[Bug bootstrap/59934] New: Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[Bug bootstrap/59934] New: Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[krebbel at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

            Bug ID: 59934
           Summary: Bootstrap fail since r206941: expmed.h:252:33: error:
                    array subscript is above array bounds
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: critical
          Priority: P3
         Component: bootstrap
          Assignee: unassigned at gcc dot gnu.org
          Reporter: krebbel at gcc dot gnu.org

Bootstrap on s390x started failing with r206941:

Author: law
Date: Wed Jan 22 21:27:38 2014
New Revision: 206941

URL: http://gcc.gnu.org/viewcvs?rev=206941&root=gcc&view=rev
Log:
    PR tree-optimization/59597
    * tree-ssa-threadupdate.c (dump_jump_thread_path): Move to earlier
    in file.  Accept new argument REGISTERING and use it to modify
    dump output appropriately.
    (register_jump_thread): Corresponding changes.
    (mark_threaded_blocks): Reinstate code to cancel unprofitable
    thread paths involving joiner blocks.  Add code to dump cancelled
    jump threading paths.

    PR tree-optimization/59597
    * gcc.dg/tree-ssa/pr59597.c: New test.

Stage 2:

In file included from /home/andreas/clean/gcc-head/gcc/expmed.c:39:0:
/home/andreas/clean/gcc-head/gcc/expmed.h: In function ‘void init_expmed()’:
/home/andreas/clean/gcc-head/gcc/expmed.h:252:33: error: array subscript is
above array bounds [-Werror=array-bounds]
   return &costs->cost[speed][idx];
                                 ^
/home/andreas/clean/gcc-head/gcc/expmed.h:252:33: error: array subscript is
above array bounds [-Werror=array-bounds]
>From gcc-bugs-return-441429-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Fri Jan 24 13:04:11 2014
Return-Path: <gcc-bugs-return-441429-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 27116 invoked by alias); 24 Jan 2014 13:04:11 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 27068 invoked by uid 48); 24 Jan 2014 13:04:08 -0000
From: "krebbel at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds
Date: Fri, 24 Jan 2014 13:04:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: bootstrap
X-Bugzilla-Version: 4.9.0
X-Bugzilla-Keywords:
X-Bugzilla-Severity: major
X-Bugzilla-Who: krebbel at gcc dot gnu.org
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Priority: P2
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: cf_gcctarget priority cc cf_gcchost cf_gccbuild bug_severity
Message-ID: <bug-59934-4-VV4f2pzlCm@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-59934-4@http.gcc.gnu.org/bugzilla/>
References: <bug-59934-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2014-01/txt/msg02571.txt.bz2
Content-length: 613

http://gcc.gnu.org/bugzilla/show_bug.cgi?idY934

Andreas Krebbel <krebbel at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |s390x-ibm-linux
           Priority|P3                          |P2
                 CC|                            |law at gcc dot gnu.org
               Host|                            |s390x-ibm-linux
              Build|                            |s390x-ibm-linux
           Severity|critical                    |major

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

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

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

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Can you please attach expmed.i and gcc options used to compile it, so that it
can be hopefully debugged using a cross-compiler?
As it is stage2, I'd think that it shouldn't be a miscompilation of cc1plus
itself.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[krebbel at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #2 from Andreas Krebbel <krebbel at gcc dot gnu.org> ---
Created attachment 31948
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=31948&action=edit
Preprocessed file

compile with: -O2 -Warray-bounds

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com
           Assignee|unassigned at gcc dot gnu.org      |law at redhat dot com

--- Comment #3 from Jeffrey A. Law <law at redhat dot com> ---
I'll take a look.  It's a couple deep in the queue, but obviously high
priority.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I think it is just warning on dead code, but GCC doesn't know it is dead code.
s390 doesn't have any partial or vector modes, so expmed_mode_index
because MIN_MODE_PARTIAL_INT and MIN_MODE_VECTOR_INT are 0 becomes:
  switch (((enum mode_class) mode_class[mode]))
    {
    case MODE_INT:
      return mode - 22;
    case MODE_PARTIAL_INT:
      return mode + 6;
    case MODE_VECTOR_INT:
      return mode + 6;
    default:
      gcc_unreachable ();
    }
}

Now, when we use the result of this in a loop with mode going from 22 to 27 or
so, for MODE_INT (which those modes are) we get correct values 0 to 5, but for
for the other cases the range of mode + 6 will be 28 to 33 which obviously
doesn't fit as a valid index into array of 6.  Now, when jump threading or what
duplicates the array accesses for the different cases, it is obvious why VRP2
complains.  But, I'd guestion the desirability to warn about it during VRP2 at
all, at least unless we have some way to mark statements that were jump
threaded and just not warn about those.  I remember various other issues like
this in the past.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #5 from Jeffrey A. Law <law at redhat dot com> ---
I'd kindof suspected something along those lines and I've had to fix problems
like this in the past.

I'll have to look at how we dealt with this in the past.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Now, if we want to "fix" this on the expmed.h side, either:
--- gcc/expmed.h    2014-01-03 11:40:57.228320531 +0100
+++ gcc/expmed.h    2014-01-24 18:30:12.513908749 +0100
@@ -221,9 +221,11 @@ expmed_mode_index (enum machine_mode mod
     case MODE_INT:
       return mode - MIN_MODE_INT;
     case MODE_PARTIAL_INT:
-      return mode - MIN_MODE_PARTIAL_INT + NUM_MODE_INT;
+      return (MIN_MODE_PARTIAL_INT == VOIDmode
+          ? 0 : mode - MIN_MODE_PARTIAL_INT + NUM_MODE_INT);
     case MODE_VECTOR_INT:
-      return mode - MIN_MODE_VECTOR_INT + NUM_MODE_IP_INT;
+      return (MIN_MODE_VECTOR_INT == VOIDmode
+          ? mode - MIN_MODE_VECTOR_INT + NUM_MODE_IP_INT);
     default:
       gcc_unreachable ();
     }

or
--- gcc/expmed.h    2014-01-03 11:40:57.228320531 +0100
+++ gcc/expmed.h    2014-01-24 18:34:01.806725352 +0100
@@ -221,8 +221,10 @@ expmed_mode_index (enum machine_mode mod
     case MODE_INT:
       return mode - MIN_MODE_INT;
     case MODE_PARTIAL_INT:
+      gcc_assert (MIN_MODE_PARTIAL_INT != VOIDmode);
       return mode - MIN_MODE_PARTIAL_INT + NUM_MODE_INT;
     case MODE_VECTOR_INT:
+      gcc_assert (MIN_MODE_VECTOR_INT != VOIDmode);
       return mode - MIN_MODE_VECTOR_INT + NUM_MODE_IP_INT;
     default:
       gcc_unreachable ();
should be enough to explain to the compiler that those cases really can't
happen and let gcc not emit the warnings.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Perhaps we could set gimple_set_no_warning on the jump threaded stmts and honor
that in VRP array checking.  Or TREE_NO_WARNING on the handled_component_p
operands of the threaded stmts.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #8 from Jeffrey A. Law <law at redhat dot com> ---
Setting TREE_NO_WARNING seems wrong to me.  That would really seem better
suited for cases where we have already warned on that expression and don't want
to warn on it again.  This case is pretty different.

In effect threading isolates paths.  Path isolation can/will expose
out-of-bounds array indexing that is missed by the insanely lame code in
tree-vrp.c.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #9 from Jeffrey A. Law <law at redhat dot com> ---
Jakub,
I think your second approach is the better solution.  It'll abort rather than
silently returning a value which may or may not be appropriate in the caller's
context.

ie, returning 0 when we don't (for example) have partial modes may avoid this
particular problem, but may cause issues elsewhere.

Granted this code only has one user, but I'd rather play things like this
defensively and just abort if we get into a situation that's not supposed to
happen.

I'll start an s390x build and see where the 2nd approach takes us...

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
But with the second patch we'll generate bigger code (because we pass arguments
to fancy_abort) and it really is a will never happen case, if you don't have
any partial (or vector) modes, then no mode will have MODE_PARTIAL_INT (or
MODE_VECTOR_INT) class.  Plus, with --disable-checking we'll trigger the
warning again.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
But it is 50 times inlined.  Anyway, supposedly the aborts would be all merged
then.  Looking at current gcc_assert definition, we actually use there
__builtin_unreachable (), at least when built with gcc 4.5+, and we only do
-Werror and care about warnings when built with recent gcc anyway, so
gcc_assert should be good enough.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 31952
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=31952&action=edit
gcc49-pr59934.patch

So what about this way?  Then I'd expect for the missing partial/vector int
the compiler could easily find out that it shouldn't do a switch with 4
different cases, but just check class against MODE_INT and otherwise abort, or
for --disable-checking don't check the class at all.

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[law at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

--- Comment #15 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Author: law
Date: Mon Jan 27 16:33:39 2014
New Revision: 207137

URL: http://gcc.gnu.org/viewcvs?rev=207137&root=gcc&view=rev
Log:
       PR bootstrap/59934
        * expmed.h (expmed_mode_index): Rework so that analysis and optimziers
        know when the MODE_PARTIAL_INT and MODE_VECTOR_INT cases can never be
        reached.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/expmed.h

[Bug bootstrap/59934] Bootstrap fail since r206941: expmed.h:252:33: error: array subscript is above array bounds

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59934

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

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

--- Comment #16 from Jeffrey A. Law <law at redhat dot com> ---
Should be fixed by commit on trunk.  Bootstrap still fails in stage3 libjava,
but that's unrelated to this fix AFAICT.