cfgloopmanip.c (fix_bb_placements): Return immediately
    if FROM is BASE_LOOP's header.

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

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

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

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

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-07 16:45:09 UTC ---
Fixed.

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |y@momonga-linux.org

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-10 15:24:33 UTC ---
*** Bug 48060 has been marked as a duplicate of this bug. ***

[Bug bootstrap/48000] [4.6 Regression] LTO bootstrap failed with bootstrap-profiled

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48000

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-21 15:51:48 UTC ---
Author: rguenth
Date: Thu Apr 21 15:51:39 2011
New Revision: 172833

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=172833
Log:
2011-04-21  Richard Guenther  <rguenther@suse.de>

    PR middle-end/48712
    Backport from 4.6 branch
    2011-03-07  Zdenek Dvorak  <ook@ucw.cz>

    PR bootstrap/48000
    * cfgloopmanip.c (fix_bb_placements): Return immediately
    if FROM is BASE_LOOP's header.

Modified:
    branches/gcc-4_5-branch/gcc/ChangeLog
    branches/gcc-4_5-branch/gcc/cfgloopmanip.c