[Bug bootstrap/52287] New: [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[Bug bootstrap/52287] New: [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

             Bug #: 52287
           Summary: [4.7 regression] ICE in ready_remove_first, at
                    haifa-sched.c:1927
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: bernds@gcc.gnu.org, ebotcazou@gcc.gnu.org
              Host: sparc-sun-solaris2.8
            Target: sparc-sun-solaris2.8
             Build: sparc-sun-solaris2.8


Between rev. 184203 and 184303, Solaris 8/SPARC bootstrap broke in stage2:

$ cc1plus -fpreprocessed reginfo.ii -quiet -mcpu=v9 -g -O2 -fno-exceptions
-fno-rtti -o reginfo.s
/vol/gcc/src/hg/trunk/local/gcc/reginfo.c: In function 'int
memory_move_secondary_cost(machine_mode, reg_class_t, bool)':
/vol/gcc/src/hg/trunk/local/gcc/reginfo.c:716:1: internal compiler error: in
ready_remove_first, at haifa-sched.c:1927

gdb finds

(gdb) up
#1  0x02ae1c64 in ready_remove_first (ready=0x346e558)
    at /vol/gcc/src/hg/trunk/local/gcc/haifa-sched.c:1927
1927      gcc_assert (QUEUE_INDEX (t) == QUEUE_READY);
(gdb) p t
$1 = (rtx) 0xf95db748
(gdb) pr
(debug_insn 49 48 50 6 (var_location:SI from (reg/v:SI 119 [ rclass ])) -1
     (nil))

I wonder if this could be related to

2012-02-14  Bernd Schmidt  <bernds@codesourcery.com>

    * haifa-sched.c (prune_ready_list): Ensure that if there is a
    sched-group insn, it either remains alone or the entire list is
    pruned.

Strangely, this doesn't occur on Solaris 9 and up.

  Rainer

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #1 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-02-16 19:47:29 UTC ---
Please attach the reginfo.ii file.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #2 from Rainer Orth <ro at gcc dot gnu.org> 2012-02-16 19:52:40 UTC ---
Created attachment 26684
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26684
preprocessed input

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #3 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-02-16 20:43:03 UTC ---
Not reproducible with a cross compiler.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #4 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-02-17 11:56:05 UTC ---
> --- Comment #3 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-02-16 20:43:03 UTC ---
> Not reproducible with a cross compiler.

I've just verified that reverting your patch allows the bootstrap to get
into stage3.

    Rainer

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.7.0

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

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> 2012-02-22 15:02:44 UTC ---
Yeah, can't reproduce this with a cross either, and even valgrind is clean on
it.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-02-22
     Ever Confirmed|0                           |1

--- Comment #6 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-02-22 17:24:53 UTC ---
I can reproduce with a native compiler.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #26684|0                           |1
        is obsolete|                            |
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |ebotcazou at gcc dot
                   |gnu.org                     |gnu.org

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-02-23 14:28:10 UTC ---
Created attachment 26734
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26734
Reduced testcase

Still reproducible on native only.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-02-23 15:23:44 UTC ---
This reproduces only on Solaris 8 because the sort of the ready list isn't
stable in the presence of debug insns, given that rank_for_schedule isn't
anti-symmetrical:

  if (MAY_HAVE_DEBUG_INSNS)
    {
      /* Schedule debug insns as early as possible.  */
      if (DEBUG_INSN_P (tmp) && !DEBUG_INSN_P (tmp2))
    return -1;
      else if (DEBUG_INSN_P (tmp2))
    return 1;
    }

The following patchlet is enough to get rid of the ICE:

Index: haifa-sched.c
===================================================================
--- haifa-sched.c       (revision 184352)
+++ haifa-sched.c       (working copy)
@@ -1647,7 +1647,7 @@ rank_for_schedule (const void *x, const
       /* Schedule debug insns as early as possible.  */
       if (DEBUG_INSN_P (tmp) && !DEBUG_INSN_P (tmp2))
        return -1;
-      else if (DEBUG_INSN_P (tmp2))
+      else if (DEBUG_INSN_P (tmp2) && !DEBUG_INSN_P (tmp))
        return 1;
     }

as it makes the sort stable and equivalent to the non-Solaris 8 case.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #9 from Bernd Schmidt <bernds at gcc dot gnu.org> 2012-02-23 15:41:49 UTC ---
(In reply to comment #8)
> This reproduces only on Solaris 8 because the sort of the ready list isn't
> stable in the presence of debug insns, given that rank_for_schedule isn't
> anti-symmetrical:

Thanks for investigating this.

> Index: haifa-sched.c
> ===================================================================
> --- haifa-sched.c       (revision 184352)
> +++ haifa-sched.c       (working copy)
> @@ -1647,7 +1647,7 @@ rank_for_schedule (const void *x, const
>        /* Schedule debug insns as early as possible.  */
>        if (DEBUG_INSN_P (tmp) && !DEBUG_INSN_P (tmp2))
>         return -1;
> -      else if (DEBUG_INSN_P (tmp2))
> +      else if (DEBUG_INSN_P (tmp2) && !DEBUG_INSN_P (tmp))
>         return 1;
>      }
> 
> as it makes the sort stable and equivalent to the non-Solaris 8 case.

Almost OK, but in the case of two debug insns, can you make it return the
difference of the INSN_UIDs here? I'd rather not fall through into the other
code for DEBUG_INSNs.

Such a patch is preapproved.

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-02-23 15:52:23 UTC ---
I don't think INSN_UID is right for that, I think you want to preserve the
original order of debug insns in that case, and INSN_UID can jump up and down.
So what about
  if (MAY_HAVE_DEBUG_INSNS)
    {
      /* Schedule debug insns as early as possible.  */
      if (DEBUG_INSN (tmp))
        {
          if (DEBUG_INSN (tmp2))
            return INSN_LUID (tmp) - INSN_LUID (tmp2);
          return -1;
        }
      else if (DEBUG_INSN (tmp2))
        return 1;
    }
?

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #11 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-02-23 16:00:58 UTC ---
Created attachment 26736
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26736
Tentative fix

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

--- Comment #12 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-02-23 22:15:53 UTC ---
Author: ebotcazou
Date: Thu Feb 23 22:15:44 2012
New Revision: 184531

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=184531
Log:
    PR bootstrap/52287
    * haifa-sched.c (rank_for_schedule): Stabilize sort for debug insns.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/haifa-sched.c

[Bug bootstrap/52287] [4.7 regression] ICE in ready_remove_first, at haifa-sched.c:1927

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52287

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

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

--- Comment #13 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-02-23 22:17:45 UTC ---
Patch applied.