[Bug bootstrap/46856] New: [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[Bug bootstrap/46856] New: [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

           Summary: [4.6 regression] internal compiler error in
                    final_scan_insn breaks m68k-linux bootstrap
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: mikpe@it.uu.se


Attempting to bootstrap gcc-4.6-20101204 natively on m68k-linux fails with:

/mnt/scratch/objdir46/./prev-gcc/xgcc -B/mnt/scratch/objdir46/./prev-gcc/
-B/mnt/scratch/crap/m68k-unknown-linux-gnu/bin/
-B/mnt/scratch/crap/m68k-unknown-linux-gnu/bin/
-B/mnt/scratch/crap/m68k-unknown-linux-gnu/lib/ -isystem
/mnt/scratch/crap/m68k-unknown-linux-gnu/include -isystem
/mnt/scratch/crap/m68k-unknown-linux-gnu/sys-include    -c   -g -O2 -gtoggle
-DIN_GCC   -W -Wall -Wwrite-strings -Wcast-qual -Wstrict-prototypes
-Wmissing-prototypes -Wmissing-format-attribute -pedantic -Wno-long-long
-Wno-variadic-macros -Wno-overlength-strings -Werror -Wold-style-definition
-Wc++-compat   -DHAVE_CONFIG_H -I. -I. -I/mnt/scratch/gcc-4.6-20101204/gcc
-I/mnt/scratch/gcc-4.6-20101204/gcc/.
-I/mnt/scratch/gcc-4.6-20101204/gcc/../include
-I/mnt/scratch/gcc-4.6-20101204/gcc/../libcpp/include
-I/home/mikpe/pkgs/linux-m68k/gmp-4.3.2/include
-I/home/mikpe/pkgs/linux-m68k/mpfr-2.4.2/include
-I/home/mikpe/pkgs/linux-m68k/mpc-0.8.2/include 
-I/mnt/scratch/gcc-4.6-20101204/gcc/../libdecnumber
-I/mnt/scratch/gcc-4.6-20101204/gcc/../libdecnumber/dpd -I../libdecnumber   
/mnt/scratch/gcc-4.6-20101204/gcc/haifa-sched.c -o haifa-sched.o
/mnt/scratch/gcc-4.6-20101204/gcc/haifa-sched.c: In function
'rank_for_schedule':
/mnt/scratch/gcc-4.6-20101204/gcc/haifa-sched.c:1293:1: internal compiler
error: in final_scan_insn, at final.c:2608
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
make[3]: *** [haifa-sched.o] Error 1
make[3]: Leaving directory `/mnt/scratch/objdir46/gcc'
make[2]: *** [all-stage2-gcc] Error 2
make[2]: Leaving directory `/mnt/scratch/objdir46'
make[1]: *** [stage2-bubble] Error 2
make[1]: Leaving directory `/mnt/scratch/objdir46'
make: *** [bootstrap] Error 2

Current gcc-4.4 and 4.5 do bootstrap on this machine so this is a regression.

I'll attach a reduced test case that ICEs a cross to m68k-linux.

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #1 from Mikael Pettersson <mikpe at it dot uu.se> 2010-12-08 23:22:01 UTC ---
Created attachment 22689
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22689
test case reduced from haifa-sched.c

> objdir/gcc/xgcc -Bobjdir/gcc -O2 -S pr46856.c 
pr46856.c: In function 'increase_insn_priority':
pr46856.c:27:1: internal compiler error: in final_scan_insn, at final.c:2608
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
> objdir/gcc/xgcc -v
Using built-in specs.
COLLECT_GCC=objdir/gcc/xgcc
Target: m68k-unknown-linux
Configured with: /tmp/gcc-4.6-20101204/configure --target=m68k-unknown-linux
--prefix=/home/mikpe/pkgs/linux-x86/cross-m68k
--with-gmp=/home/mikpe/pkgs/linux-x86/gmp-4.3.2
--with-mpfr=/home/mikpe/pkgs/linux-x86/mpfr-2.4.2
--with-mpc=/home/mikpe/pkgs/linux-x86/mpc-0.8.2 --disable-plugin --disable-lto
--disable-nls --enable-shared --disable-libmudflap --disable-multilib
--enable-threads=posix --enable-checking=release --enable-languages=c
Thread model: posix
gcc version 4.6.0 20101204 (experimental) (GCC)

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

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

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

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4
                 CC|                            |bernds at gcc dot gnu.org,
                   |                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-12-09 11:53:06 UTC ---
Seems this is caused by postreload being broken for HAVE_cc0 targets.
reload_combine_recognize_const_pattern happily moves an add after a CC0 setter
(but before CC0 user), breaking the requirement that CC0 setter immediately
preceedes CC0 user.
HAVE_cc0 targets should just be converted or die, but until that happens,
postreload should be probably fixed.

P4 as none of the primary/secondary targets is HAVE_cc0.

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #3 from Mikael Pettersson <mikpe at it dot uu.se> 2010-12-09 23:48:36 UTC ---
A bisection identified r166371 as the trigger for this bug:

Author: jakub
Date: Fri Nov  5 19:00:27 2010
New Revision: 166371

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166371
Log:
    PR target/45670
    * expr.c (expand_expr_real_1) <case MEM_REF>: Use EXPAND_SUM
    instead of EXPAND_NORMAL for base expansion.

    * gcc.target/i386/pr45670.c: New test.

Reverting that change from gcc-4.6-20101204 prevents the ICE.

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #4 from Mikael Pettersson <mikpe at it dot uu.se> 2010-12-11 12:17:46 UTC ---
Jakub's r166371 is innocent, all it did was to revert an expansion mode change
in r162618 (2nd PR44790 patch).  Trunk actually started to ICE for this test
case on m68k in r162270:

Author: bernds
Date: Fri Jul 16 23:47:46 2010
New Revision: 162270

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=162270
Log:
    PR target/42235
    * postreload.c (reload_cse_move2add): Return bool, true if anything.
    changed.  All callers changed.
    (move2add_use_add2_insn): Likewise.
    (move2add_use_add3_insn): Likewise.
    (reload_cse_regs): If reload_cse_move2add changed anything, rerun
    reload_combine.
    (RELOAD_COMBINE_MAX_USES): Bump to 16.
    (last_jump_ruid): New static variable.
    (struct reg_use): New members CONTAINING_MEM and RUID.
    (reg_state): New members ALL_OFFSETS_MATCH and REAL_STORE_RUID.
    (reload_combine_split_one_ruid, reload_combine_split_ruids,
    reload_combine_purge_insn_uses, reload_combine_closest_single_use
    reload_combine_purge_reg_uses_after_ruid,
    reload_combine_recognize_const_pattern): New static functions.
    (reload_combine_recognize_pattern): Verify that ALL_OFFSETS_MATCH
    is true for our reg and that we have available index regs.
    (reload_combine_note_use): New args RUID and CONTAINING_MEM.  All
    callers changed.  Use them to initialize fields in struct reg_use.
    (reload_combine): Initialize last_jump_ruid.  Be careful when to
    take PREV_INSN of the scanned insn.  Update REAL_STORE_RUID fields.
    Call reload_combine_recognize_const_pattern.
    (reload_combine_note_store): Update REAL_STORE_RUID field.

which supports Jakub's statement that reload_combine_recognize_const_pattern
misbehaves.

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #5 from Mikael Pettersson <mikpe at it dot uu.se> 2011-01-18 08:29:42 UTC ---
If I hack reload_combine_recognize_const_pattern to bail out early if HAVE_cc0,
then the test case doesn't ICE a cross to m68k-linux, and native bootstrap on
m68k-linux also succeeds (--enable-languages=c,c++; testsuite currently
running).

I'm not suggesting this is the proper fix, but at least it identifies roughly
were the problem lies.

--- gcc-4.6-20110115/gcc/postreload.c.~1~       2010-12-21 15:51:42.000000000
+0100
+++ gcc-4.6-20110115/gcc/postreload.c   2011-01-16 12:29:44.000000000 +0100
@@ -944,6 +944,11 @@ reload_combine_recognize_const_pattern (
   int add_moved_after_ruid = 0;
   int clobbered_regno = -1;

+#ifdef HAVE_cc0
+  /* Unbreak m68k, see PR bootstrap/46856. */
+  return false;
+#endif
+
   set = single_set (insn);
   if (set == NULL_RTX)
     return false;

[Bug bootstrap/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #6 from Mikael Pettersson <mikpe at it dot uu.se> 2011-01-26 14:43:34 UTC ---
The loop in reload_combine_recognize_const_pattern contains this check:

      /* If the add clobbers another hard reg in parallel, don't move
         it past a real set of this hard reg.  */
      if (must_move_add && clobbered_regno >= 0
          && reg_state[clobbered_regno].real_store_ruid >= use_ruid)
        break;

Since the moved add is a cc0 setter, it seems reasonable to check here that
it's not moved past another cc0 setter.  Replacing the previous patch with the
following also restores bootstrap (testsuite still running):

--- gcc-4.6-20110115/gcc/postreload.c.~1~       2010-12-21 15:51:42.000000000
+0100
+++ gcc-4.6-20110115/gcc/postreload.c   2011-01-18 20:26:39.000000000 +0100
@@ -1008,6 +1008,11 @@ reload_combine_recognize_const_pattern (
          if (must_move_add && clobbered_regno >= 0
              && reg_state[clobbered_regno].real_store_ruid >= use_ruid)
            break;
+#if defined(HAVE_cc0)
+         /* Unbreak m68k, see PR bootstrap/46856. */
+         if (must_move_add && sets_cc0_p (PATTERN (use_insn)))
+           break;
+#endif

          gcc_assert (reg_state[regno].store_ruid <= use_ruid);
          /* Avoid moving a use of ADDREG past a point where it is stored.  */

[Bug rtl-optimization/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.01.26 17:16:52
                 CC|                            |law at redhat dot com
          Component|bootstrap                   |rtl-optimization
         AssignedTo|unassigned at gcc dot       |law at redhat dot com
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #7 from Jeffrey A. Law <law at redhat dot com> 2011-01-26 17:16:52 UTC ---
More correctly, you can never move any insn between an explicit cc0 setter and
cc0 user on a cc0 target.

Regardless, I think you've nailed this one down.

The comment needs some minor work.   Rather than referring to a PR, we strongly
prefer to describe why the code works the way it does in the code itself.  We
do this so that someone reading this code later can quickly understand its
relevance without having to fire up a browser to read the bug database.

Something as simple as "do not separate a cc0 setter and cc0 user" is
sufficient.  You should also submit the testcase for the regression suite. 
Just name it pr46856.c and put it in gcc/testsuite/gcc.c-torture/compile

I don't see a login for you on gcc.gnu.org, so I'm going to assume you don't
have commit privs.  I'll go ahead and make the cleanups mentioned above and
submit the patch.  You can review the submission so that you know proper form
if you choose to submit patches again in the future.

[Bug rtl-optimization/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[law at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #8 from Jeffrey A. Law <law at gcc dot gnu.org> 2011-01-27 04:25:57 UTC ---
Author: law
Date: Thu Jan 27 04:25:49 2011
New Revision: 169321

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169321
Log:

    PR rtl-optimization/46856
    * postreload.c (reload_combine_recognize_const_pattern): Do not
    separate cc0 setter and user on cc0 targets.

    PR rtl-optimization/46856
    * gcc.c-torture/compile/pr46856.c: New test.



Added:
    trunk/gcc/testsuite/gcc.c-torture/compile/pr46856.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/postreload.c
    trunk/gcc/testsuite/ChangeLog

[Bug rtl-optimization/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #9 from Jeffrey A. Law <law at redhat dot com> 2011-01-27 04:27:14 UTC ---
Fixed.

[Bug rtl-optimization/46856] [4.6 regression] internal compiler error in final_scan_insn breaks m68k-linux bootstrap

[dnovillo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46856

--- Comment #10 from Diego Novillo <dnovillo at gcc dot gnu.org> 2011-02-02 17:56:15 UTC ---
Author: dnovillo
Date: Wed Feb  2 17:56:10 2011
New Revision: 169641

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=169641
Log:

    PR rtl-optimization/46856
    * postreload.c (reload_combine_recognize_const_pattern): Do not
    separate cc0 setter and user on cc0 targets.

    PR rtl-optimization/46856
    * gcc.c-torture/compile/pr46856.c: New test.

Added:
    branches/google/integration/gcc/testsuite/gcc.c-torture/compile/pr46856.c
Modified:
    branches/google/integration/gcc/ChangeLog
    branches/google/integration/gcc/postreload.c
    branches/google/integration/gcc/testsuite/ChangeLog