[Bug target/65171] New: [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[Bug target/65171] New: [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

            Bug ID: 65171
           Summary: [5 Regression] r214254 causes thousands of
                    std::length_errors running boost testsuite on ppc64le
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: trippels at gcc dot gnu.org
                CC: wschmidt at gcc dot gnu.org
              Host: powerpc64le-unknown-linux-gnu
            Target: powerpc64le-unknown-linux-gnu
             Build: powerpc64le-unknown-linux-gnu

Created attachment 34842
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34842&action=edit
unreduced testcase

Starting with r214254 thousands of std::length_errors are thrown
during a boost testsuite run on powerpc64le.
ppc64 big-endian is fine.

See: http://www.boost.org/development/tests/develop/developer/summary.html
trippels_power8_gcc-5.0_c++11 (big-endian) vs.
trippels_powerpc64le_gcc-5.0_c++11 (little-endian)

e.g.:

trippels@gcc2-power8 tools % g++ -w -ftemplate-depth-128 -O3 -finline-functions
-Wno-inline -Wall -fPIC -std=c++11 -mcpu=power8 -DBOOST_ALL_NO_LIB=1 -DNDEBUG
-I".." -I"../.." "../libs/sort/test/float_sort_test.cpp"
trippels@gcc2-power8 tools % ./a.out
Running 1 test case...
unknown location(0): fatal error: in "test_main_caller( argc, argv )":
std::length_error: basic_string::assign
../libs/sort/test/float_sort_test.cpp(68): last checkpoint

*** 1 failure detected in test module "Test Program"

Sorry, but reducing seems to be impossible.

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

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> ---
Not even by selectively applying __attribute__((target ("optimize-swaps")))
or __attribute__((target ("no-optimize-swaps"))) (if that is supported in
target directives) to bisect where it matters?

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |5.0

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

--- Comment #3 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Problem is an invalid swap removal in:

bool boost::test_tools::tt_detail::report_assertion(const
boost::test_tools::as\
sertion_result&, const boost::unit_test::lazy_ostream&,
boost::unit_test::const\
_string, std::size_t, boost::test_tools::tt_detail::tool_level,
boost::test_too\
ls::tt_detail::check_type, std::size_t, ...)

Something going on with subreg handling.

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

--- Comment #4 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
The logic to avoid subregs of TImode wasn't getting executed.  I have a patch
that fixes this; doing regression testing and then will get it submitted to the
list.  Still working in the background on reducing the test case. :/

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

--- Comment #5 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Patch submitted at https://gcc.gnu.org/ml/gcc-patches/2015-02/msg01550.html.

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

--- Comment #6 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
BTW I get:
FAIL: gcc.target/powerpc/swaps-p8-2.c scan-assembler-not xxpermdi
before and after your patch.

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

--- Comment #7 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Yes, this is unrelated and not really a problem.  Something has caused us to
recently start spilling a couple of nonvolatile vector registers to the stack
for this test case, resulting in the failure.  I'm going to look into why this
occurred (introduced on 2/13).

Bill

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

--- Comment #8 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Author: wschmidt
Date: Thu Feb 26 17:43:03 2015
New Revision: 221022

URL: https://gcc.gnu.org/viewcvs?rev=221022&root=gcc&view=rev
Log:
2015-02-26  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    PR target/65171
    * config/rs6000/rs6000.c (rs6000_analyze_swaps): Ensure
    instructions with TImode operands are included in the analysis.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/rs6000/rs6000.c

[Bug target/65171] [5 Regression] r214254 causes thousands of std::length_errors running boost testsuite on ppc64le

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65171

Bill Schmidt <wschmidt at gcc dot gnu.org> changed:

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

--- Comment #9 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Fixed.