[Bug regression/58221] New: [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[Bug regression/58221] New: [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

            Bug ID: 58221
           Summary: [4.9 Regression]: Immense amount of execution
                    regressions and increased test-time for cris-elf
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: regression
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hp at gcc dot gnu.org
              Host: x86_64-unknown-linux-gnu
            Target: cris-axis-elf

I see a LOT of execution regressions introduced in the interval
(201874:201887], too many to list them individually here.  At r201887 it's
regress-4951, an increase of about 4500!  Maybe every executing test.  Also, an
an increase in build-test time from to 3:39:09elapsed to 44:27:48elapsed (yup
about two days) apparently due to many tests infinitely looping and timing out.
 At r201919 it seems pretty much the same this far.

Things were bad to start with, status regress-446 at r201874, where there used
to be regress-7 as late as r201859.  Those 400+ regressions were in libstdc++
and probably the same as observed by others, due to insufficient testing of the
pass changes, but those patches were reverted at r201887.  The failing tests
are all over the supported languages for this target; C, ObjC, C++ and Fortran.

I'm entering this PR to track the issue.  Further details to follow.
First suspect is r201883, so first informed-bisect triage point will be
r201882.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

--- Comment #1 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
No significant change in results for regress-446 -> regress-444 for r201882
from r201874 (some libstdc++ changes pass again, but others now fail).  Maybe
r201883 is the winner; checking.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

Hans-Peter Nilsson <hp at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2013-08-23
                 CC|                            |tejohnson at gcc dot gnu.org

--- Comment #2 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
(In reply to Hans-Peter Nilsson from comment #1)
> Maybe r201883 is the winner; checking.

Yes, that's it.  I'll have to leave for a few hours but I'll extract a small
test-case, probably just one of the regressions as-is.  CC to committer.
N.B. cris-* does not have target support for section-switching - at least I
don't remember putting in any specific support.

Regarding the patch, I can't say at a glance what could possibly be so damaging
about *not* skipping the first non-active insn.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[kkojima at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

Kazumoto Kojima <kkojima at gcc dot gnu.org> changed:

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

--- Comment #3 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
SH also gets many execution failures at r201833.  I've reported it as
PR58220.  It looks NEXT_INSN vs. next_insn problem.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

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

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Patch to change that bogus next_insn call to NEXT_INSN is preapproved, please
put these two PRs into the ChangeLog entry.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

--- Comment #5 from Teresa Johnson <tejohnson at google dot com> ---
Thanks, and sorry for the trouble.

Kaz, are you planning to apply your patch, or do you want me to test
it and commit it? I'm kicking off x86_64 tests with it right now, but
I didn't get the failure on that target and don't have on hand yet a
reproducer.

Teresa

On Fri, Aug 23, 2013 at 6:13 AM, jakub at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221
>
> Jakub Jelinek <jakub at gcc dot gnu.org> changed:
>
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |jakub at gcc dot gnu.org
>
> --- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
> Patch to change that bogus next_insn call to NEXT_INSN is preapproved, please
> put these two PRs into the ChangeLog entry.
>
> --
> You are receiving this mail because:
> You are on the CC list for the bug.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[kkojima at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

--- Comment #6 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
(In reply to Teresa Johnson from comment #5)
> Kaz, are you planning to apply your patch, or do you want me to test
> it and commit it? I'm kicking off x86_64 tests with it right now, but
> I didn't get the failure on that target and don't have on hand yet a
> reproducer.

I've tested it on sh4-linux only.  I've just fired bootstrap
and the usual tests on i686-linux.  It will take hours on my
slow host, though.  Could you apply the patch after your x86_64
tests?

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

--- Comment #7 from Teresa Johnson <tejohnson at google dot com> ---
On Fri, Aug 23, 2013 at 6:49 AM, kkojima at gcc dot gnu.org
<gcc-bugzilla@gcc.gnu.org> wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221
>
> --- Comment #6 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
> (In reply to Teresa Johnson from comment #5)
>> Kaz, are you planning to apply your patch, or do you want me to test
>> it and commit it? I'm kicking off x86_64 tests with it right now, but
>> I didn't get the failure on that target and don't have on hand yet a
>> reproducer.
>
> I've tested it on sh4-linux only.  I've just fired bootstrap
> and the usual tests on i686-linux.  It will take hours on my
> slow host, though.  Could you apply the patch after your x86_64
> tests?

Will do, thanks.
Teresa

>
> --
> You are receiving this mail because:
> You are on the CC list for the bug.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[tejohnson at google dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

--- Comment #8 from Teresa Johnson <tejohnson at google dot com> ---
Tested on x86_64-unknown-linux-gnu, and also reproduced the failure
listed in PR rtl-optimization/58220 and verified the fix with it.
Committed as r201941:

Index: final.c
===================================================================
--- final.c     (revision 201940)
+++ final.c     (revision 201941)
@@ -1650,7 +1650,7 @@ reemit_insn_block_notes (void)
   rtx insn, note;

   insn = get_insns ();
-  for (; insn; insn = next_insn (insn))
+  for (; insn; insn = NEXT_INSN (insn))
     {
       tree this_block;

Index: ChangeLog
===================================================================
--- ChangeLog   (revision 201940)
+++ ChangeLog   (revision 201941)
@@ -1,3 +1,10 @@
+2013-08-23  Kaz Kojima  <kkojima@gcc.gnu.org>
+
+       PR rtl-optimization/58220
+       PR regression/58221
+       * final.c (reemit_insn_block_notes): Use NEXT_INSN to
+       handle SEQUENCE insns properly.
+
 2013-08-23  Gabriel Dos Reis  <gdr@integrable-solutions.net>

        * pretty-print.h (pp_newline_and_flush): Declare.  Remove macro


Thanks,
Teresa

On Fri, Aug 23, 2013 at 6:54 AM, Teresa Johnson <tejohnson@google.com> wrote:
> On Fri, Aug 23, 2013 at 6:49 AM, kkojima at gcc dot gnu.org
> <gcc-bugzilla@gcc.gnu.org> wrote:
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221
>>
>> --- Comment #6 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
>> (In reply to Teresa Johnson from comment #5)
>>> Kaz, are you planning to apply your patch, or do you want me to test
>>> it and commit it? I'm kicking off x86_64 tests with it right now, but
>>> I didn't get the failure on that target and don't have on hand yet a
>>> reproducer.
>>
>> I've tested it on sh4-linux only.  I've just fired bootstrap
>> and the usual tests on i686-linux.  It will take hours on my
>> slow host, though.  Could you apply the patch after your x86_64
>> tests?
>
> Will do, thanks.
> Teresa
>
>>
>> --
>> You are receiving this mail because:
>> You are on the CC list for the bug.
>
>
>
> --
> Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.9.0

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

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

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

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.

[Bug regression/58221] [4.9 Regression]: Immense amount of execution regressions and increased test-time for cris-elf

[tejohnson at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58221

--- Comment #10 from tejohnson at gcc dot gnu.org ---
Author: tejohnson
Date: Sat Nov 23 04:30:07 2013
New Revision: 205298

URL: http://gcc.gnu.org/viewcvs?rev=205298&root=gcc&view=rev
Log:
Backport to google/4_8 new sanity checking and some dwarf emission fixes for
-freorder-blocks-and-partition from trunk (r201883, r201941, r202125).

------------------------------------------------------------------------
r201883 | tejohnson | 2013-08-20 06:29:53 -0700 (Tue, 20 Aug 2013) | 8 lines
Changed paths:
   M /trunk/gcc/ChangeLog
   M /trunk/gcc/final.c
   M /trunk/gcc/testsuite/ChangeLog
   A /trunk/gcc/testsuite/g++.dg/tree-prof/pr57451.C

2013-08-20  Teresa Johnson  <tejohnson@google.com>

    PR rtl-optimizations/57451
    * final.c (reemit_insn_block_notes): Prevent lexical blocks
    from crossing split section boundaries.

    * testsuite/g++.dg/tree-prof/pr57451.C: New test.

------------------------------------------------------------------------
------------------------------------------------------------------------
r201941 | tejohnson | 2013-08-23 07:31:06 -0700 (Fri, 23 Aug 2013) | 7 lines
Changed paths:
   M /trunk/gcc/ChangeLog
   M /trunk/gcc/final.c

2013-08-23  Kaz Kojima  <kkojima@gcc.gnu.org>

        PR rtl-optimization/58220
        PR regression/58221
    * final.c (reemit_insn_block_notes): Use NEXT_INSN to
        handle SEQUENCE insns properly.

------------------------------------------------------------------------
------------------------------------------------------------------------
r202125 | tejohnson | 2013-08-30 18:43:33 -0700 (Fri, 30 Aug 2013) | 30 lines
Changed paths:
   M /trunk/gcc/ChangeLog
   M /trunk/gcc/basic-block.h
   M /trunk/gcc/bb-reorder.c
   M /trunk/gcc/cfg.c
   M /trunk/gcc/cfgcleanup.c
   M /trunk/gcc/cfgrtl.c
   M /trunk/gcc/predict.c

This patch sanitizes the partitioning to address issues such as edge
weight insanities that sometimes occur due to upstream optimizations,
and ensures that hot blocks are not dominated by cold blocks. This
needs to be resanitized after certain cfg optimizations that may
cause hot blocks previously reached via both hot and cold paths to
only be reached by cold paths.

The verification code in sanitize_dominator_hotness was contributed by
Steven Bosscher.

2013-08-29  Teresa Johnson  <tejohnson@google.com>
            Steven Bosscher  <steven@gcc.gnu.org>

    * cfgrtl.c (fixup_new_cold_bb): New routine.
    (commit_edge_insertions): Invoke fixup_partitions.
    (find_partition_fixes): New routine.
    (fixup_partitions): Ditto.
    (verify_hot_cold_block_grouping): Update comments.
    (rtl_verify_edges): Invoke find_partition_fixes.
    (rtl_verify_bb_pointers): Update comments.
    (rtl_verify_bb_layout): Ditto.
    * basic-block.h (probably_never_executed_edge_p): Declare.
        (fixup_partitions): Ditto.
    * cfgcleanup.c (try_optimize_cfg): Invoke fixup_partitions.
    * bb-reorder.c (sanitize_hot_paths): New function.
        (find_rarely_executed_basic_blocks_and_crossing_edges): Invoke
        sanitize_hot_paths.
    * predict.c (probably_never_executed_edge_p): New routine.
    * cfg.c (check_bb_profile): Add partition insanity warnings.

------------------------------------------------------------------------

Added:
    branches/google/gcc-4_8/gcc/testsuite/g++.dg/tree-prof/pr57451.C
Modified:
    branches/google/gcc-4_8/gcc/basic-block.h
    branches/google/gcc-4_8/gcc/bb-reorder.c
    branches/google/gcc-4_8/gcc/cfg.c
    branches/google/gcc-4_8/gcc/cfgcleanup.c
    branches/google/gcc-4_8/gcc/cfgrtl.c
    branches/google/gcc-4_8/gcc/final.c
    branches/google/gcc-4_8/gcc/predict.c