[Bug middle-end/59285] New: gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[Bug middle-end/59285] New: gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[ktkachov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

            Bug ID: 59285
           Summary: gcc.dg/builtin-unreachable-6.c:17:1: internal compiler
                    error: in rtl_verify_fallthru, at cfgrtl.c:2862
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ktkachov at gcc dot gnu.org

The gcc.dg/builtin-unreachable-6.c testcase fails on arm-none-eabi with:

$SRC/gcc/gcc/testsuite/gcc.dg/builtin-unreachable-6.c: In function 'foo':
$SRC/gcc/gcc/testsuite/gcc.dg/builtin-unreachable-6.c:17:1: error:
verify_flow_info: Incorrect fallthru 2->5
$SRC/gcc/gcc/testsuite/gcc.dg/builtin-unreachable-6.c:17:1: error: wrong insn
in the fallthru edge
(barrier 21 20 27)
$SRC/gcc/gcc/testsuite/gcc.dg/builtin-unreachable-6.c:17:1: internal compiler
error: in rtl_verify_fallthru, at cfgrtl.c:2862
0x95d645 _fatal_insn(char const*, rtx_def const*, char const*, int, char
const*)
        $SRC/gcc/gcc/rtl-error.c:109
0x69254a rtl_verify_fallthru
        $SRC/gcc/gcc/cfgrtl.c:2862
0x69254a rtl_verify_flow_info
        $SRC/gcc/gcc/cfgrtl.c:2962
0x681af7 verify_flow_info()
        $SRC/gcc/gcc/cfghooks.c:258
0xe5c24f if_convert
        $SRC/gcc/gcc/ifcvt.c:4460
0xe5d95d rest_of_handle_if_after_reload
        $SRC/gcc/gcc/ifcvt.c:4594
0xe5d95d execute
        $SRC/gcc/gcc/ifcvt.c:4625
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.

Bisection shows it started with r205074:

        * tree-ssa-threadedge.c (thread_across_edge): After threading
        through a joiner, allow threading a normal block requiring
        duplication.

        * tree-ssa-threadupdate.c (thread_block_1): Improve code to detect
        jump threading requests that would muck up the loop structures.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.9.0
            Summary|gcc.dg/builtin-unreachable- |[4.9 Regression]
                   |6.c:17:1: internal compiler |gcc.dg/builtin-unreachable-
                   |error: in                   |6.c:17:1: internal compiler
                   |rtl_verify_fallthru, at     |error: in
                   |cfgrtl.c:2862               |rtl_verify_fallthru, at
                   |                            |cfgrtl.c:2862

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2013-11-25
     Ever confirmed|0                           |1

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #1 from Jeffrey A. Law <law at redhat dot com> ---
Patch in testing.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #2 from Jeffrey A. Law <law at redhat dot com> ---
After sleeping on it, I think a different approach to the fix is going to be
better.  This is a latent bug in the RTL conditional execution/block merging
code.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #3 from Jeffrey A. Law <law at redhat dot com> ---
Working on this again.  I'm on the 4th iteration of the fix.  Bootstrapping on
ARM boxes is painfully slow :(

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[ktkachov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #4 from ktkachov at gcc dot gnu.org ---
(In reply to Jeffrey A. Law from comment #3)
> Working on this again.  I'm on the 4th iteration of the fix.  Bootstrapping
> on ARM boxes is painfully slow :(

I could bootstrap a patch on arm-none-linux-gnueabihf if you'd like.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #5 from Jeffrey A. Law <law at redhat dot com> ---
How fast is your box?  I'm using 4  processors on a calxeda system...  It's
painful, particularly when the first calxeda box's disk died, thus losing my
build trees & test results.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[ktkachov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #6 from ktkachov at gcc dot gnu.org ---
(In reply to Jeffrey A. Law from comment #5)
> How fast is your box?  I'm using 4  processors on a calxeda system...  It's
> painful, particularly when the first calxeda box's disk died, thus losing my
> build trees & test results.

I'm using a Chromebook with 2 Cortex-A15s.

I think it took about 5-6 hours to bootstrap last time I tried.
I have a feeling your system might be faster :(

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[yufeng at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

Yufeng Zhang <yufeng at gcc dot gnu.org> changed:

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

--- Comment #7 from Yufeng Zhang <yufeng at gcc dot gnu.org> ---
What is the current status?

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

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

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

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #8 from Jeffrey A. Law <law at redhat dot com> ---
I have multiple fixes.  Steven and I disagree on which is better.  

Having Richi or Jakub chime in with their opinions would help -- if they agree
with Steven, then I'll go with the majority.  If they prefer mine, then that's
what we'll go with.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

Steven Bosscher <steven at gcc dot gnu.org> changed:

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

--- Comment #9 from Steven Bosscher <steven at gcc dot gnu.org> ---
(In reply to Jeffrey A. Law from comment #8)
> I have multiple fixes.  Steven and I disagree on which is better.  
> 
> Having Richi or Jakub chime in with their opinions would help -- if they
> agree with Steven, then I'll go with the majority.  If they prefer mine,
> then that's what we'll go with.

I've proposed an alternative here:
http://gcc.gnu.org/ml/gcc-patches/2013-12/msg01871.html

It's not quite perfect, but it's conservative.

IMHO we should address the bigger issue (what does builtin_unreachable mean,
also on non-cond_exec targets?) for the next GCC stage1.

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

--- Comment #10 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Author: law
Date: Wed Jan  8 05:56:31 2014
New Revision: 206417

URL: http://gcc.gnu.org/viewcvs?rev=206417&root=gcc&view=rev
Log:
        PR middle-end/59285
        * ifcvt.c (merge_if_block): If we are merging a block with more than
        one successor with a block with no successors, remove any BARRIER
        after the second block.

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

[Bug middle-end/59285] [4.9 Regression] gcc.dg/builtin-unreachable-6.c:17:1: internal compiler error: in rtl_verify_fallthru, at cfgrtl.c:2862

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59285

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

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

--- Comment #11 from Jeffrey A. Law <law at redhat dot com> ---
Resolved on the trunk by recent commit.